Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Examining electoral accountability through the dynamics of government support : party popularity, the… Pickup, Mark 2005

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-ubc_2005-105534.pdf [ 18.48MB ]
Metadata
JSON: 831-1.0099803.json
JSON-LD: 831-1.0099803-ld.json
RDF/XML (Pretty): 831-1.0099803-rdf.xml
RDF/JSON: 831-1.0099803-rdf.json
Turtle: 831-1.0099803-turtle.txt
N-Triples: 831-1.0099803-rdf-ntriples.txt
Original Record: 831-1.0099803-source.json
Full Text
831-1.0099803-fulltext.txt
Citation
831-1.0099803.ris

Full Text

E X A M I N I N G E L E C T O R A L A C C O U N T A B I L I T Y T H R O U G H T H E D Y N A M I C S OF G O V E R N M E N T S U P P O R T : P A R T Y P O P U L A R I T Y , THE E C O N O M Y A N D POLITICAL C O N T E X T by Mark Pickup B.Sc , The University of Calgary, 1995 B.A. , The University of Calgary, 1997 M.A. , The University of Calgary, 2000 A THESIS SUBMITTED IN PARTIAL F U L F I L M E N T OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE F A C U L T Y OF G R A D U A T E STUDIES (POLITICAL SCIENCE) THE UNIVERSITY OF BRITISH C O L U M B I A June 2005 © Mark Alexander Pickup, 2005 A B S T R A C T Evidence of electoral accountability in Canada, in terms of the performance of the economy, has been contradictory and weak. This dissertation demonstrates that this confusion is largely a product of methodological flaws in the modelling of economic popularity. These flaws are corrected through the development of a unique structural approach. The basis of this structural approach is the use of the state-space form of time series modelling. In this approach, the observed values of popularity are regarded as being made up of distinct unobserved components and measurement error. The unobserved components driven by the changing political context in which economic popularity operates are each modelled separately, revealing the dynamics of this political context and accounting for the nonstationarity produced by these dynamics. The dynamics of measured popularity produced by measurement errors that correlate and cycle with time are explicitly modelled. This accounts for the nonstationarity produced by the dynamics of measurement error variances. The unobserved components also include a stationary process. This stationary component is extracted from the nonstationary dynamics by the state-space model. By modelling the impact of economic conditions directly through this component, it is appropriate to use modelling techniques that assume stationarity. Accordingly, the Box-Jenkins approach is adapted and applied to determine the correct lag and error structure for the state-space economic popularity models. When this approach is applied to Canadian federal party popularity data between 1957 and 2000, important political contextual dynamics are revealed. These dynamics include trending ii and inter-election cycling in popularity, election effects, leadership effects, national crises effects, and period-specific dynamics. Economic effects are also clearly demonstrated. Since 1984, the popularity of the party in government has been strongly affected by economic conditions - in particular, inflation and economic growth. Therefore, it is fair to conclude that governments, in a very real way, have at times been held accountable for the outcomes of their economic policies and that electoral accountability often operates as we expect it to. Suggestions are made for further analysis to examine the outstanding question: why is it that electoral accountability does not always operate as expected? T A B L E O F C O N T E N T S Abstract Table of Contents List of Tables List of Figures Acknowledgements Chapter I: Economic Conditions and Electoral Accountability in Canada Chapter II: Public Opinion and the State-Space Model Chapter III: Economic Popularity and the State-Space Model Chapter IV: Putting the State-Space Economic Popularity Model into Political Context Chapter V: Understanding Political Context and the Potential Role of the Media Conclusion Bibliography Tables Figures Appendix A: Additional Tables Appendix B: Data Sources Appendix C: Interpreting Significance Using the Bayesian Estimated Distributions of the Parameters L I S T O F T A B L E S Table 2-1: Estimated Parameters of Government Popularity State-Space Model 186 Table 2-2: Estimated Parameters for Liberal Party Popularity State-Space Model 187 Table 2-3: Estimated Parameters for PC Party Popularity State-Space Model 188 Table 2-4: Liberal and PC Party Popularity Variance, Explained and Unexplained 189 Table 3-1: Liberal Economic Popularity Estimates 190 Table 3-2: PC Economic Popularity Estimates 191 Table 3-3: Liberal Economic Popularity Estimates, No Interaction Term 192 Table 3-4: PC Economic Popularity Estimates, No Interaction Term 193 Table 3-5: Liberal Economic Popularity Estimates, Change in Unemployment 194 Table 3-6: PC Economic Popularity Estimates, Change in Unemployment 195 Table 3-7: Liberal Economic Popularity Estimates, AR(2) Model 196 Table 3-8: Modelling Economic Variables as First -Order Autoregressive Process with Drift, 197 Table 3-9: ARIMA Regression Model of Economic Variables, AR(1, 13, 25, 37) 198 Table 3-10: ARIMA Regression Model of Economic Variables, AR(1) 199 Table 3-11: ARIMA Regression Model of GDPxInflation, AR( 1, 4, 13) 200 Table 3-12: Cross-Correlation Functions for Liberal Popularity and Economic Variables 201 Table 3-13: Cross-Correlation Functions for Progressive Conservative Popularity and Economic Variables 203 Table 3-14: Box-Jenkins Progressive Conservative Economic Popularity Estimates, 1957-1975 205 Table 3-15: Box-Jenkins Liberal Economic Popularity Estimates, 1957-1975 206 Table 3-16: Box-Jenkins Progressive Conservative Economic Popularity Estimates, 1979-1993 207 Table 3-17: Box-Jenkins Liberal Economic Popularity Estimates, 1979-1993 208 Table 3-18: Box-Jenkins Liberal Economic Popularity Estimates, 1993-2000 209 Table 4-1: Estimated Political Context Effects for PC Party Popularity Model, 1957-1975 210 Table 4-2: Estimated Political Context Effects for Liberal Party Popularity Model, 1957-1975 211 Table 4-3: Estimated Political Context Effects for PC Party Popularity Model, 1979-1993 212 Table 4-4: Estimated Political Context Effects for Liberal Party Popularity Model, 1979-1993 213 Table 4-5: Estimated Political Context Effects for PC Party Popularity Model, 1993-2000 214 Table 4-6: Estimated Political Context Effects for Liberal Party Popularity Model, 1993-2000 215 Table 4-7: Estimated Political Context Effects (incl. FLQ) for PC Party Popularity, 1957-1975 216 Table 4-8: Estimated Political Context Effects (incl. FLQ) for Liberal Party Popularity, 1957-1975 217 Table 4-9: PC Economic Popularity Estimates Controlling for Political Context Effects, 1957-1975 218 Table 4-10: Liberal Economic Popularity Estimates Controlling for Political Context Effects, 1957-1975 219 Table 4-11: PC Economic Popularity Estimates Controlling for Political Context Effects, 1979-1993 220 Table 4-12: Liberal Economic Popularity Estimates Controlling for Political Context Effects, 1979-1993 221 Table 4-13: Liberal Economic Popularity Estimates Controlling for Political Context Effects, 1993-2000 222 Table 4-14: Liberal Economic Popularity Estimates - Majority Governments Only, 1957-1975 223 Table 4-15: PC Economic Popularity Estimates - Majority Governments Only, 1979-1993 224 Table 4-16: Liberal Economic Popularity Estimates - Majority Governments Only, 1979-1993 225 Table 5-1: Media Effects 226 Table 5-2: Effect of Media on Economic Evaluations 227 v L I S T OF F I G U R E S Figure 2-1: Progressive Conservative and Liberal Party Popularity, 1957-2000 ' 229 Figure 2-2: Government Popularity 230 Figure 2-3: Government Popularity Autocorrelation Functions 231 Figure 2-4: Number of Valid Decided Voters Interviewed Each Month 232 Figure 2-5: Predicted Government Popularity from Deterministic Parts of the State-Space Popularity Model 233 Figure 2-6a: Predicted Liberal Popularity from Deterministic Parts of the State-Space Popularity Model 234 Figure 2-6b: Predicted PC Popularity from Deterministic Parts of State-Space Popularity Model 235 Figure 2-7: Residual Movement within a Component of Liberal and PC Popularity, 1957-2000 236 Figure 3-1: Inflation (year-over-year percentage change in the consumer price index) 237 Figure 3-2: GDP (year-over-year percentage change in real personal income per capita) 238 Figure 3-3: Unemployment (monthly percentage, seasonally adjusted) 239 Figure 3-4a: Distributions of Estimated AR( 1) Terms, 1957-1975 240 Figure 3-4b: Distribution of Estimated Liberal AR(1) Term, 1993-2000 241 Figure 3-5: Effect of GDP and Inflation on Party Popularity, 1979-1993 242 Figure 3-3: AC & PAC of a Component of State-Space Party Popularity Models, 1957-1975 243 Figure 3-4: AC & PAC ofa Component of State-Space Party Popularity Models, 1979-1993 244 Figure 3-5: AC & PAC ofa Component of State-Space Party Popularity Models, 1993-2000 245 Figure 3-6: Distributions of Estimated First Period AR(1) and AR(2) Terms 246 Figure 3-7: AC & PAC of Economic Variables, 1957-1975 247 Figure 3-8: AC & PAC of Economic Variables, 1979-1993 248 Figure 3-9: AC & PAC of Economic Variables, 1993-2000 249 Figure 3-10: AC & PAC of GDPxInflation 250 Figure 3-11: Distributions of Estimated AR(1) Terms, 1957-1975 251 Figure 4-1: Incumbent Government Popularity Leading into Election 252 Figure 4-2: Progressive Conservatives AR(1) Term, 1957-1975 253 Figure 4-3: Liberal AR( 1) Term, 195 7-1975 254 Figure 4-4: Progressive Conservative AR(1) Term anda Component Residuals, 1979-1993 255 Figure 4-5: Liberal AR(1) Term and a Component Residuals, 1979-1993 256 Figure 4-6: Liberal AR(1) Term, 1993-2000 257 Figures 4-9a & 4-9b: Decay of a Hypothetical 1 Percent Shift in Popularity Produced by Economic Conditions in Month 1 258 Figures 4-10a & 4-10b: Contribution of Constant Economic Conditions Producing an Initial 1 Percent Shift in Popularity 259 Figure 4-11: Monthly Shift in Liberal Party Popularity due to Unemployment, 1957-1975 260 Figure 4-12: Monthly Shift in Party Popularity due to GDP and Inflation, 1979-1993 261 Figure 4-13: Monthly Shift in Liberal Party Popularity due to GDP and Inflation, 1993-2000 262 Figure 4-14: Cumulative Contribution of GDP and Inflation to Government Popularity, 1984-1993 263 Figure 5-1: GDP and Government Popularity Cycles 264 Figure 5-2: Inflation and Government Popularity Cycles 265 Figure 5-3: Unemployment and Government Popularity Cycles 266 Figure 5-4: Size of Government, 1957-2000 267 Figure 5-5: Regionalisation and Fractionalisation, 1953-2002 268 Figure 5-6: The Consequences of Defection from Government to Single Alternative, 1962 and 1980 269 Figure 5-6 (cont): The Consequences of Defection from Government to Single Alternative, 1993 and 2000 270 Figure 5-7: The Consequences of Defection from Government to All Alternatives, 1980 and 1993 271 Figure 5-8: Federal Election News Stories and Government Popularity, 1995-2002 272 vi A C K N O W L E D G E M E N T S I would like to acknowledge the contributions made to this dissertation by my advisory committee - Richard Johnston, Fred Cutler, and Angela O'Mahony. In particular, Dick has been an outstanding supervisor. He is generous with his time, guidance and research funds. Helpful comments have also been made at various conferences over the past few years by discussants of the chapters of this dissertation that I have presented. Those who had been supportive during my M.A. at the University of Calgary - Keith Archer and Anthony Sayers - have continued to be a source of support and advice throughout my Ph.D. and continue to be as I begin my postdoctoral work at the universities of Oxford and Calgary. There were also a great number of people who supported me in a less academic but equally important way. My parents were an invaluable source of support when I fell i l l physically. My fellow students and all my friends at Lickerish and Ming provided important stress relief when I found myself in danger of falling i l l mentally. Rita - thank you for having been there for me. None of the foregoing persons bears any responsibility for errors of fact or interpretation in this dissertation. Chapter I E C O N O M I C C O N D I T I O N S A N D E L E C T O R A L A C C O U N T A B I L I T Y I N C A N A D A 1.0 Introduction If an issue is important to the electorate and i f the electorate believes the government has control over the issue, the principle of electoral accountability holds that the government's record on that issue will affect the electorate's opinion of the government and ultimately whether they vote for or against it. Evidence of electoral accountability in Canada, in terms of the performance of the economy, has been contradictory and weak. This dissertation demonstrates that this confusion is largely a product of methodological flaws. By correcting these errors, it is demonstrated that Canadian governments have been held accountable for the performance of the economy. Further, it is argued that these analytical flaws are not limited to the Canadian case and that all such public opinion, time-series research can benefit from the application of the methods developed here. In his opening to The Responsible Electorate, V.O. Key argues that voters are not fools.1 They are rational and the political institutions that they have developed are also rational. In fact, Key depicts the electorate as a great god - "a rational god of vengeance and reward."2 In this role, voters assess the past performance of the incumbent government and, depending upon their assessment, use the electoral system to punish or reward it by voting for or against its return. Through this mechanism, governments in democratic systems are held accountable for the outcomes of their actions and policies. This is the principle of electoral (or democratic) accountability and it is the philosophy that forms the basis of my examination of the extent to 1 Key, V.O. Jr. 1966. The Responsible Electorate: Rationality in Presidential Voting, 1936-1960. Cambridge, Massachusetts: The Belknap Press of Harvard University Press. 2 Key, V.O. Jr. 1964. Politics, Parties, and Pressure Groups. Fifth ed. New York: Thomas Y. Crowell Company. 2 which the Canadian electorate held the federal government accountable for the performance of the national economy from 1957 to 2000. Since the work of Key, a great deal of theorising on the issue of economic performance and electoral accountability has been advanced and a vast number of studies have been undertaken. For example, in 2000 Michael Lewis-Beck and Mary Stegmaier reviewed over 150 major books and articles on the economic determinants of electoral outcomes. In their summary of the literature, Lewis-Beck and Stegmaier conclude that overall: Economics and elections form a tight weave. When anchoring economic threads snag, governments can fall....For all democratic nations that have received a reasonable amount of study, plausible economic indicators, objective or subjective, can be shown to account for much of the variance in government support....Among the issues on the typical voter's agenda, none is more consistently present, nor generally has a stronger impact, than the economy. Citizen dissatisfaction with economic performance substantially increases the probability of a vote against the incumbent.3 Lewis-Beck and Stegmaier reviewed only one Canadian study but there is every reason to believe that economic conditions are as important to voters in Canada as anywhere else. As early as the 1904 election, political observers were noting that elections were fought almost entirely on economic issues4 and more than most issues, the economy has been consistently identified as important by the Canadian electorate. In 1968, high taxes, high prices and inflation topped the list of the most urgent problems facing the country, as identified by voters.5 Unemployment was an important issue during the 1972 election, as was inflation during the 1974 election and energy pricing in 1980.6 In 1984, unemployment was again a major concern and received a great deal of 3 Lewis-Beck, Michael S., and Mary Stegmaier. 2000. Economic Determinants of Electoral Outcomes. Annual Review of Political Science 3:183-219. 4 Siegfried, Andre. 1907. The Race Question in Canada. London,: E. Nash. 5 The Gallup Report, April 6, 1968 6 Irvine, William P. 1978. An Overview of the 1974 Federal Election in Canada. In Canada at the Polls: The General Election of 1974, edited by H. R. Penniman. Washington D. C : American Enterprise Institute for Public Policy Research, Pammett, Jon H. 2001. The People's Verdict. In The Canadian General Election of2000, edited by J. H. Pammett and C. Dornan. Toronto: The Dundurn Group. However, in 1974 the Liberals attempted to divert 3 media coverage, as did the economy as a whole.7 In the 1988 election, free trade policy was the main issue. Unemployment and deficit and debt reduction dominated the issue agenda in 1993, as did jobs in the 1997 election. Even in a time of comparative economic prosperity and stability, such as during the 2000 election campaign, news articles about unemployment, debt and tax reduction were still prominent.9 Despite the evidence of the importance of the economy as an issue and further evidence that the Canadian electorate believes that the federal government has control over the domestic economy, findings regarding the impact of economic conditions on the support for Canadian federal governments/parties have been largely inconclusive and inconsistent.10 While there is general agreement that economic issues may matter on some occasions, there is little agreement over which economic issues matter, how much they matter, when they matter, or in what way they matter. This dissertation demonstrates that the Canadian findings are inconsistent, not because the principle of electoral accountability does not hold, rather the "spotty" results are the product of methodological difficulties. In general, the analysis of electoral accountability has suffered from a number of shortcomings. First, the technical foundations have been inadequate - the attention from inflation to leadership issues (Fletcher, Frederick. 1978. The Mass Media in the 1974 Canadian Election. In Canada at the Polls: The General Election of 1974, edited by H. Penniman. Washington, D. C : American Enterprise Institute for Public Policy Research.). 7 Frizzell, Alan, and Anthony Westell. 1985. The Canadian General Election of 1984: Politicians, Parties, Press and Polls. Ottawa: Carleton University press. 8 Pammett, Jon H. 2001. The People's Verdict. In The Canadian General Election of2000, edited by J. H. Pammett and C. Dornan. Toronto: The Dundurn Group. 9 Dornan, Christopher, and Heather Pyman. Ibid. Facts and Arguments: Newspaper Coverage of the Campaign. 1 0 Key studies include: Clarke, Harold D., and Gary Zuk. 1987. The Politics of Party Popularity: Canada 1974-1979. Comparative Politics:299-3l5, Monroe, Kristen, and Lynda Erickson. 1986. The Economy and Political Support: The Canadian Case. The Journal of Politics 48:616-647., Erickson, Lynda. 1988. CCF-NDP Popularity and the Economy. Canadian Journal of Political Science2\ (1):99-116, Johnston, Richard. 1999. Business Cycles, Political Cycles and the Popularity of Canadian Governments, 1974-1998. Canadian Journal of Political Science 32 (3):499-520. 4 statistical tools utilised to examine the dynamics of government and party support have been problematic and, at times, inappropriate. Second, studies of electoral accountability have only been weakly grounded in political context. Consequently, our comprehension of why electoral accountability is more effective in some times and places than in others has been restricted. These problems are not limited to Canadian research but their consequences do seem more evident in this country. This dissertation corrects these problems by employing state-space models, estimated using Bayesian analysis to separate and explicitly model the multitude of dynamic components -stationary and non-stationary - comprised by public opinion towards governments and parties over time. This allows for the estimation of economic electoral accountability, properly accounting for statistical problems such as nonstationarity and controlling for the political context in which economic conditions are translated into government and party support. The state-space approach also provides a sophisticated method by which to cope with noisy public opinion data containing errors that correlate with time. Further, a modified Box-Jenkins approach is adopted to determine the appropriate lag structure for the measures of economic conditions. These and other issues particular to the Canadian setting are dealt with in this dissertation. The result is a properly-specified model of public opinion which demonstrates the important impact of economic conditions on government/party support in Canada. 5 2.0 Past Research: Economic Determinants of Voting and Popularity In this section, I outline the key literature surrounding the economy and government support. A great number of modelling techniques have been utilised in the search for the impact of the economy on public opinion regarding the government. Government support models fall into two broad categories - voting models and popularity models. This dissertation utilises the latter of these models. However, it is important to understand both approaches and to consider the major findings produced by each. In voting models, the dependent variable is the electorate's vote decision. These models are alternatively estimated using aggregate electoral outcomes or individual level survey data as their unit of analysis. The first uses time-series modelling techniques and the second cross-sectional." In the time-series case, objective aggregate measures of economic performance such as inflation, changes in unemployment, and economic growth are commonly used, although increasingly more subjective measures based on aggregated survey responses are being utilised. In the cross-sectional case, measures of economic performance are based on individual level survey responses to questions regarding the voters' perceptions of the economy. Time-series studies are often argued to be superior to cross-sectional studies. "With the national economy being the focus of attention, the most telling variance is the movement of national aggregates over time" 1 2 and time-series models have the advantage of being able to examine the impact of variables that vary over time (e.g., the economy) but are constant across the electorate during any one election. Furthermore, Lewis-Beck and Eulau stress the importance " As will be seen, some time-series approaches use cross-sectional regional or provincial panel data. 1 2 Haller, H. Brandon, and Helmut Norpoth. 1997. Reality Bites: News Exposure and Economic Opinion. Public Opinion Quarterly 61 (4):555-575. 6 of keeping the historical context in mind. 1 3 This also requires the long-term perspective of time-series analysis. It is also argued that when examining the impact of the economy on public opinion, aggregate level studies have an advantage over those at the individual level because, by using objective aggregate measures of economic conditions, they avoid the endogeneity problem inherent in using individual level subjective measures. This is the problem of subjective perceptions of the economy being in part determined by partisan preferences14 - e.g., an individual that prefers the Conservative party may have a more positive view of the economy i f the incumbent government is Conservative rather than Liberal. Objective economic indicators are also superior because individual economic measures produced by surveys contain a degree of measurement error.15 The disadvantage of time-series voting models is a relatively small T value (allowing for very few degrees of freedom), constrained by the number of elections that occur during any particular time period. With the accumulation of cross-sectional studies over time, researchers may be able to combine the cross-sectional and time-series approaches by pooling individual level cross-sectional data sets over time and combining them with aggregate measures of economic performance. Such studies have the potential to be very powerful but pose serious data-management and statistical modelling issues. 1 3 Lewis-Beck, Michael, and Heinz Eulau. 1985. Economic Conditions and Electoral Behaviour in Transnational Perspective. In Economic Conditions and Electoral Outcomes: The United States and Western Europe, edited by M. Lewis-Beck and H. Eulau. New York: Agathon Press, Inc. 1 4 Evans, Geoffrey, and Robert Andersen. 2004. The Political Conditioning of Economic Perceptions: Evidence from the 1992-97 British Electoral Cycle. Vol. 2004-W9, Nuffield College Politics Working Paper. Oxford: University of Oxford. 1 5 At the same time, it is necessary to avoid ecological fallacy when examining results from aggregate level research. 7 When examining the degree of public support for the government, aggregate level voting models benefit from the fact that they use the most direct measure - electoral outcomes. However, since elections are relatively rare and measuring public opinion based on their outcomes is insensitive to specific government initiatives, studies regularly rely upon public opinion measures. In such models, government or party support (the dependent variable) is operationalised as 'popularity' defined as the electorate's response to a survey question inquiring into their hypothetical vote preference or their approval of the government's performance. Such popularity models are also referred to as popularity functions. An economic popularity function is an equation describing the translation of economic conditions into government or party popularity. These models are estimated using aggregated survey response data and objective measures of macro-economic performance measured over multiple time points (typically, on a monthly or quarterly basis). Popularity studies alternatively model either government or party popularity. Government popularity is the popularity of the party in government, regardless of which party that is. Party popularity is the popularity of a given party, regardless of whether it is in government or opposition. Popularity models have the advantage over voting models of using objective economic measures and public opinion time series, while not being so limited by a small T (a small degree of freedom). Moreover, voting studies can only consider the electorate's response to government policy outcomes and actions during elections. This response is, of course, vitally important. However, a great deal of change in government policies and policy outcomes occurs between elections. Popularity studies (particularly those that measure public opinion on a monthly basis) are able to detect the impact of these changes in policy outcomes on public opinion. 8 It is important to note that while voting and popularity models are usually considered alternative approaches to examining government support, they actually may measure different phenomena.16 Voting data focus exclusively on the election, while popularity data include both election and inter-election periods. Consequently, these two forms of analysis are not examining the exact same thing and differential findings can be expected as a result of differences in the focus of study as well as due to differences in technical considerations.17 For example, there is evidence from Britain that the electorate assigns greater importance to economic conditions in 1 R their vote decision during elections than in the period between elections. Given that popularity measures between elections may not be measuring exactly the same phenomenon as a vote decision during elections, it is fair to question the validity of using popularity as a diagnostic for electoral outcomes. It turns out that the response to this question differs between government popularity studies as they are performed in the US versus Canada. This is because of an important difference in the nature of the measured popularity variable. The popularity time-series used in the US are usually obtained from asking voters to evaluate their approval of the President's job performance. This makes them distinctly different from voting studies which use either actual electoral results or voter responses to a question regarding their vote-intention or behaviour. Approving of a President's job performance and voting for that President are two different things. The first is an evaluation and the second is a behaviour or intended behaviour that is influenced by the evaluation, in addition to other factors. 1 6 Kramer makes this point particularly emphatically (Kramer, Gerald H. 1983. The Ecological Fallacy Revisited: Aggregate- versus Individual-level Findings on Economic and Elections, and Sociotropic Voting. The American Political Science Review 77 (1 ):92- 111.) 1 7 Whiteley, P. 1984. Inflation, Unemployment and Government Popularity: Dynamic Models for the United States, Britain and West Germany. Electoral Studies 3 (l):3-24. 1 8 Norpoth, Helmut. 1996a. The Economy. In Comparing Democracies: Elections and Voting in Global Prospective, edited by L. LeDuc, R. Niemi and P. Norris. Thousand Oaks, California: Sage Publications. 9 By comparing measures of leadership approval and party support in Britain, Nadeau, Niemi and Amato take the position that they measure distinguishable concepts.19 Irving Crespi goes even further by suggesting that there is no straightforward relationship between the two measures in US surveys. To the extent that a-relationship does exist, we would expect the evaluation to affect the behaviour. In Canada, things are different. Canadian studies that examine government popularity use time-series constructed from survey responses to vote-intention questions. Therefore, the distinction between the dependent variables examined in popularity studies and voting studies is less important, although not insignificant. It is still quite plausible that vote-intention questions between elections tap a different phenomenon than vote-intention questions near elections. However, we know that measures of popularity based on vote-intention questions near elections 9 1 are very good predictors of electoral outcomes. This suggests that while it is important to recognise that the dynamics of popularity may vary depending on the proximity of an election, popularity measures close to elections are certainly a valid diagnostic tool for government electoral support and popularity measures between elections are a valid diagnostic tool for the type of government support that occurs when an election is not near. While the latter type of government support is not equivalent to electoral support, it is still important to understand. Vote decisions are not based purely on the events of the week leading into an election. 1 9 Nadeau, Richard, Richard Niemi, and Timothy Amato. 1996. Prospective and Comparative or Retrospective and Individual? Party Leaders and Party Support in Great Britain. British Journal of Political Science 26:245-258. 2 0 Crespi, Irving. 1980. The Case of Presidential Popularity. In Polling on the Issues, edited by A. Cantril. Washington DC: Seven Locks Press. 2 1 Beck, Nathaniel. 1991. The Economy and Presidential Approval: An Information Theoretic Perspective. In Economics and Politics: The Calculus of Support, edited by H. Norpoth, M. Lewis-Beck and J.-D. Lafay. Ann Arbor, Michigan: The University Of Michigan Press. 10 Nathaniel Beck uses a rational choice perspective of elections to describe how popularity studies may be used to understand how economic conditions affect US Presidential electoral outcomes. "If we think of elections as a principal-agent situation, where voters (principals) are choosing an agent to run the country, then the popularity polls are a measure of whether the principals wish to retain the agent at any given time."22 The argument for this connection between elections and popularity is even stronger in the Canadian case where popularity is measured using a vote-intention question. It is in light of all of these considerations that this dissertation uses popularity models to examine electoral accountability. In examining the literature on economic voting and popularity, there are at least three dimensions to consider. These are time, target, and orientation.23 Time refers to retrospective versus prospective economic considerations; target refers to the distinction between egocentric and sociotropic economic considerations; and orientation refers to whether the impact of economic conditions is incumbent oriented or party oriented. These dimensions have been a constant source of debate and must be addressed when any government support model is being considered. The first of these debates (retrospective versus prospective considerations) stems from the difference between Key's and Downs' views of voters. Key depicts them as rational gods, looking back at the past performance of the incumbent, and punishing or rewarding them accordingly.24 Downs' depiction is also of a rational being but one that looks to the future, carefully calculating the gains and losses likely to occur with each potential government, and 2 2 Ibid. 2 3 The first two dimensions are identified by Lewis-Beck and Stegmaier. They also identify a third dimension which is different from that discussed here. Their third item is context. (Lewis-Beck, Michael S., and Mary Stegmaier. 2000. Economic Determinants of Electoral Outcomes. Annual Review of Political Science 3:183-219.) 2 4 Key, V.O. Jr. 1964. Politics, Parties, and Pressure Groups. Fifth ed. New York: Thomas Y. Crowell Company. 11 voting to maximise the outcome.25 In this way, Key's voters are retrospective and Downs' are prospective. The second debate (egocentric versus sociotropic considerations) is over whether voters take into account their own personal financial situation (egocentric considerations) and/or the financial situation of the economy as a whole (sociotropic considerations) when making a vote choice. The third debate (incumbent versus party oriented effects) hinges on two different ways in which voters may translate economic conditions into a vote preference. One possibility is that when the economy is doing well, the electorate is more likely to support the incumbent government. Conversely, when the economy is doing poorly (whether in terms of inflation, GDP, unemployment or otherwise) the electorate is less likely to support the incumbent government. This outcome is based on a reward-punishment hypothesis and would be an incumbent oriented effect.26 This is the purest form of the electoral accountability principle as stated by Key. Alternatively, economic effects may be party oriented. This outcome is based on a differential partisan capability hypothesis and there are two possible variations.27 The first is based on a clientele hypothesis. When the economy is doing poorly in some regard (for example, high unemployment) there may be a particular party that is viewed as best able to handle this particular economic problem. In this case, the electorate would be more likely to support that party regardless of whether it is the incumbent or not. The second variation is based on a salient goal hypothesis. Again, a particular party may be viewed as best able to handle a particular economic problem. However instead of being rewarded whenever this economic issue 2 5 Downs, Anthony. 1957. An economic theory of democracy. New York,: Harper. 2 6 Sanders, David. 2000. The Real Economy and the Perceived Economy in Popularity Functions: How Much Do Voters Need to Know? A Study of British Data, 1974-97. Electoral Studies 19:275-294. 27Carlsen, Fredrik. Ibid.Unemployment, Inflation and Government Popularity - are there Partisan Effects? : 141-150. 12 arises as a problem, this is the issue for which that party is most held accountable. Consequently, i f the party is in government when the economic issue they are believed to be best able to handle becomes a problem, they will be punished. Moreover, they will be punished much more harshly than any other party which is not believed to have any particular skill at dealing with the specific economic problem.29 Both of these outcomes are party oriented effects. In Canada, where it has been argued that the Liberal party has 'owned' a number of economic issues, the potential for on party oriented effects is substantial. Keeping these three dimensions in mind is useful when considering the various findings regarding economic conditions and electoral accountability. The findings regarding electoral accountability utilising aggregate level, time-series voting models; individual level, cross-sectional voting models; and popularity models are now examined. 2.1 Non-Canadian Aggregate Level Time-Series Voting Literature Evidence of the importance of economic conditions on government support is stronger in voting models than in popularity functions, and stronger in time-series voting models than in cross-sectional voting models. Most voting studies focus on the US case, with some important contributions coming from Western European industrialised nations, such as Britain and France. Almost all time-series voting models use sociotropic, retrospective economic considerations. In his 1971 study of aggregate party vote outcomes for the US House of Representatives from 1896-1964, Gerald Kramer demonstrates economic growth has an impact 3 0 Belanger, Eric. 2003. Issue Ownership by Canadian Political Parties 1953-2001. Canadian Journal of Political Science 36 (3):539-558. 13 '"~ • 31 consistent with findings of Key. Economic upturns help the Congressional candidates of the incumbent party, and economic decline benefits the opposition. A 10 percent decrease in per capita real personal income costs the incumbent administration four or five percentage points in 19 the Congressional vote. In his 1975 study, Edward R. Tufte uses economic performance to predict midterm congressional election outcomes. Tufte, like Kramer, finds that a 10 percent decrease in economic growth can produce a loss of four to six percentage points for the incumbent.34 Further, in his 1978 publication, Political Control of the Economy, Tufte examines presidential and on-year congressional elections. He demonstrates that a 10 percent change in growth of real disposable income per capita in the year before a presidential election can produce a change in the congressional vote of 11 percent for the incumbent party and 13 percent for the incumbent 1S presidential candidate. In "The Effect of Economic Events on Votes for President: 1984 Update," Ray C. Fair updates his previous examination of the impact of aggregate economic conditions on presidential electoral outcomes in the US since 1916 to include the 1984 election. Fair also demonstrates that voters consider only the performance of the economy under the incumbent party and not the performance of the economy the last time the opposition party was in power. This suggests voters simply reward or punish the incumbent based on past performance, rather than compare 3 1 Kramer, Gerald H. 1971. Short-Term Fluctuations in US Voting Behavior, 1896-1964. The American Political Science Review 65 (1):131-143. 3 2 Ibid. 3 3 Tufte, Edward R. 1975. Determinants of the Outcomes of Midterm Congressional Elections. Ibid. 69 (3):812-826. 3 4 Ibid. 3 5 Tufte, Edward R. 1978. Political Control of the Economy. New Jersey: Princeton University Press. 3 6 Fair, Ray C. 1988. The Effect of Economic Events on Votes for President: 1984 Update. Political Behavior 10 (2):168-179. 14 potential alternative governments. This is consistent with Key. In fact, each of the above studies demonstrate clear and strong retrospective economic effects of the Keysian reward/punishment variety. The studies also demonstrate that aggregate level time-series models are very powerful predictors of the electoral outcomes. Since these seminal works, each US presidential election is preceded by a large number and wide variety of such economic voting models attempting to predict the outcome of the upcoming election. Preceding the 2004 election, James Campbell reviewed the various election forecasting models produced by Alan Abramowitz, Robert Erikson, Thomas Holbrook, Michael Lewis-Beck, Brad Lockerbie, Helmut Norpoth, Charles Tien, Christopher Wlezien and himself. There are seven models in all and only one did not use some measure of the state of the economy leading into the campaign. Of those that did, three used objective measures, two used subjective measures and one used both. Considering all elections since 1952, Campbell notes that several of the proposed models produce forecasts several months before election-day which are comparable in degree of accuracy to polls conducted just before or just after the actual election (2.1 to 2.4 percentage points). In a rare voting study focusing on British elections, Lewis-Beck et al produce a forecasting vote model incorporating economic performance data similar to those used in US presidential elections.39 The analysis was carried out using a time-series of 12 parliamentary elections up to and including 1997. Inflation levels are found to have important predictive power. 3 7 Campbell, James. 2004. Introduction - The 2004 Presidential Election Forecasts. Political Science and Politics 37 (4):733-735. 3 8 The median forecast from these models was 53.8 percent for George W. Bush. His actual percentage of the popular vote was approximately 51 percent. 3 9 Lewis-Beck, Michael, Richard Nadeau, and Eric Belanger. 2004. General Election Forecasts in the United Kingdom: A Political Economy Model. Electoral Studies 23:279-290. 15 A one percentage point increase in inflation is found to produce a one percentage point decrease in government support. Overall, the model is found to produce out-of-sample forecast errors of 2.3 points.40 While these studies are a very powerful predictor of electoral outcomes, the few degrees of freedom available mean they can contain only a few explanatory variables. This restricts their ability to explain the impact of economic forces. They are even worse at explaining and taking into account the effect of political context. To the extent that these studies are able to provide any nuance at all, it is primarily in terms of the finding that voter response to economic conditions is asymmetric in the face of prosperity and recession.41 Specifically, voters are more likely to respond to bad economic times by punishing the government than they are to respond to good economic times by rewarding the government. This means that economic conditions should have less of an impact on government electoral success during times of prosperity. 2.2 Non-Canadian Individual Level Cross-Sectional Voting Literature In cross-sectional, individual level voting studies, the results have been more mixed in terms of the role of economic conditions in government support. Results vary widely from one election to the next. It is difficult to get a clear sense of the role of the economy from these studies. However, important work has been done on comparing egocentric versus sociotropic economic considerations. In his examination of US election studies between 1956 and 1974, Fiorina demonstrates how a voter's personal economic condition (egocentric considerations) affects his/her presidential I IS i l l . 4 1 Bloom, Howard S., and H. Douglas Price. 1975. Voter Response to Short-Run Economic Conditions: The Asymmetric Effect of Prosperity and Recession. The American Political Science Review 69 (4): 1240-1254. 16 vote.42 In his examination of presidential and congressional elections between 1956 and 1980, Kiewiet finds both sociotropic and egocentric economic considerations affect vote choice. For both presidential and congressional races, he finds larger effects for national economic conditions than personal economic conditions.43 Similarly, Donald Kinder and Roderick Kiewiet demonstrate that sociotropic economic considerations affect congressional voting, presidential voting and party ID, while egocentric considerations have virtually no effect.44 They also make the important point that sociotropic economic evaluations are not simply a proxy for personal economic problems. In the US and Britain, there is actually very little connection between the Since these studies, it has been generally accepted that sociotropic considerations are much stronger and override any personal (egocentric) considerations. According to Lewis-Beck and Stegmaier's overview of the literature, most individual level studies find "strong collective effects and weak to nonexistent personal economic effects."46 Richard Brody makes the distinction between these considerations by arguing that voters are more spectators than participants in the economy when making their vote decision.47 He argues that this is because personal economic problems are considered just that - personal. The role of the government in 4 2 Fiorina, Morris P. 1978. Economic Retrospective Voting in American National Elections: A Micro-Analysis. American Journal of Political Science 22 (2):426-443. 4 3 Kiewiet, D. Roderick. 1983. Macroeconomics and Micropolitics. Chicago: The University Of Chicago Press. 4 4 Kinder, Donald R., and D. Roderick Kiewiet. 1981. Sociotropic Politics: The American Case. British Journal of Political Science 11 (2): 129-161. 4 5 Ibid, and Alt, James E. 1978. The Politics of Economic Decline. Cambridge: Cambridge University Press. 4 6 Lewis-Beck, Michael S., and Mary Stegmaier. 2000. Economic Determinants of Electoral Outcomes. Annual Review of Political Science 3:183-219. However, Haller and Norpoth show that personal financial experiences have a greater impact on broader economic judgments for people sheltered from mainstream news information. Haller, H. Brandon, and Helmut Norpoth. 1997. Reality Bites: News Exposure and Economic Opinion. Public Opinion Quarterly 6\ (4):555-575. 4 7 Brody, Richard. 1991. Assessing the President: The Media, Elite Opinion, and Public Support. Stanford, California: Stanford University Press. 17 the economy is unlikely to be linked to them. Consequently, personal economic problems are unlikely to become politicised. Meanwhile, information regarding national economic considerations is received through the media. This information tends to come pre-politicised. Often economic information provided by the media is already linked to government activities. Despite the substantial evidence of the dominance of sociotropic considerations, it would be unfair not to point out that some argue there are reasons to be cautious of the findings from individual level studies of egocentric economic considerations, such as these. First, measures of personal economic conditions are by their very nature subjective and they are subject to sampling error. A further problem is highlighted by Kramer. He demonstrates that the estimated effects of personal economic conditions on voting, based on individual level data, can be badly biased.48 This is because changes in personal economic conditions are only partly the product of government-induced factors. He concludes that aggregate rather than individual level data must be used and that differential findings regarding sociotropic and egocentric considerations are the artifactual product of differences in individual and aggregate level analyses.49 Notwithstanding these arguments, it is generally accepted that sociotropic considerations are more prevalent than egocentric considerations. Beyond this, in studies of electoral accountability, it is sociotropic and not personal economic considerations that are theorised to drive government support. Furthermore, this dissertation utilises popularity models and these models by their aggregate nature are designed to estimate the impact of sociotropic 4 8 Kramer, Gerald H. 1983. The Ecological Fallacy Revisited: Aggregate- versus Individual-level Findings on Economic and Elections, and Sociotropic Voting. The American Political Science Review 77 (1):92-111. 18 considerations. For all these reasons, it is the impact of sociotropic economic considerations that are modelled in this dissertation. 2.3 Non-Canadian Economic Popularity Function Literature Popularity studies have been very useful in the debate over whether voters use retrospective or prospective evaluations of the economy in their decision to support the government or not. Traditionally, retrospective evaluations have been considered likely to be the most important. Given the general lack of political or economic sophistication within the electorate, it was thought that it was more likely that voters engaged in the easier task of evaluating the past and present performance of the economy under the current government rather than the more difficult task of assessing the likely future performance of the economy under alternative governments. This traditionally held view has been challenged by those such as Michael MacKuen, Robert Erikson and James Stimson who argue that voters are more like sophisticated forward-looking bankers than self-interested peasants.50 This has been argued to be particularly true in information rich, developed countries.51 However, Norpoth convincingly demonstrates that retrospective economic evaluations are more important than prospective evaluations in US 52 presidential popularity. He does note though that this may, in part, be a consequence of the way in which presidential popularity is measured through approval rather than vote-intention questions. The former is more susceptible to influence by retrospective considerations than the latter. Where vote-intention questions are used to measure popularity (such as in Canada) prospective evaluations may play a greater role. Even if this is true though, Norpoth 5 0 MacKuen, Michael B., Robert S. Erikson, and James A. Stimson. 1992. Peasants or Bankers? The American Electorate and the US Economy. American Political Science Review 86:597-611. 5 1 Cohen, Jeffrey E. 2004. Economic Perceptions and Executive Approval in Comparative Prospective. Political Behavior 26 (\):27-43. 5 2 Norpoth, Helmut. 1996b. Presidents and the Prospective Voter. The Journal of Politics 58 (3):776-792. 19 demonstrates that prospective evaluations themselves are largely a reflection of past and current economic conditions. This suggests that using retrospective measures in popularity models does not completely restrict voters to retrospective considerations. This is consistent with the arguments made by Downs and many others that voters use past economic conditions in a prospective way. 5 4 Voters use the past performance of the economy under a particular party to predict how the economy will perform under the same party in the future. It is for all these reasons and the fact that there are few monthly measures of prospective economic considerations at the aggregate level and that electoral accountability (the focus of this dissertation) is primarily a retrospective exercise that retrospective economic conditions are used in this dissertation. While most government popularity studies find that economic conditions are important in some way, the range of variation between these studies in terms of the results and the different statistical methodologies applied is greater than for voting studies.55 Britain is responsible for a greater proportion of the significant economic popularity studies than is the case with voting studies. The earliest US research on popularity functions was done by J. Mueller. 5 6 However, Lewis-Beck and Stegmaier identify the earliest published popularly function ever as being C.A.E. Goodhart and R.J. Bhansali's British case in 1970.57 Goodhart and Bhansali examine monthly measures of British government popularity between 1947 and 1968. They find that levels of unemployment and the rate of inflation influence the government's political popularity. They also find that the strength of the impact of these economic conditions had increased over 5 3 Ibid. 5 4 For a good example see: Chappell, Henry, Jr, and William Keech. 1985. A New View of Political Accountability for Economic Performance. The American Political Science Review 79 (1): 10-27. 5 5 Lewis-Beck, Michael S., and Mary Stegmaier. 2000. Economic Determinants of Electoral Outcomes. Annual Review of Political Science 3:183-219. 5 6 Ibid. 5 7 Goodhart, C.A.E., and R.J. Bhansali. 1970. Political Economy. Political Studies 18 (1):43-106. 20 the time period under study. Further, they find that government popularity may follow what they call a "natural path" between elections.58 Such a path includes honeymoon effects, trending downwards after the honeymoon and trending upwards leading into an election. Overall, this suggests that popularity follows an inter-election cycle. This cycle is part of the political context in which economic conditions are translated into government support. Goodhart and Bhansali attempt to control for this inter-election "natural path" through the application of dummy and index variables. It is commendable that Goodhart and Bhansali so early on recognised the need to control for the political context in which economic popularity operates, even if the methods by which this was done were fairly crude. The reason it is so important is that the dynamics in government popularity produced by this political context creates the problem of nonstationarity within the popularity data. Nonstationarity is a very challenging statistical process and a problem which is discussed further in Chapter II but in short, if a model which assumes the time-series is a stationary process is used to model nonstationary popularity data, the model is seriously misspecified. Therefore, any estimated results from these models are completely unreliable and highly susceptible to spurious correlation. Very peculiar results can be obtained when nonstationary data is modelled as stationary. To model popularity, the data must be transformed so that it is stationary or a model that accounts for the nonstationarity in the data must be used. Given these considerations, it is surprising that many studies since Goodhart and Bhansali are no more sophisticated in tackling this problem. Most studies will typically use crude time-indices to control for trending before and/or after elections. Beyond this, many very recent 21 studies hardly address the issue of nonstationarity produced by political contextual forces at all. This failure renders the statistical techniques used in the studies to model popularity inappropriate. This can have drastic consequences and even the most recent work of highly respected analysts exhibits this serious oversight. For example, examining British government monthly popularity data between 1974 and 1997, David Sanders finds no evidence to indicate that objective macro-economic measures had any direct effect on government support.59 However, Sanders includes only event dummies to control for political context. Further, while he uses the differenced form of unemployment and inflation to correct for nonstationarity in these economic variables, the issue of potential nonstationarity in the popularity data is not fully considered. Political contextual factors which exhibit as trending, cycling and shifts in baseline support are left completely unaddressed.60 Sanders' modelling techniques are technically inappropriate and fail to fully take into account the dynamics of the political context in which economic conditions are translated into public opinion and the consequent nonstationarity. This puts his findings into doubt. Sanders' study also demonstrates a further shortcoming of much of the popularity research. Economic variables are entered into his model at lags of 0, 1 and 2. This is done without any particular justification. It is reasonable to suspect that there will be some lag in economic effects. Some time will be required for information regarding the state of the economy to reach the electorate and to be processed into an opinion about the government. What the appropriate lag is though is not obvious. There is no more reason to believe that it will be one or 5 9 Sanders, David. 2000. The Real Economy and the Perceived Economy in Popularity Functions: How Much Do Voters Need to Know? A Study of British Data, 1974-97. Electoral Studies 19:275-294. 6 0 Ibid. 22 two months than that it will be three or four months or even up to a year. Unfortunately, it is unhelpful to just enter all possible lags and see what falls out. Economic variables contain a great deal of autocorrelation. Including unemployment with a two-month lag (for example) will greatly affect whether the estimated effect of unemployment with a one-month lag is statistically significant. Moreover, economic variables are correlated; so, including unemployment with a two-month lag will also greatly affect whether the estimated effect of inflation lagged one-month (again, for example) will be statistically significant. A much more purposive technique is required to determine the lags at which economic variables should enter into a popularity model. There are a few studies which have begun to take the necessary steps towards addressing these important methodological issues. Paul Whiteley, in "Inflation, Unemployment and Government Popularity: Dynamic Models for the United States, Britain and West Germany," examines the impact of economic conditions on monthly measures of government popularity in the United States, Britain and West Germany.61 Whiteley uses a process developed by Box and Jenkins to specify the popularity functions. This allows him to account for autocorrelation and trending in the independent and dependent variables through the appropriate differencing of fit"? these variables. It also allows him to determine the lag structure of the independent variables. Unfortunately, the Box-Jenkins approach is not always able to fully account for the nonstationarity produced by the cycling within popularity as identified by Goodhart and Bhansali. The Box-Jenkins approach eliminates cycling by differencing the data by the cycle length. The cycling is the product of the inter-election cycle and in the US case, where elections 6 1 Whiteley, P. 1984. Inflation, Unemployment and Government Popularity: Dynamic Models for the United States, Britain and West Germany. Ibid. 3 (l):3-24. 23 are evenly spaced, the Box-Jenkins approach is appropriate. In the case of Parliamentary governments with unevenly spaced elections, the Box-Jenkins approach is unable to cope with nonstationarity produced by the cycling. There is no way to difference the data in order to eliminate it. Clarke, Stewart and Zuk use the Box-Jenkins approach in their examination of party popularity in Britain between 1979 and 1983.63 They apply this approach to an A R I M A model of popularity. They also include a number of sophisticated political event variables controlling for the effects of strikes, internal party disputes, leadership popularity, etc. Norpoth does the same for presidential popularity in the US from 1961 to 1980.64 In fact, Norpoth goes one further than Clarke et al by explicitly exploring the appropriate structure for the error process in the A R I M A model, settling on a first order moving average process. Norpoth notes that there is resistance to using approaches such as that of Box-Jenkins because it is felt that the procedure for eliminating the nonstationary in popularity, produced by political contextual forces, may throw out real economic effects - that is, "the baby may be thrown out with the bathwater."65 He further notes that this is particularly problematic with data that is noisy, such as popularity data produced by public opinion polls. Consequently, he finds it necessary to aggregate his time-series to the quarterly rather than monthly level. Norpoth is correct to identify the potential problems produced by the noise (sampling error) inherent in measures of popularity. In fact, the problem is even larger than he identifies. 6 3 Clarke, Harold D., Marianne C. Stewart, and Gary Zuk. 1986. Politics, Economics and Party Popularity in Britain, 1979-83. Ibid. 5 (2): 123-141. 6 4 Norpoth, Helmut. 1985. Economics, Politics, and the Cycle of Presidential Popularity. In Economic Conditions and Electoral Outcomes: The United States and Western Europe, edited by M. Lewis-Beck and H. Eulau. New York: Agathon Press, Inc. 24 As this dissertation demonstrates, the error within popularity time-series also tends to trend, cycle and generally be correlated with time. This is yet another source of nonstationarity. The methods developed in this dissertation explicitly account for the dynamics of measurement error within popularity. It does so by using the state-space approach to modelling popularity. This approach is also used to address each of the other methodological problems identified in this section. This is described in section 4.0. 25 3.0 Economic Determinants of Voting and Popularity in Canada Having considered the economic voting and popularity literature from the US and Britain, it is now time to turn our attention to the case study of this dissertation. Compared to the US and British studies, Canadian voting and popularity studies tend to find a less clear role for economic conditions. 3.1 Canadian Aggregate Level Time-Series Voting Literature With rare exception, national-level aggregate voting studies have been unable to find significant economic effects.66 This at first seems somewhat surprising given the strength of the results produced by these models for the US case. One explanation is that aggregate voting models are, as it has been noted, poor at accounting for political context. This context may be more complex in Canada and therefore, not controlling for it may have greater consequences.67 Due to the weak findings, Canadian time-series voting studies usually revert to using pooled time-series provincial or regional analysis. In "Voter Sensitivity to Economic Conditions: A Canadian-American Comparison," J.R. Happy examines the impact of provincial economic conditions in the year of an election on provincial level aggregate federal electoral outcomes in Canada from 1930 to 1979. Overall, Happy finds that real and nominal increases in income benefit candidates of the incumbent party. The same is true of decreases in inflation.69 Comparing his own results to those of Kramer, Happy argues that Canadians in parliamentary elections are about equally as sensitive as 6 6 The exception are: Clarke, Harold D., and Gary Zuk. 1987. The Politics of Party Popularity: Canada 1974-1979. Comparative Politics:299-215. and Nadeau, Richard, and Andre Blais. 1993. Explaining Election Outcomes in Canada: Economy and Politics. Canadian Journal of Political Science 26 (4):775-790. 6 7 Reasons for the complexity of the political context in Canada are discussed in chapter V. 6 8 Happy, J. R. 1986. Voter Sensitivity to Economic Conditions: A Canadian-American Comparison. Comparative Politics 19(l):45-56. 26 Americans in House of Representatives elections to economic conditions.m However when comparing Happy's results to those of Fair, it appears Canadians in parliamentary elections are 71 not quite as sensitive to economic conditions as Americans are in presidential elections. Happy makes two further important discoveries. First, economic conditions previous to the election year also affect Canadian electoral behaviour. This suggests economic conditions of more than one or two months previous need to be considered. Second, when only a single year passes from one 79 election to the next, economic events have less of an effect on incumbency voting. Each of these single year governments are instances of minority governments, suggesting that economic effects operate differently for minority versus majority governments. This possibility is considered and examined in this dissertation. In "Economic Conditions and the Popularity of the Incumbent Party in Canada," Calum Carmichael uses pooled time-series, regional-level electoral results from 1945 to 1988 to examine the impact of price levels, the unemployment rate and real personal disposable income per capita on the vote for the incumbent party in Canadian federal elections. Generally, he finds that from 1945 to 1972 poor economic conditions improved the government's electoral success, while from 1974 to 1988 poor economic conditions hurt the incumbent party.73 Carmichael suggests that the 1972/1974 structural split is a product of double-digit inflation, followed by stagflation that began in 1974. He argues this may have changed the electorate's response to 7 0 Kramer, Gerald H. 1971. Short-Term Fluctuations in US Voting Behavior, 1896-1964. The American Political Science Review 65 (1): 131 -143. 7 1 Fair, Ray C. 1988. The Effect of Economic Events on Votes for President: 1984 Update. Political Behavior 10 (2):168-179. 72Happy, J. R. 1986. Voter Sensitivity to Economic Conditions: A Canadian-American Comparison. Comparative Politics 19(l):45-56. 7 3 Carmichael, Calum M. 1990. Economic Conditions and the Popularity of the Incumbent Party in Canada. Canadian Journal of Political Science 23 (4):713-726. 27 economic conditions. Furthermore, he cites evidence that the positions of the Liberal and Conservative parties became more distinct in the late seventies.74 The varying impact of economic conditions during different time periods is a theme which has also emerged in the research on Canadian government popularity. This is an important dynamic to consider and is discussed further below. In "Explaining Election Outcomes in Canada: Economy and Politics," Richard Nadeau and Andre Blais argue that aggregate voting models at the national level can demonstrate economic effects i f the model controls for the origin of the party leaders (Quebec versus non-Quebec).75 Examining elections between 1953 and 1988, Nadeau and Blais argue that no effects due to inflation are evident but that changes in the rate of unemployment have an important impact on the incumbent government's vote share.76 3.2 Canadian Individual Level Cross-Sectional Voting Literature Just as in the US and U K literature, Canadian individual level voting studies have made important contributions to the debate over sociotropic versus egocentric economic considerations. In their "Support for the Canadian Federal Progressive Conservative Party since 1988: The Impact of Economic Evaluations and Economic Issues," Harold Clarke and Allan Kornberg consider both the retrospective/prospective and egocentric/sociotropic issues. They 77 examine the decline in support for the PCs from 1988 to 1990. Using individual level subjective measures of economic performance, they argue that the decline in support evident 7 5 Nadeau, Richard, and Andre Blais. 1993. Explaining Election Outcomes in Canada: Economy and Politics. Ibid. 26:775-790. 7 6 Ibid. 7 7 Clarke, Harold D., and Allan Kornberg. 1992. Support for the Canadian Federal Progressive Conservative Party since 1988: The Impact of Economic Evaluations and Economic Issues. Ibid. 25 (l):29-53. 28 during this period is largely due to the public's negative evaluation of the Canadian economy and the government's handling of it. They demonstrate that retrospective individual economic assessments do in fact divide into sociotropic and egocentric factors. However, this distinction breaks down for future-oriented assessments. Overall, this produces three distinct economic assessment factors: retrospective-sociotropic, retrospective-egocentric, and future.78 In terms of retrospective considerations, Clarke and Kornberg determine in a separate study that sociotropic considerations have a much larger impact than egocentric considerations on support for 70 incumbent governments. This is consistent with the use of retrospective-sociotropic economic considerations utilised in this dissertation. In "Inflation, Unemployment and Canadian Federal Voting Behaviour," Archer and Johnson use individual level election survey data to examine the effect of economic considerations on levels of partisan support in Canada during the 1970s and 1980s.80 They find that the effects of inflation and unemployment on partisan popularity are unstable over time. Inflation increased from approximately 3 percent in 1971 to over 11 percent in 1974. Despite the salience of inflation in the 1974 election, no party seemed to benefit from it. By the 1984 election, high unemployment had replaced inflation as the most important issue and the 81 Conservatives benefited from it. To the extent that economic voting does occur, Archer and Johnson, like Clarke and Kornberg, find it is primarily due to sociotropic considerations, rather than egocentric considerations. 7 8 ibid. 7 9 Clarke, Harold D., and Allan Kornberg. 1989. Public Reactions to Economic Performance and Political Support in Contemporary Liberal Democracies: The Case of Canada. In Economic Decline and Political Change: Canada, Great Britain, the United States, edited by H. D. Clarke. Pittsburgh: University Of Pittsburgh Press. 8 0 Archer, Keith, and Marquis Johnson. 1988. Inflation, Unemployment and Canadian Federal Voting Behaviour. Canadian Journal of Political Science 21 (3):569-584. 29 3.3 Canadian Economic Popularity Function Literature Over the past 10 to 15 years, there has been a real decline in the volume of economic popularity studies coming out of Britain and the US. To the extent that economic popularity research has been conducted in Canada, it only really began in the mid to late 1980s. Despite the recentness of this research, it has done no better at addressing the methodological issues inherent to popularity time-series research than that from the US or Britain. If anything, it has dealt even more poorly with these issues. This may explain why the inconsistency in the estimated impact of economic conditions on government/party support is even more evident in Canadian popularity studies (of which there are few) than in Canadian voting studies. There are really only four major studies of Canadian party or government popularity worth noting. Kristen Monroe and Lynda Erickson examine the impact of unemployment, inflation and changes in real personal income per capita on the popularity of the Progressive Conservatives, Liberals and N D P . 8 2 Quarterly measures of party popularity between 1954 and 1979 are used. Monroe and Erickson control for political events such as a change of Prime Minister or a change in government, the FLQ crisis, the election of the Parti Quebecois in Quebec and elections. They make no attempt to correct for problems of nonstationarity. They simply use the Cochrane-Orcutt procedure to control for first order autocorrelation. Generally, Monroe and Erickson find that support for the Liberals versus the Conservatives is relatively unaffected by economic conditions. Economic conditions do however play a role in the popularity of the N D P . 8 3 Using similar methods, Erickson in "CCF-NDP Popularity and the Economy," examines the effect of economic growth, inflation and unemployment on the 8 2 Monroe, Kristen, and Lynda Erickson. 1986. The Economy and Political Support: The Canadian Case. The Journal of Politics 48:616-647. 30 popularity of the CCF/NDP between 1954 to 1984.84 She finds that economics effects vary over time. During some periods, real personalJncome benefited the party's popularity; during other periods, it reduced the party's popularity; and at some times it had no effect at all. During the eighties, inflation improved the popularity of the CCF-NDP. During earlier periods, unemployment produced a decline in party popularity. Clarke and Gary Zuk construct Box-Jenkins-Tiao models of party popularity similar to that of Whiteley. In "The Politics of Party Popularity: Canada 1974-1979," they examine the impact of economic performance on monthly measures of party popularity. They restrict their analysis to the 1974-1979 period, arguing that the effects of economic variables may vary over time and therefore should not be estimated over an extended period.8 5 Clarke and Zuk also criticise other studies of the political economy of party popularity for not considering the political context in which economic performance is translated into vote preference. Like Monroe and Erickson, they include dummies for the occurrence of political events, such as leadership conventions, election honeymoons and "crises of Confederation" - the latter being exemplified by the 1976 election of the Parti Quebecois government in Quebec. While many of the political forces included within the model were found to be statistically significant, this was not the case for any of the variables measuring macroeconomic conditions. It is concluded that political forces rather than macroeconomic conditions were responsible for shifts in party popularity during the 1974-1979 period.8 6 8 4 Erickson, Lynda. 1988. CCF-NDP Popularity and the Economy. Canadian Journal of Political Science 21 (1):99-116. 8 5 Clarke, Harold D., and Gary Zuk. 1987. The Politics of Party Popularity: Canada 1974-1979. Comparative Politics:299-3\5. 8 6 Ibid. 31 Richard Johnston examines the impact of economic factors on the popularity of Canadian federal governments measured monthly from 1974 to 1998 in "Business Cycles, Political Cycles and the Popularity of Canadian Governments, 1974-1998. " 8 7 He uses an error-correction and a partial adjustment model to examine the data. Again, neither of these methods adequately deals with the nonstationarity within the data. In both cases he accounts for first-order autoregression using a Hildreth-Liu grid search. Johnston examines the impact of economic growth, unemployment and inflation. Like Clarke and Zuk, Johnston controls for the effect of leadership conventions.88 Johnston determines that the dynamics of public opinion differ before and after 1993. Economics have a statistically significant effect on government popularity before 1993 but not after. Before 1993, real income growth benefited the popularity of the government as did an on increase in inflation. Furthermore, an electoral cycle appears to exist pre- but not post-1993. Nowhere has it become more evident than in Canada that both the impact of economic conditions on popularity and the dynamics of popularity itself can be very distinct from one time period to the next. Consequently, this dissertation pays close attention to these varying dynamics both controlling for them and attempting to understand them. 8 7 Johnston, Richard. 1999. Business Cycles, Political Cycles and the Popularity of Canadian Governments, 1974-1998. Canadian Journal of Political Science 32 (3):499-520. 8 8 Ibid. 8 9 Ibid. 32 4.0 The Task at Hand Having examined the literature on the economy and government support, it is clear that there are two separate important puzzles posed by the research to date. The first is embedded in the literature as a whole. That puzzle is to determine why the findings regarding the impact of economic conditions on government support is stronger and so much more consistent in aggregate voting models than in popularity models. The second puzzle is related but specific to the Canadian case. That puzzle is understanding why it has been such a challenge to determine the impact of macro-economic conditions on aggregate government support in Canada - using either voting or popularity models - given the strength of the relationship in other nations such as the US and Britain. The primary focus of this dissertation is the second puzzle utilising popularity models and it is demonstrated that the key to solving this puzzle is to apply appropriate statistical modelling techniques and to properly account for the political context in which economic conditions are translated into government support. In solving this puzzle, it becomes apparent that much of the economic popularity research conducted outside of Canada is subject to the same methodological problems as that within Canada. Therefore, a great deal of progress on the first puzzle may be made by applying the techniques developed to solve the Canadian puzzle to economic popularity research more broadly. In solving the Canadian puzzle, I use popularity rather than voting models. Beyond the enticement presented by the fact that this is the greater challenge, the practical reasons for this decision have already been outlined. In its use of popularity models, this dissertation considers both government and party popularity but ultimately examines the impact of economic conditions on government support using party popularity models, rather than government ' 33 popularity models. In a party system where the two major parties do not completely dominate and where support for different third parties rises and falls significantly, the popularity of one major party will not be a mirror reflection of the other. Moreover, i f voters respond to economic conditions in a party-oriented rather than an incumbent-oriented way then the impact of economic conditions on one party in government will differ (possibly have a completely different direction of impact) from the other. This is argued to be the case even in the two-party system in the U S . 9 0 The work done by Eric Belanger on issue ownership by Canadian political parties suggests this is also true in Canada.91 For these reasons, it is necessary to model the popularity of each major party separately - in this case, the popularity of the Liberal and Progressive Conservative parties. Economic conditions considered in this dissertation are levels and changes in unemployment, inflation rates and real economic growth. These are the standard conditions considered in much of the economic popularity literature. Unlike in many economic popularity studies, this dissertation also carefully considers the complex interaction between different economic conditions that likely exists within the minds of the electorate. The measures of economic conditions used in the party popularity models are sociotropic, retrospective measures. This is also a fairly standard approach. The rationale for this has been given. As indicated above, this dissertation employs the state-space approach in the modelling of economic popularity. The name "state-space" originates with the initial use of these models to represent the state of a physical system (position, velocity, momentum, etc) in spatial form. 9 0 Brody, Richard. 1991. Assessing the President: The Media, Elite Opinion, and Public Support. Stanford, California: Stanford University Press. 9 1 Belanger, Eric. 2003. Issue Ownership by Canadian Political Parties 1953-2001. Canadian Journal of Political Science 36 (3):539-558. 34 When applied to time-series public opinion data, state-space models are distinguishable from other modelling approaches in that observations are regarded as made up of distinct elements, such as trend, cycling, regression, measurement error and disturbance components, each of which is modelled separately. "The techniques that emerge from this approach are very flexible and are capable of handling a much wider range of problems than the main analytical system currently in use for time series analysis, the Box-Jenkins A R I M A system." It is the state-space approach that is the basis of the Kalman filter. The Kalman filter is recommended by Donald Green, Alan Gerber, and Suzanna De Boef to reduce sampling error in public opinion time-series such as popularity and by Nathaniel Beck to do the same in modelling US presidential popularity.93 This dissertation goes far beyond the Kalman filter. Not only does it use the state-space approach to reduce sampling error, it employs this approach to account for errors correlating with time and to control for nonstationary dynamics within popularity produced by political contextual forces by explicitly modelling them. The use of the state-space approach as it is developed in this dissertation is uniquely able to properly account for both unevenly spaced inter-election cycling within popularity and measurement error variances correlating with time. Having developed a technique that fully accounts for nonstationarity, the Box-Jenkins technique is modified so that it can be applied to the modelling of the impact of economic conditions on party popularity through the state-space approach. In this way, the proper lag structure of 9 2 Durbin, James. 2004. Introduction to State Space Time Series Analysis. In State Space and Unobserved Component Models: Theory and Applications, edited by A. C. Harvey, S. J. Koopman and N. Shephard. Cambridge: Cambridge University Press. 9 3 (Green, Donald, Alan Gerber, and Suzanna De Boef. 1999. Tracking Opinion over Time: A Sampling Method for Reducing Sampling Error. Public Opinion Quarterly 63:178-192.) and (Beck, Nathaniel. 1990. Estimating Dynamic Models Using Kalman Filtering. Political Analysis 1.) 35 economic variables is determined. Special attention is also paid to the structure of the error process in the state-space model as was done by Norpoth using A R I M A modelling. No other approach developed to date is able to handle each and every one of these methodological challenges. Overall, the methods and statistical techniques developed in this dissertation are very successful at determining how economic conditions have translated into party and government support in Canada between 1957 and 2000, and how different economic variables have mattered to different degrees at different times. My findings support some of the claims of some Canadian popularity research but overturn many others. I demonstrate that economic conditions do play an important role in shaping party popularity and certainly since 1984, Canadian federal governments have been held accountable for key economic conditions such as economic growth and inflation. I also reveal a number of dynamics within popularity - many of which had not previously been discovered, such as the persistence of an inter-election popularity cycle, coupled with the strategic timing of elections. My examination of economic performance and electoral accountability proceeds as follows. In this chapter, I have highlighted the key findings from the vast literature on economic performance, government popularity and voting, placing this dissertation in context. In Chapter II, I describe the dependent variables - government and party popularity - and discuss the statistical difficulties inherent in modelling such political public opinion data. To address these issues, I propose a time-series, state-space model of aggregate government and party popularity, estimated using Bayesian analysis. I present the results of modelling both government and party popularity using such an approach. In Chapter III, I outline how this modelling technique can be 36 used to estimate the impact of economic conditions on party popularity controlling for the political context in which economic conditions are translated into public opinion. I continue to build upon this model by determining the appropriate autocorrelation and error process structure for party popularity and using a modified Box-Jenkins approach to determine the most effective lag structure for the economic variables. Finally, I discuss the resulting estimates of the effects of economic conditions on party popularity using the state-space model. In Chapter IV, I test the robustness of the estimated impact of economic conditions on party popularity by including historical and political contextual controls and put it all together providing a model that puts the dynamics of macroeconomic conditions and party popularity in Canada over the past half century in historical, political and technical perspective. In Chapter V , I discuss other potentially important political contextual variables that may account for variations in economic effects and variations in popularity dynamics between different time periods. I suggest directions for further research in this area. I also examine the potential role of the media in translating economic conditions into party popularity and shaping the political context in which this occurs. I suggest directions for further research in this area as well. 37 C H A P T E R II P U B L I C O P I N I O N A N D T H E S T A T E - S P A C E M O D E L 1.0 Introduction This dissertation considers both government and party popularity time-series as measures of government support. Most analysts using such data do not spend enough time trying to understand the problems which such public opinion time-series processes can pose. This chapter examines the many potential difficulties of working with aggregate, public opinion time-series data. I suggest an analytical approach which alleviates many of these challenges and provides a framework for examining the impact of macro-economic conditions on public opinion. This chapter suggests a different path of analysis for students of government and party popularity from that traditionally taken, suggesting that popularity must be understood in terms of its component dynamics. Many of these components represent the political context in which economic conditions are translated into government support. The contextual dynamics include shifts in baseline support produced by political events, a cycling in popularity which corresponds with the timing of elections, trending in popularity, unequal variances in popularity over time, and varying dynamics during distinct time periods. Other dynamics, such as errors correlating with time, are a product of the method of measuring popularity - the public opinion poll. The various aggregate popularity time-series used in this dissertation cover the period from 1957 to early 2000. They were constructed from individual level survey data collected by the Canadian Institute of Public Opinion (CIPO or Gallup Canada). The CIPO surveys are the most consistent measure of Canadian government/party popularity publicly available. For more than 50 years Gallup has been asking Canadians: "If a federal election were held today, which party's candidate do you think you would favour?" Government popularity is defined as the percentage that indicate they would vote for the party in government. Party popularity is defined 39 as the percentage that indicate they would vote for a particular party. (For the purposes of this dissertation, the calculation is made excluding those that indicate they do not know for whom they would vote - Appendix C describes data sources and data compilation.) Since the early 1970s, Gallup has run surveys including vote-intention questions monthly. Before this time, survey data is available on a less consistent but still highly regular basis. This chapter is outlined as follows: in part 2.0,1 identify the challenges posed by the party and government popularity time-series processes. I also introduce the state-space model and outline how it can be used to work with arid around such problems. In part 3.0,1 examine the outcome of this modelling technique applied to government popularity. I discover that government popularity since the early fifties can be divided into three distinct periods and that during these periods, government popularity has followed three distinct patterns of inter-election cycles and trends. In part 4.0,1 identify the need to model party popularity, rather than government popularity. I suggest a party popularity model and present the modelling results, demonstrating that the dynamics present in government popularity are reflected in party popularity. Further, I demonstrate that the state-space model effectively solves the problems identified in this chapter and provides a method by which the impact of economic conditions on party popularity can be properly modelled. This is done in the following chapter. 40 2.0 Government and Party Popularity Figure 2-1 is a plot of Liberal and Conservative party popularity from 1957 to 2000. Some of the largest shifts in party popularity coincide with fairly clear historical events. One of the most popular times for the Tories since 1957 was experienced from April-October, 1958. This is a product of Diefenbaker's electoral break into Quebec. The Liberals experienced the opposite fortune in popularity, losing much of Quebec for the first time since the 1930 election. The popularity of the Diefenbaker government quickly dissipated. This may in part have been a consequence of rising unemployment in the late 50s-early 60s. It may also have been in part reflective of the difficulty in holding a Quebec/Western Canadian coalition together. In terms of popularity, PC gains and losses were somewhat more dramatic than Liberal losses. The spike in popularity for the Liberals in December, 1970 corresponds with the FLQ Crisis. This is likely the product of a "rallying around the flag" effect, similar to that apparently experienced by the Liberals between March, 1977 and March, 1978 in response to the election of a separatist Parti Quebecois in the November, 1976 Quebec provincial election. In each case, the Liberal's gain was notably greater than the opposition PC's loss. Liberal government popularity declined sharply in the early 1980s. A n economic explanation would attribute this decline to simultaneously rising inflation and unemployment. Just previous to the 1984 election, Liberal government popularity spiked upwards in response to John Turner's selection as leader. The Liberals were unable to sustain this rise in popularity through to the election. In the 1984 election, Mulroney produced a coalition between Quebec and the West, similar to that of Diefenbaker. This gave the PC party a surge of popularity between October 1984 and April 1985. This surge was foreshadowed by the popularity of the PC party in 41 opposition in 1983. Subsequent to 1985, the popularity of the Tories dropped significantly. This corresponds with the debate over the FTA. PC popularity rebounded from this low just in time for the 1988 election. The record low popularity for the Tories between February, 1990 and January, 1991 was likely driven by the unpopularity produced by the Meech Lake Accord and Brian Mulroney's personal unpopularity - his approval rating declined to 15 percent. It could also have been in part the product of the early 1990s recession. The Liberal gains during this period were not nearly so extreme. The fate of the PC party became even worse in October, 1993 during Kim Campbell's election campaign. On a much smaller scale, the Liberals experienced a similar election-time downward spike in popularity during Jean Chretien's 1997 election campaign. However since the 1993 election, gains and losses by the Liberals are not obviously reflected in PC party popularity. Many (although not all) economic popularity studies use and focus on government popularity, rather than party popularity. (Figure 2-2 provides plots of government popularity.) This is done because the electoral accountability debate is ultimately about support for the governing party. As was explained in Chapter I, I find it is prudent to model the popularity of the Liberal and Conservative parties separately, regardless of whether they are in government or opposition. However before focusing on individual party popularity, the next section builds a model of government popularity so that the result can be compared with those from previous studies and so that the estimated dynamics of government popularity may be later compared with the estimated dynamics of party popularity. 42 3.0 Difficulties with Modelling Government Popularity When examining time-series data (such as government popularity) it is necessary to account for statistical challenges, such as autoregression and nonstationarity in the series. With the exception of those studies that use the Box-Jenkins approach, the studies conducted thus far fail to really consider nonstationarity. The best they do is attempt to control for nonstationarity produced by trending by using arbitrary time-count variables. They account for first-order autoregression through methods such as Cochrane-Orcutt or by including a lagged dependent variable but these methods are only effective to the extent that the time-series data is stationary (or at least, weak stationary). A time-series process (v,, v 2,..., v 7 ) is said to be weak stationary if: 1) E(y]) = E{y2) = ... = E(yt) = M; 2) Var(yx) = Var(y2) =... = Var(yt) = a1; 3) Cov(yl,y,_]) = Cov(y,,y,_T) = yT; whereju and a2 are the mean and variance of(yvy2,...,yr). yT =Cov(y,,y,_T) is called the autocovariance at lag r . Just as the autocovariances only depend on the lag, so do the 7 Y autocorrelations p(r) = —y = — . °- n The first condition for weak stationarity will be violated i f the mean of the time-series is correlated with time - that is, the series trends up or down. The second condition will be violated if the variances are correlated with time. This may occur if the underlying variance of the process and/or the variance in the measurement process itself systematically changes over time. The third condition will be violated if the autocorrelations are correlated with time. This will occur in data that contains cycles. 43 The Canadian government popularity time-series is, at times, clearly subject to each of these violations. Examining figure 2-2, we see a general downward trend from around 1970 until 1993 when government popularity surges to near record heights.94 Furthermore, an inter-election cycle appears present from 1975 on, although the post-1993 cycle differs from that before it. Such trending and cyclicity are violations of the first and third conditions of weak stationarity. One way to further explore these potential violations of weak stationarity is to examine the autocorrelation (AC) and partial autocorrelation functions (PAC) for government popularity. The autocorrelation of a time-series process at lag A: is a measure of correlation between time-series observations k units apart and the partial autocorrelation is a "measure of correlation between time-series observations k units apart after the correlation at intermediate lags have been controlled or 'partialed out'."9 5 This raises an additional potential complication in the Canadian government popularity time-series. Johnston observes that there appears to be three distinct time periods since World War II, in which the dynamics of government popularity are unique.96 The earliest period extends back before the beginning of our time-series to the end of the War and ends during the mid-seventies (shortly after the 1974 election). The second period continues from the mid-seventies until the 1993 election and the most recent period picks up from there. Johnston describes the first period as exhibiting no special inter-election rhythm. In the second, he notes a consistent cycle. Each election is followed by a honeymoon period in which popularity increases. 9 4 Figure 2-2 uses Kalman filtered data, making the earlier period of the time-series easier to interpret by including interpolated values for the months in which no poll was reported. Harvey, A. C. 1993. Time Series Models. 2nd ed. Cambridge, Mass.: MIT Press. 9 5 McCleary, Richard, Richard Hay Jr., Errol Meidinger, and David McDowall. 1980. Applied Time Series Analysis for the Social Sciences. London: Sage Publications. 9 6 Johnston, Richard. 1999. Business Cycles, Political Cycles and the Popularity of Canadian Governments, 1974-1998. Canadian Journal of Political Science 32 (3):499-520. 44 Subsequently, popularity drops below the level of the government's election return and bottoms out. Popularity then begins to recover as the government enters the next election. Underlying these cycles is a long downward trend. In the third period (after 1993), this downward trend ceases and government popularity surges up beyond the 50 percent level. According to Johnston, this level of popularity is largely sustained for the entire period except during election campaigns when popularity temporarily spikes downwards to produce a vote return within the forties.97 In order to accommodate the possibility of distinctive government popularity periods, the autocorrelation functions for each period are examined separately (figure 2-3). Note that a 1975/1979 separation between the first and second periods is used. This necessity is explained further below. Both the autocorrelation functions for the 1957-1975 and 1979-1993 periods indicate some autoregressive process at work but they decay far too slowly to be simple AR(1) processes. The partial autocorrelation functions for these first two periods also suggest government popularity cannot be modelled as a stationary first-order autoregressive process. If this was the case, the partial autocorrelation would simply show a single large value at lag 1 - both PACs have substantial values at lags two and three. The complexity of the autocorrelation and partial autocorrelation functions is almost certainly the product of trending. It is likely also the product of cycling. However, observations regarding cyclicity are inconclusive. The autocorrelation and partial autocorrelation functions do not show classical indications of fixed period cycling (this would exhibit itself as autocorrelation function peaks at regular intervals) but they would not be expected to because an inter-election government popularity cycle would have an uneven 45 frequency - one based on the inconsistent timing of elections. It is not clear what the partial autocorrelation for such a process would look like. The slow decay within the autocorrelation functions also suggests that while part of the dynamics of government popularity is stationary, part of it is likely integrated. A n integrated public opinion process is one in which all shocks carry over from one period to next. The value of the time-series at any time t is equal to the sum of all previous shocks in addition to the current shock. An integrated process within party popularity may represent permanent shifts in baseline support. In a stationary process a shock to the time-series will decay over the following periods. The speed at which the shock decays depends upon the degree of memory in public opinion. Because all shocks decay, the value of the time-series tends towards some equilibrium value. A stationary process within party popularity may represent the temporary and decaying impact of events such as changes in economic conditions. A time-series process which contains both integrated and stationary components is called fractionally integrated. Fractional integration is a lesser problem than straightforward nonstationarity but it is a complication that ought to be addressed. The autocorrelation function for the 1993-2000 period contains little evidence of either an autoregressive process or cyclicity and does not present itself as a simple stationary process. It is less straightforward to visually determine the potential for uneven variances over time and the violation of the second condition of weak stationarity. From a conceptual point of view, however, the violation of this condition is very plausible. Given the length of the time-series (about 43 years), it is risky to assume that the dynamics of government popularity, and 9 8 Wlezien, Christopher. 2000. An Essay on 'Combined' Time Series Processes. Electoral Studies 19:77-93. 46 therefore the variance of the error term in whatever model we use to represent those dynamics, will remain constant." It is for this very reason that each of the three periods identified above are considered separately. There is the further problem that the measurement error component of our model's error term may be correlated with time. Since 1974, Gallup has regularly used sample sizes of just over 1000 respondents. Before that time, many of the Gallup poll results used much smaller sample sizes (although, sometimes much larger). Moreover, the fifties and sixties component of the time-series contains a number of missing values at the monthly level of measurement. This means more values in an analysis must be interpolated. These interpolated values will, of course, contain larger errors than those which were directly measured.100 These circumstances could possibly produce greater variances in the earlier part of the time-series, compared to the later.101 A trend which runs counter to this but which may also produce complications is the increasing number of respondents since the early 1990s (except during election months) that indicate they do not know for whom they would vote. Since the measure is of decided voters, the increase in "don't know" respondents may produce greater variances in the latter part of the time-series, compared to the middle. Figure 2-4 graphs the number of decided voters interviewed each month. In addition to the increased frequency with which public opinion is measured over time, the figure also reveals how spikes in measurement accuracy occur around elections. This is produced by the combination of two phenomena. Leading into an election, there is an increase in the number of 9 9 Consequently, some analysts suggest only modelling short periods of time. Clarke, Harold D., and Gary Zuk. 1987. The Politics of Party Popularity: Canada 1974-1979. Comparative Politics:299-3\5. 1 0 0 Aggregating the data to a quarterly level does not solve this problem. Quarterly measurements made later in the time-series will still be more accurate and have smaller variances than those earlier in the time-series. 47 polls, while at the same time the number of undecided voters drops significantly. It is generally known that non-response rates are greater between elections than closer to them. 1 0 2 These non-response rates can be substantial, ranging from 25 to 30 percent between elections and dropping down to around 10 to 20 percent during elections. This has the potential to produce a cycling in the measurement accuracy and possibly even contribute to cycling in the popularity series, violating both the second and third conditions of weak stationarity. It is also interesting to note the general decline in the number of valid respondents indicating a vote preference since the early 1990s. This is the product of the noted increase in the proportion of survey respondents indicating that they do not know whom they would vote for i f an election were held. The potential for the problem of variances varying with time can be explored by applying the Breusch-Pagan test for heteroskedasticity to the model expressing government popularity as a function of time (r): govvote, = /?0 + /3xt +error . The calculated L M statistic is 152.375. With a x] distribution, the BP test results in a p value of 0.000. This finding suggests that the model's error term varies with t, confirming the very real potential for unequal variances in measured popularity over time. Overall, the evidence suggests great potential for the violation of the stationarity assumption in time-series models of Canadian popularity data. The problem of nonstationarity produced by inter-election cycles and trending at least has not gone unnoticed and various solutions have been employed. As mentioned, the most advanced solution has been the Box-Jenkins approach. However, this solution to cycling and trending is to eliminate them through differencing. While this can be a statistically acceptable procedure, it does not allow us to 1 0 2 Penniman, Howard Rae. 1981. Canada at the polls, 1979 and 1980 : a study of the general elections, AEI studies. Washington, D.C.: American Enterprise Institute for Public Policy Research. 48 examine these components. In public opinion series, these components are not only statistical challenges, they are also substantively interesting. Trending and cycling are very real dynamics within public opinion, with important political origins and consequences - simply eliminating them forfeits the possibility of understanding them. Therefore, the Box-Jenkins approach is inadequate i f we wish to fully understand the dynamics of public opinion. James Durbin, in his comparison of the Box-Jenkins A R I M A and state-space approaches, also notes that the Box-Jenkins approach does not necessarily lead to a unique model. Very different models can appear to fit the data equally wel l . 1 0 3 Moreover even if the Box-Jenkins approach was considered adequate, when cycling is driven by the timing of elections, as it appears to be here, it will not have a constant frequency. In which case, differencing the series will not eliminate the cycle and this nonstationary component will remain. US and European studies will often control for the election cycle, by employing an arbitrary election cycle count variable and a length in office trending variable. 1 0 4 The difficulty with including an election count variable is that it is arbitrary. This is evident in the wide variety of count variables employed in different studies.105 Samuel Kernell argues that such variables do nothing but measure time and are inappropriately used in such models. 1 0 6 Moreover, these counts usually only take into account the election period and do not account for the potential cyclicity of the data during times far from an election. As we shall see, this can be misleading. 1 0 3 Durbin, James. 2004. Introduction to State Space Time Series Analysis. In State Space and Unobserved Component Models: Theory and Applications, edited by A. C. Harvey, S. J. Koopman and N. Shephard. Cambridge: Cambridge University Press. 1 0 4 Lewis-Beck, Michael. 1988. Economics and Elections. Ann Arbor: University of Michigan Press. 1 0 5 Ibid. 1 0 6 Kernell, Samuel. 1978. Explaining Presidential Popularity. How Ad Hoc Theorizing, Misplaced Emphasis, and Insufficient Care in Measuring One's Variables Refuted Common Sense and Led Conventional Wisdom Down the Path of Anomalies. The American Political Science Review 72 (2):506-522. 49 A less common method is to construct models that explicitly include the election cycle as part of an underlying dynamic, such as modelling the honeymoon and decline as a product of an explicit retrospective comparison of the incoming government with the past government. This solution is as effective as the assumptions used to model the election cycle dynamics. Unfortunately, these assumptions are difficult to test directly. . None of these methods is able to control for variances in measurement error. Nor do they account for the problem that popularity is likely a fractionally integrated process. So, what is the solution? In order to truly understand the dynamics of public opinion, it is necessary to model explicitly the various components of which it comprises. In addition to the cycling and trending components, it is necessary to model fully integrated shifts in baseline support (e.g., those produced by political events) and stationary shocks/deviations (e.g., those produced by changing economic conditions) with decaying effects. At the same time, this has to be done in a way that accounts for the nonstationarity in the error component of the popularity time-series and the changing dynamics of popularity over time. This dissertation suggests an inductive approach, providing a great deal of flexibility in modelling different types of effects. It is a structural approach in that it builds models which explicitly include components representing the various dynamics of public opinion. The basis of the methodological option proposed is to express the time-series model in state-space form and include separate shocks, cycling, trending, baseline shifts and measurement error components. The basic model that contains each of these components is: 1 0 7 Hibbs, Douglas, Jr. On the Demand for Economic Outcomes: Macro Economic Performance and Mass Political Support in the United States, Great Britain, and Germany. 50 The Government Popularity State-Space Model GOWOTE, = a, + B, + eye, + v, Observation Equation a, = pat_x + Y\LIBl + y2PC, + s" Stationary Component B, = pxLIB, + (32PC, + rjibtrend, + T2pctrend, + ef B a s d i m C o m p o n m t eye, = ©! sm(A0) + 0 2 cos(^ 6>) + s'yc Cycling Component where • t = \,...,T at monthly increments; . *(a~(0,<rj,), ^ B ~ ( ° > < B ) ' ^ c ~ ( 0 , < , J , V , ~ ( 0 , < ) and COV{e„v,) = Q av is the standard deviation of the estimated sampling error calculated as yJpl (1 -/?,)/N, , where p{ is the proportion of valid respondents supporting the government at time = t and N, is the sample size. The sample size is calculated as the number of decided voters polled in each survey. If more than one poll was performed in any given month, the individual responses were combined and overall aggregate popularity values were calculated. The sample sizes in these cases would be the total number of decided voters obtained from combining the polls. 1 0 8 X is the frequency (1/wavelength) of the popularity cycle and is defined by the length of the inter-election period. (Note: fixing the wavelength to the inter-election period means that it varies from one election to the next.) 1 0 8 Including the separate measurement error term ( V , ) is closely related to Nathaniel Beck's use of the Kalman filter to estimate presidential popularity. Beck, Nathaniel. 1990. Estimating Dynamic Models Using Kalman Filtering. Political Analysis 1. 51 • p,yx,y2, / ? , , / ? 2 , r , , r 2 , 0 , , and 0 2 are parameters to be estimated. In the state-space model, empirical values of party popularity ( G O W O T E ) are considered the sum of structural elements B , , eye,, and a and measurement error vt. G O W O T E t is the time series that we observe and therefore the equation describing it (first line of the government popularity state-space model) is called the observation equation. The sum of the three structural components (second, third and fourth lines of the government popularity state-space model) represents the state of the system, or more precisely the state of government popularity. These components cannot be directly observed but in order to understand the dynamics of popularity we need to infer their behaviour. We can estimate the precision of each observation within the G O W O T E t time series. This precision is the inverse of the variance of the measurement error term. Given the precision of each observation and given the hypothesised structure of the structural components, Bayesian methods can be used to estimate the parameters of the structural components, such that they maximise the likelihood of each observation being made. Because the model contains a memory component, the estimation also maximises the likelihood of each observation being made given all other observations that were made - that is, the estimation of the state of popularity at a particular time t takes into account the estimated state of popularity at all other time points, giving greatest weight to those time points closest to t. This can be done because we know that government popularity in a given month is not independent of government popularity in the months immediately preceding it. This is particularly useful for those months in 52 which surveys contain few valid respondents or no survey at all was conducted and popularity has to be interpolated.109 Each of the structural components represent different dynamics. The cyct component explicitly accounts for any inter-election cycling that may exist within the government popularity series. Estimated parameters 0, and 0 2 can be used to calculate the cycle amplitude = and phase = cos'1 (a m^ i l u J e) for the inter-election cycle. The component a captures the effects of variables (presently unmeasured) not related to baseline'shifts or cycling. Once included, these effects would produce deviations (shocks) in popularity from its baseline. The p term represents and controls for the first order autoregression [AR (1)] within government popularity and determines the length of time an event which produces a deviation from baseline (and is modelled through the a component) continues to have an effect on popularity. The fading impact of these events describes a stationary process. Ultimately, it is through the a component that economic conditions will enter into the state-space model. Economic effects are modelled this way because they are expected to have long-term effects which are cumulative but not permanent. In other words, the electorate responds to the state of the economy over the long-term but a party is not punished or rewarded for the state of the economy in any given month forever after. B, is a measure of baseline support for the party excluding cycling. Political variables producing shifts in baseline support will eventually enter through this term. This component 1 0 9 Jackman, Simon. 2000. Estimation and Inference via Bayesian Simulation: An Introduction to Markov Chain Monte Carlo. American Journal of Political Science 44:369-398. Missing values are handled by using interpolation and then assigning large standard errors to the interpolated value. Within the Bayesian framework, this allows the surrounding time points which have more certain measurements to be used to fill in for the missing data. 53 includes two constants (/?, and /? 2). The first relates to the popularity of the Liberals in government and the second to the popularity of the Conservatives in government. The magnitude of these constants reflects the underlying support with which the Liberals or Conservatives begin their term in government. Once included, the dynamics modelled by the B, component can be designed to contain memory or not. If they are designed to contain memory, they are modelled as an integrated process. Such dynamics would represent an immediate shift in popularity in response to the occurrence of some political event. This shift remains forever after or until some other occurrence. In other words, the impact of the event becomes fully integrated. An occurrence can include the end of an event, in which case the impact of the event lasts only for the duration of the event. This is generally the type of effect that political events such as national crises are expected to have on popularity. Popularity responds immediately to the crisis and that response is sustained for the duration of the event but once it is over, popularity almost immediately returns to previous levels. The proposed method of separately modelling integrated and stationary processes is an alternative solution to the problem of fractional integration within the popularity time-series from those suggested by Christopher Wlezien or Janet Box-Steffensmeier and Andrew Tomlinson. 1 1 0 However i f fully integrated processes contained in the popularity series are not explicitly modelled in the B, component, they are likely to be captured by the residuals of the a component. This is because while the B, component can be used to model fully integrated political dynamics, this is done through the way in which the political variables are constructed. 1 1 0 (Box-Steffensmeier, Janet, and Andrew Tomlinson. 2000. Fractional Integration Methods in Political Science. Electoral Studies 19:63-76, Wlezien, Christopher. 2000. An Essay on 'Combined' Time Series Processes. Electoral Studies 19:77-93.) 54 The component itself is not fully integrated. It contains no memory term whatsoever. Therefore, the a component may remain somewhat fractionally integrated. The impact of two different forms of trending can be calculated by the y and z terms, r, and z2 measure long-term trending in the government popularity series, depending upon whether the Liberals or PCs are in power. The trend is modelled simply as a linear increase or decrease in popularity. The trend would generally be expected to be negative. Its structure is based on the idea that a government is formed through a coalition of interests. The longer a party is in power, the harder it becomes to hold this coalition together. Parameters y] and y2 measure memory-based, short-term trending. It is theoretically based on the notion that a party may gain or lose popularity by virtue of the fact that it is in government. On one hand, the party in government may be unable to avoid decisions that are inherently unpopular. On the other hand, it may have the resources to implement popular programs, which the other parties do not have. This trending component assumes that the impact of these actions produces a shift in popularity only so long as they are remembered. This memory is measured by p . Unlike the long-term trending, short-term trending is not ever-increasing or decreasing. Eventually, the shift in popularity produced by recent actions will be offset by the diminishing impact of past actions, as they are forgotten. The total drift that would occur in a party's popularity due to these actions i f they were to remain in government indefinitely is calculated as y/(\ - p). This value is the equilibrium level for the short-term trend - the total increase or decrease in popularity produced by the short-term trend. 55 The sum of B , , eye, and a (the state of government popularity) represents "filtered" values of government popularity, in that they exclude v,, the "noise" produced by survey measurement error.111 Including the measurement error term and estimating its standard deviation based on the number of valid respondents each month allows us to explicitly account for the variations in measurement accuracy produced by fluctuating sample sizes, previously described - that is, increased accuracy over time with rising numbers of polls, decreased accuracy over time with increasing numbers of undecided voters, spikes in accuracy near and during election months and cycling accuracy between elections. In order to account for the potential of three distinct government popularity periods, the time-series model is estimated separately for each proposed period. As previously noted, the model for the second period has to start in 1979. This is a consequence of including the memory-based, short-term trending term. It requires that the time-series begin the month after an election and the first election to follow the one held in 1974 occurred in 1979. With monthly measures of popularity, the first period contains 222 time points, the second period contains 213 time points and the third period contains 84 time points. The next section discusses the results of this modelling exercise. 1 1 1 Harvey, A. C. 1993. Time Series Models. 2nd ed. Cambridge, Mass.: MIT Press. 56 4.0 Government Popularity Dynamics Table 2-1 contains median values of the Bayesian-estimated distributions of the parameters for the estimated cycle, trend and memory components from the state-space model of government popularity. Figure 2-5 is a plot of the original government popularity data and the predicted government popularity based on the deterministic parts of these components. Parameter estimations were made using Winbugs. Winbugs performs Bayesian analysis of statistical models using Markov Chain Monte Carlo (MCMC) methods. See Appendix B for a guide to interpreting the results of such an analysis. During the first period (1957-1975), there is a statistically significant cycle but its amplitude is only 1.4 percent. This suggests that Johnston's observation that before 1974 government popularity did not follow any inter-election dynamic is not completely accurate.114 However, one can hardly be blamed for failing to visually catch such a small fluctuation. The autoregressive factor is about 0.82, suggesting a large amount of public opinion memory in the government popularity process from one month to the next. In a stationary process, the AR(1) is greater than -1 and less than 1. If it is between 0 and 1, then values closer to 1 indicate longer memory than values closer to 0. The short-term trend for the PCs is positive, while the long-term trend is negative. This is consistent with the initial increase in popularity of the Diefenbaker government and the subsequent long-term decline in popularity. The long-term trend for Liberal governments is much smaller but still negative. The short-term trend is not 1 1 2 Note: popularity dependent variable is entered into the model as a proportion rather than a percentage. 1 1 3 All models were estimated using two chains with varying initial values. 1 1 4 Johnston, Richard. 1999. Business Cycles, Political Cycles and the Popularity of Canadian Governments, 1974-1998. Canadian Journal of Political Science 32 (3):499-520. 57 significant. It should be noted that as expected, the long-term trend for both Liberal and PC governments in all periods is estimated to be downward. The difference between the Liberal and PC constant values is an indicator of the difference between the underlying popularity of Liberal and PC governments before trending occurs. In the first period, the difference is over 22 percentage points in favour of PC governments. This reflects the difference between the large majority initially put together by Diefenbaker and the initial minority government of Lester Pearson. During the second period (1979-1993), a clear cycle is evident. The amplitude is 6 percentage points, meaning that government popularity fluctuated by 12 percent throughout the cycle. With an autoregressive factor of 0.89, the time-series is nearly fully integrated - a government's popularity at each time point was roughly equal to its popularity at the previous time point plus an update of opinion. The time shift (phase) of the cycle indicates that when elections occurred, popularity was 8 percent of the cycle away from reaching its maximum. This means that in a full five-year inter-election period, an election occurs almost 5 months before popularity reaches its maximum in the cycle. It is incorrect to assume that i f an election occurs before popularity reaches a maximum in its cycle, the government would have received a greater proportion of the vote if the election was delayed. The rise in popularity subsequent to an election is probably a function of the election having occurred (a honeymoon). What is being described here is simply the position of elections along the popularity cycle, which may occur for a number of reasons. These will be considered in the concluding chapter - Chapter V . 58 The findings regarding the second period cycle are consistent with Johnston's observations.115 Following an election, government popularity trends upwards through a traditional honeymoon period. Popularity then proceeds to decline to its nadir before beginning to make a recovery. Soon after popularity has risen to average or just above average levels for the government, an election is called. 1 1 6 A similar pattern is observed in the first period cycling. Underlying second period cyclicity is a downward trend. The long-term decline in popularity during the second period is larger for Liberal than PC governments. This is a somewhat surprising result given the very large decline in popularity experienced by the Mulroney PC government. Although, it may be explained by the fact that the larger long-term Liberal government decline is partially offset by a positive and significant short-term trend. This short-term trend is insignificant for PC governments. During the third period (1993-2000), the amplitude of the cycle is less than that of the second period but not insignificant at 2.6 percentage points. The time shift suggests that elections occur when popularity is around 30 percent of the cycle away from reaching its maximum - that is, elections occur over 18 months before popularity reaches its maximum in a full five-year inter-election time period. (Of course, the inter-election period never reached a full five years between 1993 and 2000.) As was evident in the autocorrelation function, memory within the time-series for this period is low. The autoregressive factor is much less than that of the second period at 0.66. 1 1 5 Ibid. 1 1 6 The election called by Prime Minister Turner in 1984 appears to be an exception, in that the month in which the election was called Turner's popularity had just dipped below its average. However, the poll results just previous to the month in which the election was called were above-average and it was likely this information that Turner had available when making his decision. 59 Although the amplitude is smaller and elections are held earlier in the cycle, the government popularity cycle since the 1993 election is similar to that of the second period. Distinctive features of the third period are the particularly high constant for the Liberals (57 percent) and the insignificant trend. This is consistent with the sustained popularity of the Chretien government during this time. Another distinctive feature is that the election campaign period is marked by a sharp downward spike in popularity, followed by an equally quick recovery. While this downward campaign shock is most evident in the third period, it is not unique. During both the second and third periods, elections have represented deviations from the underlying cycle, and for the Liberals, almost always in a negative direction. This view of campaigns as producing deviations from the inter-election popularity cycle, rather than as being part of it, is a different perspective from the norm. This demonstrates the problem with using arbitrary election count variables to control for cyclicity. An election count variable is more likely to capture the negative effect of campaigns than the upward swing in popularity that most governments experience just prior to the campaign. It is this second phenomenon that constitutes the cyclicity which must be controlled and not the first. Modelling the effect of campaigns will be addressed in Chapter IV. Having modelled the various components of government popularity, we can now examine the estimated stationary a component to determine i f any nonstationarity remains. Remember the a component is assumed to be weakly stationary and this assumption needs to be confirmed before economic effects are modelled through it. Figure A - l (Appendix A) contains the autocorrelation functions for the a components of the first and second periods. In both cases, the autocorrelation functions suggest that the cyclicity and the long-term trending has been 60 largely removed (the a component does still contain short-term trending). The potential for unaccounted cycling or trending will continue to be monitored throughout the modelling performed in this dissertation. We are also now able to test for the potential of variances correlating with time within the a component by applying the Breusch-Pagan test to the model: govvote, = /?„ + /?,f •+ error using the a component in place of govvote. The calculated L M statistic is 0.575. With a zf distribution, the BP test results in a p value of 0.4482. This indicates that explicitly modelling the cyclical and trending components of the time-series and explicitly accounting for variations in measurement accuracy also removes the problem of variances correlating with time from the a component. 61 5.0 Party Popularity, 1957-2000 As noted, in many economic popularity studies the dependent variable is the popularity of the government, regardless of the party in power. This is what has been examined here, up to this point. Other studies model party popularity. These models model the impact of economic conditions on each party separately, regardless of whether they are in government or opposition. Modelling party rather than government popularity is important in situations that are not simple two-party races like they are in the US. Even in a two-party system, it is important if there is reason to believe the electorate's response is party-oriented, rather than incumbent-oriented.117 In Canada, there have been periods when third parties have played an important role. Consequently, the dynamics of the Liberal party's popularity has not simply been a mirror reflection of the PC's and vice versa. The dynamics of Liberal government popularity cannot be treated as equivalent to PC government popularity. Moreover, the Liberals have often been argued to "own" certain economic issues.118 Therefore, the impact of economic conditions on Liberal government popularity cannot be assumed to be equivalent to the impact of these same conditions on PC government popularity. Accordingly, it is prudent to examine separately the popularity of the Liberal and Conservative parties over the full 1957-2000 period. This requires using party popularity as the dependent variable in the popularity model. The proposed state-space model for Liberal party popularity is as outlined below. This equation is for the Liberal party only and, for illustrative purposes, it is the one that is described. It is important to keep in mind though that an equivalent and symmetrical equation was used for the PC party. The estimation results for both 1 1 7Brody, Richard. 1991. Assessing the President: The Media, Elite Opinion, and Public Support. Stanford, California: Stanford University Press. 1 1 8 Belanger, Eric. 2003. Issue Ownership by Canadian Political Parties 1953-2001. Canadian Journal of Political Science 36 (3):539-558. 62 parties are always presented. As will become evident, the dynamics for each party can be quite distinct from each other. Liberal Popularity State-Space Equation LIBVOTE, =at+B, + cyc, + vt a, = pa,_x + y1LIB, +y2PC, +s" B, = pxLIBt + ft2PCl + Txlibtrendt +r2pctrendl +£? eye, = 0 , sm(A0)LIB, + 0 2 cos(A0)LIBt + 0 3 sm(A0)PC, + 0 4 cos(A0)PC, +ectyc where • As in the government model, yx, y2, /?,, p2, r,, r 2 ,0,, © 2, ©3 and 0 4 are parameters to be estimated. However, these parameters represent slightly different dynamics than before. • Estimated parameters 0 , and 0 2 can be used to calculate the cycle amplitude = ^Q\+Q\ and phase = cos~] ( ^'Mt/e) for the Liberals when they are in government. Parameters 0 3 and 0 4 can be used to calculate the cycle amplitude and phase for the Liberals when they are in opposition. This is different from the government model where one set of parameters reflected the amplitude and phase for Liberal governments and the other set of parameters the amplitude and phase for PC governments. • The impact of trending on Liberal popularity while in government and opposition can be estimated by the y and r terms. This differs from the government model in the same way that the cycling parameters do. 63 • PC and LIB are dummy variables. PC is 1 when the Conservatives are in government and 0 otherwise. LIB is 1 when the Liberals are in government and 0 otherwise. In this state-space model, empirical values of party popularity (LIB VOTE) are considered the sum of structural elements B ( and eye,, a and measurement error v,. The PC and LIB dummy variables allow for the dynamics of Liberal popularity to depend on whether the PCs or Liberals are in government - that is, whether the Liberals are in government or opposition.1 1 9 Finally, in order to account for the potential of three distinct government popularity periods, the time-series models for both the Liberal and PC parties are again estimated separately for each proposed period - 1957-1975, 1979-1993, 1993-2000. Tables 2-2 and 2-3 contain median values of the Bayesian-estimated distributions of the parameters from the PC and Liberal state-space models of party popularity during the three periods under consideration. This includes estimates of the cycling, trending and autoregressive components of PC and Liberal party popularity. When plots of the density estimates for each parameter are examined, they are unimodal and appear to be roughly normally distributed. For example, figure A2 (appendix A) displays the distributions for the estimated AR(1) terms for each model. Note that to the extent that the distributions do deviate from normality, they tend to include long tails in the negative direction. This accounts for the fact that the AR(1) term for Liberal popularity during the third period appears to be insignificant despite the fact that its median estimated value is clearly different from 0. (Again, refer to Appendix C for a description of how these values were interpreted.) Both Liberal and PC dummies can appear together because there is no constant in the a component. 64 It has been demonstrated that the cycling component for the Canadian government popularity time-series is greatest during the 1979-1993 period, smaller for the 1993-2000 period and even smaller for the 1957-1975 period. This same pattern roughly holds for Liberal and PC party popularity. For both parties, the amplitude of the cycling terms are greater in the second than in the first period. A l l amplitudes during the 1957-1975 period are around 2 percent or less. During the 1993-2000 period, the Liberals were never in opposition and the PCs were never in government and therefore, it is reassuring that there is little difference between the calculated amplitudes for Liberal party and government popularity in the third period. Overall, the amplitudes for the government popularity cycles are generally closer to those calculated for Liberal governments and PC oppositions, rather than the other way round. Possibly, the cycling dynamics of government popularity are driven more by Liberal party than PC party popularity. For the sake of parsimony, only those long-term trends that were found to be significant were kept in the party popularity models. The long-term trends found to be significant for the Liberal party were for Liberal governments in the second and third periods. For the 1979-1973 period, popularity trended upwards 0.14 percent a month and for the 1993-2000 period, it trended downwards 0.02 percent a month (although, it started from a much higher level). While second period Liberal governments experienced long-term upward trending, they also experienced short-term downward trending. Chretien's 1993-2000 governments did not experience short-term trending in any direction. For the PC party, popularity trended (long-term) downwards for PC governments during the first and second periods (0.43 and 0.15 percent a month respectively). The popularity of the PC party in opposition also trended downwards during the second period, approximately to the 65 same degree as PC governments (0.14 percent per month) - although, the PC party in opposition during this period did experienced some short-term upward trending. With the exception of Liberal governments between 1979 and 1993, there is a great deal of consistency between the trending of government popularity and the trending of the PC and Liberal party popularity while in government. (The direction of short- and long-term trending is reversed for second period Liberal popularity in the party popularity model compared to the government popularity model.) Looking at the autoregressive components for both the Liberals and PCs, memory is the greatest within the 1979-1993 period. For the Liberal party, it is smallest for the 1993-2000 period (0.61). This is the smallest first order autoregressive value for either party in any period. This is consistent with the estimated dynamics of government popularity. There are some important differences between the PC and Liberal party popularity dynamics. However, there is a great deal of consistency in the estimated dynamics between the models utilising government popularity and the models utilising party popularity. This suggests that using party popularity will capture many of the same dynamics as government popularity, while accounting for the fact that dynamics between parties differ. Figures 2-6a & 2-6b plot the predicted popularity of the PC and Liberal parties based on the deterministic parts of the party popularity models - that is, the cycling, trending, and baseline components. Many of the largest movement in party popularity are predicted by these components. However, there is clearly still a great deal of residual movement to be explained. Part of this movement will , of course, be measurement error but a great deal of it can be attributed to economic conditions and political events not explicitly included in the model so far. 66 An approximation of the degree of movement that may be explained by such factors can be obtained by comparing the residual movement not explained by the cycling, trending and baseline components and that which cannot be attributed to measurement error to the total variance of the original popularity time-series. These values are presented in Table 2-4. For the Liberal party, the percentage of the total variance that remains to be explained is 52, 62 and 47 for the first, second and third periods respectively. For the PC party, it is 19, 26 and 55 percent. It is within this remaining movement that we expect to find the impact of economic conditions. Specifically, it is within the residual movement of the PC and Liberal party popularity a components that we expect to find the impact of economic conditions. This residual movement is plotted in figure 2-7. Notice these plots have the appearance of residuals from stationary processes, in that they appear to describe white noise processes. That the residuals are a white noise process can be confirmed by applying the Q-test to them. The results of these tests were presented in tables 2-2 & 2-3. The null hypothesis is that the residuals are white noise. We are unable to reject the null for the models from any of the three periods. The autocorrelation & partial autocorrelation functions for the a component residuals are presented in figure A-3 (Appendix A). These also indicate that the residuals are a white noise process. Based on this evidence, the a is a stationary process, meaning that the nonstationarity in the popularity time-series, produced by political contextual variables and correlated measurement error, has been . removed and so it is appropriate to model the impact of economic conditions through it. This is the first time that the statistical problems posed by the inter-election cycle or measurement errors correlating with time have been adequately addressed. Furthermore, this approach provides one 67 of the most sophisticated methods to date to account for the fact that popularity contains both stationary and integrated processes. As indicated, it is through the a components which have had the inter-election cycles removed that the impact of economic conditions will be estimated. Consequently, it would be fair to ask whether the inter-election cycle was itself not in part a product of economic forces. This may be true and would be consistent with a political business cycle hypothesis. Because the cycle is explicitly modelled by the state-space approach, this hypothesis can be explored. This highlights just how powerful the state-space approach can be. Even if the Box-Jenkins approach could contend with a cycle with an uneven frequency, it would control for it by simply eliminating it. In which case, there would be no hope of understanding what drives the cycle. The exploration of the estimated cycle is carried out in Chapter V . For now, the stage has been set to consider the impact of economic conditions on party popularity through the a component. This is the focus of the next chapter. 68 Chapter III E C O N O M I C P O P U L A R I T Y A N D T H E S T A T E - S P A C E M O D E L 1.0 Base Economic Popularity Model, 1957-2000 The sources of deviations in party popularity to be examined in this chapter are those - produced by economic conditions. There is substantial residual movement in party popularity that is not explained by the larger popularity dynamics discussed in Chapter II and that cannot be attributed to measurement error. It is quite reasonable to believe part of this unaccounted movement is produced by changing economic conditions. After all, the Canadian electorate often focuses on economic issues. As described in Chapter I, economic issues routinely top the voters' list of the most urgent problems facing the country and there is a great deal of Gallup poll evidence that Canadians feel the government can and should manage the economy.1 2 0 This chapter describes the economic variables included in the party popularity models. It builds Base economic party popularity state-space models, according to many of the most common assumptions regarding economic popularity functions found in the electoral accountability literature. It then proceeds to test many of these assumptions - in particular, those regarding the autocorrelation structure, error process, and the economic variables used. Once these issues have been settled, the Box-Jenkins approach is adapted to determine the appropriate lag structure for the economic variables. The Base and Box-Jenkins state-space models are compared and the optimal economic popularity function is determined. The economic variables included in the models of this dissertation are the three most commonly used and have been found to have the greatest impact on public opinion. These are inflation, changes in real per capita GDP (change in real income or 'economic 1 2 0 Clarke, Harold D., and Gary Zuk. 1987. The Politics of Party Popularity: Canada 1974-1979. Comparative Politics:299-315. 70 growth') and unemployment levels. Figures 3-1, 3-2 and 3-3 provide plots of each of these variables between 1957 and 2000. Inflation is the year-over-year change in the consumer price index, GDP is year-over-year percentage change in real personal income per capita and unemployment is the monthly percentage, seasonally adjusted. Changes in real per capita GDP and levels of unemployment were only measured quarterly up until 1976. After 1976 they were measured monthly, as was inflation throughout the 1957-2000 period. This in part determined the endpoint for the first period. Having a single economic indicator measured on two different timescales would have complicated things more than was necessary. As it is, the fact that the first period contains quarterly economic measurements, while the second and third only contain monthly measurements, may result in weaker estimated economic effects for the former. It is constructive to consider some of the larger shifts in these economic variables over the past half century. GDP throughout the 1957-1974 period exhibited nearly uniform positive growth. Unemployment was relatively low with its peak occurring in the late fifties and early sixties. This small spike was likely a reflection of the Eisenhower recession occurring at the time in the US. The small but steady increase in inflation from 1962 to 1970 was simply the product of rising aggregate demand. Compared to later patterns, inflation was relatively low. However, at the time it caused Canadians concern. In 1969 and 1970, the Bank of Canada controlled the money supply quite tightly. This was an anti-inflationary manoeuvring that may have started the drop in inflation in 1970. In May of 1970, the Canadian dollar was allowed to float. Where the Canadian dollar was strong in 1970-1971 making imports cheaper, inflationary pressures were moderated. After 1971, the 71 monetary system expanded quickly, the value of the dollar dropped and inflation again began to rise. 1 2 1 Around 1973, inflation spiked to a record high as a result of increasing oil prices produced by the Yom Kippur War and the Arab Oil embargo. Unemployment rates increased slightly at the same time and there were long-term consequences for economic growth. While increasing oil prices were outside of the control of the Canadian government, it has been argued that inflation was exacerbated in the latter half of the seventies by federal government deficit spending. Alternatively, wage push is held by * 122 some to be responsible for the skyrocketing inflation. The second spike in inflation during the late seventies/early eighties coincided with the overthrow of the Iranian Shah in 1979 and rising world oil prices. Around 1982, US Federal Reserve chairman Paul Volcker implemented economic policies that put an end to rising inflation. These policies simultaneously resulted in negative economic growth and skyrocketing unemployment in Canada, during the early eighties. They did however reduce inflation levels to those of the late sixties. The economy in the latter part of the eighties began to recover. Unemployment and GDP levels returned to those of the late seventies, prior to Volcker's anti-inflationary economic policies. GDP did experience a quick downturn in growth in 1987 but recovered in time for the 1988 federal election. In the early nineties, the world economy experienced a recession. In Canada, GDP grew on average as slowly as it did in the 1930s and unemployment was just as high. Inflation, on the other hand, was lower than it had been in 60 years. While the recession was a worldwide phenomenon, produced in part by a US recession and a sharp 1 2 1 Bothwell, Robert, Ian Drummond, and John English. 1989. Canada Since 1945: Power, Politics, and Provincialism. Revised ed. Toronto: University of Toronto Press. 1 2 2 Ibid. 72 decline in commodity prices, it was felt particularly strongly in Canada due to the government's restrictive monetary policies that began in the late eighties and was made worse by a large accumulated debt. Canada finally recovered from the recession in 1997. The rest of the nineties witnessed a steady decline in unemployment and economic growth cycled at reasonably high levels. Having reviewed some of the larger trends in these economic variables over the past 50 years, it is now appropriate to consider how they may have affected party popularity. Based on the state-space models developed thus far and the need to allow the impact of economic conditions to vary according to whether the party is in government or opposition, the following model of Liberal Party economic popularity is proposed. This is a "Base" economic popularity model predicated upon a number of the most common assumptions regarding the relationship between economic conditions and government support. An equivalent model is proposed for the popularity of the PC party. Once these Base economic models have been estimated and examined, the underlying assumptions will be tested and a modified Box-Jenkins approach will be used to further build upon these models. 1 2 3 Fortin, Pierre. 2001. Interest Rates, Unemployment and Inflation: The Canadian Experience in the 1990s. In The Review of Economic Performance and Social Progress The Longest Decade: Canada in the 1990s, edited by K. Banting, A. Sharpe and F. St-Hilaire. Montreal: Centre for the Study of Living Standards and the Institute for Research on Public Policy and distributed by McGill-Queen's University Press. 73 Liberal Popularity State-Space Equation L I B V O T E , = a, + B , + eye, + v, a, =pat_x + y]LIB, +y2PC, + 5lGOVtINFl_l + S 2 O P P t I N F l _ i + S 3 G O V , G D P l _ i + S 4 O P P l G D P t _ l + S 5 G O V t U N E M P t _ l + S 6 O P P l U N E M P , _ l + S 1 G O V t G D P t _ J N F t _ , + S s O P P , G D P l _ ] I N F l _ ] +e B, = fiiLIBt + (32PCt +Tllibtrend, +r2pctrendt + sf cyct = 0 , sm(XO)LIBt + 0 2 cos(W)LIBt + 0 3 sm(A6)PCt + 0 4 cos(XO)PCt + efc where • p,5i,52,SJ,Si,S5,56,S7,S8,yl,y2 are parameters to be estimated. • /?,,/? 2,r,,T 2 ,0,, 0 2 , 0 3 and 0 4 have already been estimated in the state-space model excluding economic variables and are fixed at these estimated values in this model. • G O V and O P P are dummy variables. G O V is 1 when the Liberals were in government and 0 otherwise. O P P is 1 when the Liberals were in opposition and 0 otherwise. Both GO V and O P P are 0 the first month after a new party became the government. • INF is the year-over-year change in the consumer price index. • G D P is year-over-year percentage change in real personal income per capita. • UNEMP is the monthly percentage, seasonally adjusted. The eye, and B, components have already been described. The component a acts as the dependent variable in the equation which estimates the impact of economic conditions on popularity. The p term within the a equation represents the first order autoregression within party popularity. The equation describing component a is similar to the partial 74 adjustment model recommended by Nathaniel Beck and used by Johnston.124 The inclusion of only first order autoregression is consistent with all Canadian studies of popularity and most non-Canadian studies.125 It is based on the assumption that the AR(1) term will capture all autoregression within the popularity timeseries. The GOV and OPP dummy variables allow for the impact of economic conditions on party popularity to vary across the party's position as government or opposition. In terms of economic effects, this is clearly necessary i f they are incumbent oriented. If this is the case, good economic performance will benefit the Liberals when they are in government and hurt them when they are in opposition. The GO V and OPP dummies are also necessary i f economic effects are party-salient goal oriented, although the economic issues that matter for the Liberals may differ from those that matter for the Conservatives. If economic effects are party-clientele oriented then the economic issues which the Liberals own should have a similar effect whether or not they are in government, in which case the GO I 7 and OPP dummies may be unnecessary. Whatever the case may be, the above party popularity equation is flexible enough to cope with any one of them. In fact, it gives us an opportunity to determine just how economic effects are oriented. By constructing GO V and OPP such that they are 0 the first month after a new party becomes the government allows for the probable fact that a new party in government will not be held accountable for the condition of the nation (economic or otherwise) in the month of or previous to the election they just won. 1 2 4 Beck, Nathaniel. 1991. The Economy and Presidential Approval: An Information Theoretic Perspective. In Economics and Politics: The Calculus of Support, edited by H. Norpoth, M. Lewis-Beck and J.-D. Lafay. Ann Arbor, Michigan: The University Of Michigan Press, Johnston, Richard. 1999. Business Cycles, Political Cycles and the Popularity of Canadian Governments, 1974-1998. Canadian Journal of Political Science 32 (3):499-520. 1 2 5 For the exception see Whiteley, P. 1984. Inflation, Unemployment and Government Popularity: Dynamic Models for the United States, Britain and West Germany. Electoral Studies 3 (l):3-24. 75 The component of the model estimating the impact of economic conditions includes an interaction term between changes in real per capita income and inflation. This is done for theoretical reasons. It is quite possible that in the minds of the electorate, the gains or losses in popularity produced by changes in real income are mediated by the degree of concurrent inflation. For example, when inflation is high the popularity gains to be made by a party in government from economic growth may not be as great as when inflation is more moderate. It is important to note that this interaction does not represent a real economic interaction between changes in real GDP and inflation. The interaction is hypothesised to be purely within the opinions of the electorate. If this were an economic rather than a public opinion model, the structure would certainly be different. Each economic term is entered into the model with a one-month lag. The one-month lag is consistent with most economic popularity models but is based on an 1 96 untested assumption. It is based on the idea that economic conditions have a relatively immediate impact on public opinion, yet some minimal time may be required for voters to become aware of changes in economic conditions and because polls may occur either at the beginning or end of each month. If the poll occurs at the beginning then that month's economic conditions will not have had an opportunity to influence it. Again, the time-series model is estimated separately for each period - 1957-1975, 1979-1993, 1993-2000. Tables 3-1 and 3-2 present the estimated parameters for the economic state-space models predicting Liberal and PC party popularity. Before examining the estimated parameters, it is important to note that the Q-Tests suggest that the residuals from the a components of the estimated state-space models are not Again, for the exception see P. Whiteley, 1984. 76 significantly different from white noise. 1 2 7 This is a necessary condition in order to argue that the stationarity assumption is not violated by these models. The particular stationarity assumption made in these models is that a components are modelling stationary processes. This is required to argue that the estimated economic effects estimated through the a components are reliable and not spurious. It is also important to note that including economic variables changes the estimated parameters for the autoregressive and short-term trending components of the model. First order autoregression changes little for the second period. For the 1957-1975 period, the AR(1) term becomes negative for both the PCs and Liberals. It is unclear why this occurs. It should be noted that, unlike the distribution of most other estimated parameters, the AR(1) term for the first period Liberal popularity becomes bimodal once economic variables are included (figure 3-4a). As can be seen, the distribution contains two distinct peaks. The bimodal distribution may suggest that autoregression is different for one part of the a component time series than another. However, the AR(1) term distribution was approximately normal before economic variables were included. This may suggest that economic conditions affect Liberal popularity at a different rate for one part of electorate than another. This may be evidence that the a component has remained fractionally integrated to some degree. Although the sizeable standard deviations for the estimated AR(1) terms in the first period suggests an insignificant AR(1) term for the Liberal party, it is clear from its distribution that the autoregression is in fact different from 0. The third period Liberal 1 2 7 The null hypothesis for the Q-Test is that the time series (in this case the residuals from the a component) is white noise. 1 2 8 Wlezien, Christopher. 2000. An Essay on 'Combined' Time Series Processes. Electoral Studies 19:77-93. 77 autoregression term is also estimated to be insignificant due to a large standard deviation. As can be seen in figure 3-4b, this is largely the consequence of the estimated distribution for the AR(1) term being negatively skewed. In fact, the first order autoregression for Liberal popularity in the third period appears substantively different from 0 and positive. Once economic variables are included, few short-term trending parameters remain statistically significant. First period Liberal government popularity experienced positive short-term trending, as did second period PC opposition popularity. The estimated parameter for first period Liberal government popularity short-term trending (0.2969) is misleadingly large because of the negative AR(1) term estimated for this period. The interpretation of parameters in a time-series model with negative autoregression is discussed further in chapter IV. Examining now the estimated impact of economic conditions, there is little evidence that party popularity was affected during the 1957-1975 or 1993-2000 periods. However during the 1979-1993 period, economic growth and inflation played a statistically significant role. The significance and direction of the estimated parameters in the second period suggest that while in government, the PCs benefited from increases in economic growth, just as the Liberal opposition was disadvantaged by it. The PC party in opposition was punished for both increases in GDP and high inflation. The GDP/inflation interaction term for PC governments and Liberal oppositions are significant. The interaction between GDP and inflation make straightforward interpretation of the impact of these economic factors difficult. In order to get a picture of the magnitude of the impact of GDP and inflation on party popularity, the combined month-to-month immediate contribution of these economic conditions is plotted in figure 3-5. This is the 78 shift away from baseline support produced by changing economic conditions each month. Baseline support is the sum of the baseline shift, cycling and trending components. It is important to note that the plotted values do not represent the total contribution of economic conditions in a given month, because the estimated models include first order autocorrelation terms. The residual impact of past economic effects will also contribute to popularity. In addition to the immediate contribution of economic conditions to party popularity in a particular month, there are two other ways of thinking about the potential magnitude of economic effects. The first is to consider the cumulative contribution that the economic conditions in a particular month make to a party's popularity in the months to follow. The cumulative contribution depends upon the degree of memory. For example in the 1979-1993 popularity models, the AR(1) term for the PCs and Liberals have a value of 0.82. This high level of memory means that if economic conditions produced a 1 percent shift in popularity away from baseline support in a given month (month 1), approximately 82 percent of this shift will continue to be felt in the next month (month 2). If the individual monthly contributions to popularity are added together, the economic conditions that produced the 1 percent immediate shift produces overall a cumulative contribution of 5.6 percentage points. Smaller estimated AR(1) terms mean smaller cumulative effects for economic conditions over time because the effects dissipate more quickly. Another perspective on the magnitude of economic effects is obtained by considering the shift that would be produced by economic conditions held at a constant level. As will become apparent, this is actually just a different perspective on the same 79 calculation. For example i f the economic conditions that produced an initial 1 percent shift in second period Liberal or PC popularity were held constant, eventually about 5.6 percent of each month's party popularity would be the cumulative consequence of these constant economic conditions. This 5.6 percent contribution (as an increase or decrease) would remain until economic conditions changed. In a model with a positive autoregressive term, considering only the immediate impact of economic conditions tends to underestimate their overall effect. The over time impact of economic conditions must also be kept in mind, although it is the immediate impact that will be referenced in the following discussion. When growth in real per capita GDP hit a high of 5.71 percent under the Liberal government during the last quarter of 1983 and the first quarter of 1984, inflation varied between 4 and 5.5 percent. Liberal government popularity was unaffected but the Tories lost as much as 4.2 percent. Previous to this point in time in the fourth quarter of 1982, economic growth was down to -4.81 percent and inflation varied between 9 and 10 percent. Despite poor economic growth, the Liberals still were not penalised for economic conditions. The PCs were penalised by approximately 3 percent. This was due to the effect of high inflation. It would seem that during times of high inflation, voters were reluctant to turn to the PC party and punish the Liberals for poor economic growth. Put another way, the gains to be made by a PC opposition by a downturn in GDP were offset by a reluctance to vote PC during times of high inflation. Between 1984 and 1993, PC governments benefited from economic growth and the Liberal opposition was hurt. These gains and losses were moderated by inflation. Voters were again less likely to reward PC governments for economic growth when 80 inflation was high. When economic growth hit 5.29 percent in the first quarter of 1988, inflation hovered just above 4 percent. The PC government at the time only gained approximately 0.8 percent and the Liberal opposition only lost approximately 1 percent. Overall during this period, the party in power benefited from economic growth and the party in opposition was penalised. However when inflation was high, voters were reluctant to vote for the PC party regardless of whether or not they were in government or opposition. Given the relatively high inflation levels throughout this period, this gave the Liberal party an intrinsic advantage. (Average growth in GDP during the second period was 1.5 percent and average inflation levels were 9.2 percent.) One interpretation of this effect is that the Liberal party "owned" the inflation issue. However, the Liberal party was never really directly rewarded for inflation. Therefore, it could alternatively be argued that the PC party was simply an unattractive option during times of high inflation. So far, it would seem that the success of the Canadian electorate at holding the federal government accountable for the performance of the economy has been mixed. Growth in GDP, followed by inflation, is most consistently translated into party popularity but only during the 1979-1993 period. Governments seem to have been immune to judgment based on the performance of the economy before and after this period, and no party at any time appears to have been held responsible for levels of unemployment. However, the Base economic models presented in this chapter are based upon a number of theoretically derived assumptions. In addition to testing these , assumptions, alternative forms of the state-space economic popularity model must be considered. This is done in the next two sections of this chapter. 81 2.0 Testing the Base Economic Popularity Model The Base economic popularity model estimated above makes a number of assumptions in its construction. These assumptions must be examined further, as is done in this section. 2.1 The GDP/Inflation Interaction Term An argument was made for including an interaction term. This argument was that in the minds of the electorate, the gains or losses in popularity produced by changes in real income are mediated by the degree of concurrent inflation and vice versa. In order to illustrate the impact of this interaction term, the economic popularity models were rerun without it (tables 3-3 & 3-4). The Base economic popularity model found no economic effects in the first and third periods. This is not changed by eliminating the interaction term. When the second period state-space models are run without a GDP/inflation interaction term, the negative impact of GDP on Liberal party popularity while in opposition loses its significance and becomes much smaller. For PC governments during the second period, inflation becomes significant but is of a similar magnitude, while the impact of GDP on PC governments during the second period is reduced by over 70 percent. The impacts of inflation and GDP on the PC party, while in opposition during the second period, are reduced by a third and three-quarters respectively. Both become statistically insignificant. The fact that there is likely an interaction between economic growth and inflation within the mind of the electorate suggests that it is important to include an interaction term in the state-space models. Given that it is statistically significant during the 1979-1993 period for the Liberals in opposition and the Progressive Conservatives in government and that many economic effects become insignificant when the interaction 82 term is removed, it is clear that previous models which did not include this term would have missed important economic effects. 2.2 The Unemployment Variable The measures of inflation and GDP for the basic popularity model are calculated in terms of year-over-year changes. For unemployment, the value used is simply the monthly percentage. Consequently, it is a useful exercise to re-estimate the basic model, replacing the monthly unemployment percentage with the year-over-year change in the monthly unemployment percentage. The results from this estimation are found in tables 3-5 and 3-6. For both the Liberals and Progressive Conservatives, the impact of unemployment remains insignificant in all three periods, suggesting the change of variable fails to improve the models. Replacing level of unemployment with change in unemployment does cause the impact of inflation and its interaction with GDP on the Liberal party while in opposition during the first period to become statistically significant. The effect on the significance of other economic variables is not sufficient grounds to change the form of the unemployment variable. However, this effect is something that should be considered in further models. 2.3 The Autoregression Structure The basic economic popularity function includes an AR(1) term. The appropriateness of this lag dependent variable structure can be explored by examining the autocorrelation and partial autocorrelation functions for the PC and Liberal popularity time series for each time period. The autocorrelation and partial autocorrelation functions are presented in figures 3-3 through 3-5. These autocorrelation and partial autocorrelation functions are for the estimated a component, rather than the original measured popularity series. This is 83 done because it is the autoregression structure of this component that is actually being modelled. Examining the autocorrelation functions, the decay from lag to lag in all three periods for both the Liberals and Progressive Conservatives suggests at least first order autocorrelation. The decay is not exponential and therefore suggests that the a component continues to retain the dynamics of a partially integrated process. This may also suggest a high order moving average process. However, the partial autocorrelation functions show no evidence of any such process. (This would present itself as a slow decay in the partial autocorrelations from lag to lag.) The partial autocorrelation functions also show no evidence of trending or cycling. As might be expected, the first order autocorrelation is positive. For the first two periods, the partial autocorrelation functions provide evidence of possible second order 1 autocorrelation (positive in the first period and negative in the second). The potential of negative second order autoregression is also suggested for third period PC popularity. Despite the potential second order autoregression, the fact that the residuals for the component using only first order autoregression were white noise (tables 3-1 & 3-2) suggest that including this second order term is unnecessary. As a rule, time series 130 models should include as few autoregression and moving average terms as possible. A 1 2 9 The negative second order autocorrelation suggests that significant increases or decreases in a party's popularity may partially be countered within two months. Keep in mind, the model already contains constant and trending terms. The terms are often significant and model the tendency of party popularity to equilibrate at some base level. Therefore, this potential second order autocorrelation would represent a response to shifts in party popularity in addition to the natural tendency of popularity to drift towards its base level. Substantively, this may be produced by a party taking measures to increase its own popularity when down in the polls and taking measures to decrease is opponents popularity when the opponent is up in the polls. 1 3 0 McCleary, Richard, Richard Hay Jr., Errol Meidinger, and David McDowall. 1980. Applied Time Series Analysis for the Social Sciences. London: Sage Publications. 84 possible exception to this conclusion is first period Liberal popularity. As was demonstrated, the estimated first order autoregression term is bimodal. It is a useful exercise to rerun the state-space model for Liberal popularity during this period containing both AR(1) and AR(2) terms to examine if it accounts for this distribution. This model appears as follows: Liberal AR(2) Popularity State-Space Equation LIB VOTE, =a,+B,+cycl+vt a, = p\at_x + p1at_2 + yxLIBt + y2PCt + 8lGOV,INFt_]+82OPPlINF,_] + 5,GOVtGDPs_,+ 5fiPPtGDPt_x + S5GOVtUNEMP,_x + 8bOPPtUNEMP,_, + S.GO^GDP^INF,^ + 5&OPP,GDPt_,INFt_x + ef Bt=pxLIBt+P2PCl+e? cy^ = 0, sin(A0)Lffi, + 0 2 <zos(X6)LIBt + 0 3 sm(W)PC, + 0 4 cos(A0)PC, + The results for this first period Liberal AR(2) model are presented in table 3-7. The first order autoregression term is estimated to be negative and the second order autoregression term to be positive. This is not what would be expected from the partial autocorrelation function, which indicates positive first order autoregression. Examining the estimated distributions of the autoregression terms reveals that both terms are bimodal, making the median a poor measure (see figure 3-6). This may explain the discrepancy between the partial autocorrelation function and the estimated autoregression parameters. Clearly, the inclusion of a second order autoregressive term does not account for the bimodal distribution of the estimated AR(1) term in the original 1957-1975 Liberal economic popularity models and is therefore unhelpful. 85 Overall, the evidence suggests that the Inflation/GDP interaction term is an important component of the state-space economic popularity model and that the use of unemployment levels (rather than changes) and first order autoregression is appropriate and adequate. Next, it is important that the lag structure of the economic variables be considered further. The lag structure simply refers to the lag at which each economic variable enters into the model. The Base model includes economic variables with a one-month lag. As previously noted, this is one of the most popularly employed structures. It is a choice based upon theoretical considerations but is rarely empirically substantiated. Despite the popularity of the one-month lag, scholars such as Paul Whiteley have argued that a systematic approach is required to determine the appropriate lag-structure. To do this, a more inductive approach to model construction than that employed for the Base model is required. The Box-Jenkins approach is a well tested method for doing just this 131 for A R I M A models. This approach is adapted here for the state-space model. 1 3 1 Ibid. 86 3.0 Box-Jenkins Economic Popularity Model, 1957-2000 The first step in the Box-Jenkins approach is to determine the univariate model (or transfer function) for all time series. In this case that would be for the party popularity and economic variables. If nonstationarity is evident then the series is differenced appropriately. This strategy is appropriate for the economic variables but cannot be applied to the party popularity time series. As noted before, i f nonstationarity is in part produced by cycling with an inconsistent frequency (based on the timing of elections), differencing the series will not correct the problem. Fortunately, the state-space approach allows us to extract the stationary component of party popularity in which we expect to find the effects of economic conditions. This is the a, component. It is this component that we utilise rather than the differenced popularity series usually used in the Box-Jenkins approach. This is the first way in which the model construction method used here differs from the traditional Box-Jenkins approach. Once the transfer function is determined for each stationary economic series, it can be inverted and applied to both the economic series and the a, component of the party popularity series estimated in Chapter II excluding economic variables. The resulting series are referred to as prewhitened. The cross-correlation function between each prewhitened series (economic and popularity) is calculated in order to identify potential lags at which each economic series should enter the popularity model. Having identified the appropriate lag structure, the model is estimated and the resulting residuals are examined. If the residuals are not white noise, an appropriate model for the noise component may be identified. If this is necessary, the model is reestimated and the residuals are examined again. The method employed here deviates 87 from the Box-Jenkins approach in a second way. Once the Box-Jenkins models have been identified, the statistically significant elements of the theoretically constructed Base economic popularity models are combined with those from the Box-Jenkins models and the state-space models are reestimated. The results are compared to determine the optimal model. Once the appropriate models have been identified, the new model residuals can be cross-correlated with the prewhitened economic variables to identify any lagged relationships which might have been missed. To begin, the transfer function of the economic variables is considered. The seasonal-cycling components have already been removed from the unemployment series at the source by Statistics Canada and eliminated from the GDP and Inflation variables through yearly differencing. This differencing occurs as a consequence of calculating GDP and inflation as year-over-year changes. The potential trending component of these economic variables can be explored through direct estimation. Modelling each economic variable within each time period as a first order autoregressive process with trending (table 3-8), it becomes evident that the trending component is not statistically different from zero in each case and is unnecessary. The AR(1) term in each case is approximately 1 suggesting a fully integrated process, otherwise known as a random walk. The Q-tests indicate that each model's residuals are a white noise process, suggesting the AR(1) model accounts for the autoregression within the economic series. The autoregression structure of each economic variable can be further examined through plots of their autocorrelation and partial autocorrelation functions within each time period. These are presented in figures 3-7 through 3-9. 88 The autocorrelation functions suggest first order autoregression with the potential of higher order autoregression. The partial autocorrelation functions confirm the evidence of first order autoregression and, in many cases, indicate small but significant autoregression at lags of 13 and 25 - this is most evident for inflation and GDP. In some instances, there is evidence of autoregression of other orders. There is no evidence of a moving average process.132 If there had also been statistically a significant spike at lag 12 of the partial autocorrelation functions, then a first order autoregressive cyclical process may be suspected. However, this is not the case and there is no theoretical reason to expect that any seasonality remains within the economic data. More likely, the statistically significant lags are a byproduct of removing seasonality from the economic data - in the case of inflation and change in GDP, through the differencing involved in their calculation. Given the likely stationarity (absence of cycling or trending) of the economic variables, the observations made examining the autocorrelation and partial autocorrelation functions can be used to determine the transfer function for the economic variables. Given the persistence of first, thirteenth and twenty-fifth order autocorrelation AR(1, 13, 25, 37) models were estimated (without a trending term). The AR(37) term was included because of the potential pattern exhibited by the consistency of 12 months between 1,13 and 25 lags. 1 3 3 The estimated parameters for these models are presented in table 3-9. These models were estimated for each economic variable across the entire 1957-2000 time period. 1 3 2 There is some indication of statistically insignificant cycling within 1993-2000 GDP. 1 3 3 Note: this is not evidence of seasonality. Evidence of seasonality (without autocorrelation) would be lags at 12, 24, 36, etc. 89 In each case, the estimated AR(1) parameter is again very close to 1. The higher order lag term parameters are always either statistically insignificant or substantively inconsequential. This is consistent with the fact that the simple AR(1) models appear to account for all autoregression. The AR(1) model was also estimated for each economic variable for the purposes of comparison. The estimated AR(1) term changed very little (table 3-10). Plots of the autocorrelation and partial autocorrelation functions for the GDP/Inflation interaction were also examined. These are presented in figure 3-10. Based on these plots, the appropriate transfer function appears to be a AR(1, 4, 13) process. The estimated parameters for this transfer function are presented in table 3-11. In accordance with the Box-Jenkins approach, the estimated transfer functions for each economic variable were used to prewhiten the economic and the estimated a component of the public opinion data series. The AR(1) parameters from the estimations of the full AR(1, 13, 25, 37) models were used in this prewhitening procedure for the GDP, inflation, unemployment and party popularity variables.1 3 4 The AR(1) and AR(4) parameters were used for the interaction term. As an example, unemployment was prewhitened as follows: prewhitened(unemployment), = unemployment\ - .9997 * unemployment(_,. Cross-correlation functions for each economic and public opinion prewhitened data series were calculated. These were calculated for up to 20 lags. These values are presented in tables 3-12 and 3-13. The cross-correlation functions between the public opinion and economic variables are expected to change depending on whether the party is 1 3 4 It is unnecessary to include the AR(13), AR(25) and AR(37) terms as they are substantively insignificant. 90 in government or opposition. Consequently for the first period, the cross-correlation functions were calculated separately for the 1957-1963 and 1963-1975 time segments. The PCs were in government during the 1957-1963 segment and the Liberals were in government during the 1963-1975 segment. For the second period, cross-correlation functions were calculated separately for the 1980-1984 and 1984-1993 segments. The Liberals were in government for the first and the PCs for the second. The cross-correlation values can be used to determine the potential lag-structures of the economic variables within the at components of the state-space models. By convention in order for a cross-correlation value to be statistically significant, it must be greater in absolute value than two times its standard error. At lag the standard error is calculated as SE(k) = 1 , where N is the total number of points in the y/N-k time series.135 These values are also presented in tables 3-12 & 3-13. The bolded values in the tables are clearly statistically significant and the bolded + italicised values are within 0.01 of being statistically significant. In the 1957-1963 segment, none of the economic variables appeared to be cross-correlated with the popularity of Diefenbaker's PC government. However, GDP, lagged 6 months, does appear to be cross-correlated (negatively) with the popularity of the Liberal opposition at the time. The cross-correlation between unemployment, also lagged 6 months, and Liberal opposition popularity is on the edge of being statistically significant. During the 1963-1975 period, GDP, lagged 2 and 3 months, is negatively cross-correlated , with PC opposition popularity, as is the interaction between GDP and inflation, lagged 3 1 3 5 McCIeary, Richard, Richard Hay Jr., Errol Meidinger, and David McDowall. 1980. Applied Time Series Analysis for the Social Sciences. London: Sage Publications. 91 months. Inflation lagged, 10 months, is cross-correlated with Liberal government popularity during this time. Based on these observations, a first period PC party state-space economic popularity model with inflation, GDP and their interaction, lagged both 2 and 3 months, was estimated. A first period Liberal party economic popularity model with inflation, GDP and their interaction, lagged both 6 and 10 months, and unemployment, lagged 6 months, was also estimated.1361 refer to these models as "Box-Jenkins" state-space models. Tables 3-14 & 3-15 present the estimated parameters for these models. In the case of the PC model, none of the estimated coefficients appear to be significant. While the distribution of the AR(1) term is unimodal in the Base economic popularity model, it is bimodal in the Box-Jenkins model (figure 3-12). Neither the Base or Box-Jenkins state-space models find any impact of economic conditions on PC party popularity during the first period. When the Liberal model is estimated, only the impact of unemployment is significant. This suggests that the popularity of the Liberal party, while in opposition during this period, benefited from increases in unemployment six months prior. Given the insignificance of all economic variables lagged 10 months, the Liberal model was re-estimated without them. Unemployment continues to have a significant impact on the popularity of the Liberal party while in opposition, although the magnitude of the effect is smaller. The distribution of the AR(1) term also continues to be bimodal in the model excluding the economic variables lagged 10 months (figure 3-12). The modified Box-Jenkins model including economic variables, lagged 6 months, is an 1 3 6 Unemployment was only entered into the opposition side of the model. A model with unemployment entered into the government side was also estimated - the effect of unemployment on Liberal government popularity was insignificant. 92 improvement over the Base model, which suggested that the economy had no impact on Liberal party popularity between 1957 and 1975. Turning now to the second period, the popularity of Mulroney's PC government between 1984 and 1993 is cross-correlated with GDP and the interaction between GDP and inflation, lagged 13 months. While in opposition, the popularity of the PC party between 1980 and 1984 is cross-correlated with unemployment, lagged both 9 and 10 months. It is also cross-correlated with GDP, lagged 8 months, and inflation, lagged 14 months. Based on these observations, a Box-Jenkins second period PC party state-space economic popularity model was estimated. This model included the impact of GDP, inflation and their interaction, lagged 13 months, on popularity while the Tories were in government. The model also included the impact of GDP, lagged 8 months, inflation, lagged 14 months, and unemployment, lagged both 9 and 10 months, on PC popularity while in opposition (table 3-16). Inflation, GDP and their interaction, lagged 13 months, all have a statistically significant impact on PC government popularity. Each of the economic variables entered into the opposition part of the model is statistically insignificant. Due to the significance of a number of the one-month lagged economic variables from the Base economic popularity model, a second PC popularity model was run including these variables with the statistically significant 13 month lagged variables from the Box-Jenkins model. I refer to this as a "combined model." In this combined economic popularity model, GDP, inflation and their intercept, lagged 13 months, continues to have a statistically sufficient impact on PC government popularity - as does inflation and the intercept between GDP and inflation, lagged only one month. On the opposition side of the model, only inflation, 93 lagged one month, is statistically significant. The combined economic popularity model is an improvement on both the Base and Box-Jenkins models. Examining the cross-correlation table for Liberal government popularity between 1980 and 1984, the only significant cross-correlation is with unemployment, lagged 0 months. The popularity of the Liberal party while in opposition between 1984 and 1993 is also cross-correlated with unemployment, lagged 0 months, and with inflation, lagged 11 months. A model including each of these terms for the full 1979-1993 second period was estimated (table 3-17). None of the economic variables included in the Box-Jenkins model appears to have an impact on Liberal Party popularity. The Base economic popularity model performed better, demonstrating a statistically significant impact on Liberal opposition popularity by GDP and its interaction with inflation, lagged one month. During the 1993-2000 period, PC party popularity does not appear to be cross-correlated with any of the economic variables. This is consistent with the Base model finding of no economic effects for third period PC popularity. Liberal government popularity, on the other hand, is cross-correlated with inflation and the interaction between GDP and inflation at lag 5. A new Liberal party economic popularity model was estimated. It included inflation, GDP and their interaction, lagged 5 months (table 3-18). Only inflation proves to be statistically significant, suggesting Liberal government popularity was vulnerable to raising inflation between 1993 and 2000. 1 3 7 This model appears to be an improvement over the original Base model, which found no significant economic effects. 1 3 7 This model was rerun two more times - once including unemployment lagged one month and once including unemployment lagged five months. In neither case was unemployment statistically significant or did its inclusion substantively change the estimated parameters for other economic variables. 94 Based on the results of estimating the Base, Box-Jenkins and combined models for all three periods, a tentative final form for the a, component of each model may be proposed (next page). The results are tentative because it is still necessary to check if the residuals from the estimation of the a, component are white noise and it is necessary to cross-correlate the residuals with the prewhitened economic series to capture any missed lagged economic variables. Before this is done and the final form for the state-space economic popularity model is established, there are some important political contextual variables that need to be included in theB, component of the model. This is the task of the next chapter. 95 at Components of State-Space Models 1957-1975 PC Party Popularity a, - pa,_x + y^IB, + y2PC, + s" 1957-1975 Liberal Party Popularity a, = pa,_x + y{LIB, + y2PC, + 5xOPP,UNEMP,^ + e" 1979-1993 PC Party Popularity a, =pa,_l+ylLIBt + y2PC, + 8xGOVtINFt_x + 82GOV,GDP,_x + S3GOV,GDP,_xINFt_x .+StGOV,lNFl_l3 + S5GOV,GDP,_l3 +S6GOVtGDPl_l3INF,_n+S7OPPlINFl_l+£? 1979-1993 Liberal Party Popularity a,=p\a,_x +yxLIB, +y2PC, + 8fiOV,GDP,_x + S2GOV,GDP,_xINF,_x + 83OPP,GDP,_xINF,_x + s" 1993-2000 PC Party Popularity a, -pa,_x + yxLIB, + y2PC, + s" 1993-2000 Liberal Party Popularity a, = pa,_x+yxLIB,+y2PC, + 8xGOV,INF,_5 + 82GOV,GDP,_5 + S.GOV.GDP^INF,^ + s? Chapter IV P U T T I N G T H E S T A T E - S P A C E E C O N O M I C P O P U L A R I T Y M O D E L I N T O P O L I T I C A L C O N T E X T 1.0 Theoretical Impact of Political Context The proper structure of the economic popularity models including GDP, inflation and unemployment have now been tentatively determined. However, the influence of economic conditions on party popularity does not occur in a political vacuum. Consequently, this dissertation has so far attempted to control for intrinsic political contextual dynamics, such as cycling and trending, by modelling them endogenously. This chapter further considers some important political contextual variables by exogenously modelling their impact through the B component. Once this has been done, the tentative final party popularity models are re-estimated controlling for these contextual factors and the appropriate tests are applied to the updated state-space models, demonstrating that the statistical issues outlined in Chapter II have.been addressed. At that point, the final form of the party popularity state-space models can be declared. It has been argued that contextual factors can affect the relationship between economic performance and vote decision. 1 3 8 Norpoth suggests the key contextual factors are the clarity of responsibility and the alternatives for dissent.139 Similarly in his comparative study, Anderson found that economic assessments have stronger effects on government popularity when the target of credit and blame is clear and sizeable, and when voters have more viable alternatives for which to vote. 1 4 0 In their comparative study, Powell and Whitten found that economic voting depends on political context 1 3 8 Anderson, Christopher. 2000. Economic Voting and Political Context: A Comparative Perspective. Electoral Studies 19:151-170, Paldam, Martin. 1991. How Robust Is the Vote Function?: A Study of Seventeen Nations over Four Decades. In Economics and Politics: The Calculus of Support, edited by H. Norpoth, M. Lewis-Beck and J.-D. Lafay. Ann Arbor, Michigan: The University Of Michigan Press, Powell, G. Bingham, Jr., and Guy D. Whitten. 1993. A Cross-National Analysis of Economic Voting: Taking Account of the Political Context. American Journal of Political Science 37 (2):391-414. Powell and Whitten, 1993 (see Lewis-Beck, 209). 1 3 9 Norpoth, Helmut. 1996a. The Economy. In Comparing Democracies: Elections and Voting in Global Prospective, edited by L. LeDuc, R. Niemi and P. Norris. Thousand Oaks, California: Sage Publications. 1 4 0 Anderson, Christopher. 2000. Economic Voting and Political Context: A Comparative Perspective. Electoral Studies 19:151-170. 98 variables, such as minority government status and coalition government status.141 These contextual factors alter the logic of economic voting, strengthening or weakening the influence of economic evaluations. Within Canada, many contextual factors such as government size, opposition fragmentation and the.number of viable parties have varied greatly between the three periods examined here. These variables may provide important clues as to why the dynamics of economic popularity vary so much between periods. Consideration is given to this possibility in the concluding chapter - Chapter V. In this chapter, those contextual factors that may impact party popularity and alter the logic of economic popularity within each period are examined. Surveying the relevant literature, it would seem that any model measuring changes in party popularity over time should include variables that can measure and control for the impact of election campaigns and party leadership effects. Other political variables, such as the impact of unique political events, should also be tested. Moreover, the varying impact of economic conditions under minority versus majority governments ought also be considered. These things are done in this chapter. 1 Powell and Whitten, 1993 (see Lewis-Beck, 209. 99 2.0 Controlling for Political Context 2.1 Election Campaigns It has now been repeatedly noted that government popularity deviates from the inter-election cycle during election campaigns.142 Incumbent governments have increasingly experienced downward turns in their popularity during the campaign period. The media are the likely source of this phenomenon. Johnston has argued that trends within party popularity during election campaigns can be tied to media effects.143 John Zaller, with his "receive, accept and sample" theory of mass opinion formation describes a process by which information, such as that provided by the media, can influence public opinion. 1 4 4 Thomas Patterson and John Zaller have both noted an increase over time in the negativity of the media toward politicians during election campaigns.145 Zaller attributes this to "the escalating struggle between politicians and journalists to control the content of political communication."146 If these things are true - events within the media can affect public opinion during election campaigns, and media coverage of politicians has become increasingly negative - we have a potential explanation for the increasingly consistent downward trend in government popularity during elections. The negative impact that elections have on incumbent government popularity is clearly evident in figure 2-2 for the 1993, 1997 and 2000 elections. By examining the 1 4 2 Monroe and Erickson control for months in which elections occur in their longitudinal analysis of party support, for this very reason. Monroe, Kristen, and Lynda Erickson. 1986. The Economy and Political Support: The Canadian Case. The Journal of Politics 48:616-647. 1 4 3 Johnston, Richard. 1992. Letting the People Decide: Dynamics of a Canadian Election. Montreal: McGill-Queen's University Press. 1 4 4 Zaller, John. 2001. The Rule of Product Substitution in Presidential Campaign News. In Election Studies: What's Their Use?, edited by E. Katz and Y. Warshel. Colorado: Westview Press. 1 4 5 Patterson, Thomas. 1993. Out Of Order. New York: Knopf, Zaller, John. 2001. The Rule of Product Substitution in Presidential Campaign News. In Election Studies: What's Their Use?, edited by E. Katz and Y. Warshel. Colorado: Westview Press. 1 4 6 Zaller, John. 2001. The Rule of Product Substitution in Presidential Campaign News. In Election Studies: What's Their Use?, edited by E. Katz and Y. Warshel. Colorado: Westview Press. 100 popularity data for the eight weeks leading into each election since 1965 (figure 4-1), we see the election campaigns produced a drop of approximately 14 percentage points during the 1990s and that such a drop had often occurred previously.1 4 7 In fact, only two of the eleven elections held since 1965 did not witness a downturn in popularity for the incumbent government - Prime Minister Charles Joseph Clark's minority government leading into the 1980 election and Prime Minister Pierre Elliott Trudeau's minority government leading into the 1974 election. The ability of the Liberal government in 1974 to produce favourable media coverage, despite a potentially hostile press, has been attributed to their campaign strategy.148 The net loss of popularity for the incumbent government in the remaining elections ranged from 3 percent during the 1968 and 1988 elections to 20 percent during the 1984 election. Elections are generally negative events for the incumbent government, and have become much more so in recent years. In order to test the theory that elections increasingly produce a negative shock for the incumbent government's popularity, both Liberal election and Progressive Conservative election dummy variables were added to the economic popularity models. The Liberal election variable takes on a value of 1 for any month in which an election is held and the Liberals are the incumbent government. The PC election variable does the same for any month in which an election is held and the PCs are the incumbent government. 1 4 7 Johnston shows this to have also been true during the 1957, 1962 and 1963 elections. Johnston, Richard. Ibid.Capturing Campaigns in National Election Studies. Boulder, Colorado. 1 4 8 Fletcher, Frederick. 1978. The Mass Media in the 1974 Canadian Election. In Canada at the Polls: The General Election of1974, edited by H. Penniman. Washington, D. C : American Enterprise Institute for Public Policy Research. 101 2.2 Leadership Effects Those who have attempted to model the impact of economic conditions on party popularity and electoral outcomes have argued for the inclusion of a variety of variables controlling for potential leadership effects. Nadeau and Blais include a "party leader from Quebec" variable in their model of election outcomes in Canada. 1 4 9 They note that past research has demonstrated a substantial boost in popularity for the Liberal party when the party leader comes from Quebec.1 5 0 Their own analysis demonstrates that the transition from a Quebec Liberal leader and non-Quebec PC leader to a non-Quebec Liberal leader and a Quebec PC leader in 1984 precipitated a substantial loss for the Liberals. 1 5 1 Monroe and Erickson suggest that support will rally around the party in government, regardless of economic conditions, during the six months following a change of Prime Minister or a change in government.152 The six months following a change in government is the equivalent of a honeymoon effect and is already accounted for by the models built in Chapter III - within the cycling component. A change in Prime Minister not related to a change in government is not accounted for and should be examined. In fact, the effect of a change in leadership of any party - governing or not -should be considered. Richard Johnston argues that shifts in popularity due to changes in leadership are less about the actual change and more about the increased media coverage produced by 1 4 9 Nadeau, Richard, and Andre Blais. 1993. Explaining Election Outcomes in Canada: Economy and Politics. Canadian Journal of Political Science 26 (4):775-790. l50Lemieux, Vincent, and Jean Crete. 1981. Quebec. In Canada at the Polls, 1979 and 1980, edited by H. R. Penniman. Washington: American Enterprise Institute. 1 5 1 Nadeau, Richard, and Andre Blais. 1993. Explaining Election Outcomes in Canada: Economy and Politics. Canadian Journal of Political Science 26 (4):775-790. 1 5 2 Monroe, Kristen, and Lynda Erickson. 1986. The Economy and Political Support: The Canadian Case. The Journal of Politics 48:616-647. 102 1 f 1 the leadership convention. He cites evidence from the US which demonstrates that presidential leadership conventions increase the volume of media coverage of a party and makes the tone more positive. 1 5 4 Accordingly, Johnston recommends controlling for the period during which a leadership convention is held, rather than the period following a change in leadership. These three potential leadership effects were tested using various dummy variables. The "Leader from Quebec" variable simply takes on a value of 1 when the leader of the party is from Quebec and 0 otherwise. The "New Leader" variable is coded as 1 for the six months following the selection of a new party leader and 0 otherwise.155 The "Leadership Convention" variable is coded as 1 for the months following the resignation of the party leader and preceding the selection of a new leader at a leadership convention.156 In order to directly compare competing theories of leadership effects, only one leadership variable was included at a time. 2.3 "Rally around the Flag" Events Beyond regularly occurring political events, there have been individual political crises that may have had an impact on the dynamics of party popularity. It is commonly argued that US presidents receive a substantial boost in popularity during times of international crises.1 5 7 This is referred to as a "rally around the flag" effect. British studies argue that a similar effect is responsible for the finding that the Falklands War dwarfed economic 1 5 3 Johnston, Richard. 1999. Business Cycles, Political Cycles and the Popularity of Canadian Governments, 1974-1998. Canadian Journal of Political Science 32 (3):499-520. 1 5 4 Holbrook, Thomas M. 1996. Do campaigns matter?, Contemporary American politics. Thousand Oaks: Sage Publications. 1 5 5 This is the construction used by Monroe and Erickson. Monroe, Kristen, and Lynda Erickson. 1986. The Economy and Political Support: The Canadian Case. The Journal of Politics 48:616-647. 1 5 6 This is the construction used by Johnston. Johnston, Richard. 1999. Business Cycles, Political Cycles and the Popularity of Canadian Governments, 1974-1998. Canadian Journal of Political Science 32 (3):499-520. 1 5 7 Brody, Richard. 1991. Assessing the President: The Media, Elite Opinion, and Public Support. Stanford, California: Stanford University Press. 103 effects for a time. 1 5 8 Within Canada, the FLQ crisis and the initial election of the PQ in Quebec are considered to be two of the most important political crises. 1 5 9 Rather than being international crises, these were domestic and constitutional. It seems reasonable that, just as Americans may rally around their president when the nation is under threat from external forces during times of war, Canadians may rally around their government when their nation is under threat from internal forces during times of constitutional crises. This rally could, in theory, boost government popularity and/or override economic considerations. Brody identifies the US presidential rally phenomenon as a media effect produced by a lack of elite dissent160 - that is, certain international crises change the political incentives of opinion leaders. When information is low and the public mood tends towards patriotism, most of the government opposition elite will choose to remain silent or be vaguely supportive of the President. Media coverage consequently contains an unusual volume of bipartisan support. It is this unusually unbalanced media coverage that produces a public opinion rally. Brody argues that only international crises in the US can produce such a rally. However, since it is crises which cause opposition leaders to lose their usual incentive to criticise the government that produce rallies, it seems plausible that in Canada certain constitutional crises qualify - possibly more than many international crises for Canada. The FLQ crisis is an obvious candidate. In October, James Cross, British Trade Commissioner and Pierre Laporte, Quebec provincial Minister of Labour were kidnapped 1 5 8 For example Norpoth, 1987 (see Lewis-Beck 205). 1 5 9 Monroe, Kristen, and Lynda Erickson. 1986. The Economy and Political Support: The Canadian Case. The Journal of Politics 48:616-647. 1 6 0 Brody, Richard. 1991. Assessing the President: The Media, Elite Opinion, and Public Support. Stanford, California: Stanford University Press. 104 by the Front de Liberation du Quebec (FLQ), a terrorist group that was part of the Quebec sovereignty movement. A list of demands including the release of 23 "political prisoners", the broadcast and the publication of the FLQ manifesto, and an aircraft to take the kidnappers to a safe haven in Cuba or Algeria, in exchange for the safe release of the men was given. In response, Prime Minister Trudeau announced the imposition of the War Measures Act, essentially establishing martial law in Quebec. The police carried out 1,628 raids by October 20. By October 31, the number arrested passed 400. 1 6 1 On October 18, the body of Pierre Laporte was found in the trunk of a car, on the south shore of Montreal. On December 3, James Cross was freed by police action. Following the work of Brody, an examination of the media coverage of Canada's largest constitutional crises during the 1957-2000 period may provide insight into the dynamics behind the rallying of public opinion around the Canadian government during specific events. For now, a (much simpler media count variable is employed. To test and control for the impact of the FLQ crisis on party popularity, a variable was constructed based on the number of pages of the Globe and Mail per month that contained either "Front de Liberation du Quebec" or "FLQ." This variable is designed to measure the extent to which the FLQ crisis was being covered by the media each month and therefore could potentially influence public opinion. Using media counts is similar to how rally around the flag effects are operationalised in some US studies. The count was actually restricted to the 1970-1972 period in which mention of the FLQ would have definitely been about the crisis in Quebec. It is unnecessary to control for the impact of the initial 1 6 1 "Chronology of the October Crisis, 1970, and its Aftermath," Claude Belanger, Department of History, Marianopolis College, <http://www2.marianopolis.edu/quebechistory/chronos/october.htm> 105 election of the PQ in Quebec. This event occurred in 1976, which falls within the 1976-1979 gap between the first and second periods analysed. 106 3.0 Modelling Political Context Clearly, the robustness of the economic popularity models determined in Chapter III must be tested by including variables that control for the potential impact of election campaigns, the FLQ crisis and the various leadership effects. These variables also provide a test of the propositions that electoral campaigns, constitutional crises and party leadership matter to party popularity. Unlike economic effects, political context variables do not enter into the model through the a component. Political events such as election campaigns, the FLQ crisis and changes in leadership are proposed to impact party popularity as immediate shifts in baseline support that last as long as the event. Hence, they enter the model through the B component, which is structured to model these types of effects. For example, political context variables (PCELEC, FLQ, LIBELEC and LEADER) would enter into the B component of the Liberal party popularity equation as follows: Liberal Popularity State-Space Equation LIBVOTE, = at + B, + eye, + v, a< = pa,-i +7 ,1 /5 ,+ y2PC, + ECONEFFECTS + e" B, = BXLIB, + B2PC, + &PCELEC + BALIBELEC + J34 LEADER + J35FLQ + sf eye, = 0 , s,m(XG)LIB, + 0 2 cos(A0)LIB, + 0 3 sm(A0)PC, + 0 4 cos(X6)PCt + s'yc where • ECONEFFECTS are the economic models built in chapter III. • PCELEC and LIBELEC are election dummy variables as described above and L E A D E R is one of the three leadership effect dummy variables, also described above. • FLQ is the FLQ media count variable. 107 The results of estimating these models are presented in tables 4-1 through 4-8. Tables 4-1 through 4-6 consider election and leadership effects only. Tables 4-7 and 4-8 also include the impact of the FLQ crisis in the first period. 108 4.0 Estimation Results Examining tables 4-1 and 4-2, we see that during the 1957-1975 period, the PC and Liberal election variables fail to be significant for either PC or Liberal party popularity. Despite a lack of significance, the effect of elections on incumbent Liberal governments is estimated to be negative, regardless of the leadership variable employed. Similarly, the effect of elections on incumbent PC governments is also estimated to be negative. This is consistent with the notion that elections produce downswings in popularity for the incumbent party. Tables 4-3 and 4-4 demonstrate the impact of election campaigns during the 1979-1993 period. Regardless of the leadership variable included in the model, elections produced downswings in popularity for PC incumbent governments. The magnitude of this effect is approximately 10 percentage points. During these same elections, the Liberal opposition gained over 6 percentage points in popularity. The remaining 3 to 4 points may have gone to the NDP. The Liberal incumbent governments, during this period, experienced an even larger election campaign downturn in popularity -approximately 12 percentage points. The PC opposition during these elections gained over 6 percentage points, just as the Liberal opposition did during PC incumbent elections. In the 1993-2000 period, no election was held with a PC incumbent government. Each election was held with a Liberal incumbent government. Table 4-6 illustrates an election effect of approximately negative 8 percentage points for Liberal governments. This is of a similar magnitude to PC governments in the previous period. It would appear that since 1979, elections do produce a downswing in popularity for the incumbent government. Therefore, this is an important contextual variable to 109 include in any economic popularity model. It should be noted that based on this analysis election effects are clearly larger in the second and third periods than the first but there is no discernible increase from the second to the third period. This suggests that elections have a larger impact now than in the past but that these effects are not necessarily growing, as we might have expected from the ever-increasing media coverage of them. The failure of the analysis to support the theory that elections are producing even larger downswings in the popularity of the incumbent government in the third compared to the second period may be a product of the monthly measures of popularity. Figures A-4a & A-4b (Appendix A) plot PC and Liberal popularity against actual electoral outcomes. It is clear that there is a substantial gap between the measures of Liberal popularity for the months of the 1997 and 2000 elections and their actual share of the vote in the two elections. This is not because public opinion polls have become any less accurate than in the past. The polls conducted in the days leading up to each election very accurately predicted the election outcome. The large gaps are a consequence of the much larger drops in popularity for the Liberal party in these two elections than in any preceding election. The popularity measures are an average measure for the month. The large Liberal popularity drops during the campaign periods of the 1997 and 2000 elections mean this average measure is less accurate than for preceding elections when the popularity shift was much smaller. Popularity, to some extent, smooths out the larger campaign effects. Consequently, the increasing impact of election effects can be missed in an analysis using popularity data. The impact of leadership effects, controlling for election effects, are now examined. For the 1957-1975 period, the PC party never had a leader from Quebec. It did 110 experience a change in leadership. In 1967, the leadership of the PC party was transferred from John Diefenbaker to Robert Stanfield. The circumstances of this particular leadership convention drew a great deal of attention. It followed from open dissension within the party and a call for a leadership review by party president Dalton Camp - at the time, an unprecedented event. Whether measured as the effect of a new leader or as the media coverage of a leadership convention, the change in leadership substantially boosted PC party popularity - 10 to 15 percentage points (table 4-1). The largest leadership effect is that of the leadership convention. This variable should be retained in any future economic popularity model. During the same period, none of the leadership variables produced significant results for Liberal party popularity (table 4-2). During the 1979-1993 period, PC popularity was positively affected by having a new leader. This variable is in effect picking up on the increase in PC popularity produced by the transition from Joe Clark to Brian Mulroney and the transition from Brian Mulroney to K i m Campbell. The statistically significant effect is positive, as predicted and has a magnitude of roughly 4 percentage points. The impact of having a leader from Quebec is not statistically significant. This result fails to confirm the findings of Nadeau and Blais that transitioning to a party leader from Quebec could produce a 5 to 6 percent boost in popularity.1 6 3 Rather, it suggests that the transition to any new leader could temporarily produce this result. For Liberal party popularity during this time, it is the occurrence of a leadership convention that positively impacts popularity (table 4-4) -5.5 percentage points. The two leadership conventions held during this period produced 1 6 2 The relevant leadership transitions for the Liberals was from Louis St-Laurent to Lester Pearson and from Pearson to Pierre Trudeau. 1 6 3 Nadeau, Richard, and Andre Blais. 1993. Explaining Election Outcomes in Canada: Economy and Politics. Canadian Journal of Political Science 26 (4):775-790. I l l first John Turner and second Jean Chretien as leader of the Liberal party. For the second period overall, it would seem that economic popularity models for the PC party should include a control for a new leader and economic popularity models for the Liberal party should include a control for leadership conventions. During the 1993-2000 period, the Liberals never experienced a change in leadership and they consistently had a leader from within Quebec - leadership was a constant. Therefore, there are no leadership effects to examine. For the PC party though, there was a change in leadership from Jean Charest to Joe Clark (a Quebec to non-Quebec leader). However, the impact of this change in leadership is not evident in the results presented in table 4-5. Overall, there are election effects but no leadership effects to control for in the third period. This is not to say that Chretien's leadership of the Liberals was inconsequential. It may have been a very important factor in the difference between second and third period Liberal popularity dynamics. However, in the present setup, period effects are endogenously controlled through separate estimations. Only political contextual variables that change within a given period can be exogenously modelled. It has been determined that election and leadership variables should be included in the 1957-1975 PC and Liberal economic popularity models. The additional inclusion of the FLQ variable in these models suggests that Canadians did rally around the Liberal government during the crisis (tables 4-7 and 4-8). Liberal party popularity increased approximately 0.2 percentage points per page of the Globe & Mail that the FLQ story appeared on, while PC party popularity dropped a little more than 0.1 percentage points per page. At the peak of the crisis in October 1970, with 146 pages of the newspaper 112 mentioning the FLQ crisis in a single month, Liberal popularity was boosted 24.8 percentage points. This is consistent with the spike in Liberal popularity during October 1970 that is evident in the plot of party popularity (figure 2-1). Clearly, the impact of the FLQ crisis should be controlled for in economic popularity models for this period. The state-space model used in this dissertation makes the structural assumption that economic conditions shift party popularity away from some underlying baseline support. At the same time, this baseline support is somewhat of a moving target. That is, it trends downwards or upwards over the time in which a party is in government or opposition and it cycles between elections. With the inclusion of exogenous political contextual variables, these popularity components must be re-estimated. Therefore, it is necessary now to re-examine the structure of the cycling and trending components of the popularity models, controlling for these political contextual variables. Leaving cycling aside for the moment, the estimated state-space models are structured to allow two types of trending - linear long-term and equilibrating short-term. The popularity of PC governments underwent downward, long-term trending in both the first and second periods. This reflects the declining popularity of both the Diefenbaker and Mulroney governments over their terms in office (-0.43 percentage points per month for Diefenbaker and -0.19 percentage points per month for Mulroney), as they struggled to hold together Quebec/Western Canada coalitions. However, both first and second period PC governments experienced some initial benefit from being in office. Short-term trending produced an overall gain of 12 percentage points for Diefenbaker and 19 percentage points for Mulroney (table 4-9 - Box-Jenkins and table 4-11). 1 6 4 1 6 4 In order to calculate the effects of short-term trending, it is necessary to use the estimated values from the tables including economic variables (tables 4-9 through 4-13). The inclusion of these variables can alter 113 Although Liberal governments in the first period experienced no long-term trending, they did experience an initial gain of 11 percentage points (table 4-10 - refined Box-Jenkins). Liberal governments during the second period experienced positive long-term trending (0.22 percentage points per month) and negative short-term trending. The short-term trending produced an overall loss of 18 percentage points for each Liberal government (table 4-12 - refined Base model). Clearly, the short-term loss dominated any long-term gains. This reflects the fact that each time Trudeau gained control of Parliament, he tended to take on controversial issues losing support for the Liberal party. Neither short-term nor long-term trending is apparent for the Liberal government or PC opposition in the third period. The starting point for underlying baseline support (the constant within the baseline component) is largest for the Liberal governments in the third period (55.5 percent), followed by PC governments in the first period (49.5 percent) and Liberal governments in the second period (46.9 percent). These reflect the initial popularity of the Chretien, Diefenbaker and Trudeau governments respectively. The lowest starting point for underlying baseline support is not surprisingly for the PC opposition in the third period (around 8 percent). Looking now at cycling, the amplitudes are clearly greatest for both parties within the second period. However, the amplitude of the cycle for the Liberal party in opposition during the second period is comparable to that of both parties in the first and third periods. Cycling amplitudes are somewhat larger for each party in the third compared to the first period. The largest cycle amplitude is found for the Liberal government in the somewhat the magnitude of estimated short-term trending, as economic conditions may account for part of the trending. 114 second period. It is 9.6 percent, meaning that from peak to trough, the inter-election cycle shifts baseline popularity almost 20 percent. The smallest cycle amplitudes are for the Liberal party (within government or opposition) and the PC party (in opposition) within the first period - between 1 and 2 percent. As explained in Chapter II,.the wavelength of the cycle is set so that one complete cycle occurs per inter-election period. However, the elections do not necessarily fall on a peak or trough in the cycle. This is left to be estimated. As it turns out, elections generally fall on the upswing of the cycle some months before the cycle's peak. The timing of elections for Liberal and PC governments within the first and second periods are between 10 and 20 percent away from the cycle's maximum (preceding it). In the third period, elections occur when the cycle is approximately 30 percent away from reaching its maximum. For all periods the cycling component are statistically significant. The consistency with which an inter-election cycle can be found is compelling. As the concluding chapter shall demonstrate, there is some circumstantial evidence that the cycle is in part related to economic conditions. With various relevant political contextual events controlled, the role of economic conditions in party popularity can be re-evaluated. For PC party popularity during the 1957-1975 period this is done both for the Base and Box-Jenkins models. Neither had found any significant role for economic conditions in PC party popularity during the this period. Controlling for election, FLQ and leadership effects does not change this finding (table 4-9). Interestingly having now controlled for these contextual variables, the distribution of the estimated AR(1) term within the Box-Jenkins model is unimodal -previously it was bimodal. The AR(1) term within the Base model is also unimodal, as 115 before (figure 4-2). The Q-tests for the residuals from both models suggests that each is a white noise process and that the estimated a components are stationary. Controlling for election and FLQ effects also does little to change the estimated impact of economic conditions on Liberal party popularity during the 1957-1975 period. Unemployment continues to have a positive impact on the popularity of the Liberal party in opposition (table 4-10). The magnitude of this effect within the Box-Jenkins model (0.026) does not change significantly. The Box-Jenkins model (with election and FLQ controls) was rerun including only the impact of unemployment on the Liberals in opposition.165 The estimated parameter decreases significantly to 0.0036. However, the AR(1) term goes from being negative to positive (and large - approximately 0.85). So in fact, the estimated impact of unemployment is actually larger in the refined Box-Jenkins model - this is discussed further below in the discussion of cumulative economic effects. As with the PC Box-Jenkins model, controlling for economic and F L Q effects has resulted in a unimodal distribution for the AR(1) term (figure 4-3). Again, the Q-test suggests the estimated residuals from each model form a white noise process and that the potential for spurious correlations is.minimal. Including election and leadership effects in the combined model from Chapter III does little to change the significance of the effect of economic variables on PC government popularity in the second period or the estimated magnitude of these effects (table 4-11). GDP, inflation and their intercept (all lagged 13 months) continue to have statistically significant effects, with GDP being the largest. GDP and its intercept with 1 6 5 The Box-Jenkins model was also rerun including unemployment only but within both the opposition and government parts of the model. There was no significant difference from the model including unemployment only on the opposition side. 116 inflation (only lagged one-month) also continued to have statistically significant effects. However, the significant negative impact of inflation on PC opposition popularity becomes insignificant and the magnitude of the estimated term becomes half as large. This leaves no remaining economic effect for PC party popularity while in opposition during the second period. The combined model was rerun without economic variables on the opposition side of the model. The estimated parameters for economic effects on the government side remain virtually identical - as one might expect. However, Q-tests for the residuals from each model produce significantly different results. In the first model (with economic variables on the opposition side), we can reject the null-hypothesis that the residuals are white noise. In the second model, we cannot. If one examines the autocorrelation functions for the residuals from each model, there is no evidence of any autocorrelation, trending, cycling, etc. (figure 4-4) - it is not clear why the Q-test suggests the residuals from the first model are not white noise. Either way, since residuals distributed as white noise are desirable, the second model is preferable. As in the previous models, the AR(1) has a unimodal distribution, with a positive median in either model (approximately 0.73). Without controlling for election and leadership effects, the second period Liberal popularity Base'model suggested a significant role for GDP and its interaction with inflation (lagged one month) for the party in opposition. The Base model including such controls suggests the same thing (table 4-12). The magnitudes of these effects are only reduced slightly. With the AR(1) term remaining fairly constant, the magnitude of the estimated parameters change from 0.0017 to 0.0015 for the interaction term and from -0.0089 to -0.0074 for the GDP term. 117 The Base model including election and leadership controls was rerun excluding economic variables on the government side of the model -1 refer to this as the refined Base model. The consequence for economic effects is insubstantial. However, the Q-test applied to the residuals from each model again produce different results. For the refined Base model, the null hypothesis that the residuals are white noise can not be rejected. For the Base model, the null hypotheses can be rejected at the 95 percent confidence level but not the 90 percent confidence model. Similar to the second period PC model, neither of the autocorrelation functions for the residuals shows much evidence of anything but a white noise process (figure 4-5). Once again, it is unclear what produces the discrepancy but since the refined Base model produces clearly uncorrected errors, this is the model that is retained. For this model and the Base model, the AR(1) terms remain positive and unimodal (approximately 0.82). No model up to this point has found any role for economic conditions in PC party popularity during the third period. The inclusion of political context variables does not change this finding (not shown). As for Liberal governments between 1993 and 2000, the inclusion of political context variables has minimal consequences for the Liberal economic popularity model. Without the election control variable only inflation (lagged five months) was found to have a statistically significant impact. With the election control, both inflation and its intercept with GDP (lagged five months) are statistically significant, with virtually unchanged coefficients (around -0.03 for inflation and 0.007-0.008 for its intercept). The AR(1) term is relatively small and negative (-0.035). Before controlling for election effects it was also small but positive. 118 With the exception of eliminating economic effects for PC party popularity in opposition during the second period, controlling for exogenous political events does little to our conclusions regarding the impact of economic conditions on party popularity. The next section discusses what those conclusions are exactly. 119 5.0 Discussion: Final Form of the State-Space Economic Popularity Models Having now put the economic popularity models into political context, it is necessary to cross-correlate the residuals from the estimated a component from each model with the prewhitened economic variables to identify any lagged relationships which may have been missed. The estimated cross-correlation functions are presented in tables A-5 & A-6 (Appendix A). The cross-correlation functions suggest the following potential relationships: • between first period Liberal opposition popularity and inflation, lagged three months; • between first period Liberal government popularity and inflation, lagged two months; • between first period Liberal government popularity and unemployment, lagged in six months; • between first period PC government popularity and GDP, lagged zero months; • between second period Liberal government popularity and the GDP/Inflation intercept, lagged three months; and • between second period PC government popularity and the GDP/inflation intercept, lagged four months. None of these variables produced statistically significant effects once included in the appropriate models. 1 6 6 Therefore, there is no reason to believe that any lagged economic variables have been inappropriately excluded. It is possible now to declare the final form of the economic popularity state-space models incorporating the impact of changes in GDP, unemployment rates and inflation. These are as follows: 1 6 6 When either GDP, inflation or their intercept were included in a model for a particular lag, all three were included. 120 State-Space Economic Popularity Models"" 1957-1975 PC Party Popularity PCVOTE, =a,+B,+ eye, + v, a, = 0.6676a,_, - 0.0228L75, + 0.040IPC, + e" B, = 0.3756Z/5, + 0.4954PC, -0.0008LIBELEC -0.0213PCELEC-0.0010FLQ+ 0A5LEADERCONV + eye, = 0.0005 sm(A0) LIB, -0.006 cos(W) LIB, + 0.0245sin(A6>)PC, - 0.004lcos(A6»)PC, +s?c 1957-1975 Liberal Party Popularity LIBVOTE = a + B + eye + v t t i s t t a = 0.8529a , + 0.0M92LIB - 0.01186PC + 0.003634* OPPUNEMP c + s" B = 0.3346L/5 + 0.3291PC - 0.0266LIBELEC - 0.0M3PCELEC + 0.00UFLQ + / eye eye = 0.0\\1 sm(X6)LIB( + 0.0088cos(-W)LIBi - 0 . 0 0 0 6 s i n ( ^ ) P C ; - 0.0011 cos(A.9)PCi + e{ 1979-1993 PC Party Popularity PCVOTE] - a, + B, +cyc, +v, a =0.73ke,_, + 0.0523L/5, + 0.0397PC, -0.001322*GOV,INFt_xGDP,_, -0.00265XX)V,INF,_X + 0.006635*GOV,GDP,_x -0.002377*GOV^GDP^INF,^ -0.004604*GOV,_nINF,_n + 0.01 \23'GOV,_uGDP_X3 +s" B, =0.23641/5, +0.3209PC, -0.\034PCELEC+0.066524LIBELEC+0.03933NEWLEADER+£? eye, =-0.0096hin(Aff)LIB, -0.0643cos(Xff)LIBl + 0.053llsm(X0)PCt + 0.02977cos(/t<9)PC, + £c,yc (cont.) * indicates statistical significance for economic variables only. 121 1979-1993 Liberal Party Popularity LIBVOTE, =at + B, + cyc, +vt a, = 0.8268a,_, -0.030901/5, -0.0005PC, + 0.0014*OPP,GDP,_xINF,_x -0.0010OPPtINF,_X -0.0069*OPP,GDP,_, + s" B, = 00.469 ILIB, + 0A327PC, + 0.06369PCELEC - 0A243LIBELEC + 0.05454LEADERCONV + eye, = 0.03067sm(A6)LIB, + 0.08591 cos(A0)LIB, -O.O1794sin(A0).PC, -0.01629 cos(/K9)PC, +s 1993-2000 PC Party Popularity PCVOTE, =a,+B, + cyc, + v, a, = 0.8504a,_, - 0.009885Z/5, + e ° B, -0.31681/5, +sf eye, = -0.0094 sm(X9)LIB, -0.01747 cos(A0)LIB, +e?c 1993-2000 Liberal Party Popularity LIB VOTE, = a, + B, + eye, + v, a =-0.0351 l a j +0.04631Z/5, + 0.007703* GOV,_5GDP,_5INF,_5 - 0.0327* GOV,_5INF,_5 -0.00S6GOV,_5GDP,_5 + s" B, =0.5551Z/5, -0.0S32LIBELEC + S? eye, = -0.0075 sin(W)LIB, + 0.0208 cos(X0)LIB, +s'yc where • L E A D E R C O N V is the leadership convention dummy variable; and • N E W L E A D E R is the dummy variable for the six months after a leadership change. 122 Having declared the final form of these models, it is time to discuss the magnitude of economic effects on the dynamics of party popularity throughout the entire 1957-2000 time-span. In order to get a picture of the magnitude of the impact of economic conditions on party popularity, the month-to-month immediate contribution of economic conditions is plotted in figures 4-11 through 4-13. This is the shift away from baseline support produced by changing economic conditions each month. It is important to note that the plotted values do not represent the total contribution of economic conditions in a given month, because the estimated models include first order autocorrelation terms. The residual impact of past economic effects will also contribute to popularity levels and therefore as described in Chapter III, the over time impact of changes in economic conditions must also be considered. The nature and magnitude of the over time contribution of a given month's economic conditions depends upon the degree and type of memory. In a model with a positive autoregressive term, considering only the immediate impact of economic conditions tends to underestimate their overall effect. A positive parameter suggests that i f popularity is moved above the party's baseline in a given month, it likely will continue to be above the baseline in the following month. Figure 4-9a plots the decaying impact of the economic conditions from a given month if the popularity function has an autoregressive coefficient such as that for the Liberals in the second period (0.83). This assumes that the economic conditions produced a one percentage point shift in the first month and that there are no further contributions to popularity from economic conditions following that month. As this figure demonstrates, the impact of a given month's economic conditions decays exponentially over time when memory is positive. In this sense, popularity corrects itself after receiving a shock by returning to its baseline. Figure 4-10a plots the accumulating contribution of economic conditions if they are held 123 at some constant level. This assumes that the constant level produced a 1 percentage point shift in popularity in the first month and that economic conditions remain at that level following. This figure demonstrates that i f economic conditions are held at a constant level, their contribution to popularity quickly reaches some equilibrium. This equilibrium depends upon the degree of memory and the level of economic conditions. In this case, i f the economic conditions that produced an initial 1 percent shift in Liberal popularity were held constant, eventually about 5.6 percent of each month's party popularity would be the consequence of these constant economic conditions. This 5.6 percent contribution (as an increase or decrease) would remain until economic conditions changed. The dynamics of a series with a negative autocorrelation parameter, such as that for the Liberals in the third period, are somewhat different. The negative parameter suggests that if the party's popularity is moved above its baseline one month, it is likely to be under the baseline the next. Not only does popularity fluctuate around a baseline (as in all stationary models), it also tends to overcorrect for deviations from the baseline. Consequently, considering only the immediate monthly impact of economic conditions in models with negative autocorrelation terms tends to exaggerate economic effects. Figure 4-9b contains a plot of the decaying impact of a 1 percent shift, given the AR(1) parameter estimated for the third period, Liberal party popularity model (-0.035). The impact decays away almost immediately, returning to baseline levels. In fact, it initially overcorrects and produces a small negative contribution. If the magnitude of the AR(1) term had been greater, this overcorrection would have been larger and the shock would have oscillated around the baseline before quickly settling down. Figure 4-10b plots the accumulating contribution of economic conditions if they are held at some constant level which produced an initial 1 percent shift. It 124 demonstrates that these constant economic conditions would eventually contribute only 0.97 of a percentage point of popularity each month. The dynamics of shifts in party popularity produced by economic conditions are different for the Liberals in the third period than those for the Liberal and PC parties in the second period. In both cases, a shock from a single month will decay and popularity will return to its baseline. However the decay in the second period is slow and steady, unlike that in the third period. It is slow because the memory is greater. It is steady because it does not contain a negative autocorrelation term. Therefore, it does not overcorrect itself by oscillating back and forth around the baseline. In both the second and third periods, economic conditions held at a constant level will eventually reach some equilibrium level of contribution. In the case of a positive A R term, that equilibrium will be greater than the initial shock. In the case of a negative A R term, the equilibrium will be less than the initial shock. Keeping in mind the dynamics produced by the autocorrelation components of each model, it is time to now turn to the impact of economic conditions. In the first period, only Liberal opposition popularity was significantly affected by economic conditions. The Liberal opposition benefited from relatively high unemployment rates (six months prior). With rates hitting around 7 percent, the one-month impact of unemployment is estimated to be around 2.5 percent. It was during 1958, 1960 and 1961 that unemployment hovered around the high of 7 percent and consequently it is from mid 1958 to mid 1959 and mid 1960 to mid 1962 that Liberal opposition popularity received this boost. The above does not mean that only 2.5 percentage points of Liberal popularity were attributable to unemployment in a given month. There were also lingering contributions from the impact of unemployment in previous months, as the final Liberal party popularity model 125 estimated for the first period has a positive AR(1) term. If unemployment remained at 7 percent throughout the entire period, 17 percent of popularity in a given month would be attributable to unemployment - this is quite a substantial contribution. More realistically, an average unemployment rate (5.5 percent for the period) would have a cumulative contribution of approximately 13.5 percentage points of Liberal opposition party popularity. Unemployment was at its lowest in 1957 and therefore during the first half of 1958, the Liberal opposition only received a one-month boost of 1.7 percentage points. While the Liberal party in opposition benefited from unemployment rates, the PC government was not penalised. This can be explained by and is consistent with the finding of Erickson, which suggests that NDP popularity was hurt by unemployment during this time. 1 6 8 When unemployment increased, popularity shifted from the NDP opposition to the Liberal opposition. No party has been affected by levels of unemployment since 1963. During the second period, GDP significantly (statistically speaking) benefited the PC party while in government when accompanied by low inflation (1 and 13 months prior). In November of 1984, the PC opposition gained a modest 0.12 percentage points because economic growth was good a month earlier (5 percent) but inflation was not particularly low (4 percent). In August of 1990, the PC government suffered a 4 percentage point drop in popularity due to high inflation levels of 5.4 percent (13 months prior) and 4.1 percent (one month prior) and a poor growth in GDP of only 1.3 percent (13 months prior) and -1.2 percent (one month prior). In July of 1993, the effect of GDP and inflation again marginally benefited the PC government (0.5 of a percentage point) as both reached levels of about 1 percent (1 and 13 months prior). With average inflation levels being relatively high and GDP being only moderate, the PC government 1 6 8 Erickson, Lynda. 1988. CCF-NDP Popularity and the Economy. Canadian Journal of Political Science 21 (1):99-116. 126 between 1984 and 1993 was primarily penalised for economic conditions. Average inflation was 4.0 percent and average economic growth was 1.1 percent. At these levels, PC popularity would take a hit of 4.5 percentage points in a given month or a cumulative hit of 16.8 percentage points. Liberal governments experienced no clear economic effects in the second period but the popularity of the Liberal party in opposition was affected by GDP and inflation. Moderate levels of GDP growth hurt the Liberal opposition and high inflation made matters worse for the Liberals, not better. The Liberals in opposition experienced their worst drop due to economic conditions in February of 1991 - (-1.9 percentage points). This is a relatively small drop. The Liberals should have benefited from poor economic growth (-4.6 percent) at this time but were negatively affected by high inflation (6.9 percent) in the previous month. Economic conditions helped the Liberal opposition most in September 1992. The Liberals experienced a one-month gain of 0.7 percentage points due to poor economic growth (-1.5 percent) and low inflation (1.1 percent). Overall, the impact of economic conditions on the popularity of the Liberals in opposition was small. Again, average inflation was 4.0 percent and average economic growth was 1.1 percent during this period. These economic conditions in any given month would produce a shift of approximately -0.55 percentage points in Liberal opposition popularity. This one month shift would translate into a cumulative impact of -3.2 percentage points. Examining the second period as a whole, the Liberal governments preceding 1984 escaped being punished for high inflation levels. Even during the negative growth of 1982, the Liberals escaped unharmed by economic conditions. After the 1984 election, Liberal opposition popularity was hurt marginally during good economic growth and helped marginally during poor economic growth (particularly in 1991). Although lower than before 1984, high inflation kept the 127 PC government from really capitalising on good economic growth. At the same time, the Liberal opposition did not benefit from high inflation. This raises the question as to who may have benefited. Erickson's work suggests the NDP benefited from high inflation during the 1980s.169 Her work only extends until 1984 but if this trend continued into the second half of the 1980s, this would explain where the popularity lost due to inflation by the PC government went. During the 1993-2000 period, inflation did not help Liberal governments any more than it did PC governments in the second period. For the most part, Liberal governments were penalised for inflation (five months prior). This effect was moderated by positive economic growth (also five months prior). The direction of the combined impact of these variables is consistent with expectations. In November of 1995 when growth in GDP five months earlier was around 1 percent and inflation was 2.6 percent, the Liberals were punished by approximately 7.3 percentage points. Over a year earlier, Liberals were only punished by 2 percentage points, with growth in GDP five months previous at approximately 2 percent and inflation lower than 0.2 percent. These effects are largely driven by changes in inflation. This is consistent with the estimated magnitudes of the coefficients for the Liberal popularity model (table 4-13). As inflation increases, Liberal government popularity decreases. This is as expected. What is a little surprising is that 0.2 percent inflation still wasn't low enough for the Canadian electorate - at least not with only 2 percent growth in GDP. It is surprising enough to suggest that the estimated impact of GDP and inflation is incorrect for low levels of inflation. This may have occurred because the impact of inflation may not be monotonic (at low levels inflation may have a much smaller effect i f any) and yet the model estimates it as such. In terms of the overall magnitude of economic effects during this period, average growth in GDP was 2.7 percent and average inflation was 1.6 percent. In a given month, these economic conditions would produced a 4.2 128 percent decline in popularity. The cumulative impact of these average GDP and inflation levels in any given month is minus 4.1 percentage points (remember, the estimated AR(1) term was small and the negative). 129 6.0 Majority versus Minority Governments Before discussing the impact of economic conditions on government support more generally, it is important to note that the final form of the economic popularity models were developed by treating the effects of economic conditions in majority and minority governments as equivalent. Minority governments have been formed in six elections since 1954 - these were in 1957, 1962, 1963, 1965, 1972, and 1979. We would expect economic voting to be suppressed by some degree during these minority governments because, as Anderson has suggested, economic assessments have stronger effects on government popularity when the target of credit and blame is clear and sizeable. In a minority situation, the government can feasibly argue when things go awry that they do not have complete control over their own policies. Alternatively when things go well, the opposition can (and likely will) argue that they are partly responsible for the outcome. Consequently, the target of credit and blame is obscured. The type of minority governments that have formed in Canada - those that must seek support from opposition parties to pass any legislation - are the truest form according to Powell and Whitten and should strongly disperse responsibility for economic outcomes. Therefore, the assumption that economic effects in majority and minority governments differ must be tested. The next section does just this. The 1957, 1962, 1963, 1965 and 1972 minority governments all took place within the first period. The first two were PC minority governments and the last three were Liberal. The first period Box-Jenkins models were rerun for the Liberal and PC parties, testing for economic effects only during majority governments. If economic effects are stronger during majority governments, then one would expect to see stronger relationships between economic variables and party popularity within these models than those estimating effects for the entire period. 1 7 0 Powell, G. Bingham, Jr., and Guy D. Whitten. 1993. A Cross-National Analysis of Economic Voting: Taking Account of the Political Context. American Journal of Political Science 37 (2):391-414. 130 For the PC party, the Box-Jenkins models failed to find any economic effects in the first period. This turns out to be true even when the estimation is limited to only majority 171 governments. For the Liberals, it is not clear whether economic effects are larger for majority governments alone. Looking first at the refined Box-Jenkins model (the Box-Jenkins model including only economic effects on the opposition side), the fact that the effects of unemployment become statistically insignificant when the estimation is limited to times of majority governments suggests that economic effects are not stronger during majority governments (table 4-14). With the full Box-Jenkins model, unemployment remains significant but it is necessary to take into account the drastically different AR(1) terms. It is negative for the full period and positive (and insignificant) when economic effects are limited to times of majority governments. Considering the cumulative over-time effects, a 1 percent change in unemployment is estimated to produce a 1.6 percentage point shift for the Liberal opposition during all PC governments and a 2.0 percentage point shift for the Liberal opposition during majority PC governments alone. This suggests that economic effects may be a little larger during majority governments. It also needs to be noted that when economic effects are limited to times of majority ' governments, the interaction term for GDP and inflation on the opposition side of the full Liberal Box-Jenkins model becomes statistically significant. Although not significant, the estimated parameters for inflation and GDP (lagged six months) are both negative and of greater magnitude than in the Box-Jenkins model estimated assuming economic effects for the entire period. These particular economic effects may be stronger during majority governments. 1 7 1 The estimation limiting effects to majority governments was also run for the Base model. Again, no economic effects were detected. 131 For the 1979-1993 period, there was only one minority government - that of Joe Clark's PC government between 1979 and 1980. For the PC party, the refined combined model (the combined model including only economic effects on the government side) was rerun assuming economic effects only for the post 1980 period. The consequence is that the impact of economic variables lagged 13 months on PC government popularity becomes statistically insignificant. However, those variables, lagged only one month, remain statistically significant and their 172 magnitude increases substantially (table 4-15). A sense of the magnitude of economic effects on the PC party during majority governments relative to the whole second period can be obtained by looking at the distribution of the estimated economic effects in each case. For the whole second period, the mean economic effect on PC governments (produced by GDP and inflation) is -2.3 percentage points (SD: 1.2). During majority governments only, the mean economic effect is -1.0 percentage points (SD: 2.1). 1 7 3 This suggests that economic effects are more variable but not larger when restricted to times of majority governments. For the Liberals in opposition during the second period, re-estimating the Base model assuming economic effects only during majority governments results in all economic effects becoming statistically insignificant.. This is also true for the refined Base model (table 4-16). Overall, the evidence that economic effects are stronger for majority governments than otherwise is contradictory. It is clearly not the case in the second period. As for the first period, the evidence is somewhat more unclear. Some economic variables, such as inflation and GDP may be more relevant (although difficult to explain) for the Liberal opposition during majority PC governments. There certainly does not appear to be any evidence to suggest that the 1 7 2 The Combined model was also run producing similar results 1 7 3 The mean economic effects were calculated using the same average economic conditions (those for the full period). 132 economic popularity models are improved by restricting them to times of majority governments. Therefore, it is appropriate to interpret the impact of economic conditions on party popularity using the final models developed in the last section. 133 7.0 Conclusions It is now possible to make some generalisations about the impact of economic conditions on government support. Figure 4-14 is a plot of the cumulative contribution of economic conditions on government party popularity from 1984 to 2000. This allows us to get a sense of the magnitude of the impact of economic conditions across this time-span, accounting for the effects of memory within popularity. This plot does not contain the first period because no economic effects on the popularity of the party in government were found before the 1979 Clark government. It does not include the effect of economic conditions on the 1979-1980 Clark government because, given the brevity of this government, it is not clear i f the estimated economic effects can be attributed to it as clearly as they can be to the 1984-1993 Mulroney governments. Looking at the plot, it becomes clear that economic conditions had a greater cumulative impact on PC governments in the second period than on Liberal governments in the third. While the initial impact is of a similar magnitude to that of economic conditions in the third period, the shock to government popularity remains longer in the second. This produces a greater cumulative effect. Economic effects in the third period dissipate very quickly (exhibited by the low AR(1) value) suggesting economic memory is shorter than in the second period. Overall, the party in government is primarily punished for economic conditions rather than rewarded and the cumulative contribution of economic conditions to party popularity can be quite large. The greatest cumulative loss experienced by the PC government due to economic conditions was in August of 1990 - 14.6 percentage points. The greatest cumulative loss experienced by the Liberal government in the third period occurred in December of 1995 and was only 7.1 percentage points. Because these are cumulative effects, they do not suggest PC 134 government popularity dropped by almost 15 percent or that Liberal government popularity dropped by over 7 percent in one month. Rather in these extreme months, PC government popularity was almost 15 percent lower and Liberal government popularity was over 7 percent lower than they would have been i f economic conditions had no impact on the electorate's opinion of the governing party. During the second period, the popularity of the PC government was more sensitive than the popularity of the Liberal party in opposition. This is consistent with the idea that it is the government that is being held accountable for the state of the economy. It also highlights the importance and justifies the approach of considering the popularity of each party separately. The impact of the economy on the party of the official opposition is certainly not a mirror image of that for the governing party. Considering all three periods, PC government popularity is more sensitive than Liberal government popularity to inflation. Conversely, the Liberal party has been more sensitive to levels of unemployment then the PC party - although, only in the first period and only when in opposition. Overall, GDP tends to benefit parties in government and hurt parties in opposition. Neither the Liberals nor the Conservatives - in opposition or government - have benefited from high levels of inflation. Based on my findings, I agree with Johnston that economic growth (moderate as it was) was a positive force on the popularity of the Mulroney/Campbell government between 1984 and 1993. Unlike Johnston, I argue that Liberal governments in the late seventies/early eighties avoided being punished for poor economic conditions during this period. 1 7 4 On this, I am in agreement with the findings - or the lack of the findings - from the rest of the Canadian 1 7 4 The differentia] effects of economic conditions on the Liberals and Conservatives during this period is a distinction that Johnston's method cannot make. 135 literature - in particular, those of Clarke and Zuk. 1 7 5 Furthermore, I do not agree with Johnston's somewhat perverse finding that governments benefited from inflation between 1974 and 1993. 1 7 6 Rather, I find that the PC government between 1984 and 1993 was penalised for inflation rates, just as it was rewarded for economic growth. I also find that both economic growth and inflation continued to have an impact on the Chretien government's popularity, although in a somewhat more muted manner. Moreover, I argue that the electoral cycle so evident in the second period also operates in the first and third periods but with smaller amplitudes. In their voting study, Archer and Johnson found that the PC party benefited from high unemployment in the 1984 election. This is consistent with the important effects that Nadeau and Blais attribute to unemployment during elections between 1953 and 1988. It is difficult to compare election voting studies to a popularity study but these results are inconsistent with my assertion of no unemployment effects after 1963. That I do find economic effects prior to 1963 is also somewhat in disagreement with the findings of Monroe and Erikson which suggest that national economic conditions had no impact on support for either of the major political parties 177 between 1954 and 1979. However, the general conclusion that economic effects were weak prior to 1979 is substantiated. Chapter III built economic popularity models which rigorously minimised the potential of spurious correlation produced by the nonstationarity inherent in almost any party popularity time-series. This chapter goes further and explicitly models the impact of some of the political contextual events which may in part contribute to this nonstationarity. In doing so, it has been 1 7 5 Clarke, Harold D., and Gary Zuk. 1987. The Politics of Party Popularity: Canada 1974-1979. Comparative Politics:299-3\5. 1 7 6 Johnston, Richard. 1999. Business Cycles, Political Cycles and the Popularity of Canadian Governments, 1974-1998. Canadian Journal of Political Science 32 (3):499-520. 1 7 7 Monroe, Kristen, and Lynda Erickson. 1986. The Economy and Political Support: The Canadian Case. The Journal of Politics 48:616-647. 136 demonstrated that the estimated economic effects are not only robust to the inclusion of political context variables, they are sometimes even strengthened by it. Only the finding that inflation affects PC opposition popularity within the second period is determined to be spurious. Based on the findings of this chapter, there is little question that the economy matters to party popularity. Even more importantly, it is clear that many governments have been held accountable for economic conditions. In particular, PC governments from 1979 to 1980 and 1984 to 1993 and Liberal governments from 1993 to 2000 have been held accountable for economic growth and/or inflation levels. It is also evident that the impact of economic conditions on the popularity of the party in opposition is not a mirror reflection of their impact on the party in government and that economic effects for Liberal governments differ from economic effects for PC governments. What still remains to be answered is why some governments seem to have escaped judgment for growth in GDP, inflation levels and unemployment levels - specifically, Diefenbaker's PC government and Pearson and Trudeau's Liberal governments. During Diefenbaker's government, the Liberal party in opposition benefited from high unemployment rates but the PC government was not punished. During Pearson and Trudeau's governments, neither the popularity of the Liberals in government or the PCs in opposition was affected by economic conditions. It would seem suggestive that all of Pearson's and one of Trudeau's governments were minority governments. Maybe economic conditions just do not play as clear a role in popularity when the government is a minority. However, the comparative analysis of economic effects on majority governments versus all governments in the previous section finds at best contradictory evidence to this effect. The issue of why some governments are not held accountable for economic conditions is considered further in the next and concluding chapter of 137 this dissertation - as is the question of why the dynamics of economic popularity vary so much between periods. Before examining these questions further, it should be noted that this is by no means the end of the economic popularity story.. Changes in GDP, inflation and unemployment are certainly the most commonly analysed economic conditions within the context of party popularity but there are many other potential candidates. Further research should consider other economic performance variables. In his model of U K government popularity, Hibbs suggests ) 78 using the US dollar-British pound exchange rate as a measure of economic performance. It certainly seems reasonable to include the Canadian-US dollar exchange rate in the Canadian case. On a similar note, Nathaniel Beck emphasises the importance of putting economic performance into an international context.179 G. Bingham Powell Jr and Guy Whitten find it important to consider economic performance in the US relative to the performance of other industrialised democracies.180 In the case of Canada, it could be argued that the performance of the economy relative to that of the US matters more than absolute performance. Other types of fiscal variables could also be considered, such as government spending and government deficit levels, as suggested by Lowry et al in their analysis of US state spending and voting. 1 8 1 Powell and Whitten also suggest that it may be necessary to take account of the performance of the economy within sub-regions. These all represent directions in which 1 7 8 Hibbs, Douglas, Jr. On the Demand for Economic Outcomes: Macro Economic Performance and Mass Political Support in the United States, Great Britain, and Germany. 1 7 9 Beck, Nathaniel. 1991. The Economy and Presidential Approval: An Information Theoretic Perspective. In Economics and Politics: The Calculus of Support, edited by H. Norpoth, M. Lewis-Beck and J.-D. Lafay. Ann Arbor, Michigan: The University Of Michigan Press. 1 8 0 Powell, G. Bingham, Jr., and Guy D. Whitten. 1993. A Cross-National Analysis of Economic Voting: Taking Account of the Political Context. American Journal of Political Science 37 (2):391-414. 1 8 1 Lowry, Robert C , James E. Alt, and Karen E. Ferree. 1998. Fiscal Policy Outcomes and Electoral Accountability in American States. The American Political Science Review 92 (4):759-774. 1 8 2 Powell, G. Bingham, Jr., and Guy D. Whitten. 1993. A Cross-National Analysis of Economic Voting: Taking Account of the Political Context. American Journal of Political Science 37 (2):391-414. 138 further research into economic popularity can proceed. In doing so, the analytical framework established in this dissertation will be invaluable. 139 Chapter V U N D E R S T A N D I N G P O L I T I C A L C O N T E X T A N D T H E P O T E N T I A L R O L E O F T H E M E D I A 1.0 Political Context and the Media During the analysis performed in this dissertation, a disparate collection of questions have been raised. While these questions are widely dispersed, most pertain to the forces which drive the contextual conditions which in turn account for: 1) the varying dynamics of party popularity; and 2) the varying connection between the economy and party popularity over time. The importance of controlling for political contextual factors when estimating the effect of economic conditions on party popularity has been stressed throughout the dissertation. The analytical framework designed in Chapters II and III does just this. It has also been noted that these contextual factors are not only statistical challenges but are of substantive interest, being of political origin. In order to not just control but also better understand these contextual factors, some of the forces such as elections and party leadership which drive the political context are explicitly modelled in Chapter IV. Building on this work, this chapter considers other forces that may drive contextual conditions, and thereby may be responsible for the changing dynamics of party popularity and economic popularity between 1957 and 2000. Directions for further research on these forces are suggested. Addressing these questions regarding political contextual factors raises another question: to the extent that changes in political context do account for variations in party popularity dynamics and the connection between the economy and party popularity, what is the mechanism by which this occurs? The role of the media in shaping the political context is suggested as one possible answer. Moreover, the media as part of the mechanism by which economic conditions are translated into party popularity itself forms part of the changing context in which economic popularity operates. And so, this chapter ends with a discussion of how to proceed with an examination of the role of the media as the mechanism by which economic conditions are 141 communicated to voters and translated into party popularity, and how the media shapes the political context in which this occurs. 142 2.0 Political Context 2.1 Inter-Election Cycling First, the forces driving the dynamics of inter-election cycling are considered. Cycles, as contextual components, are one of the most statistically important for which to control. They are also one of the most interesting. They are persistent and can have substantial effects. Yet, there is little known about them. There is a great deal of work that needs to be done to understand what drives this particular dynamic of party popularity. In addition to representing part of the political context in which economic popularity operates, there is the potential that the cycling is itself at least in part driven by economic conditions. In order to obtain a preliminary sense of the extent to which this may be true, it is necessary to determine i f economic conditions themselves contain a cyclicity which corresponds with the timing of elections (that is, an economic inter-election cycle) and the relationship between this cycling and popularity cycling. In order to test for this potential economic cycling, the following model was estimated for each economic variable: ECONVAR = 6 + a,+ 0 , sin(W) + 0 2 cos(^6>) + e, a, = pECONVAR^ + e? where • t = l,...,T at monthly increments; . e?~(0,crlr); *,~(o,<) and COV(e?,e,) = 0 • p,,B,®x,and 0 2 are parameters to be estimated; and • A is the frequency (1/wavelength) of the inter-election cycle and is defined by the length of the inter-election period, varying from one election to the next. 143 The first line of the model is the structural equation and it contains both cycling terms and a constant. As with the government/party popularity equations, the 0, and 0 2 parameters can be used to estimate the amplitude of the cycle. The second line models the disturbance (or error process). It is modelled as an ARMA(1,0) process. For each economic variable, this equation was estimated separately for each of the three periods. When this was done, a statistically significant cycle became evident for all economic variables in all periods. The magnitude of these cycles, however, were relatively small. For unemployment the magnitudes are 0.05, 0.08 and 0.32 percent for the first, second and third periods respectively. These values represent levels of unemployment expressed as percentages. They suggest that the unemployment cycles within the first, second and third periods are responsible for a movement of 0.1, 0.16 and 0.62 of a percentage point of unemployment from cycle trough to cycle peak. For inflation the amplitude magnitudes are 0.10, 0.32 and 0.16 percent. And for GDP, 0.10, 0.31 and 0.19 percent. It is interesting to note that for GDP and inflation the amplitude of the cycle in the first period relative to the second and relative to the third corresponds with the relative amplitudes found within the government and party popularity cycles. The amplitudes are largest for the second period, smaller for the third period and even smaller for the first. Figures 5-1, 5-2 and 5-3 plot the cycling components for each economic variable relative to the estimated cycling component for government popularity. In each case, the amplitude of the cycle is standardised so that it equals 1 for both economic and popularity cycles within the second period. This allows for a direct comparison of relative amplitude magnitudes. The plots also indicate the points at which elections were held. The observable fact that each economic variable contains a cycle which corresponds with the inter-election cycle found within the government and party popularity data suggests that popularity cycling and economic forces 144 are related. The fact that the relative amplitudes between periods is so similar for GDP, inflation and popularity is further evidence and it may be this relationship that explains the varying amplitudes of the cycles across periods. As a preliminary analysis, it is useful to compare the phases of the economic and popularity cycles. The easiest way to do this is to just compare the distance between the relative positions of the economic and popularity cycles peaks. Because each inter-election cycle is of a different length, the distance between cycle peaks is expressed as a percentage of an inter-election cycle. For example, i f the distance between peaks is 10 percent of an inter-election cycle than this would represent 4.8 months i f the inter-election cycle was four years long or alternatively six months if the inter-election cycle was five years long. Within the first period, GDP peaks about 4 percent of the inter-electibn period after popularity peaks and inflation peaks about 6 percent of the inter-election period afterwards. In the second period, GDP peaks 17 percent before government popularity and inflation 18 percent afterwards. In the third period, GDP peaks 30 percent before and inflation 41 percent afterwards. It is clear that rises in inflation follow rises in GDP and declines in inflation followed declines in GDP. Economically, this is reassuring. The fact that in the second and third periods GDP precedes the popularity peak and inflation follows suggests that upswings in popularity may be driven by economic growth and subsequent downturns are the product of declining GDP and/or increased inflation produced by the earlier economic growth. In the first period economic growth lags popularity increases but only marginally. In fact, they are almost coincident, consistent with the notion that GDP drives popularity. Of concern though is that in the third period GDP leads popularity by such a large margin (30 percent of the cycle). With third period inter-election lengths of 41 to 44 months, this 145 suggests that the GDP cycle takes approximately a year to be reflected in popularity. Intuitively, this seems a little long. Turning now to unemployment, in the first period its cycle is at its nadir 14 percent of the inter-election cycle before government popularity reaches its peak. In the second period it reaches its nadir 5 percent of the cycle before the popularity peak. In the third period, the unemployment nadir is 22 percent before the popularity peak. This suggests that upswings in the government popularity cycle correspond with downturns in unemployment. Again in the third period, the lag between the lowest point in unemployment and the highest point in popularity is longer than one might intuitively expect. This all could be interpreted to mean that at least part of the popularity cycle is driven by economic conditions. This would be an economic effect beyond that which has so far been modelled. Whether or not economic forces are in fact driving popularity cycling, there does appear to be a logically consistent relationship between economic cycles and popularity cycles. In the traditional literature on political-economic cycling, it is not only argued that economic conditions can drive government popularity but also that cycling within economic conditions are the product of politically driven government decisions. Empirical evidence of this relationship in the US context is most famously presented by Tufte in his publication "Political Control of the Economy." 1 8 3 Soon after Tufte's work, Bruno Frey and Friedrich Schneider developed their reaction 184 function which describes how government uses policy instruments to influence the economy. This function is based on the theory that when government popularity is below some threshold required to win an election and when an election is imminent, the government will use these 1 8 3 Tufte, Edward R. 1978. Political Control of the Economy. New Jersey: Princeton University Press. 1 8 4 Frey, Bruno, and Friedrich Schneider. 1978a. An Empirical Study of Political-Economic Interaction in the United States. The Review of Economics and Statistics 60 (2): 174-183. 146 policy instruments to improve the economy and subsequently improve its popularity. When an election is not near and the level of government popularity is adequate, the function states that parties will follow their ideological predispositions in shaping the economy. Frey and Schneider I O C originally developed the reaction function for the US and it has been replicated for a number of countries showing strong electoral cycle effects on government economic policy. This research has come under heavy scrutiny and rejected by many on both statistical and theoretical grounds. Statistical advances since have allowed analysts to improve upon the original reaction function and the contemporary consensus is that even if the original research overstated the impact of the electoral cycle, governments do manipulate economic conditions for electoral purposes. The relationship between the electoral cycle and government economic policy has been demonstrated in the Canadian context at the federal level by Francois Petry and Howard Harmatz and at the provincial by Blais and Nadeau. 1 8 7 These analyses use economic policy (inputs) rather than economic conditions (outputs) because until now no connection between popularity cycling and economic outputs has been demonstrated in Canada. To get a sense of the extent to which the cyclicity demonstrated by this dissertation within GDP, unemployment and inflation may be a product of government policy based on electoral opportunism, it is useful to compare the phase of the cycles to the timing of elections. Again, the easiest way to do this is to just compare the distance between the relative positions of the economic cycle peaks and the occurrence of elections. Also as before, the distances are expressed as percentages of an inter-election cycle. 1 8 6 Frey, Bruno, and Friedrich Schneider. 1978b. A Politico-Economic Model of the United Kingdom. Economic Journal 88:243-254, Frey, Bruno, and Friedrich Schneider. 1979. An Econometric Model with an Endogenous Government Sector. Public Choice 34:25-45. 1 8 7 Blais, Andre, and Richard Nadeau. 1992. The Electoral Budget Cycle. Public Choice 74:389-403, Perry, Francois, and Howard Harmatz. 1995. Politico-Economic Interactions in Canada: An Empirical Assessment. Public Finance Quarterly 23 (3):305-335. 147 For GDP, its cycle peaks 2 percent of the inter-election cycle before an election within the first period, 13 percent of the inter-election cycle before an election within the second period, and 2 percent of the inter-election cycle before an election within the third period. The fact that GDP peaks just before or almost just before an election is strongly suggestive. The potentially detrimental consequence of peaks in economic growth (a peak in inflation) occurs well after elections have been won or lost within the second and third periods - 24 and 25 percent of the inter-election cycle afterwards respectively. Within the first period, the peak in inflation occurs within the election month, potentially offsetting any electoral gains that may have been produced by a cyclical peak in economic growth. As for unemployment, the evidence is more varied. In the first period the unemployment cycle is at its nadir 6 percent of the cycle after an election. For the second period, the unemployment cycle nadir occurs during the election month and in the third period it occurs 43 percent of the electoral cycle before the election. This suggests governments may receive electoral gains from the unemployment cycle during the first and second periods. During the third period, quite the opposite is suggested. Unemployment is close to its peak around the time of elections. This all provides circumstantial evidence for the theory that governments manipulate the economy for electoral purposes, inducing a cyclicity in economic conditions. However, the causal direction is not necessarily as clear as the reaction function would suggest. In addition to manipulating the economy to correspond with the timing of elections, where election timing is not fixed, governments may time elections to take optimal advantage of economic conditions. This last phenomenon is referred to as "surfing." Surfing would produce the same 148 correspondence between economic conditions and election timing as would manipulation.185 There is however an important difference in terms of the cause of this relationship. In the case of economic manipulation, economic cycling and election timing correspond because governments manipulate the economy to peak just before an election. In the case of surfing, economic cycling and election timing correspond because governments choose to call an election when the economy is peaking. There are also differences in the economic consequences of manipulation and surfing. In the first case, the economy is manipulated for political purposes. This may be detrimental to the economy. In the case of surfing, politics passively responds to the economy and there is no intervention. From an economic point of view, this may be preferential. Based on the preliminary analysis of economic and popularity cycling presented here, it is not possible to determine whether Canadian governments manipulate or surf. It is likely that they do a bit of both. It has been noted that elections are almost inevitably called as government popularity is on the popularity cycle's upswing and has surpassed the government's average since the last election and this dissertation has demonstrated that economic conditions can affect government popularity. Based on these observations, one possible scenario is as follows. Once the honeymoon component of the cycle is long gone (the potential origin of the honeymoon component is discussed in section 3.0 on the role of the media) and government popularity is plumbing the depths of the downswing, the government may engage in policies that increase its popularity - e.g., manipulating policies that increase economic growth. This produces a popularity upswing, which the government closely follows until it has surpassed some threshold that the government's political strategists feel is sufficiently high for an election to be called -i.e., surfing. 1 8 8 A useful analysis of manipulation and surfing is provided by Mark Kayser. Kayser, Mark. 2005. Who Surfs, Who Manipulates? The Determinants of Opportunistic Election Timing and Electorally Motivated Economic Intervention. American Political Science Review 99 (1): 17-27. 149 This is just one possible scenario that could explain the cycling in popularity and its relationship to economic cycling. Clearly, there remains a great deal of work to be done to understand what drives this dynamic component of popularity and thanks to the contribution this dissertation has made to the modelling of party popularity dynamics, this analysis can be more effectively pursued. 2.2 Distinct Period Dynamics Now a number of potential factors that might account for the distinct popularity dynamics found in each period are considered. The fact that party popularity dynamics and the link between economic conditions and party popularity are distinctly different between the three periods identified in Chapter II, has been firmly established. What has been left to explain is why these periods exist - why are there such variations? It has already been argued that political contextual factors can affect the relationship between economic performance and party popularity by filtering voters' efforts to assign responsibility.189 These contextual factors alter the logic of economic popularity, strengthening or weakening the influence of economic evaluations. Chapter IV controlled and tested for the impact of events that drive political contextual factors which occur within each period. What remains is an examination of those factors which vary across periods. In addition to their potential impact on economic popularity, these same factors may provide us with more general explanations for the varying dynamics of party popularity between the three periods. The clearest period difference is between the second and third and so much of this preliminary analysis focuses on this. 1 8 9 Anderson, Christopher. 2000. Economic Voting and Political Context: A Comparative Perspective. Electoral Studies 19:151-170, Paldam, Martin. 1991. How Robust Is the Vote Function?: A Study of Seventeen Nations over Four Decades. In Economics and Politics: The Calculus of Support, edited by H. Norpoth, M. Lewis-Beck and J.-D. Lafay. Ann Arbor, Michigan: The University Of Michigan Press, Powell, G. Bingham, Jr., and Guy D. Whitten. 1993. A Cross-National Analysis of Economic Voting: Taking Account of the Political Context. American Journal of Political Science 37 (2):391-414. 150 Some potentially important contextual factors identified in the literature as importance to the assignment of responsibility, such as the nature of the committee system and the strength of the bicameral opposition, are not pertinent to the Canadian case.1 9 0 Canada's Upper House has been consistently weak during the time period under study and the committee system has been consistently dominated by the government party. Furthermore, no formal coalition governments have existed at the federal level in Canada during the 1957-2000 time-span. However, the size of government (the target of credit and blame) and the number of viable alternatives to which voters may throw their support have varied greatly. Looking first at government size: measured as the proportion of parliamentary seats controlled by the government party, government size has varied from 78.5 percent of the seats during Diefenbaker's 1958 government to 41.3 percent during Trudeau's 1972 minority government (figure 5-4). Chretien's 1997 government has been the smallest majority government within this time period with only 51.5 percent of the parliamentary seats. Larger governments have a greater capacity to affect the economy. Accordingly, voters may attribute greater responsibility to them. Based on this proposition, we might expect economic evaluations to have had less of an impact on government popularity between 1997 and 2000 than during other majority governments. It certainly is true that the magnitude of economic effects in this third period for Liberal governments were found to be smaller than those for PC governments during the second period - although as was discussed,, this is largely due to the smaller cumulative effects produced by a shorter memory in third period public opinion. 1 9 0 Anderson, Christopher. 2000. Economic Voting and Political Context: A Comparative Perspective. Electoral Studies 19:151-170. 151 In addition to the potential impact of these factors on economic popularity, they may have an independent effect on government popularity regardless of economic evaluations.191 It is argued that bigger governments experience greater losses than smaller governments and regardless of size, the governing party experiences greater losses when citizens have fewer but larger alternatives. Considering these effects, the Diefenbaker government was clearly the largest and exhibited the greatest long-term decline in popularity. The second-biggest government was Mulroney's 1984-1988 government. It also experienced a strong decline in popularity (although it generally recovered by 1988). This is consistent with the proposition that bigger governments experience larger losses. Minority governments - of which there have been many - are a special case of government size. In addition to the already explored possibility that minority governments obscure economic responsibility, the prevalence of minority governments may also affect the general popularity of the government. For example, in 1974, the incumbent Liberals benefited from the electorate's predisposition toward majority governments.192 A number of voters which may have preferred a Conservative government are argued to have supported the Liberals because a Conservative win was seen as unlikely and a majority Liberal government was seen as preferable to a minority Liberal government propped up by the NDP. This is a variation of the type of strategic voting discussed below. The issue of viable alternatives may provide us with an explanation of another difference between periods. A clear difference between the second and third periods is the much higher sustained equilibrium level for the government's popularity since 1993. The level of popularity 1 U 1 U . 192Fletcher, Frederick. 1978. The Mass Media in the 1974 Canadian Election. In Canada at the Polls: The General Election of 1974, edited by H. Penniman. Washington, D. C : American Enterprise Institute for Public Policy Research. 152 that the Liberal government maintained between 1993 and 2000 (an equilibrium of about 55 percent) is largely a product of its sweep of Ontario, bolstered by an ability to maintain a level of popularity in Quebec, despite its losses to the Bloc Quebecois. This is somewhat analogous to the level of popularity that both Mulroney and Diefenbaker's governments attained when they made their breakthroughs into Quebec. However, the popularity of both Conservative governments declined to more traditional levels fairly rapidly, while that of Jean Chretien's Liberal government did not. Was Liberal popularity maintained at such levels for almost a decade because of a lack of viable alternatives? There are two immediately obvious factors that may affect the number of viable options available to voters unsatisfied with the government. These are the regionalisation and fractionalisation of the opposition. If the opposition is highly fractionalised, none of the alternatives to the government will be in a position to replace the government. None of the opposition parties will have enough support to be the governing party. As the opposition becomes more regionalised, again none of the alternatives will be in a position to replace the government. None of the opposition parties will have enough nationwide support to become the governing party. Examining figure 5-5, it is evident that the fractionalisation of the opposition since 1993 has been unusually high but only from 1997 onward has it exceeded 1957 levels. As for regionalisation of the opposition, since 1993 it has been substantially higher than any time prior. The Liberals in the third period compared to other governments in other periods faced a much more regional opposition (and in particular a fractionalised regional opposition). This degree of opposition fragmentation and regionalization is a consequence of the rise of the Reform and Bloc parties. With the opposition in such a state, voters may conclude that there is no viable national 153 alternative to the Liberals, thus maintaining an artificially high level of support for the government. This is consistent with the theory that the governing party experiences greater losses when citizens have fewer but larger alternatives. The third period is one of multiple small regional alternatives. This is the consequence of a change in the party system and is very indicative of the difference between the second and third periods. It is, however, only an indicator. Further research on the impact of the fractionalisation and regionalisation of the opposition on government popularity is much needed. With the dynamics of party popularity now properly modeled by the methods developed in this dissertation, this important issue can be addressed fully. Strongly related to the issue of viable alternatives is the issue of strategic voting. In a ' first past the post' electoral system like Canada's, some voters may vote for a party other than the one they really prefer because they do not want to ' waste ' their vote on a party that has little chance of winning. 1 9 3 When a voter's first preference is unlikely to win and the voter has a preference between the top two candidates, it is strategically logical for the voter to vote for his/her preference of the two front-runners, rather than his/her overall favourite.194 Blais et al demonstrate that in the 2000 election, "3 percent of voters outside Quebec cast a strategic vote for a party that was not the one they most preferred. Typically, these strategic voters were people who preferred the Conservatives or the NDP, but decided to vote for the Liberals or the Alliance." 1 9 5 Like fractionalisation and regionalisation, strategic voting may also have contributed to the difference between the second and third periods. 1 9 3 Blais, Andre, Elizabeth Gidengil, Richard Nadeau, and Neil Nevitte. 2002. Anatomy of a Liberal Victory: Making Sense of the Vote in the 2000 Canadian Election. Peterborough: Broadview Press. 1 9 4 Blais, Andre, Richard Nadeau, Elizabeth Gidengil, and Neil Nevitte. 2001. Measuring Strategic Voting in Multiparty Plurality Elections. Electoral Studies 20:343-352. 1 9 5 Blais, Andre, Elizabeth Gidengil, Richard Nadeau, and Neil Nevitte. 2002. Anatomy of a Liberal Victory: Making Sense of the Vote in the 2000 Canadian Election. Peterborough: Broadview Press. 154 In a parliamentary system, such as that in Canada, strategic considerations may occur both at the constituency and national levels - that is, in addition to considering the likely outcome of the local constituency race a voter may also take into account the parties that have the greatest chance of forming government nationally. Johnston et al suggest such considerations may have affected voter choice in the 1988 federal election.1 9 6 In order to consider the extent to which the strategic considerations of those that voted for the government but may have preferred voting for a different party inflated Liberal government popularity in the third period, compared to governments in other periods, we begin by considering the impact of national level considerations on the decisions of those that voted for the government. This is done by examining their potential alternative vote preferences and the consequences of such preferences on government formation. We initially perform this exploration for the 1962, 1980 and 2000 elections. First, we must speculate whom these voters would vote for if they chose to no longer support the government - that is, i f the government was no longer their first choice and the voter chose to defect from the government. For the 2000 and 1980 Liberal governments, we have the National Election Studies to assist us. In 2000, 27 percent of Liberal voters indicated that their second choice was PC, while 22 percent indicated NDP. Only 17.5 percent suggested the Alliance was their second choice (Canadian Election Study, 2000 - see Appendix B). In 1980, 51 percent of Liberal voters indicated that they had also considered voting PC. About 33 percent indicated they considered voting NDP and only 8.2 percent indicated Social Credit (Canadian National Election Panel, 1980 - see Appendix B). It is reasonable to assume that a Liberal voter in either the 1980 or 2000 election choosing to defect from the government would have voted for 1 9 6 Johnston, Richard. 1992. Letting the People Decide: Dynamics of a Canadian Election. Montreal: McGill-Queen's University Press. 155 their second choice. Along similar lines, we can use the 1965 National Election Study to estimate whom 1962 supporters of the Conservative government would have voted for i f they chose to defect. Of those who indicated they voted Conservative in the 1965 election, 42 percent also indicated that their second choice of federal party was the Liberals. Twenty-two and a half percent indicated their second choice was the NDP and almost 14 percent indicated their second choice was Social Credit (Canadian National Election Study, 1965 - see Appendix B). Figure 5-6 provides plots of the outcome of various levels of defection from the government to the government supporters' various alternatives in the 1962, 1980 and 2000 elections. What is being plotted is the number of seats each party would win given a particular percentage defection of government voters from the government to a specific alternative party. (Defection is assumed to be uniform across all constituencies.) We see that defection from the PC government to the Social Credit Party in 1962 and defection from the Liberal government to the NDP in 1980 and 2000 produces similar results. In each case, a minimum to moderate level of defection from the government not only lowers the number of seats won by the government but also increases the number of seats won by a third party - the Liberals in 1962, the PC in 1980 and the Alliance in 2000. Moreover, as defection levels rise, it is these third parties that will eventually form government - not the voter's alternative. This holds true until defection reaches a threshold, at which point the alternative choice may form government. In most cases this is a very high threshold and remains unrealistic. So, the consequence of a government supporter defecting to Social Credit in 1962 or NDP in 1980 or 2000 (unless defection rates are very high) is to assist a party which is neither the government nor the voter's alternative choice. This acts to make defection to Social Credit in 1962 or NDP in 1980 or 2000 an unviable alternative. 156 Now we turn to the Liberals and Conservatives as alternatives. We see that in 1962, i f a PC government voter chose to defect to the Liberals, the defection would assist the Liberals - the intention of the defection - and only the Liberals. In 1980, i f a Liberal voter chose to defect to the Conservatives, the defection would assist the Conservatives - again the intention. So in 1962, the 42 percent of government supporters that would have voted Liberal i f they chose to defect had a viable alternative. In 1980, the 51 percent that would have voted PC i f they chose to defect also had a viable alternative. However, in the 2000 election, the Conservatives no longer offered a viable alternative to Liberal government voters. As with defection to the NDP, defection to the Conservatives would have primarily benefited the Alliance. They were not a viable alternative. Only the less than 18 percent of Liberal government voters which indicated their second choice was Alliance had a viable alternative. This would seem to suggest that national level strategic considerations may have artificially inflated Liberal government popularity. Neither the 27 percent of the Liberal government voters which might have preferred voting PC or the 22 percent which might have preferred voting NDP had a viable alternative. Unfortunately, this argument does not hold for the 1993 and 1997 elections. It turns out that in 1993 (figure 5-3) defecting to the PC's, the Alliance, or the NDP were all strategically logical options. The Alliance strength was still tenuous enough and that of the NDP and PC parties was strong enough that defecting from a Liberal vote to any one of these three would have benefited the intended party. In 1997 (not shown) these dynamics remained the same. National level strategic considerations as we have structured them here cannot explain the high levels of Liberal support in the first part of the 1993-2000 period. Before we abandon national level strategic considerations, we should consider re-evaluating our view of national level strategic voting. So far, we have only considered the impact 157 of government voters defecting to one party at a time. However, by the time 10 (for example) percent of Liberal government voters have defected to another party (e.g. the PC party) many other government voters will have also defected to the other parties. To test the impact of government defection, we might consider the potential impact on electoral outcomes of a defection from the government party to all other parties. In doing so, we can either assume uniform defection to each party or more realistically, defection in proportion to the distribution of the second choices of government voters. (In either case, defection is again assumed to be uniform across constituencies.) In examining the cases of 1980 and 1993, the distinction between these two assumptions produces little difference in the outcome so the simpler case of uniform defection is presented (figure 5-7). What is being plotted in this figure is the number of seats each party would win given a particular percentage defection of government voters from the government to each of the alternative parties equally. It is clear that in 1980 a defection from the winning Liberal party to each of the other parties would have quickly led to a PC victory. Those who might consider defecting to either the NDP or Social Credit had reason to think twice. However, the 51 percent that might have considered defecting to the PC's had no such considerations. There was a viable alternative for many government voters. In 1993, we see that defection from the winning Liberal party would have eventually lead to a Reform victory. Therefore, the 37.8 percent that might consider defecting to the PC's, the 23.2 percent that might have considered defecting to the NDP, and the 3.5 percent that might have considered defecting to the Bloc had reason to remain with the Liberal party for strategic reasons. However, given that 25 percent of Liberal voters would have to defect before such a Reform victory would occur, such strategic considerations were limited (although, the fear of a Liberal minority with a Reform opposition would increase strategic 158 considerations). This modifies our previous finding that the 1993 election would have been potentially winnable by any party (except the Bloc) i f the Liberals were less popular. First, the Liberals would have to have been much less popular and second, it is the Reform party that would likely have emerged the winner. Therefore, most Liberal government voters may not have felt they had an available defection option. A similar scenario existed for the 1997 election and therefore, Liberal support may have been inflated in the first part of the third period, as it was in the last by national level strategic considerations. Of course, these conclusions are based only on the potential strategic voting produced by national level considerations. Constituency level dynamics may also lead to strategic voting. This is beyond the scope of this preliminary analysis but is an important direction for future research into the changing dynamics of party popularity in Canada. Theories of strategic voting all assume that government voters have some sense of how the parties stand in the polls and that they use this information. Blais et al show that the more votes a party had obtained in a constituency in 1997, the better its chances of winning in the 2000 election were perceived to be by voters in that constituency. This would suggest that voters' expectations are rational and based in part on electoral results.197 If this is true, voters must be receiving information regarding electoral and polling results from somewhere. The media is certainly a likely candidate. The potential for the media to influence public opinion is discussed further in the next section. Beyond the size of government and the number of viable alternatives to which voters may throw their support, there are a number of other potentially important political factors that should be considered in any further research. Powell and Whitten have suggested that leadership is one 1 9 7 Blais, Andre, Elizabeth Gidengil, Richard Nadeau, and Neil Nevitte. 2002. Anatomy of a Liberal Victory: Making Sense of the.Vote in the 2000 Canadian Election. Peterborough: Broadview Press. 159 such factor. Extremely popular and extremely unpopular leaders can have important effects on party popularity.1 9 9 There are a number of examples of this at work in Canada. Clarke et al. find that perceptions of Joe Clark may have prevented the Conservatives from forming a majority in 1979 and conversely, the popularity of Trudeau, compared to Clark, may have saved the Liberals from a bigger defeat.200 Data on prime ministerial popularity is not as abundant as that on their government. However, to the extent that it is available, the effect of leadership on government popularity should be tested, particularly during periods of great prime ministerial popularity and unpopularity. Media coverage (again, discussed in the next section) of particularly popular or unpopular leaders should also be explored. In addition to directly increasing or decreasing party popularity, particularly popular or unpopular leaders may weaken economic effects. Extreme sentiment regarding the leader may override economic considerations. The importance of leadership may explain the finding that economics did not play as big a role in the 1979 election as it did in the previous 1974 election despite higher inflation rates. Economic considerations may also be overridden in response to the occurrence of specific and notable political events. Hetherington shows that during the 1984 and 1988 United States Presidential elections, the electorate was focused on foreign and domestic 201 • policy issues and despite the efforts of the media the economy was not a central issue. It is this potential distraction from economic issues that is the basis for accounting for significant political events - such as the FLQ crisis - within the state-space models developed in this dissertation. 1 9 8 Powell, G. Bingham, Jr., and Guy D. Whitten. 1993. A Cross-National Analysis of Economic Voting: Taking Account of the Political Context. American Journal of Political Science 37 (2):391 -414. 1 9 9 For example, Margaret Thatcher's popularity between 1979 and 1983 had a significant impact on the popularity of the Conservative government in Britain. Clarke, Harold D., Marianne C. Stewart, and Gary Zuk. 1986. Politics, Economics and Party Popularity in Britain, 1979-83. Electoral Studies 5 (2): 123-141. 2 0 0 Clarke, Harold D., Jane Jenson, Lawrence LeDuc, and Jon Pammett. 1982. Voting Behaviour and the Outcome of the 1979 Federal Election: The Impact of Leaders and Issues. Canadian Journal of Political Science 15 (3):518-552. 2 0 1 Hetherington, Marc. 1996. The Media's Role in Forming Voters' National Economic Evaluations in 1992. American Journal of Political Science 40 (2):372-395. 160 Other sources of distraction ought to be considered. Relevant sources will be those that are heavily covered by the media. Lewis-Beck and Stegmaier suggest other contextual factors which require further investigation when considering the dynamics of economic popularity. These include the impact 202 of income inequality and insecurity in society, and the effects of increasing globalisation. Greater levels of income equality and insecurity may result in greater sensitivities to economic conditions. Meanwhile, Lewis-Beck and Heinz Eulau argue that globalisation (the increase in a • 203 country's dependence on the international economy) will reduce economic voting. Measuring globalisation as the sum of imports and exports as a proportion of GDP, Canada's economic global dependence has varied substantially enough that it should be examined as a contextual factor.204 Incumbency is a further contextual factor that may affect economic voting. Nadeau and Lewis-Beck show that in US presidential elections economic voting is mostly retrospective when the incumbent candidate is an elected president. When the incumbent candidate is not an elected president, economic voting is mostly prospective.205 In the Canadian context, a similar effect would suggest that when the incumbent party is led by a Prime Minister that led the party through an election, economic voting is mostly retrospective and when the incumbent party is led by a Prime Minister that has come to power since the last election (e.g., Turner in 1984 and 2 0 2 Lewis-Beck, Michael S., and Mary Stegmaier. 2000. Economic Determinants of Electoral Outcomes. Annual Review of Political Science 3:183-219. 2 0 3 Lewis-Beck, Michael, and Heinz Eulau. 1985. Economic Conditions and Electoral Behaviour in Transnational Perspective. In Economic Conditions and Electoral Outcomes: The United States and Western Europe, edited by M. Lewis-Beck and H. Eulau. New York: Agathon Press, Inc. 2 0 4 Pickup, Mark. 2003. Globalisation, Politics, and Provincial Government Spending in Canada. Paper read at Canadian Political Science Association Meeting, at Halifax. 2 0 5 Nadeau, Richard, and Michael Lewis-Beck. 2001. National Economic Voting in US Presidential Elections. The Journal of Politics 63 (1): 159-181. James Campbell also provides some evidence of this effect. Campbell, James. 2001. An Evaluation of the Trial-Heat and Economic Forecast of the Presidential Vote in the 2000 Election. American Politics Research 29 (3):289-296. 161 Campbell in 1993), economic voting is mostly prospective. These Canadian examples are similar to Nadeau and Lewis-Beck's examples of Presidents Ford and Johnson who found themselves overseeing economic programs which they had inherited from their processors.206 In terms of the economic popularity of the government party, this hypothesis would suggest that the Turner and Campbell governments were less held accountable for the past performance of the economy and more likely judged on how the economy was expected to perform under them in the future. Again, this is an important direction for future research. As with other potential directions for further research on the political context in which economic popularity operates, the role of the media in shaping incumbency effects needs to be considered. It is to the needed analysis of the media as the mechanism by which economic conditions are translated into party popularity and by which the political context in which economic popularity operates is shaped that the next section addresses. 2 0 6 Nadeau, Richard, and Michael Lewis-Beck. 2001. National Economic Voting in US Presidential Elections. The Journal of Politics 63 (1): 159-181. 162 3.0 The Role of the Media 3.1 Perception vs. Reality Due to the inconsistent results produced by previous analyses of the impact of objective economic conditions on government support, the focus of much of the recent work is on the relationship between perceptions of the economy and government support.207 Some suggest that perceptions of the economy matter more than objective reality. "The electoral battle over the economy has become one of credit-taking and blame-denial. What counts is how voters perceive the real economy and whose partisan claims about the economy they believe or are persuaded to 90S believe as a result of campaigns and media coverage." Hence, some studies suggest the use of measures of consumer attitudes and perceptions in place of objective economic indicators.209 In the US case, Nadeau and Lewis-Beck have created an aggregate national business index using the question "Would you say that at the present time business conditions are better or worse than a year ago?" This is an aggregate measure of the electorate's retrospective economic evaluation.210 They have also created an economic future index from the question "Now turning to business conditions as a whole - do you think that during the next 12 months we'll have good times financially, or bad times financially?" They show that these measures are more effective than objective, macroeconomic ones in predicting 212 presidential vote. In Canada, the Canadian Conference Board has measured consumer attitudes since 1960. A quarterly index of consumer attitudes is available since 1962. The difficulty with using such 2 0 7 Lewis-Beck, Michael S., and Mary Stegmaier. 2000. Economic Determinants of Electoral Outcomes. Annual Review of Political Science 3:183-219. 2 0 8 Norpoth, Helmut. 1996a. The Economy. In Comparing Democracies: Elections and Voting in Global Prospective, edited by L. LeDuc, R. Niemi and P. Norris. Thousand Oaks, California: Sage Publications. 2 0 9 Nadeau, Richard, and Michael Lewis-Beck. 2001. National Economic Voting in US Presidential Elections. The Journal of Politics 63(1):159-181. 2 1 0 Ibid. 2 1 1 Ibid. 2 1 2 Ibid. 163 perceptions of the economy is that they are to some degree endogenous to evaluations of the government. When a government is popular for any number of reasons," the electorate may be more inclined to positively evaluate the performance of the economy. Moreover, Norpoth demonstrates that the relationship between economic perceptions and presidential popularity may at times be spurious. Events which may buoy presidential popularity may also do the same for economic expectations. Also of importance is that examining the association between economic perceptions and government popularity does little to increase understanding of how objective economic conditions are translated into popularity and hence, the mechanism of electoral accountability. David Sanders copes with this problem by using a two-stage analysis.2 1 4 He first examines the relationship between objective economic conditions and individual perceptions of the economy. The second stage examines the relationship between those perceptions and support for the government. This certainly allows for a proper focus on the issues that matter but does not 215 solve the endogeneity or spurious correlation problems. A great number of analysts, including Sanders, suggest that the most important source of economic perceptions is the media. The media also potentially have a strong influence on the electorate's view of political parties in general. If voters do play rational gods at the polls, they are not omniscient gods. Voters require information i f they are going to make rational assessments of the performance of the economy and the government. This information is available in spades from the news media. 2 1 6 However, there is no guarantee that news media 2 1 3 Norpoth, Helmut. 1996b. Presidents and the Prospective Voter. Ibid. 58 (3):776-792. 2 1 4 Sanders, David, and Neil Gavin. 2004. Television News, Economic Perceptions and Political Preferences in Britain, 1997-2001. Ibid. 66 (4): 1245-1266. 2 1 5 Sanders uses individual-level data where endogeneity is particularly prevalent. 2 1 6 Within Canada, Dornan, Christopher, and Heather Pyman. 2001. Facts and Arguments: Newspaper Coverage of the Campaign. In The Canadian General Election of2000, edited by J. H. Pammett and C. Dornan. Toronto: The Dundurn Group, Frizzell, Alan, and Anthony Westell. 1989. The Media and the Campaign. In The Canadian General Election of 1988, edited by A. Frizzell, J. H. Pammett and A. Westell. Ottawa: Carleton University Press, 164 information is accurate, that voters will access it, or that it will be correctly incorporated into their assessments. Each voter's evaluation of the government may be more subjective than the rational gods metaphor may suggest. The role of the media in the electorate's evaluation of politicians will likely only increase as party mobilisation, civic engagement and other traditional sources of campaign information decrease. Therefore, understanding the role of the media in translating economic conditions into economic perceptions, economic perceptions into party popularity, and shaping the political context in which this occurs may be the key to understanding how electoral accountability truly operates. Fortunately for analysts, unlike other measures of perceptions, media coverage of the economy and political parties is not strongly endogenous to public opinion. Therefore it is an ideal way to take into account the subjective nature of the evaluations of the electorate regarding the economy and the government without having to rely upon subjective and endogenous measures. Theories on the role of the media and areas of needed research are discussed below. 3.2 Election Campaigns One of the Driving forces of political contextual conditions previously described is election campaigns. Elections produce downswings in the popularity of the incumbent party. In Chapter IV it was suggested that this was the product of media forces. Zaller's "receive, accept and sample" process of mass opinion formation is a useful tool in understanding how the media may produce such a campaign effect.218 At the risk of oversimplifying a relatively sophisticated idea, Zaller's theory contains a number of propositions regarding how individuals receive information, accept it and sample from it when deciding their position on an issue. These propositions suggest Frizzell, Alan, and Anthony Westell. 1994. The Press and the Prime Minister. In The Canadian General Election of 1993, edited by A. Frizzell, J. H. Pammett and A. Westell. Ottawa: Carleton University Press. 2 1 7 Shah, Dhavan, Mark Watts, David Domke, David Fan, and Michael Fibison. 1999. News Coverage, Economic Cues, and the Public's Presidential Preferences, 1984-1996. The Journal of Politics 61 (4):914-943. 2 1 8 Zaller, John. 1992. The Nature and Origins of Mass Opinion. Cambridge (England): Cambridge University Press. 165 that different types of people receive different types and amounts of information on an issue, such as their vote preference. Acceptance of this information also varies. Generally, those that are most interested in the issue and have previously received and accepted the most information are the most resistant to accepting new information. Those that are the least interested are the most susceptible to accepting information but will receive the least. Those interested enough in the issue to receive information but not so interested as to have already settled on their opinion are most susceptible to receiving and accepting new information. When choosing a position, such as whether the government has performed well, the respondent samples from the information that he/she has received and accepted, choosing a position somewhere in the middle of the information sampled. The sampling, however, is not completely random. Information that is ' top of mind ' is more likely to be sampled. The probability that information is 'top of mind ' will depend upon factors such as the recentness with which the respondent has accepted it or been reminded of it. On a decision such as whom to vote for, the media provides a great deal of the information that respondents receive. The media also can bring to ' top of mind ' information that the respondent has already accepted. There is evidence that in US presidential elections the incumbent receives greater media coverage.219 In Canada, the pattern is similar. The government party has been shown to have 9 9 0 greater access to the media outside electoral campaigns than its opponents. During campaigns, this difference diminishes but it is still present. During the 1984 election, the incumbent Liberal party and its leader, John Turner, received a greater volume of overall attention and a 2 1 9 Shah, Dhavan, Mark Watts, David Domke, David Fan, and Michael Fibison. 1999. News Coverage, Economic Cues, and the Public's Presidential Preferences, 1984-1996. The Journal of Politics 61 (4):914-943. 2 2 0 Moniere, Denis, and Julie Fortier. 2000. Radioscopie de Vinformation televisee au Canada. Montreal: Presses de l'Universite de Montreal. 2 2 1 Blais, Andre, Elizabeth Gidengil, Richard Nadeau, and Neil Nevitte. 2002. Anatomy of a Liberal Victory: Making Sense of the Vote in the 2000 Canadian Election. Peterborough: Broadview Press. 166 greater volume of negative attention than the Progressive Conservative party and its leader, Brian 999 Mulroney. This pattern was reversed in 1988 with Mulroney and the PC government receiving both more overall attention and more unfavourable attention.223 During the 1993 election, Frizzell and Westell demonstrate that the leader of the incumbent Progressive Conservatives, • 994 Kim Campbell, received more coverage than her opponents and that more of it was negative. Dornan and Pyman show that during the 2000 election campaign, newspaper stories were more negative toward Liberal Prime Minister Jean Chretien than any of the other leaders.225 In fact, of those stories which were determined as having a direct attack on a leader, "Chretien had the highest proportion of stories at 53 percent, followed by [Stockwell] Day at 32 percent and the rest of the leaders below 5 percent."226 With so much of the media attention being focused on the government leader and with so much of it being negative, it is in this way that the media may bring negative evaluations of the incumbent party ' top of mind ' during election campaigns, producing the increasingly downward swing in popularity. Further evidence of the potential media impact on government popularity during election campaigns is found in the relationship between the number of stories about the federal election in the media and the popularity of the incumbent government in the following month (each are plotted in figure 5-8). During the 1995-2002 period, the correlation is -0.58. Further, we can estimate the A R I M A model: 2 2 2 Frizzell, Alan, and Anthony Westell. 1985. The Canadian General Election of 1984: Politicians, Parties, Press and Polls. Ottawa: Carleton University press. 2 2 3 Frizzell, Alan, and Anthony Westell. 1989. The Media and the Campaign. In The Canadian General Election of 1988, edited by A. Frizzell, J. H. Pammett and A. Westell. Ottawa: Carleton University Press. 2 2 4 Frizzell, Alan, and Anthony Westell. 1994. The Press and the Prime Minister. In The Canadian General Election of 1993, edited by A. Frizzell, J. H. Pammett and A. Westell. Ottawa: Carleton University Press. 2 2 5 Dornan, Christopher, and Heather Pyman. 2001. Facts and Arguments: Newspaper Coverage of the Campaign. In The Canadian General Election of2000, edited by J. H. Pammett and C. Dornan. Toronto: The Dundurn Group. 167 \ Media Effects Model (1) govpop = a,+j30+ pumA * MEDIA,_X a , = pa,_x + s" where • t = \,...,T at monthly increments; • p,,P0and PmD1A are parameters to be estimated; and • MEDIA is a simple media count variable equal to the number published news stories about the federal election each month. The number of stories was calculated by a simple search on ' federal election ' with the Canadian Newsdisk (see Appendix B) for each month. The first line of the model is the structural equation and it contains both a constant and the effects of the number of stories about the federal election in the media in the preceding month. The second line models the disturbance as an ARMA(1,0) process. The results presented in table 5-1 suggest that for every additional news story on the federal election, government popularity dropped by 0.03 percent. Of course, this could be a spurious result. It is not surprising that the numbers of stories about federal elections increases during elections and there is no evidence of a direct causal link between this and incumbent government popularity. A proper examination requires a coding of media stories regarding the federal election in terms of their negativity towards the government over the entire 1957-2000 period. It must be demonstrated both that media negativity is correlated with government popularity and that the volume of negative media about the government during elections has increased relative to other parties. 3.3 Inter-Election Cycling In addition to explaining why campaigns produce negative election shocks for the incumbent government, the relationship between the media and government popularity may provide insight into part of the inter-election cycle which was only briefly discussed in the previous section. 168 Information effects produced by the media may contribute to the honeymoon period. The initial honeymoon that governments experience immediately following an election may be a product of stories in the media about the government's electoral victory combined with a lack of negative coverage. Richard Brody suggests this occurs because the political elite are hesitant to criticise a new government, seeing no political value in it, leaving the media without a source of anti-government sentiment.227 Moreover, unusually positive coverage of the government may occur because of the opposition's difficulty in obtaining its own media attention while Parliament is not in session. Once again, a proper coding of media coverage following elections is required to further analyze this potential source of the honeymoon effect. 3.4 Economic Perceptions Turning now from the general impact of the media on government popularity, we examine the potential for the media to intervene in government support decision by affecting the voters' economic perceptions. Hetherington shows that media coverage of the 1992 election campaign had a greater impact on voters' evaluations of the economy than did their party identification and 228 the impact of media coverage was more than half that of personal financial experience. We can test for the potential of a similar phenomenon in Canada by reproducing Hetherington's model for a few select Canadian elections using data from the 1997 Canadian election survey and the 1993 Canadian election study. The individual level model Hetherington uses is: 227Brody, Richard. 1991. Assessing the President: The Media, Elite Opinion, and Public Support. Stanford, California: Stanford University Press. 2 2 8 Hetherington, Marc. 1996. The Media's Role in Forming Voters' National Economic Evaluations in 1992. American Journal of Political Science 40 (2):372-395. 169 Media Effects Model (2) ECONEVAL = B0 + B m m * MEDIA + BPERFIN * PERFIN * PARTY ID + BPMRATING * PMRATING where • ECONEVAL is the individual's rating of the national economy (change in national economic conditions over the past year are rated on a seven-point scale from "gotten much worse" to "gotten much better"); • MEDIA is a dummy variable equalling 1 i f the respondent indicated they pay attention to the media during the election; • PERFIN is the individuals rating of their personal financial situation (personal finances compared to a year ago is rated on a seven-point scale from "much worse off " to "much better off"); • TALK is a dummy variable equalling 1 if the respondent indicated that they talked to others about the election; • PARTYID is a dummy variable equalling 1 if the respondent indicated their party identification was that of the governing Conservatives in 1993 or the governing Liberals in 1997;and • PMRATING is a measure of the respondent's feeling towards the Conservative Prime Minister in 1993 and the Liberal Prime Minister in 1997 measured on a thermometer scale from zero to 100. Applying Hetherington's model to the 1993 and 1997 Canadian federal elections, it is apparent that the degree to which voters pay attention to the media affects their evaluation of economic performance (table 5-2). In 1993, those that paid more attention to the media had a less positive evaluation of the economy. In 1997, those that paid more attention to the media had a more positive evaluation of the economy. The economy was relatively weak in 1993 and significantly stronger in 1997. It cannot be affirmed that those that scored high on media consumption had a , more accurate assessment of the economy. However, it can be stated that they had a more extreme evaluation. This is consistent with the findings of Haller and Norpoth in the US that 22 news exposure affects the strength of voters' opinions about the economy but not the direction. That is, economic news produces a more pronounced assessment. 2 2 9 Haller, H. Brandon, and Helmut Norpoth. 1997. Reality Bites: News Exposure and Economic Opinion. Public Opinion Quarterly 61 (4):555-575. 170 Those with higher education levels are generally greater consumers of the news. Greater education levels may also lead to more accurate assessments of economic performance. If the more extreme opinions of media viewers described above are in fact a reflection of a more accurate assessment, then the relationship between media consumption and economic evaluation may be spurious. To control for this possibility, Hetherington's model was rerun including an education dummy variable indicating those with a university education. As is evident, the effect of media consumption on economic evaluations is changed little by the inclusion of this additional control. It would appear that the media does have a measurable impact on the strength of voters' economic evaluations. Beyond strengthening the voter's opinion about the economy, there are number of other potential ways in which the news media may affect the electorate's evaluation of the economy and the government's responsibility for it. In terms of the information provided by the news media, there is evidence from the US that the volume, tone and focus of media coverage varies with the context. The news media has been found to respond to economic downturns by increasing the volume of coverage on economic issues and by painting the incumbent in an unfavorable light. 2 3 0 Norpoth argues that "the economy is likely to be in the news when it is ailing, not when it is healthy... According to this hypothesis, the popular standing of the governing party suffers from bad economic performance but benefits much less, if at all, from good performance."231 In contradiction to this, Shah finds that in the US an incumbent does benefit from positive economic performance. When a popular incumbent oversees a strong 2 3 0 Shah, Dhavan, Mark Watts, David Domke, David Fan, and Michael Fibison. 1999. News Coverage, Economic Cues, and the Public's Presidential Preferences, 1984-1996. The Journal of Politics 61 (4):914-943. 2 3 1 Norpoth, Helmut. 1996a. The Economy. In Comparing Democracies: Elections and Voting in Global Prospective, edited by L. LeDuc, R. Niemi and P. Norris. Thousand Oaks, California: Sage Publications. 171 economy, the total volume and tone of coverage increases in favour of the incumbent/ J Z "News media, like the public, reward incumbents for economic upturns and punish them for 911 downturns." Whether coverage of economic conditions is asymmetrical in good versus bad times and/or biased is a debate that continues in the US context and has barely been addressed in the Canadian context. Clearly it has important consequences for the electorate's perception of the economy and how economic conditions are translated into party popularity. Hetherington demonstrates that voters may be more likely to underestimate economic performance when growth is positive but not dramatic.234 This phenomenon is argued to be produced by the media. The media will report little on an economy that is neither particularly good nor particularly bad and therefore, not especially interesting. This may have been in effect during the 1992 American presidential election. Despite moderate economic growth, attention to the campaign through the news media negatively influenced voters' retrospective assessments of 91S the economy, reducing support for Bush. A media effects argument is also used by Thomas Holbrook in explaining why the usual economic voting models overpredicted Gore's share of the popular vote in 2000. He notes that the media may have played a role in reducing the positive benefits Gore would have normally 9 1 6 expected from the strong economy. Accordingly, Holbrook tested two new economic vote forecasting models. The first replaces economic performance with consumer reported recollection of economic news (positive versus negative). The second weights economic 2 3 2 Shah, Dhavan, Mark Watts, David Domke, David Fan, and Michael Fibison. 1999. News Coverage, Economic Cues, and the Public's Presidential Preferences, 1984-1996. The Journal of Politics 61 (4):914-943. 2 3 3 Ibid. 2 3 4 Hetherington, Marc. 1996. The Media's Role in Forming Voters' National Economic Evaluations in 1992. American Journal of Political Science 40 (2):372-395. 2 3 5 Ibid. 2 3 6 Holbrook, Thomas. 2001. Forecasting with Mixed Economic Signals: A Cautionary Tale. Political Science and Politics 34 (l):39-44. 172 performance by consumer reported economic news. Both improve the model substantially. The second makes the greatest overall improvements.237 A l l of these various debates regarding the information provided by media coverage of the economy and the government, and how voters incorporate this information into their decision whether or not to support the government are important. They are still very contentious debates within the US literature. Within Canada, they have only begun to be addressed. This is equally true of debates regarding the media's role in shaping the political context in which economic popularity operates. This is in part because the first step in addressing these issues is to be able to properly model the dynamics of party popularity and therefore government support. This requirement has been taken care of by this dissertation. The next step is to perform an analysis of media coverage of relevant events - economic and political. This must be done through the detailed coding of media coverage. Once this has been accomplished, the questions raised in this chapter regarding the effects of the media on party popularity and the relationship between party popularity and the economy can be addressed. The role of the media in shaping the political context - campaign effects, leadership affects, strategic voting considerations, etc - in which economic popularity operates can also be examined. Ultimately, this will reveal a great deal about how democratic accountability operates in Canada. 2 3 7 Ibid. 173 C O N C L U S I O N 174 This dissertation endeavors to make a substantial methodological contribution to the modelling of difficult time-series public opinion data. It proposes and demonstrates the utility of a unique structural approach to account for the great number of sources of nonstationarity within such data. The basis of this structural approach is the use of the state-space form of time series modelling. In this approach, the observed values of popularity are regarded as being made up of distinct unobserved components and measurement error. The unobserved components include nonstationary processes such as trending, cycling and fully-integrated baseline shifts. These components represent the political context in which economic popularity operates and are each modelled separately. This allows me to explore the dynamics of this political context. Another important nonstationary dynamic of measured popularity is measurement error. By explicitly modelling this error, the state-space model accounts for error variances which correlate and cycle with time. The unobserved components also include a stationary process. This stationary component is extracted from the nonstationary dynamics by the state-space model and is also modelled separately. Consequently, it is appropriate to use modelling techniques that assume stationarity to model the impact of economic conditions through it. Accordingly, I adopt a modified version of the Box-Jenkins approach to determine the correct lag structure for the state-space economic popularity models. When the state-space approach developed in this dissertation is applied to Canadian federal party popularity data between 1957 and 2000, important dynamics are revealed. The dynamics include trending and inter-election cycling in popularity, election effects, leadership effects, national crises effects, and period specific dynamics. As stated, these dynamics are the political context in which economic popularity operates. And operate it does. Economic effects are demonstrated clearly. Since 1984, the popularity of the party in government has been 175 strongly affected by economic conditions - in particular, inflation and economic growth. These effects have not been symmetrical for the Liberal and Conservative parties. Conservative governments before 1993 benefited from economic growth but only when it was accompanied by low inflation. Otherwise, the Conservatives were penalised. Liberal government after 1993 were also penalised for inflation but this penalty was tempered by good economic growth. The impact of economic conditions is stronger for PC governments before 1993 than Liberal governments after 1993 and generally governments are punished rather then rewarded for economic conditions but overall, economic effects have been substantial. In one extreme month, the popularity of the Conservative government was 15 percent lower than it would have been if the Canadian electorate did not hold the government accountable for the performance of the economy. Therefore, it is fair to conclude that governments in a very real way have at times been held accountable for the outcomes of their economic policies - but not always. We can tentatively conclude that electoral accountability operates as we expect but it is a tentative conclusion because before we can be fully comfortable with it we must understand why it is that some governments have not been held accountable. To do this, it is necessary to understand the mechanism that links economic conditions to party popularity. I have suggested that this understanding can be obtained through a careful analysis of the role of the media and have demonstrated the potential for the media to influence voters' evaluations of the economy. It is also necessary to understand how the changing political context may contribute to variations in economic effects. This dissertation has gone a long way in controlling for political contextual factors and has begun to explicitly model some of the forces driving this political context (elections, party leadership, national crises) but a great deal of further analysis has been suggested as necessary. The forces driving the inter-election cycle are of particular interest. This 176 cycling is persistent and is potentially related to economic conditions and the timing of elections. The amplitude of the cycle differs from period to period but its relationship to the timing of elections is consistent. This suggests that the cycle is the product of forces which have been at work throughout the 1957-2000 time-span. There is strong circumstantial evidence that this cycle is related to cycling within economic conditions which is also persistent. The varying amplitudes of the economic cycling corresponds with the varying amplitudes of the political cycling. Moreover, the timing of economic cycling relative to political cycling is consistent with a strong relationship between the two: peaks in GDP tend to precede peaks in popularity; peaks in inflation tend to follow peaks in popularity; and, in the first period, unemployment lows precede peaks in popularity. Clearly, there is a great deal more work to be done but this dissertation provides some of the strongest evidence to date of a political business cycle. Distinct period dynamics are also of interest. It has been suggested that the distinct party popularity dynamics within each of the three periods described is a product of varying political forces and that these forces are ultimately responsible for varying economic effects in different periods. Yet this remains a puzzle to be solved. Again, a careful analysis of the role of the media in shaping the political context would go a long way to contributing to our understanding of the changing political context and its potential impact on economic effects. The methodological advances made in this dissertation have a great number of potential payoffs. A great deal more can now be said about the relationship between the performance of the Canadian economy and support for the federal government than at any time previous. This greatly enhances our understanding of how electoral accountability operates within the Canadian context. Beyond the contributions made by this dissertation, a great number of interesting and exciting directions for further research are revealed. None of these suggested analyses could be 177 pursued without the methods developed in this dissertation. In this way the work done here lays the foundation for a great deal of needed research. 178 B I B L I O G R A P H Y 179 Alt, James E. 1978. The Politics of Economic Decline. Cambridge: Cambridge University Press. Anderson, Christopher. 2000. Economic Voting and Political Context: A Comparative Perspective. Electoral Studies 19:151-170. Archer, Keith, and Marquis Johnson. 1988. Inflation, Unemployment and Canadian Federal Voting Behaviour. Canadian Journal of Political Science 21 (3):569-584. Beck, Nathaniel. 1990. Estimating Dynamic Models Using Kalman Filtering. Political Analysis 1. Beck, Nathaniel. 1991. The Economy and Presidential Approval: A n Information Theoretic Perspective. In Economics and Politics: The Calculus of Support, edited by H. Norpoth, M . Lewis-Beck and J.-D. Lafay. Ann Arbor, Michigan: The University Of Michigan Press. Belanger, Eric. 2003. Issue Ownership by Canadian Political Parties 1953-2001. Canadian Journal of Political Science 36 (3):539-558. Blais, Andre, Elizabeth Gidengil, Richard Nadeau, and Neil Nevitte. 2002. Anatomy of a Liberal Victory: Making Sense of the Vote in the 2000 Canadian Election. Peterborough: Broadview Press. Blais, Andre, and Richard Nadeau. 1992. The Electoral Budget Cycle. Public Choice 74:389-403. Blais, Andre, Richard Nadeau, Elizabeth Gidengil, and Neil Nevitte. 2001. Measuring Strategic Voting in Multiparty Plurality Elections. Electoral Studies 20:343-352. Bloom, Howard S., and H . Douglas Price. 1975. Voter Response to Short-Run Economic Conditions: The Asymmetric Effect of Prosperity and Recession. The American Political Science Review 69 (4): 1240-1254. Bothwell, Robert, Ian Drummond, and John English. 1989. Canada Since 1945: Power, Politics, and Provincialism. Revised ed. Toronto: University of Toronto Press. Box-Steffensmeier, Janet, and Andrew Tomlinson. 2000. Fractional Integration Methods in Political Science. Electoral Studies 19:63-76. Brody, Richard. 1991. Assessing the President: The Media, Elite Opinion, and Public Support. Stanford, California: Stanford University Press. Campbell, James. 2001. An Evaluation of the Trial-Heat and Economic Forecast of the Presidential Vote in the 2000 Election. American Politics Research 29 (3):289-296. Campbell, James. 2004. Introduction - The 2004 Presidential Election Forecasts. Political Science and Politics 37 (4): 73 3-73 5. Carlsen, Fredrik. 2000. Unemployment, Inflation and Government Popularity - are there Partisan Effects? Electoral Studies 19:141-150. Carmichael, Calum M . 1990. Economic Conditions and the Popularity of the Incumbent Party in Canada. Canadian Journal of Political Science 23 (4):713-726. Chappell, Henry, Jr, and William Keech. 1985. A New View of Political Accountability for Economic Performance. The American Political Science Review 79 (1): 10-27. Clarke, Harold D., Jane Jenson, Lawrence LeDuc, and Jon Pammett. 1982. Voting Behaviour and the Outcome of the 1979 Federal Election: The Impact of Leaders and Issues. Canadian Journal of Political Science 15 (3):518-552. Clarke, Harold D., and Allan Kornberg. 1989. Public Reactions to Economic Performance and Political Support in Contemporary Liberal Democracies: The Case of Canada. In Economic Decline and Political Change: Canada, Great Britain, the United States, edited by H. D. Clarke. Pittsburgh: University Of Pittsburgh Press. 180 Clarke, Harold D., and Allan Kornberg. 1992. Support for the Canadian Federal Progressive Conservative Party since 1988: The Impact of Economic Evaluations and Economic Issues. Canadian Journal of Political Science 25 (l):29-53. Clarke, Harold D., Marianne C. Stewart, and Gary Zuk. 1986. Politics, Economics and Party Popularity in Britain, 1979-83. Electoral Studies 5 (2):123-141. Clarke, Harold D., and Gary Zuk. 1987. The Politics of Party Popularity: Canada 1974-1979. Comparative Politics:299-315. Cohen, Jeffrey E. 2004. Economic Perceptions and Executive Approval in Comparative Prospective. Political Behavior 26 (l):27-43. Crespi, Irving. 1980. The Case of Presidential Popularity. In Polling on the Issues, edited by A. Cantril. Washington DC: Seven Locks Press. Dornan, Christopher, and Heather Pyman. 2001. Facts and Arguments: Newspaper Coverage of the Campaign. In The Canadian General Election of2000, edited by J. H. Pammett and C. Dornan. Toronto: The Dundurn Group. Downs, Anthony. 1957. An economic theory of democracy. New York,: Harper. Durbin, James. 2004. Introduction to State Space Time Series Analysis. In State Space and Unobserved Component Models: Theory and Applications, edited by A. C. Harvey, S. J. Koopman and N . Shephard. Cambridge: Cambridge University Press. Erickson, Lynda. 1988. CCF-NDP Popularity and the Economy. Canadian Journal of Political Science 21 (1):99-116. Evans, Geoffrey, and Robert Andersen. 2004. The Political Conditioning of Economic Perceptions: Evidence from the 1992-97 British Electoral Cycle. Vol . 2004-W9, Nuffield College Politics Working Paper. Oxford: University of Oxford. Fair, Ray C. 1988. The Effect of Economic Events on Votes for President: 1984 Update. Political Behavior 10 (2):168-179. Fiorina, Morris P. 1978. Economic Retrospective Voting in American National Elections: A Micro-Analysis. American Journal of Political Science 22 (2):426-443. Fletcher, Frederick. 1978. The Mass Media in the 1974 Canadian Election. In Canada at the Polls: The General Election of1974, edited by H. Penniman. Washington, D. C : American Enterprise Institute for Public Policy Research. Fortin, Pierre. 2001. Interest Rates, Unemployment and Inflation: The Canadian Experience in the 1990s. In The Review of Economic Performance and Social Progress The Longest Decade: Canada in the 1990s, edited by K. Banting, A . Sharpe and F. St-Hilaire. Montreal: Centre for the Study of Living Standards and the Institute for Research on Public Policy and distributed by McGill-Queen's University Press. Frey, Bruno, and Friedrich Schneider. 1978a. An Empirical Study of Political-Economic Interaction in the United States. The Review of Economics and Statistics 60 (2): 174-183. Frey, Bruno, and Friedrich Schneider. 1978b. A Politico-Economic Model of the United Kingdom. Economic Journal 88:243-254. Frey, Bruno, and Friedrich Schneider. 1979. A n Econometric Model with an Endogenous Government Sector. Public Choice 34:25-45. Frizzell, Alan, and Anthony Westell. 1985. The Canadian General Election of1984: Politicians, Parties, Press and Polls. Ottawa: Carleton University press. Frizzell, Alan, and Anthony Westell. 1989. The Media and the Campaign. In The Canadian General Election of1988, edited by A. Frizzell, J. H. Pammett and A . Westell. Ottawa: Carleton University Press. 181 Frizzell, Alan, and Anthony Westell. 1994. The Press and the Prime Minister. In The Canadian General Election of1993, edited by A . Frizzell, J. H . Pammett and A . Westell. Ottawa: Carleton University Press. Goodhart, C.A.E. , and R.J. Bhansali. 1970. Political Economy. Political Studies 18 (1):43-106. Green, Donald, Alan Gerber, and Suzanna De Boef. 1999. Tracking Opinion over Time: A Sampling Method for Reducing Sampling Error. Public Opinion Quarterly 63:178-192. Haller, H. Brandon, and Helmut Norpoth. 1997. Reality Bites: News Exposure and Economic Opinion. Public Opinion Quarterly 61 (4):555-575. Happy, J. R. 1986. Voter Sensitivity to Economic Conditions: A Canadian-American Comparison. Comparative Politics 19 (l):45-56. Harvey, A . C. 1993. Time Series Models. 2nd ed. Cambridge, Mass.: MIT Press. Hetherington, Marc. 1996. The Media's Role in Forming Voters' National Economic Evaluations in 1992. American Journal of Political Science 40 (2):372-395. Hibbs, Douglas, Jr. On the Demand for Economic Outcomes: Macro Economic Performance and Mass Political Support in the United States, Great Britain, and Germany. Holbrook, Thomas. 2001. Forecasting with Mixed Economic Signals: A Cautionary Tale. Political Science and Politics 34 (l):39-44. Holbrook, Thomas M . 1996. Do campaigns matter?, Contemporary American politics. Thousand Oaks: Sage Publications. Irvine, William P. 1978. A n Overview of the 1974 Federal Election in Canada. In Canada at the Polls: The General Election of1974, edited by H. R. Penniman. Washington D. C : American Enterprise Institute for Public Policy Research. Jackman, Simon. 2000. Estimation and Inference via Bayesian Simulation: An Introduction to Markov Chain Monte Carlo. American Journal of Political Science 44:369-398. Johnston, Richard. 1992. Letting the People Decide: Dynamics of a Canadian Election. Montreal: McGill-Queen's University Press. Johnston, Richard. 1999. Business Cycles, Political Cycles and the Popularity of Canadian Governments, 1974-1998. Canadian Journal of Political Science 32 (3):499-520. Johnston, Richard. 2001. Capturing Campaigns in National Election Studies. In Election Studies: What's Their Use?, edited by E. Katz and Y . Warshel. Boulder, Colorado: Westview Press. Kayser, Mark. 2005. Who Surfs, Who Manipulates? The Determinants of Opportunistic Election Timing and Electorally Motivated Economic Intervention. American Political Science Review 99(1): 17-27. Kernell, Samuel. 1978. Explaining Presidential Popularity. How A d Hoc Theorizing, Misplaced Emphasis, and Insufficient Care in Measuring One's Variables Refuted Common Sense and Led Conventional Wisdom Down the Path of Anomalies. The American Political Science Review 72 (2):506-522. Key, V.O. Jr. 1964. Politics, Parties, and Pressure Groups. Fifth ed. New York: Thomas Y . Crowell Company. Key, V.O. Jr. 1966. The Responsible Electorate: Rationality in Presidential Voting, 1936-1960. Cambridge, Massachusetts: The Belknap Press of Harvard University Press. Kiewiet, D. Roderick. 1983. Macroeconomics and Micropolitics. Chicago: The University Of Chicago Press. Kinder, Donald R., and D. Roderick Kiewiet. 1981. Sociotropic Politics: The American Case. British Journal of Political Science 11 (2): 129-161. 182 Kramer, Gerald H. 1971. Short-Term Fluctuations in US Voting Behavior, 1896-1964. The American Political Science Review 65 (1): 131-143. Kramer, Gerald H. 1983. The Ecological Fallacy Revisited: Aggregate- versus Individual-level Findings on Economic and Elections, and Sociotropic Voting. The American Political Science Review 11 (1):92-111. Lemieux, Vincent, and Jean Crete. 1981. Quebec. In Canada at the Polls, 1979 and 1980, edited by H. R. Penniman. Washington: American Enterprise Institute. Lewis-Beck, Michael. 1988. Economics and Elections. Ann Arbor: University of Michigan Press. Lewis-Beck, Michael, and Heinz Eulau. 1985. Economic Conditions and Electoral Behaviour in Transnational Perspective. In Economic Conditions and Electoral Outcomes: The United States and Western Europe, edited by M . Lewis-Beck and H . Eulau. New York: Agathon Press, Inc. Lewis-Beck, Michael, Richard Nadeau, and Eric Belanger. 2004. General Election Forecasts in the United Kingdom: A Political Economy Model. Electoral Studies 23:279-290. Lewis-Beck, Michael S., and Mary Stegmaier. 2000. Economic Determinants of Electoral Outcomes. Annual Review of Political Science 3:183-219. Lowry, Robert C , James E. Alt, and Karen E. Ferree. 1998. Fiscal Policy Outcomes and Electoral Accountability in American States. The American Political Science Review 92 (4):759-774. MacKuen, Michael B., Robert S. Erikson, and James A. Stimson. 1992. Peasants or Bankers? The American Electorate and the US Economy. American Political Science Review 86:597-611. McCleary, Richard, Richard Hay Jr., Errol Meidinger, and David McDowall. 1980. Applied Time Series Analysis for the Social Sciences. London: Sage Publications. Moniere, Denis, and Julie Fortier. 2000. Radioscopie de Vinformation televisee au Canada. Montreal: Presses de l'Universite de Montreal. Monroe, Kristen, and Lynda Erickson. 1986. The Economy and Political Support: The Canadian Case. The Journal of Politics 48:616-647. Nadeau, Richard, and Andre Blais. 1993. Explaining Election Outcomes in Canada: Economy and Politics. Canadian Journal of Political Science 26 (4):775-790. Nadeau, Richard, and Michael Lewis-Beck. 2001. National Economic Voting in US Presidential Elections. The Journal of Politics 63 (1): 159-181. Nadeau, Richard, Richard Niemi, and Timothy Amato. 1996. Prospective and Comparative or Retrospective and Individual? Party Leaders and Party Support in Great Britain. British Journal of Political Science 26:245-258. Norpoth, Helmut. 1985. Economics, Politics, and the Cycle of Presidential Popularity. In Economic Conditions and Electoral Outcomes: The United States and Western Europe, edited by M . Lewis-Beck and H. Eulau. New York: Agathon Press, Inc. Norpoth, Helmut. 1996a. The Economy. In Comparing Democracies: Elections and Voting in Global Prospective, edited by L. LeDuc, R. Niemi and P. Norris. Thousand Oaks, California: Sage Publications. Norpoth, Helmut. 1996b. Presidents and the Prospective Voter. The Journal of Politics 58 (3):776-792. Paldam, Martin. 1991. How Robust Is the Vote Function?: A Study of Seventeen Nations over Four Decades. In Economics and Politics: The Calculus of Support, edited by H. 183 Norpoth, M . Lewis-Beck and J.-D. Lafay. Ann Arbor, Michigan: The University Of Michigan Press. Pammett, Jon H. 2001. The People's Verdict. In The Canadian General Election of2000, edited by J. H. Pammett and C. Dornan. Toronto: The Dundurn Group. Patterson, Thomas. 1993. Out Of Order. New York: Knopf. Penniman, Howard Rae. 1981. Canada at the polls, 1979 and 1980 : a study of the general elections, AEI studies. Washington, D.C.: American Enterprise Institute for Public Policy Research. Petry, Francois, and Howard Harmatz. 1995. Politico-Economic Interactions in Canada: A n Empirical Assessment. Public Finance Quarterly 23 (3):305-335. Pickup, Mark. 2003. Globalisation, Politics, and Provincial Government Spending in Canada. Paper read at Canadian Political Science Association Meeting, at Halifax. Powell, G. Bingham, Jr., and Guy D. Whitten. 1993. A Cross-National Analysis of Economic Voting: Taking Account of the Political Context. American Journal of Political Science 37(2):391-414. Sanders, David. 2000. The Real Economy and the Perceived Economy in Popularity Functions: How Much Do Voters Need to Know? A Study of British Data, 1974-97. Electoral Studies 19:275-294. Sanders, David, and Neil Gavin. 2004. Television News, Economic Perceptions and Political Preferences in Britain, 1997-2001. The Journal of Politics 66 (4):1245-1266. Shah, Dhavan, Mark Watts, David Domke, David Fan, and Michael Fibison. 1999. News Coverage, Economic Cues, and the Public's Presidential Preferences, 1984-1996. The Journal of Politics 61 (4):914-943. Siegfried, Andre. 1907. The Race Question in Canada. London,: E. Nash. Tufte, Edward R. 1975. Determinants of the Outcomes of Midterm Congressional Elections. The American Political Science Review 69 (3):812-826. Tufte, Edward R. 1978. Political Control of the Economy. New Jersey: Princeton University Press. Whiteley, P. 1984. Inflation, Unemployment and Government Popularity: Dynamic Models for the United States, Britain and West Germany. Electoral Studies 3 (l):3-24. Wlezien, Christopher. 2000. A n Essay on 'Combined' Time Series Processes. Electoral Studies 19:77-93. Zaller, John. 1992. The Nature and Origins of Mass Opinion. Cambridge (England): Cambridge University Press. Zaller, John. 2001. The Rule of Product Substitution in Presidential Campaign News. In Election Studies: What's Their Use?, edited by E. Katz and Y. Warshel. Colorado: Westview Press. 184 J T A B L E S 185 Table 2-1: Estimated Parameters of Government Popularity State-Space Model 1957-1975 1979-1993 1993-2000 Standard Deviation Median Standard Deviation Median Standard Deviation Median Memory AR(1) 0.0708 0.8200 0.0525 0.8916 0.2618 0.6576 Amplitude 0.0076 0.0137 0.0157 0.0601 0.0123 0.0261 Phase 0.1798 0.0797 0.3358 Liberals Trend - Short 0.0222 0.0306 0.0113 0.0152 0.0218 -0.0050 Trend - Long 0.0003 -0.0001 0.0014 -0.0052 0.0011 -0.0001 Constant 0.0867 0.2871 0.0680 0.4130 0.0415 0.5715 Progressive Conservatives Trend - Short 0.0113 0.0184 0.0086 0.0046 Trend - Long 0.0009 -0.0045 0.0006 -0.0018 Constant 0.0338 0.5181 0.0378 0.3180 Q-Test P-value Q-Test P-value Q-Test P-value Residuals 18.4530 0.9986 31.9544 0.8138 49.4384 0.1221 Notes: 1) Phase is calculated post-Bayesian estimation and so no SE for its distribution is provided but the statistical significance of the parameters used in the calculation indicate the significance of phase. 2) The popularity dependent variable is entered into the model as a proportion rather than a percentage. 3) Bolded values are determined to be statistically significant based on Bayesian estimated distribution of parameters. 2 3 8 Details on the guidelines used to determine significance of Bayesian estimated parameters can be obtained from the author at mapickup@interchange.ubc.ca. 186 Table 2-2: Estimated Parameters for Liberal Party Popularity State-Space Model 1957-1975 1979-1993 1993-2000 Standard Deviation Median Standard Deviation Median Standard Deviation Median Memory 0.0496 0.8855 0.0420 0.8949 0.2435 0.6137 Constant 0.0500 0.3049 0.0639 0.4161 0.0335 0.5681 Governmenl Trend - Long 0.0012 0.0014 0.0005 -0.0002 Governmenl Trend - Short 0.0107 0.0172 0.0094 -0.0127 0.0180 -0.0035 Governmenl Phase 0.2109 0.1857 0.3319 Governmenl Amplitude 0.0093 0.0164 0.0332 0.0739 0.0103 0.0254 Constant 0.0347 0.3209 0.0385 0.4173 c .© Trend - Long '5 o Trend - Short 0.0061 0.0085 0.0055 -0.0007 0 . 0 Phase 0.2935 0.3332 Amplitude 0.0117 0.0209 0.0129 0.0235 Q-Test P-value Q-Test P-value Q-Test P-value Residuals 18.0886 0.9989 27.6009 0.9314 33.6974 0.7101 187 Table 2-3: Estimated Parameters for PC Party Popularity State-Space Model 1957-1975 1979-1993 1993-2000 Standard Deviation Median Standard Deviation Median Standard Deviation Median Memory 0.2006 0.7942 0.0442 0.8988 0.0934 0.8054 Constant 0.0366 0.5172 0.0365 0.3297 jrnmenl Trend - Long 0.0009 -0.0043 0.0009 -0.0016 • • • • • H l i i H i i i l jrnmenl Trend - Short 0.0275 0.0193 0.0077 0.0016 Gov< Phase 0.1578 0.1050 Gov< Amplitude 0.0115 0.0215 0.0163 0.0477 ^^ ^^ ^^ ^^ ^^ ^^ ^^ ^^ ^^ ^^ ^^ ^^ ^^ ^^ ^^  Constant 0.1264 0.3236 0.0601 0.2627 0.0299 11.11— s Trend - Long • • • • • • • Opposit Trend - Short 0.0497 -0.0057 0.0116 0.0203 0.0091 0.0099 Opposit Phase 0.2206 0.2383 0.3168 Amplitude 0.0081 0.0142 0.0324 0.0677 0.0104 0.0231 Q-Test P-value Q-Test P-value Q-Test P-value Residuals 35.1737 0.6871 54.7514 0.0601 53.4751 0.0612 188 Table 2-4: Liberal and PC Party Popularity Variance, Explained and Unexplained 1957-1975 1979-1993 1993-2000 Liberals Total Variance 36.4 50.7 15.6 Variance Explained by Deterministic Components 12.4 17.1 3.8 Variance Attributable to Measurement Error 5.0 2.2 4.5 Unexplained Variance 18.9 31.4 7.3 Progressive Conservatives Total Variance 87.9 156.7 12.4 Variance Explained by Deterministic Components 71.1 121.1 3.5 Variance Attributable to Measurement Error 3.3 3.8 2.1 Unexplained Variance 13.5 31.8 6.8 Popularity is expressed as percentage for the purposes of this table Table 3-1: Liberal Economic Popularity Estimates 1957-1975 1979-1993 1993-2000 Standard Deviation Median Standard Deviation Median Standard Deviation Median Memory AR(1) 0.4692 -0.5651 0.0531 0.8243 0.3253 0.1961 GDP x Inflation 0.0016 0.0001 0.0006 -0.0002 0.0051 0.0038 c e Inflation 0.0060 -0.0013 0.0020 0.0027 0.0161 -0.0142 Govern n GDP 0.0132 -0.0059 0.0053 0.0031 0.0083 -0.0024 Govern n Unemployment 0.0094 -0.0043 0.0024 -0.0001 0.0056 0.0013 Govern n Trend - short 0.1033 0.2969 0.0340 -0.0437 0.0545 -0.0072 GDP x Inflation 0.0094 0.0131 0.0009 0.0017 e Inflation 0.0175 -0.0194 0.0019 -0.0001 o GDP 0.0179 0.0001 0.0046 -0.0089 a o Unemployment 0.0212 0.0233 0.0020 0.0010 Trend - short 0.1409 -0.0399 0.0242 -0.0096 Q-Test P-value Q-Test P-value Q-Test P-value Residuals 25.9024 0.9586 32.6373 0.7893 33.7412 0.7082 190 Table 3-2: PC Economic Popularity Estimates 1957-1975 1979-1993 1993-2000 Standard Deviation Median Standard Deviation Median Standard Deviation Median Memory AR(1) 0.2055 -0.7557 0.0486 0.8214 0.1151 0.7504 Government GDP x Inflation 0.0076 0.0049 0.0007 -0.0012 Government Inflation 0.0172 0.0044 0.0017 -0.0020 Government GDP 0.0169 -0.0042 0.0035 0.0079 Government Unemployment 0.0207 0.0247 0.0019 0.0000 Government Trend - short 0.1305 -0.0063 0.0217 0.0098 Opposition GDP x Inflation 0.0017 -0.0006 0.0005 0.0007 0.0031 -0.0005 Opposition Inflation 0.0066 -0.0036 0.0019 -0.0032 0.0097 0.0030 Opposition GDP 0.0135 -0.0028 0.0047 -0.0081 0.0049 0.0000 Opposition Unemployment 0.0087 0.0057 0.0023 0.0001 0.0032 0.0014 Opposition Trend - short 0.0549 -0.0359 0.0311 0.0659 0.0299 -0.0014 Q-Test P-value Q-Test P-value Q-Test P-value Residuals 25.8381 0.9595 51.2324 0.1098 33.3899 0.7231 191 Table 3-3: Liberal Economic Popularity Estimates, No Interaction Term 1957-1975 1979-1993 1993-2000 Standard Deviation Median Standard Deviation Median Standard Deviation Median Memory AR(1) 0.6486 0.6729 0.0446 0.8766 0.3127 0.2190 Government Inflation 0.0024 -0.0006 0.0018 0.0025 0.0060 -0.0032 GDP 0.0048 -0.0023 0.0015 0.0017 0.0037 0.0030 Unemployment 0.0063 -0.0017 0.0022 0.0010 0.0048 -0.0006 Trend - short 0.1290 0.0726 0.0332 -0.0471 0.0542 -0.0054 Opposition Inflation 0.0117 -0.0064 0.0017 0.0015 GDP 0.0133 0.0061 0.0011 -0.0003 Unemployment 0.0192 0.0129 0.0020 0.0007 Trend - short 0.1080 -0.0553 0.0238 -0.0140 192 Table 3-4: PC Economic Popularity Estimates, No Interaction Term 1957-1975 1979-1993 1993-2000 Standard Deviation Median Standard Deviation Median Standard Deviation Median Memory AR(1) 0.4056 -0.7420 0.0480 0.8314 0.1064 0.7609 Government Inflation 0.0148 0.0101 0.0016 -0.0030 • H H H H H H I GDP 0.0084 0.0049 0.0012 0.0021 Unemployment 0.0193 0.0258 0.0018 0.0004 Trend - short 0.1152 -0.0379 0.0217 0.0109 Opposition Inflation 0.0038 -0.0049 0.0017 -0.0021 0.0033 0.0017 GDP 0.0059 -0.0061 0.0015 -0.0019 0.0018 -0.0006 Unemployment 0.0074 0.0063 0.0022 -0.0009 0.0026 0.0016 Trend - short 0.0478 -0.0284 0.0312 0.0597 0.0293 -0.0023 193 Table 3-5: Liberal Economic Popularity Estimates, Change in Unemployment 1957-1975 1979-1993 1993-2000 Standard Standard Standard Deviation Median Deviation Median Deviation Median Memory AR(1) 0.5252 0.1073 0.0482 0.8245 0.2986 0.3487 GDPx Inflation 0.0010 0.0004 0.0006 -0.0002 0.0041 0.0029 jrnment Inflation 0.0034 -0.0022 0.0021 0.0028 0.0141 -0.0128 jrnment GDP 0.0083 -0.0050 0.0055 0.0030 0.0072 -0.0013 Govt Change in Unemployment 0.0249 0.0221 0.0199 -0.0003 0.0257 0.0240 Trend - short 0.1008 0.1518 0.0220 -0.0449 0.0247 0.0060 GDP x Inflation 0.0091 0.0092 0.0009 0.0017 s _o Inflation 0.0153 -0.0200 0.0017 -0.0006 — . • josit GDP 0.0121 -0.0064 0.0045 -0.0090 o. O Change in Unemployment 0.2653 -0.0129 0.0191 -0.0050 Trend - short 0.0504 0.0746 0.0082 0.0020 194 Table 3-6: PC Economic Popularity Estimates, Change in Unemployment 1957-1975 1979-1993 1993-2000 Standard Deviation Median Standard Deviation Median Standard Deviation Median Memory AR(1) 0.6366 -0.3685 0.0414 0.8228 0.1118 0.7438 Government GDPx Inflation 0.0071 0.0015 0.0006 0.0008 Government Inflation 0.0139 0.0110 0.0015 -0.0020 Government GDP 0.0132 0.0031 0.0035 0.0077 Government Change in Unemployment 0.0527 0.0041 0.0170 -0.0034 Government Trend - short 0.0738 0.0711 0.0076 0.0100 Opposition GDPx Inflation 0.0012 -0.0005 0.0007 -0.0012 0.0025 -0.0011 Opposition Inflation 0.0037 0.0000 0.0019 -0.0036 0.0090 0.0046 Opposition GDP 0.0089 0.0000 0.0050 -0.0087 0.0046 0.0002 Opposition Change in Unemployment 0.0201 -0.0167 0.0181 0.0066 0.0200 -0.0162 Opposition Trend - short 0.0392 -0.0125 0.0204 0.0701 0.0151 0.0099 195 Table 3-7: Liberal Economic Popularity Estimates, AR(2) Model 1957-1975 Standard Deviation Median Memory AR(1) 0.4881 -0.4517 Memory AR(2) 0.3174 0.0908 Government GDP x Inflation 0.0015 0.0004 Government Inflation 0.0052 -0.0017 Government GDP 0.0114 -0.0060 Government Unemployment 0.0086 -0.0029 Government Trend - short 0.1325 0.2490 Opposition GDP x Inflation 0.0101 0.0098 Opposition Inflation 0.0169 -0.0176 Opposition GDP 0.0164 -0.0012 Opposition Unemployment 0.0218 0.0186 Opposition Trend - short 0.1207 -0.0385 Q-Test P-value Residuals 25.1538 0.9677 Table 3-8: Modelling Economic Variables as First -Order Autoregressive Process with Drift, yt =pyt-X+P + £t 1957-1975 1979-1993 1993-2000 Standard Standard Standard Deviation Median Deviation Median Deviation Median GDP AR(1) 0.02052 0.9515 0.02213 0.9455 0.03761 0.9421 Unemployment AR(1) 0.01677 0.9714 0.009722 0.9908 0.01777 0.9737 Inflation AR(1) 0.008063 1.009 0.01485 0.9816 0.04358 0.9526 GDP Trend 123.9 2.73 103 1.136 92.03 2.729 Unemployment Trend 82.66 5.895 333.9 10.48 885.2 7.012 Inflation Trend 229 -0.1011 166.8 3.512 301.8 1.945 P-Value Q-Test P-Value Q-Test P-Value Q-Test GDP Residuals 0.7613 33.37 0.1890 47.67 0.4488 40.49 Unemployment Residuals 0.1319 50.08 0.4650 40.12 0.4453 40.57 Inflation Residuals 0.5488 38.26 0.3559 42.70 0.6493 36.04 197 Table 3-9: ARIMA Regression Model of Economic Variables, AR(1,13, 25, 37) Unemployment GDP Inflation Coefficient Standard Error Coefficient Standard Error Coefficient Standard Error AR(1) 0.9997 0.0012 0.9734 0.0150 0.9973 0.0030 AR(13) -0.0000062 0.0000025 0.0000000 0.0135 -0.0000033 -0.0000109 AR(25) 0.0000050 0.000002 0.0000000 0.0134 -0.0000099 0.0000110 AR(37) -0.0000009 0.000003 0.0000000 0.0146 -0.0000010 0.0000072 Log Likelihood P-value Log Likelihood P-value Log Likelihood P-value 25.02 0.0000 -594.15 0.0000 -307.40 0.0000 198 Table 3-10: ARIMA Regression Model of Economic Variables, AR(1) Unemployment GDP Inflation Coefficient Standard Error Coefficient Standard Error Coefficient Standard Error AR(1) 0.9996 0.0012 0.9734 0.0115 0.9971 0.0032 Log Likelihood P-value Log Likelihood P-value Log Likelihood P-value 20.02 0.0000 -594.15 0.0000 -313.40 0.0000 199 Table 3-11: ARIMA Regression Model of GDPxInflation, AR(1, 4,13) GDPxInflation Coefficient Standard Error AR(1) 0.9818 0.0138 AR(4) -0.0560 0.0148 AR(13) 0.0000005 0.0107 Log Likelihood P-value -1604.63 0.0000 ©1 o o bo 0 0 N O o 2 3 ! UJ 1 ss Xcorr Unemployment H is O o " i re 5" 5" s o © UJ ON O © © 4^ o © o tO o © o o Xcorr Inflation o o oo oo o © o 4^ ON N O 4^ oo o t o o o o © © o ON UJ UJ o 1—1 UJ O N ON NO Xcorr GDP N © ©N. s n o N O UJ © © o oo to o © o o o ON NO o © U1 oo J> Xcorr GDPx Inflation tO O l u> o UJ o UK O tO NO 0 0 I o t o oo oo 2 S E M3 I© o O N o © U"l 0 O NO © t o Si Xcorr Unemployment • < a 3 O. PI r> o 3 O 3 < 5" ST o t o NO 1^ o © o t o o t o t o o o NO t o NO 0 0 ON o © 0 0 oo Xcorr Inflation N O S O o © © ON 4i. o NO I NO u/i Xcorr GDP o o UJ un o oo o o oo o o © o © 0 0 t o o © 1^ Xcorr GDPx Inflation 2 S E 2 1^ o NO U J - J t o o t o O N O N l o 4^ © NO o © o © o NO oo 4^ Xcorr Unemployment o © o oo © o o © UJ o © oo oo oo Xcorr Inflation N £ 00 o I h-k N O 00 NO ON ON © o © UJ •o UJ © © un t o o N O t o Xcorr GDP o © oo 1^ UJ UJ UJ UJ © UJ t o oo oo © UJ 4^. UJ t o © UJ © UJ O N t o o © u i 4^ t o © UJ t o © t o oo oo © i o Xcorr GDPx Inflation 2 S E Table 3-12: Cross-Correlation Functions for Liberal Popularity and Economic Variables (cont) 1984-1993 1993-2000 Lag Xcorr Unemployment Xcorr Inflation Xcorr GDP Xcorr GDPx Inflation 2SE Xcorr Unemployment Xcorr Inflation Xcorr GDP Xcorr GDPx Inflation 2SE 0 -0.1959 -0.0572 0.0993 0.0997 0.1451 -0.2112 -0.0869 0.0011 -0.1367 0.217 1 -0.0313 -0.0591 0.0381 0.0489 0.1455 0.061 -0.0553 0.0538 0.0146 0.218 2 -0.0083 0.058 0.0017 -0.0168 0.1459 0.1842 0.0239 -0.0866 -0.0518 0.220 3 0.0389 0.081 -0.0693 -0.091 0.1463 0.0274 0.1159 -0.0265 0.127 0.221 4 -0.0432 -0.1151 0.0402 0.087 0.1466 -0.0425 -0.0172 0.1074 0.095 0.222 5 -0.0481 0.0866 -0.1428 -0.1187 0.1470 -0.1005 -0.3136 0.0981 -0.2409 0.224 6 -0.0819 0.0361 -0.0102 -0.0114 0.1474 0.0119 -0.1218 -0.0423 -0.084 0.225 7 0.0716 0.0408 0.0579 0.0485 0.1478 -0.0003 0.0283 -0.0301 0.088 0.226 8 0.0405 -0.0033 0.0482 0.0433 0.1482 0.0478 -0.0132 0.0329 0.0532 0.228 9 0.0525 -0.0166 0.1356 0.1023 0.1487 -0.1085 0.02 0.0545 0.062 0.229 10 -0.0365 -0.0467 0.004 0.0063 0.1491 0.0243 -0.069 0.0111 -0.0368 0.231 11 0.0308 -0.1567 -0.0287 -0.0167 0.1495 -0.0001 -0.1285 -0.0297 -0.1441 0.232 12 0.0194 -0.0081 -0.0135 0.0211 0.1499 0.1072 0.1171 0.0742 0.1103 0.234 13 -0.0769 -0.0221 -0.0159 -0.0166 0.1503 -0.0747 0.1851 -0.0553 0.1303 0.236 14 0.0477 -0.0157 0.0051 0.0004 0.1508 -0.0114 0.061 0.0488 0.0914 0.237 15 0.0954 0.0914 -0.1516 -0.1497 0.1512 -0.1243 -0.0121 -0.0344 -0.0178 0.239 16 -0.0132 0.0386 -0.0074 -0.0113 0.1516 0.0785 -0.0602 -0.0696 -0.1557 0.241 17 0.0022 -0.096 0.1236 0.1187 0.1521 0.1105 0.1566 0.0201 0.0443 0.243 18 0.0331 0.0631 -0.0551 -0.072 0.1525 0.0437 0.1952 -0.0359 0.1246 0.244 19 -0.0195 0.073 -0.0959 -0.1184 0.1529 -0.0168 0.136 -0.0712 0.0465 0.246 20 0.0484 0.0396 0.0127 0.008 0.1534 -0.081 0.0632 -0.0071 0.0276 0.248 o 4^. O s © b O s ko o b o © o O s O s t o o © o © oo o 0 0 o © o 0 0 S O o b o b t o o so S O 4^ © oo O s © S O S O o b o b 3 o b © © o b O s O s o b o b o b tO S, © o oo O s © o o ' lO oo oo ^1 o o b S O O b 3M o b t o 351 o b o o S O o b O s o b t o oo S O O b oo o b o to o S O 4^ t o o o S O © b o S O o b t o so © b oo t o ^1 O b | © t o -0 o l O S O o b oo so o t O o b o b o oo o b 2 o u>| O s OO O s t o O s 4^ 4* O s t o 0 0 o 3 o b oo oo t o S O t o O S O 4^ OO tO t o S O t o O io § o b t o O b © t O © OO oo 4^ o b ho o Si 4^ O IO o b oo on S O 4*. t o O s S O t o oo © © LO LO oo O LO t o oo oo © L/ l S O LO t o © © LO O s I t o © S O Ln O s © ^1 2 © © t o I Si © t o < l 4*. TO Xcorr Unemployment Xcorr Inflation Xcorr GDP Ul I N O ON Xcorr GDPx Inflation H S3 a ST • O n o 2. 5" 5' s TI B 3 re 2SE Xcorr Unemployment Xcorr Inflation Xcorr GDP N O O N N © Ul Xcorr GDPx Inflation 2SE < re n o s Ire o II 65 a a. W re o s o 3 < 69 Xcorr Unemployment S3 re" Xcorr Inflation Xcorr GDP N O 00 o I I—' NO 00 Xcorr GDPx Inflation 2SE ©1 © © ho o © 0 0 oo © © J> NO to © © I © O N © © 89 OTQ © N O to Xcorr Unemployment H ss © -M © to O N U-I t o I © © UJ O N O © OO U« © © t o O N t o © oo 4^ Xcorr Inflation © t o N O O O © © © © ON © © O N t o IS, © t o oo Xcorr GDP N O 00 I H* N © N © W © U*l N O UJ © © © to t o t o © © oo © O N © © 0 0 NO J> Xcorr GDPx Inflation < n o s N O t o © t o t o I © N O t o © O N 2SE N O O N N O t o © © © © 3 © © © to © N O © © N O © © © © © © O N t o Xcorr Unemployment n *0 o •o e © © t o © © UJ © © © t o O N O N © N O O O © © Xcorr Inflation N as W n o s o 3 ST © UJ O N © O 0 O N © O O N ^1 © N O ^1 © © © © © © © © UJ O N I N O © © UJ UJ Xcorr GDP N © N © W O © o © © O N oo t o -M oo © © © t o © © o o t o UJ © t o © Xcorr GDPx Inflation 2SE Table 3-14: Box-Jenkins Progressive Conservative Economic Popularity Estimates, 1957-1975 Memory AR(1) SE 0.5782 Median -0.6241 e E > o InflationxGDP(t-2) 0.0082 Inflation (t-2) 0.0152 GDP (t-2) 0.0198 InflationxGDP(t-3) 0.0082 Inflation (t-3) 0.0172 GDP (t-3) 0.0210 Trend - short 0.0622 -0.0013 0.0179 0.0048 0.0037 0.0142 0.0024 0.0479 o D. Q. o InflationxGDP(t-2) 0.0016 Inflation (t-2) 0.0134 GDP (t-2) 0.0106 InflationxGDP(t-3) 0.0018 Inflation (t-3) 0.0124 GDP (t-3) 0.0117 Trend - short 0.0447 0.0000 0.0041 -0.0107 -0.0010 -0.0070 0.0065 -0.0034 Table 3-15: Box-Jenkins Liberal Economic Popularity Estimates, 1957-1975 Model 1 Model 2 SE Median SE Median Memory AR(1) 0.2230 -0.6387 0.5898 -0.6481 InflationxGDP(t-6) 0.0016 0.0011 0.0014 0.0013 Inflation (t-6) 0.0112 -0.0014 0.0042 -0.0022 S GDP (t-6) 0.0104 -0.0077 0.0098 -0.0089 £ InflationxGDP(t-lO) 0.0014 0.0015 c Inflation (t-10) 0.0104 -0.0041 > O GDP (t-10) 0.0085 -0.0092 O Trend - Short 0.0598 0.2999 0.1093 0.2805 InflationxGDP(t-6) 0.0097 0.0022 0.0066 0.0017 Inflation (t-6) 0.0139 -0.0156 0.0129 -0.0147 GDP (t-6) 0.0171 0.0187 0.0120 0.0104 5 5 Unemployment (t-6) 0.0089 0.0307 0.0114 0.0252 O InflationxGDP(t-lO) 0.0106 0.0059 o Inflation (t-10) 0.0206 -0.0169 a. a. GDP (t-10) 0.0153 -0.0122 O Trend - Short 0.0420 -0.0507 0.0421 -0.0385 Table 3-16: Box-Jenkins Progressive Conservative Economic Popularity Estimates, 1979-1993 Model 1 Model 2 SE Median SE Median AR(1) 0.0567 0.8108 AR(1) 0.0552 0.7614 intercept 13 0.0014 -0.0031 intercepts 13 0.0014 -0.0027 inflation 13 0.0020 -0.0034 inflation 13 0.0023 -0.0049 GDP 13 0.0068 0.0148 a GDP 13 0.0067 0.0133 £ E intercepts (t-1) 0.0007 -0.0017 e c u inflation (t-1) 0.0019 -0.0028 > © > o GDP (t-1) 0.0036 0.0090 O Trend - Short 0.0080 0.0131 O Trend - Short 0.0149 0.0298 inflation 14 0.0011 0.0002 intercepts (t-1) 0.0005 0.0006 c _o GDP 8 0.0015 -0.0002 B O inflation (t-1) 0.0017 -0.0031 '-*-» Q Unemployment (t-9) 0.0072 0.0059 'v} o GDP (t-1) 0.0048 -0.0068 Q. a. Unemployment (t-10) 0.0073 -0.0050 a a. O Trend - Short 0.0094 0.0247 O Trend - Short 0.0195 0.0749 Table 3-17: Box-Jenkins Liberal Economic Popularity Estimates, 1979-1993 Model 1 SE Median Memory AR(1) 0.0428 0.8702 Government Unemployment (t-0) 0.0019 -0.0012 Trend - Short 0.0183 -0.0026 Opposition Inflation (t-11) 0.0014 -0.0001 Unemployment (t-0) 0.0018 -0.0007 Trend - Short 0.0180 0.0049 Table 3-18: Box-Jenkins Liberal Economic Popularity Estimates, 1993-2000 SE Median Memory AR(1) 0.2915 0.1003 InflationxGDP(t-5) 0.0049 0.0070 Inflation (t-5) 0.0182 -0.0310 GDP (t-5) 0.0086 -0.0084 Trend - Short 0.0297 0.0323 Table 4-1: Estimated Political Context Effects for PC Party Popularity Model, 1957-1975 Leader = Leader New leader Leadership Convention SE Median SE Median Memory AR(1) 0.2346 0.7797 0.1605 0.7870 Liberal election 0.0199 0.0039 0.0197 0.0005 PC election 0.0260 -0.0192 0.0256 -0.0186 Leader 0.0322 0.1024 0.0453 0.1507 Government Constant 0.0358 0.4929 0.0349 0.4942 Trend - Long 0.0008 -0.0047 0.0008 -0.0047 Trend - Short 0.0368 0.0307 0.0247 0.0296 Phase 0.1116 0.1152 Amplitude 0.0129 0.0263 0.0128 0.0259 Opposition Constant 0.0842 0.2288 0.0825 0.2281 Trend - Short 0.0365 0.0100 0.0291 0.0105 Phase 0.2256 0.1850 Amplitude 0.0074 0.0128 0.0075 0.0130 Note: popularity dependent variable is entered into the model as a proportion rather than a percentage. 210 Table 4-2: Estimated Political Context Effects for Liberal Party Popularity Model, 1957-1975 Leader - Leader= Leader Leader from New Leader Leaders!) ip Quebec Convention SE Median SE Median SE Median Memory AR(1) 0.0490 0.8954 0.0549 0.8747 0.0526 0.8784 Liberal election 0.0240 -0.0247 0.0244 -0.0258 0.0243 -0.0252 PC election 0.0274 -0.0167 0.0281 -0.0158 0.0285 -0.0173 Leader 0.0332 0.0230 0.0289 -0.0117 0.0406 0.0138 Government Constant 0.0702 0.3185 0.0711 0.3433 0.0686 0.3474 Trend - Short 0.0108 0.0121 0.0127 0.0141 0.0118 0.0133 Phase 0.2143 0.1937 0.2095 Amplitude 0.0095 0.0170 0.0096 0.0174 0.0096 0.0172 Opposition Constant 0.0538 0.3016 0.0394 0.3301 0.0373 0.3312 Trend - Short 0.0071 0.0099 0.0072 0.0080 0.0066 0.0075 Phase 0.2319 0.2437 0.2528 Amplitude 0.0124 0.0212 0.0121 0.0209 0.0118 0.0208 Table 4-3: Estimated Political Context Effects for PC Party Popularity Model, 1979-1993 Leader = Leader = Leader -Leader from Quebec New leader Leadership Convention SE Median SE Median SE Median AR(1) 0.0428 0.8986 0.0455 0.8843 0.0432 0.8950 Liberal election 0.0375 0.0630 0.0373 0.0665 0.0376 0.0638 PC election 0.0211 -0.1010 0.0210 -0.1034 0.0211 -0.1018 Leader 0.0250 0.0120 0.0213 0.0393 0.0189 0.0048 Government Constant 0.0353 0.3210 0.0370 0.3209 0.0359 0.3197 Trend - Long 0.0008 -0.0016 0.0007 -0.0019 0.0008 -0.0016 Trend - Short 0.0079 0.0032 0.0082 0.0068 0.0078 0.0048 Phase 0.0951 0.0897 0.0956 Amplitude 0.0149 0.0641 0.0146 0.0636 0.0150 0.0638 Opposition Constant 0.0609 0.2393 0.0609 0.2364 0.0595 0.2315 Trend - Short 0.0124 0.0210 0.0135 0.0241 0.0127 0.0231 Phase 0.2735 0.2716 0.2614 Amplitude 0.0291 0.0748 0.0268 0.0711 0.0291 0.0745 Table 4-4: Estimated Political Context Effects for Liberal Party Popularity Model, 1979-1993 Leader = Leader= Leader = Leader from Quebec New leader Leadership Convention SE Median SE Median SE Median Memory AR(1) 0.0419 0.8815 0.0414 0.8886 0.0436 0.8786 Liberal election 0.0457 -0.1215 0.0452 -0.1199 0.0443 -0.1243 PC election 0.0238 0.0621 0.0235 0.0631 0.0233 0.0637 Leader 0.0238 0.0160 0.0207 -0.0243 0.0213 0.0545 Government Constant 0.0847 0.4347 0.0734 0.4420 0.0705 0.4691 Trend - Long 0.0012 0.0023 0.0012 0.0023 0.0012 0.0022 Trend - Short 0.0133 -0.0198 0.0125 -0.0172 0.0140 -0.0224 Phase 0.1896 0.1824 0.1983 Amplitude 0.0331 0.0984 0.0323 0.1003 0.0312 0.0960 Opposition Constant 0.0493 0.4137 0.0390 0.4293 0.0379 0.4327 Trend - Short 0.0066 -0.0022 0.0060 -0.0031 0.0065 -0.0043 Phase 0.3629 0.3704 0.3549 Amplitude 0.0133 0.0278 0.0135 0.0296 0.0130 0.0293 Table 4-5: Estimated Political Context Effects for PC Party Popularity Model, 1993-2000 Leader - Leader = Leader= Leader from Quebec New leader Leadership Convention SE Median SE Median SE Median Memory AR(1) 0.0924 0.8301 0.0905 0.8287 0.0957 0.8072 Liberal election 0.0219 0.0006 0.0218 0.0007 0.0220 0.0004 Leader 0.0203 0.0040 0.0220 -0.0203 0.0206 0.0132 Constant 0.0344 0.0792 0.0270 0.0866 0.0271 0.0839 Trend - Short 0.0078 0.0078 0.0071 0.0073 0.0078 0.0082 Phase 0.3239 0.3386 0.3110 Amplitude 0.0119 0.0245 0.0119 0.0255 0.0115 0.0252 Table 4-6: Estimated Political Context Effects for Liberal Party Popularity Model, 1993-2000 SE Median Memory AR(1) 0.3024 0.5399 Liberal election 0.0289 -0.0832 Constant 0.0394 0.5551 Trend - Long 0.0004 -0.0002 Trend - Short 0.0258 0.0013 Phase 0.3004 Amplitude 0.0095 0.0242 Table 4-7: Estimated Political Context Effects (incl. FLQ) for PC Party Popularity, 1957-1975 FLQ SE Median Memory AR(1) 0.4402 0.6676 Liberal election 0.0208 -0.0008 PC election 0.0269 -0.0213 Leader Convention 0.0458 0.1522 FLQ 0.0004 -0.0010 Government Constant 0.0382 0.4954 Trend - Long 0.0007 -0.0043 Trend - Short 0.0650 0.0401 Phase 0.0903 Amplitude 0.0131 0.0291 Opposition Constant 0.1476 0.3756 Trend - Short 0.1396 -0.0228 Phase 0.2434 Amplitude 0.0069 0.0119 216 Table 4-8: Estimated Political Context Effects (incl. FLQ) for Liberal Party Popularity, 1957-1975 FLQ SE Median Memory AR(1) 0.0548 0.8811 Liberal election 0.0227 -0.0266 PC election 0.0265 -0.0173 FLQ 0.0004 0.0017 Government Constant 0.0628 0.3346 Trend - Short 0.0114 0.0138 Phase 0.1498 Amplitude 0.0101 0.0199 Opposition Constant 0.0344 0.3291 Trend - Short 0.0061 0.0076 Phase 0.2552 Amplitude 0.0109 0.0193 Table 4-9: PC Economic Popularity Estimates Base SD Median Memory AR(1) 0.2191 -0.4755 GDP x Inflation 0.0070 0.0052 £ 3 g Inflation 0.0141 0.0141 B GDP 0.0136 0.0047 > O Unemployment 0.0148 0.0147 O Trend - short 0.1093 0.0520 s GDP x Inflation 0.0012 -0.0004 o Inflation 0.0041 -0.0030 o GDP 0.0095 -0.0004 B. c Unemployment 0.0072 0.0090 O Trend - short 0.0512 -0.1382 Q-Test P-value Residuals 25.8862 0.9589 for Political Context Effects, 1957-1975 Box-Jenkins SD Median Memory AR(1) 0.2353 -0.4848 InflationxGDP(t-2) 0.0080 -0.0008 B Inflation (t-2) 0.0146 0.0151 E GDP (t-2) 0.0192 0.0042 B >- InflationxGDP(t-3) 0.0068 0.0058 > o Inflation (t-3) 0.0127 0.0189 U GDP (t-3) 0.0195 0.0065 Trend - short 0.0488 0.1092 InflationxGDP(t-2) 0.0014 0.0000 Inflation (t-2) 0.0095 -0.0017 e GDP (t-2) 0.0090 -0.0083 .© •*-» InflationxGDP(t-3) 0.0014 -0.0011 '35 o Inflation (t-3) 0.0093 0.0001 a. a. GDP (t-3) 0.0099 0.0072 O Trend - short 0.0390. -0.0879 Q-Test P-value Residuals 28.6946 0.9084 ( ( Table 4-10: Liberal Economic Popularity Estimates Controlling for Political Context Effects, 1957-1975 Box-Jenkins Refined Box-Jenkins SE Median SE Median Memory AR(1) 0.2299 -0.6398 0.0575 0.8529 Government InflationxGDP(r-6) 0.0013 0.0005 Inflation (t-6) 0.0047 0.0021 j GDP (t-6) 0.0092 -0.0011 Unemployment (t-6) 0.0097 -0.0098 Trend - Short 0.0690 0.2335 0.0072 o.o P9 Opposition InflationxGDP(<-6) 0.0065 0.0010 Inflation (t-6) 0.0118 -0.0167 GDP (t-6) 0.0109 0.0154 Unemployment (t-6) 0.0074 0.0257 0.0017 0.0036 Trend - Short 0.0381 -0.0538 0.0096 -0.0119 Q-Test P-value Q-Test P-value Residuals 26.2501 0.9538 50.0617 0.1323 219 Combined Refined Combined SE Median SE Median AR(1) 0.0654 0.7320 0.0622 0.7314 Government GDP x Inflation (M3) 0.0014 -0.0024 0.0013 -0.0024 inflation (t-13) 0.0022 -0.0046 0.0022 -0.0046 GDP (t-13) 0.0065 0.0112 0.0063 0.0112 intercepts (t-1) 0.0007 -0.0013 0.0007 -0.0013 inflation (t-1) 0.0019 -0.0027 0.0019 -0.0027 GDP (t-1) 0.0034 0.0067 0.0033 0.0066 Trend - Short 0.0164 0.0396 0.0158 0.0397 Opposition GDP x Inflation-(t-1) 0.0005 0.0003 inflation (t-1) 0.0016 -0.0015 GDP (t-1) 0.0045 -0.0030 Trend - Short 0.0194 0.0671 0.0118 0.0523 Q-Test P-value Q-Test P-value Residuals 55.9053 0.0487 31.4470 0.8311 220 Table 4-12: Liberal Economic Popularity Estimates Controlling for Political Context Effects, 1979-1993 Base Refined Base SE Median SE Median Memory AR(1) 0.0527 0.8172 0.0440 0.8268 Government GDP x Inflation (t-l) 0.0005 0.0000 Inflation (t-l) 0.0019 0.0012 GDP (t-l) 0.0047 0.0005 Unemployment (t-l) 0.0022 0.0002 1111 Trend - Short 0.0304 -0.0453 0.0076 -0.0309 Opposition GDP x Inflation (t-l) 0.0008 0.0015 0.0007 0.0014 Inflation (t-l) 0.0017 -0.0010 0.0016 -0.0010 GDP (t-l) 0.0040 -0.0074 0.0037 -0.0069 Unemployment (t-l) 0.0018 0.0003 0.0017 0.0004 Trend - Short 0.0212 -0.0047 0.0204 -0.0050 Q-Test P-value Q-Test P-value Residuals 51.8502 0.0993 39.5046' 0.4924 221 Table 4-13: Liberal Economic Popularity Estimates Controlling for Political Context Effects, 1993-2000 Box-Jenkins SE Median Memory AR(1) 0.2957 -0.0351 InflationxGDP(t-5) 0.0050 0.0077 Inflation (t-5) 0.0181 -0.0327 GDP (t-5) 0.0085 -0.0086 Trend - Short 0.0303 0.0463 Q-Test P-value Residuals 33.7664 0.7071 222 Table 4-14: Liberal Economic Popularity Estimates - Majority Governments Only, 1957-1975 Majority SE Median SE Median Memory AR(1) 0.4765 0.5681 0.0466 0.8874 Government InflationxGDP(f-6) 0.0019 0.0019 Inflation (t-6) 0.0037 -0.0016 GDP (t-6) 0.0090 -0.0049 Unemployment (t-6) 0.0053 0.0004 Trend - Short 0.0564 0.0510 0.0059 0.0137 Opposition InflationxGDPi>6) 0.0231 0.0321 Inflation (t-6) 0.0218 -0.0232 GDP (t-6) 0.0442 -0.0489 Unemployment (t-6) 0.0075 0.0086 0.0009 0.0010 Trend - Short 0.0239 0.0212 0.0055 0.0030 Q-Test P-value Q-Test P-value Residuals 26.3246 0.9528 49.9911 0.1338 Table 4-15: PC Economic Popularity Estimates - Majority Governments Only, 1979-1993 Majority SE Median AR(1) 0.0445 0.8136 Government GDP x Inflation (M3) 0.0013 -0.0006 inflation (t-13) 0.0019 0.0013 GDP (t-13) 0.0060 0.0037 intercepts (t-1) 0.0012 -0.0034 inflation (t-1) 0.0022 -0.0044 GDP (t-1) 0.0058 0.0169 Trend - Short 0.0074 0.0155 Opposition Trend - Short 0.0087 0.0372 Q-Test P-value Residuals 31.5886 0.4951 224 Table 4-16: Liberal Economic Popularity Estimates - Majority Governments Only, 1979-1993 Majority SE Median Memory AR(1) 0.0440 0.8264 Government Trend - Short 0.0076 -0.0310 Opposition GDP x Inflation (t-l) 0.0012 0.0017 Inflation (t-l) 0.0016 -0.0010 GDP (t-l) 0.0057 -0.0086 Unemployment (t-l) 0.0009 -0.0002 Trend - Short 0.0083 0.0012 Q-Test P-value Residuals 39.4440 0.4951 Table 5-1: Media Effects Government Popularity Lag dependent 0.312** (0.106) MEDIA -0.0277** (0.005) Constant 55.22 Log likelihood -208.05 ** significant at 5% level * significant at 10% level t values are presented in parentheses Table 5-2: Effect of Media on Economic Evaluations 1993 1997 1993 1997 Media consumption -0.047** 0.060** -0.044* 0.054** (-2.71) (3.22) (-1.96) (2.89) Personal finances 0.198** 0.205** 0.177** 0.200** (11.81) (12.26) (8.18) (11.88) Talk about politics -0.008 0.000 -0.009 0.003 (-0.48) (0.007) (-0.38) (0.18) Government party 0.114** 0.042** 0.121** 0.054** Identification (6.78) (2-34) (5.58) (2.98) Prime Minister feeling 0.015 0.181** 0.020 0.163** thermometer (0.92) (9.95) (0.93) (8.81) University education 0.028 0.111** (1.27) (6.56) standard deviations are in parentheses 227 FIGURES 2 2 8 Figure 2-1: Progressive Conservative and Liberal Party Popularity, 1957-2000 L i b e r a l a n d P C P o p u l a r i t y 1 9 5 7 - 2 0 0 0 PCelec PCelecs Libelee Libelee Libelee Libelee Libelee PCelec Libelee PCelec 111 1111 | I 111 I 1111 11 I 11111111 I 111 I 1111 I I 11 11 1111111 I I 11 I 1111111 I | 111111 II11111111 11 11 111111 III 11 Ml 1111 11 I 111 111 11111III | 111 11111111 111II 11111 PCelec Libelee Libelee Y i i ' i , , , •, | | | I I I I I 1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 Year Libelee: Liberal Incumbant Election PCelec: Conservative Incumbant Election 229 Figure 2-2: Government Popularity Government Popularity 1957-2000 with Elections Demarcated 0.70 - i n 1 n — i — i p n 1 1 — i — n 1 1 — n 1 r _ i — 1 1 I 1 r 0.10 -0.05 -0.00 -I H 1 r 1 —'—i 'l r 1 1 1 — 1 r-1 1 1 I 1 1 - 1 1 1 — r 1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982-1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 Year 230 Figure 2-3: Government Popularity Autocorrelation Functions Autocorrelation Function 1957-1975 I Partial Autocorrelation Function 1957-1975 231 Figure 2-4: Number of Valid Decided Voters Interviewed Each Month N u m b e r o f D e c i d e d V o t e r s I n t e r v i e w e d E a c h M o n t h 7000 6000 5000 4000 A 3000 H 2000 1000 0 • i- - i 1 i 1 1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 Y e a r (t ick mark ident i f ies beg inn ing of year ) 232 Figure 2-5: Predicted Government Popularity from Deterministic Parts of State-Space Popularity Model P r e d i c t e d G o v e r n m e n t P o p u l a r i t y 95% Confidence 233 Figure 2-6a: Predicted Liberal Popularity from Deterministic Parts of the State-Space Popularity Model P r e d i c t e d L i b e r a l P o p u l a r i t y 0.7 0.6 >, 0.5 -\ -4—* 1 £ 0.4 H i_ . a - 1 0.3 -I "r 0.2 H 0.1 ~i 1 1 1 1 1 1 1 r n 1 1 1 1 1 1 1 1 1 1 r i — 1 — i— 1 — i— 1 — i— 1 — i—|— i—>— i — 1 — i— 1 — i— i— i—•— i—'— i— i— i— i— i— i— i— i— i—>— i—'— i—'— i — ' — i — > — i — ' — i—r -1956195819601962196419661968197019721974197619781980198219841986198819901992199419961998 2000 2002 Year Predicted Measured 95% Confidence 234 0.0 A I I 1 ' I ' I ' I ' I 1 I — 1 — I — 1 — I — 1 — 1 — 1 — I — ' — I — ' — 1 — ' — I — ' — I — 1 — I — ' — I — 1 — I — ' — I — ' — I — 1 — I — 1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 Measured — Predicted 95% Confidence Year 235 Figure 2-7: Residual Movement within a Component of Liberal and PC Popularity, 1957-2000 Residual Movement within a Component , 1957-2000 0.0004 0.0003 co 3 0.0002 a . D- 0.0001 >. •c C D 0.0000 Q _ 2 -0.0001 CD —' -0.0002 -0.0003 1960 1970 1980 1990 2000 0.0004 0.0003 % 0.0002 -| Q. O D_ •c to D_ O 0_ 0.0001 0.0000 -I -0.0001 -0.0002 -| -0.0003 1960 1970 1980 Year 1990 2000 236 Figure 3-1: Inflation (year-over-year percentage change in the consumer price index) 1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 Year 237 Figure 3-2: GDP (year-over-year percentage change in real personal income per capita) 3-1 , , . , . , • , • , • , • , • , • , ' , ' 1 ' , ' 1 ' . ' . 1 ' 1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 Year 238 Figure 3-3: Unemployment (monthly percentage, seasonally adjusted) 2 -| • , • , • , • , ' 1 ' 1 ' 1 ' 1 ' 1 ' 1 1 1 ' 1 ' ' ' 1 ' ' ' ' 1956 1958' 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 Year 239 ) Figure 3-4a: Distributions of Estimated AR(1) Terms, 1957-1975 Liberal AR(1): PC AR(1) only: p1 chains 1:2 sample: 80002 p1 chains 1:2 sample: 80002 3.0 4.0 2.0 A 3.0 A / \ 2.0 1.0 / V ^ 1.0 / V 0.0 J v , , / \ _ 0.0 —i 1 1 1 -2.0 -1.0 0.0 1.0 -2.0 -1.0 0.0 240 Figure 3-4b: Distribution of Estimated Liberal AR(1) Term, 1993-2000 Liberal AR(1): p1 chains 1:2 sample: 80002 1.5 -1.0 0.5 / \ 0.0 —s v_ • i i i i i -2.0 -1.0 0.0 1.0 Figure 3-5: Effect of GDP and Inflation on Party Popularity, 1979-1993 Ef fec ts o f G D P a n d Inf lat ion on Par ty Popu lar i t y , 1 9 7 9 - 1 9 9 3 242 Figure 3-3: AC & PAC of a Component of State-Space Party Popularity Models, 1957-1975 1957-1975 - AC LIB I 1957-1975 - PAC LIB '111 Lag Barttett's formula for MA(q) 95% confidence bands 1957-1975-AC PC Lag Barttott's formula for MA(q) 95% confidence bands ° 3 n ci Lag 95% Confidence bands [se • 1/sqrt(n)] 1957-1975-PAC PC Lag 95% Confidence bands [se = 1/»qrt(n)] 243 Figure 3-4: AC & PAC of a Component of State-Space Party Popularity Models, 1979-1993 1979-1993-AC LIB It 0 10 20 Lag Barltott'a formula for MA(q) 95% confidence bands 1979-1993-AC PC g o Lag Bartletl's formula for MA(q) 95% confidence bands 1979-1993-PAC LIB • J j — J -1 95% Confidence bands [se • 1/sqrt(n)} 1979-1993-PAC PC J , 95% Confidence bands [se * 1/sqrt(n)] 244 Figure 3-5: AC & PAC of a Component of State-Space Party Popularity Models, 1993-2000 1993-2000-AC LIB JiJiJjJJiiJJj|| 0 10 2 0 L a g Bartlett's formula for MA(q) 95% confidence bands 1993-2000-AC PC Bart left's formula for MA(q) 95% confidence bonds 1993-2000-PAC LIB 1 i • r'-i1 1 i 1 1 ' 1 * ] ' ' ll i 95% Confidence bands [se = 1/sqrt(n)] 1993-2000-PAC PC 1 , ' t . 11,1 0 10 95% Confidence bands [se • l/sqrtfn)] 245 Figure 3-6: Distributions of Estimated First Period AR(1) and AR(2) Terms 1.5 1.0 0.5 0.0 p1 chains 1:2 sample: 80002 -2.0 -1.0 0.0 1.0 p2 chains 1:2 sample: 80002 246 Figure 3-7: AC & PAC of Economic Variables, 1957-1975 1957-1975 - AC Unemployment 8-1 a> in 0 5 10 15 Lag Bartlett's formula for MA(q) 95% confidence bands 1957-1975-AC Inflation 8-i « Is-l 4) o Q | 0 5 10 15 Lag Bartlett's formula for NlA(q) 95% confidence bands 1957-1975-AC GDP 8. • B o 0 5 10 15 Lag Bartlett's formula for Mft(q) 95% conf idence bands l l l l l l 1957-1975 - PAC Unemployment 8-1% 8 0 5 10 15 Lag 95% Confidence bands [se = 1/sqrt(n)] 1957-1975 - PAC Inflation in 5? I T * • i 1 1 I T | * ' * i * ' i 7 • • r J 0 5 10 15 Lag 95% Confidence bands [se » 1/3qrt(n)] 1957-1975-PAC GDP 8 . o 0 5 10 15 Lag 95% Confidence bands {se « 1/Bqrt(n}] 247 Figure 3-8: AC & PAC of Economic Variables, 1979-1993 1979-1993 - AC Unemployment 8-4 Is g a £8 l l 0 5 10 15 Lag Bartlett 's formula Tor MA(q) 95% conf idence bands 1979-1993-AC Inflation 8 J TO 8 0 5 10 15 Lag Bartlett 's formula for MA(q) 9 5 % con f idence bands 25 1979-1993 - PAC Unemployment 8J o d 15 b CO o (5 1 n r • i * i1 * * ' • i! * * ) 5 10 15 Lag 5% Conf idence bands [ se = 1/sqrt(n)] 20 1979-1993 - PAC Inflation 8. 8 . I r i • i M l j1 T • * . T 0 5 10 9 5 % Con f idence bands [ se = 1/sqrt{n)j 25 Lag 1979-1993-AC GDP 8-J "as •Bo 1979-1993-PAC GDP 84 '111! I H i m t ^8 CD O i I i i i i i * * * * * * * * * **^ 0 5 10 15 Lag Bartlett 's formula for MA{q) 9 5 % con f idence bands 20 0 5 10 15 Lag 95% Conf idence bands [se = Vsqrt (n) ] 25 248 Figure 3-9: A C & P A C of Economic Variables, 1993-2000 1993-2000 - A C Unemployment E d 8 •:-:-).:. : S8 5 § 0 5 10 15 Lag Bartlett 's formula for MA(q) 9 5 % con f idence bands 1993-2000 - A C Inflation 8. I S to Is < Lag Bartlett 's formula for MA(q) 9 5 % con f idence bands 15 1993-2000-AC G D P 8. 8 <8 IT ' 1 I 0 5 10 15 Lag Bartlett 's formula fo r MA(qJ 9 5 % conf idence bands 1111 25 l l lH i J l 1 1 1 * 20 25 1993-2000 - P A C Unemployment 8. U 8 5 I Is 11 i V l l 0 5 10 9 5 % Conf idence bands [ se » 1/Bqrt(n)] 20 25 1993-2000 - P A C Inflation <fl iB go o 3 is - n •Vf ' i r ' ' . I -0 5 10 15 Lag 9 5 % Conf idence bands [ se - 1/sqrt[n)] 20 25 1993-2000-PAC G D P SA O S 99 TO o 11 11 11 i i * . I 1 i 1 ' • * 1 0 5 10 15 Lag 9 5 % Conf idence bands [ se = 1/sqrt{n)] 25 249 Autocorrelations of gdpxinf -0.50 0.00 0.50 SO so loo (5 i J o o o I > n o o x 3 c 3 Partial autocorrelations of gdpxinf 0.00 0.50 1.00 sO I t o e o © I > n o o -o x e" to o Autocorrelations of gdpxinf 0.00 0.50 Partial autocorrelations of gdpxinf -0.50 0.00 0.50 1.00 Figure 3-11: Distributions of Estimated AR(1) Terms, 1957-1975 PC AR(T) Liberal Model 2 ARH) 3.0 2.0 1.0 0.0 p1 chains 1:2 sample: 80002 : 3.0 2.0 1.0 0.0 p1 chains 1:2 sample: 80002 : A_ i i i i -2.0 -1.0 0.0 - 1.0 -2.0 -1.0 0.0 1.0 25.1 Figure 4-1: Incumbent Government Popularity Leading into Election Incumbent Government Popularity Leading into Election Incumbent Government Popularity Leading into Election 30 20 Days before Election 2000-L ibera I -Chre t ien 1997-L ibera l -Chre t ien 1 9 9 3 - P C C a m p b e l l Incumbent Government Popularity Leading into Election 40 30 20 Days before Election 4 0 30 20 Days before Election • 1 9 8 0 - P C Minor i ty -C lark o 1984-L ibera l -Turner • 1 9 8 8 - P C - M u l r o n e y Change in Incumbent Government's Popularity 1979-L ibera l -Trudeai i 1974-L ibera l Minor i ty -Trudeau 1972-L ibera l -Trudeau 1968-Liberal Minor i ty -Trudeau 1965-Liberal Minor i ty -Pearson Incumbent government Year Net Change Liberal 1965 -4 percent Liberal 1968 -3 percent Liberal 1972 -5 percent Liberal 1974 3 percent Liberal 1979 -5.5 percent PC 1980 1 percent Liberal 1984 -20 percent PC 1988 -3 percent PC 1993 -14 percent Liberal 1997 -14 percent Liberal 2000 -14 percent 252 Figure 4-2: Progressive Conservatives AR(1) Term, 1957-1975 Base Box-Jenkins pi chains 12 sample: 80002 3.D 2.0 : A ' 1.0 • I V 0.0 . J ^-1—J , 1 1 1 •1.0 -0.5 D.D 0.5 253 Figure 4-3: Liberal AR(1) Term, 1957-1975 Box-Jenkins p1 chains 1:2 sample: 80002 Revised Box-Jenkins 10.0 7.5 5.0 2.5 0.0 p1 chains 1:2 sample: 80002 -i 1 1 r— 0.2 0.4 0.6 0.8 1.0 l r 254 Figure 4-4: Progressive Conservative AR(1) Term and a Component Residuals, 1979-1993 Combined Model, AR(1) 8.0 6.0 4.D 2.0 0.D p1 chains 1 2 sample: S0002 A 0.2 0.4 0.6 o.s Refined Combined Model, AR(1) Autocorrelation function for Combined Model Residuals Autocorrelation function for Refined Combined Model Residuals E "I o d * 4 i *} 1 1 • 1 I*' ft « 1 • 0 10 20 Lag Bartlett's formula for MA(q) 95% confidence bands 0 10 2 0 Lag Bartlett's formula for MA(q) 95% confidence bands 255 Figure 4-5: Liberal AR(1) Term and a Component Base Model, AR(1) Residuals, 1979-1993 Autocorrelation function for Base Model Residuals 11 t 5 3 o 0 10 20 Lag Bartlett's formula for MA(q) 95% confidence bands Refined Base, AR(1) Autocorrelation function for Refined Base Model Residuals a. •8 8 So is 0 10 20 Lag Bartlett's formula for MA(q) 95% confidence bands 2 5 6 Figure 4-6: Liberal AR(1) Term, 1993-2000 Box-Jenkins, AR(1) p1 chains 1:2 sample: 80002 1.5 -1.0 0.5 0.0 —y v — • i i i i -2.0 -1.0 0.0 V 257 Figures 4-9a & 4-9b: Decay of a Hypothetical 1 Percent Shift in Popularity Produced by Economic Conditions in Month 1 Figure 4-9a E ^ 0.8 J >• 0.6 20 30 40 Months fol lowing initial shift AR(1) = 0.83 Figure 4-9b 10 20 30 40 Months fol lowing initial shift AR(1) =-0.035 258 Figures 4-10a&4-10b: Contribution of Constant Economic Conditions Producing an Initial 1 Percent Shift in Popularity Figure 4-1 Oa 5 2 Months following initial shift AR(1) = 0.83 Figure 4-10b 20 30 40 Months following initial shift AR(1) = -0.035 259 Figure 4-11: Monthly Shift in Liberal Party Popularity due to Unemployment, 1957-1975 Effect of Unemployment on Liberal Party Popularity (while in opposit ion) 0.028 -r c sel 0.026 -ro 00 E o 0.024 ->> ro Q_ 0.022 -O 0. >. tr 03 D_ 0.020 -TO <D .O _J 0.018 -CO 0.016 -1957 1958 1959 1960 1961 Year 1962 1963 1964 Figure 4-12: Monthly Shift in Party Popularity due to GDP and Inflation, 1979-1993 | Effect of GDP and Inflation on Party Popularity Figure 4-13: Monthly Shift in Liberal Party Popularity due to GDP and Inflation, 1993-2000 Q. O 0_ >> •e 03 Q. Effect of GDP and Inflation on Liberal Party Popularity -0.02 -0.04 H to -0.06 JO (D -0.08 Baseline -0.02 -0.04 -0.06 -0.08 1993 1994 1995 1996 1997 1998 1999 2000 Year (tick marks beginning of year) 2001 2002 Figure 4-14: Cumulative Contribution of GDP and Inflation to Government Popularity, 1984-1993 \ Effect of G D P and Inflation on Government Popularity 263 Figure 5-1: GDP and Government Popularity Cycles G D P versus G o v e r n m e n t Popular i ty Cycle (Cyc le Ampl i tudes Standard ised to 1979-1993 Per iod) -T—•—i—•—r——i "i 1 — — i — ' 1 — " — i - 1 - ' — i i f r 1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 Year Government Cycle GDP Cycle 264 Figure 5-2: Inflation and Government Popularity Cycles Inflation versus Government Popularity Cycle (Cycle Amplitudes Standardised to 1979-1993 Period) -1.0 A | 1 r 1— 1 1 ' r 1 — " — i 1 —h ' r^-1 1 1 1 — " r — 1 1 1 i i i i i r • • • • • 1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 Year — Government Cycle Inflation Cycle 265 Figure 5-3: Unemployment and Government Popularity Cycles U n e m p l o y m e n t v e r s u s G o v e r n m e n t Popu la r i t y Cyc le (Cyc le A m p l i t u d e s S t a n d a r d i s e d to 1 9 7 9 - 1 9 9 3 Per iod) 1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 Year Government Cycle Unemployment Cycle 266 Figure 5-4: Size of Government, 1957-2000 Government Size 0.9 - i — i — i — i — | — i — i — i — i — | — i — i — i — i — | — i — i — i — i — | — i — i — i — i — i — i — i — i — i — i — < — 1 1 1 I 1 1 1 1 r (A 0-<H -*-» 03 0) CO -£ CTE" 0.7 c E a) a) =c ^ 0.6 CL -C o 2 0.5 •+= <u o X CL O ^ 0.4 0.3 1960 1965 1970 1975 1980 Y e a r i i i i 1985 1990 1995 2000 267 Figure 5-5: Regionalisation and Fractionalisation, 1953-2002 Opposit ion Fractionalisation 0.7 -c o isat 0.6 • ional "8 ro 0.S -LL O idex 0.4 • i — i — i — i i i i i i 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 Year Opposition Regionalisation ra 0.4 c o ' c n <D or •S 0.3 X <D 1L 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 Year Index of Fractionalisation (opposition): F = \ -1-1 , where *. is the number of seats received by the ;th opposition party; x is the total number of parliamentary seats occupied by opposition parties; and N is the total number of opposition parties with seats in Parliament. This produces a measure of the degree of fractionalisation of the opposition in Parliament. Index of Regionalisation (opposition): n2=-£U—hi—f—, where p. is the proportion of seats received by the rth opposition party in j,i N the jtb province; pj is the proportion of opposition seats received by the ith party nationally; w. is a weighting factor based on the proportion of opposition held parliamentary ridings that reside in theyth province; and N is the total number of opposition parties with seats in Parliament. 268 Figure 5-6: The Consequences of Defection from Government to Single Alternative, 1962 and 1980 Hypothetical Defection from PC Government to Liberal in 1962 Election • Liberal - CCF • SC % Liberal Defection Hypothetical Defection from Liberal Governemnt to PCs in 1980 Election - • — Liberal O P C —T— N D P 10 15 % PC Defection Hypothetical Defection from PC Government to Social Credit in 1962 Election • O • Liberal — • — C C F - V SC 5 10 15 20 % Social Credit Defection Hypothetical Defection from Liberal Governemnt to NDP in 1980 Election % NDP Defection - NDP defect ion vs NDP-Libera l PC • N D P defect ion vs N D P - N D P (figure continued on next page) 269 Figure 5-6 (cont): The Consequences of Defection from Government to Single Alternative, 1993 and 2000 _ _ Hypothetical Defection from Liberal Government to NDP in 2000 Election Hypothetical Defection from Liberal Government to PCs in 2000 Election 270 Figure 5-7: The Consequences of Defection from Government to All Alternatives, 1980 and 1993 Hypothetical Defection from 1980 Liberal Government Vote 160 140 120 c o g 100 I 80 OO ° 60 0) _Q E ACi , — - T Liberal • O Social Credit —T— PC — v - NDP 10 15 Defection % Hypothetical Defection from 1993 Liberal Government Vote 2 0 0 180 160 c 140 o | 1 2 " 0 ) too (O O 8 0 <D c 60 • Liberal • Re fo rm N D P Bloc 10 15 Defection % 271 Figure 5-8: Federal Election News Stories and Government Popularity, 1995-2002 Canadian Federal Election News and Government Popularity 500 , ! , 1 1 1 • 1 1 1 1 i 1 , 1 1 ' ' ' 1 1 I 1997 1996 1999 2000 Year 272 Appendix A A D D I T I O N A L T A B L E S A - l : Decycled/Detrended Autocorrelation Functions 1957-1975 o 0 o S o .2 o to ° <D i_ o o o o o 0! s. lim 'I: •' 20 Lag Bartlett's formula for MA(q) 95% confidence bands 40 — r -60 1979-1993 o o o 3 <D O C | to M — o in 5 o o o « b i» o o o o o llllttt*.... Milium], 20 Lag 40 60 Bartlett's formula for MA(q) 95% confidence bands A-2: Distributions of Baysian Estimated AR(1) Terms Liberal AR(1), 1957-1975: PC AR(1), 1957-1975: p1 chains 1:2 sample: 80002 p1 chains 1:2 sample: 80002 10.0 6.0 7.5 5.0 2.5 4.0 2.0 0.0 : A 0.0 -—i 1- i i i 0.4 0.6 0.8 1.0 1 1 1 1 I -1.0 -0.5 0.0 0.5 Liberal AR(1), 1979-1993: PC AR(1), 1979-1993: p1 chains 1:2 sample: 80002 p1 chains 1:2 sample: 80002 10.0 15.0 7.5 5.0 2.5 • 10.0 5.0 0.0 0.0 -• - I 1 1 1— 0.6 0.8 1.0 i i i i < • 0.6 0.7 0.8 0.9 1.0 Liberal AR(1), 1993-2000: PC AR(1), 1993-2000: p1 chains 1:2 sample: 80002 p1 chains 1:2 sample: 80002 2.0 6.0 -1.5 1.0 0.5 : J\ 4.0 2.0 0.0 0.0 * —\ 1 i r — -1.0 0.0 1.0 0.0 0.25 0.5 0.75 1.0 r 275 T3 O Autocorrelations of l ibart -0.10 0.00 0.10 o O s o vO oo - 0 so oo oo V 3 B CD o 3 — s pa c o J> 3 o •—* * o IT. g 1 JO H a* O * o II 1! o b> U l * oo U l £" 3 — 1— -0.20 Partial autocorrelations of libarl -0.10 0.00 0.10 1 3 > 0.20 O 2"1 •3 o cr a-Autocorrelat ions of pear l -0.10 0.00 0.10 0.20 ~4 to" -0.20 Partial autocorrelations of pearl -0.10 0.00 0.10 > 73 > 0.20 O n ~0 10 3 2 9 3 V 1 § • 1 e g so g t o • £ b\ I ' o o -0.20 Autocorrelations of l ibar l -0.10 0.00 0.10 0.20 -0.20 Partial autocorrelations of libarl -0.10 0.00 0.10 > 0.20 O -o o 3 5 PS 3 -0.20 Autocorrelations of pearl -0.10 0.00 0.10 CD •d n > -0.20 Partial autocorrelations of pearl -0.10 0.00 0.10 > 0.20 P — L *C O > 2 JO ft w to - J 1 3 1 3 3 2 f 1 v I s f o Autocorrelations of l ibar l -0.20 -0.00 0.20 0.40 13 O 3. -0.40 Autocorrelat ions of pear l -0.20 -0.00 0.20 <3\ OS -0.30 Partial autocorrelations of l ibarl -0.20 -0.10 0.00 0.10 0.20 o 3 H". 8 to g r ET -0.30 Partial autocorrelations of pearl -0.20 -0.10 0.00 0.10 0.20 O re •• B 1 3 o > r. I' ta 33 s re sO w 1 3 > — o > 2 53 ft to A-4a: PC Party Popularity and Electoral Results P C Popu la r i t y a n d Electora l Resu l t s A-4b: Liberal Party Popularity and Electoral Results L i b e r a l P o p u l a r i t y a n d E l e c t o r a l R e s u l t s 0.1 H— i— i— i— i— i— i— i— i— i— i— i— i— '— i—'— i— '— i—'—i—'—i— i— .— i— i— i— i— i—'—i—'—i—'— i— 1 i 1 i 1 i 1 r 1956195819601962196419661968197019721974197619781980198219841986198819901992199419961998 2000 2002 Liberal Popular i ty • L iberal V o t e in L ibera l I n c u m b e n t Elect ion -A— L iberal V o t e in PC I n c u m b e n t Elect ion Year H c o J>. to o d o o * . to o d to Xcorr Unemployment > in © to oo s o O s I U J N O os os o d o d oo oo 0 0 o U>1 MS o d Xcorr Inflation n o " I 5" 5' s a n o to U* o d © d 3 8 to Xcorr GDP N O Wl 1^ i NO ON o N O o o oo| U J o d o O s Ufl OO o to O s to o d o d o Xcorr GDPx Inflation to o © UJ UJ o oo O S o to N O © to 0 0 2SE O 1 o O S I UJ N O © d oo o d UJ oo u-i o UJ ^1 © © to UJ O N © d 2 Xcorr Unemployment re © © N O oo © © oo O s S O © O S 0 0 to © d © O s N O to © to © © t o ION NO Xcorr Inflation © O s UJ © © un oo © d to © © © © 3 Xcorr GDP NO ON U J I NO - 4 <Jl W n o a I ° 3 © © N O © N O © © © © © © d Xcorr GDPx Inflation 2SE 2 © N O O S N O © O s un O s © © I ur, UJ N O to © © © © © © © d to to O s © N O © UJ O s Xcorr Unemployment © © un 0 0 © © © © U/1 © Ufl © © Xcorr Inflation © © 2 © to Uh © © OO © © © d 0 0 to N O © © © d Xcorr GDP N O 00 o I NO 00 © O s 2 © to © UJ to oo 0 0 © to N O © to to O s © © © UJ © Uh © © 23 © to I oo © Xcorr GDPx Inflation 2SE Table A-5: Cross-Correlation Functions for Liberal Popularity a Residuals and Economic Variables (cont) 1984-1993 1993-2000 Lag Xcorr Unemployment Xcorr Inflation Xcorr GDP Xcorr GDPx Inflation 2SE Xcorr Unemployment Xcorr Inflation Xcorr GDP Xcorr GDPx Inflation 2SE 0 -0.0295 -0.0344 0.0172 0.031 0.1916 0.0024 -0.1583 0.0859 0.2195 1 -0.0321 0.2038 0.0849 0.0732 0.1925 -0.2327 -0.038, 0.0202 0.2209 • 2 0.0123 -0.0425 -0.0575 -0.0499 0.1933 0.1954 0.0374 0.0096 0.2222 3 0.0617 0.0935 0.1604 0.1339 0.1943 0.2135 0.0853 -0.03 0.2236 4 0.0909 0.0197 -0.0032 -0.0362 0.1952 -0.0914 0.2389 0.0425 0.2250 5 -0.0138 -0.1363 0.1047 0.0494 0.1961 -0.1711 0.0077 -0.1142 0.2265 6 -0.0154 0.0252 -0.0356 -0.0027 0.1971 0.0865 -0.0697 -0.0152 0.2279 7 -0.0204 -0.0746 -0.0927 -0.0798 0.1980 0.1248 0.0739 -0.083 0.2294 8 0.1233 -0.205 0.1138 0.116 0.1990 0.082 0.0989 -0.1434 0.2309 9 0.0012 0.0881 -0.0116 -0.0402 0.2000 0.0002 0.1389 0.196 0.2325 10 -0.0129 -0.0562 -0.0566 0.0116 0.2010 -0.0982 0.0029 -0.114 0.2341 11 -0.0744 -0.0384 -0.0227 -0.0151 0.2020 0.0069 -0.0455 -0.0154 0.2357 12 0.0862 -0.0648 0.043 0.0203 0.2031 0.1179 0.1153 0.2199 0.2374 13 -0.0298 -0.1749 0.092 0.0872 0.2041 0.0623 0.0009 -0.0552 0.2390 14 0.0175 0.0647 -0.0987 -0.0984 0.2052 0.0119 -0.1076 0.0215 0.2408 15 -0.0111 -0.1165 -0.0127 -0.0101 0.2063 -0.2207 -0.1349 0.0763 0.2425 16 -0.1102 -0.1325 0.0614 0.0565 0.2074 -0.1104 -0.0978 -0.0589 0.2443 17 0.0521 0.1092 -0.1097 -0.1468 0.2085 0.049 -0.1691 -0.0009 0.2462 18 -0.0358 0.0971 -0.0519 -0.0776 0.2097 0.1486 0.0971 0.0871 0.2481 19 0.0787 0.0422 0.0843 0.1087 0.2108 -0.1006 0.0073 0.0205 0.2500 20 0.0838 0.1071 -0.0424 -0.1279 0.2120 -0.0752 -0.0807 -0.0308 0.2520 282 Table A-6: Cross-Correlation Functions for Progressive Conservative Popularity a Residuals and Economic Variables 1957-1963 1963-1975 1980-1984 Lag Xcorr Unemployment Xcorr Inflation Xcorr GDP Xcorr GDPx Inflation 2SE Xcorr Unemployment Xcorr Inflation Xcorr GDP Xcorr GDPx Inflation 2SE Xcorr Unemployment Xcorr Inflation Xcorr GDP Xcorr GDPx Inflation 2SE 0 -0.1578 0.1019 0.3472 0.0618 0.2604 0.0267 -0.125 -0.1308 0.1147 0.1622 0.099 -0.0909 -0.1864 -0.1255 0.2697 1 0.124 -0.0553 0.1107 -0.1018 0.2626 -0.0209 0.0123 0.0749 -0.0745 0.1628 -0.0733 -0.0371 -0.0466 -0.0531 0.2722 2 -0.1672 0.1607 -0.0638 0.1018 0.2649 0.0449 0.0449 -0.0298 0.1322 0.1633 -0.0527 0.0553 0.0134 0.0113 0.2747 3 -0.0625 -0.2431 0.0983 -0.28 0.2673 -0.1253 0.075 -0.0077 -0.0172 0.1638 0.0768 0.2272 0.0642 0.1443 0.2774 4 -0.0246 -0.0808 -0.013 0.1863 0.2697 -0.0004 0.0516 0.041 0.0046 0.1644 -0.114 0.0864 0.0514 0.0785 0.2801 5 -0.1544 0.0184 -0.1417 -0.0859 0.2722 -0.0682 0.1448 0.0302 0.0347 0.1650 0.0131 -0.0062 -0.0761 -0.0769 0.2828 6 0.16 0.0568 0.0338 -0.0074 0.2747 -0.0814 -0.0562 -0.0812 0.0154 0.1655 -0.0789 0.0344 -0.1049 -0.1145 0.2857 7 0.1379 -0.0471 0.0898 -0.1332 0.2774 0.0871 -0.0673 -0.0834 -0.107 0.1661 -0.2277 -0.0366 0.0677 0.0625 0.2887 8 -0.0439 0.088 0.0933 -0.0193 0.2801 -0.1074 -0.0272 -0.0137 -0.0555 0.1667 -0.1543 0.1618 0.0254 0.0349 0.2917 9 0.0171 -0.0824 0.1204 -0.1843 0.2828 0.0812 0.0345 -0.1274 0.0131 0.1672 -0.0392 -0.1401 -0.0104 -0.0449 0.2949 10 -0.0037 -0.0638 -0.0385 0.0723 0.2857 -0.0081 -0.0042 -0.0308 0.0477 0.1678 -0.17 -0.0352 0.1425 0.0395 0.2981 11 -0.0456 0.0972 0.0414 -0.0262 0.2887 0.1544 0.0174 -0.014 -0.0699 0.1684 0.0867 -0.2562 0.0164 0.004 0.3015 12 0.1204 -0.2074 0.1524 -0.1188 0.2917 -0.0145 0.0415 0.0832 0.0122 0.1690 -0.0914 0.0609 0.1988 0.1749 0.3050 13 0.0587 0.098 0.1089 0.0658 0.2949 0.0485 -0.1404 0.0034 -0.0404 0.1696 -0.1354 -0.0864 0.2261 0.202 0.3086 14 0.1595 -0.0378 0.0743 -0.0343 0.2981 -0.033 0.0554 0.0007 -0.0795 0.1703 -0.0553 -0.0564 0.0221 -0.026 0.3123 15 0.1085 -0.0134 0.062 0.0562 0.3015 -0.0086 -0.1722 0.1382 0.1068 0.1709 -0.0491 -0.2195 0.1566 0.1627 0.3162 16 0.0297 -0.0738 -0.0409 -0.0742 0.3050 -0.0536 -0.0474 -0.0324 -0.0434 0.1715 -0.2419 -0.1601 0.0079 -0.0692 0.3203 17 0.0042 -0.1596 0.0471 0.0831 0.3086 0.007 -0.012 0.0401 0.033 0.1721 -0.1036 -0.1135 0.0397 0.0661 0.3244 18 0.1152 -0.0318 0.0276 -0.0811 0.3123 -0.0932 -0.035 0.1055 0.0909 0.1728 0.0104 -0.0432 0:0377 0.0294 0.3288 19 -0.0101 0.1101 0.0016 0.0473 0.3162 0.0461 0.0099 0.0156 0.0104 0.1734 -0.0334 -0.1003 0.075 0.0508 0.3333 20 0.042 -0.0681 -0.0093 -0.0901 0.3203 0.0141 0.0048 0.1423 -0.0453 0.1741 0.1114 -0.1779 -0.1259 -0.1619 0.3381 283 Table A-6: Cross-Correlation Functions for Progressive Conservative Popularity a Residuals and Economic Variables (cont) 1984-1993 Lag Xcorr Unemployment Xcorr Inflation Xcorr GDP Xcorr GDPx Inflation 2SE 0 -0.0598 -0.0393 0.0658 0.0188 0.1916 1 -0.1822 -0.0516 -0.1199 -0.0804 0.1925 2 -0.0687 0.0373 0.0572 0.0505 0.1933 3 0.1076 -0.0922 -0.0479 0.013 0.1943 4 0.0916 0.0114 -0.1757 -0.2027 0.1952 5 -0.1329 -0.0061 0.0459 0.055 0.1961 6 -0.1718 -0.0009 0.0438 0.061 0.1971 7 0.0399 -0.0653 -0.0441 0.0382 0.1980 8 -0.0717 0.1015 0.0591 0.0331 0.1990 9 -0.0898 0.0624 0.1037 0.0939 0.2000 10 0.0659 0.1325 -0.0021 -0.0078 0.2010 11 0.0216 0.0977 -0.1133 -0.0633 0.2020 12 0.0514 -0.0608 -0.0466 -0.0755 0.2031 13 0.0285 -0.0266 0.0458 0.0414 0.2041 14 -0.09 0.1141 -0.0366 -0.047 0.2052 15 -0.0739 0.1771 -0.0769 -0.0665 0.2063 16 0.0588 0.0066 -0.0057 0.0148 0.2074 17 0.0168 0.0317 -0.0071 0.0339 0.2085 18 -0.0657 -0.0207 -0.0373 -0.02 0.2097 19 0.0226 0.0678 -0.0063 -0.0006 0.2108 20 -0.0489 -0.0553 0.0455 0.0773 0.2120 284 Appendix B D A T A S O U R C E S Inflation 1914-2003 C A N S I M II SERIES V735319 T A B L E N U M B E R : 3260001 T A B L E TITLE: C O N S U M E R PRICE INDEX (CPI), 1996 B A S K E T CONTENT Data Sources: IMDB (Integrated Meta Data Base) Numbers: #2301 - Consumer Price Index SERIES TITLE: INDEX; C A N A D A ; ALL-ITEMS CANSIM I Series Number: PI00000 SERIES FREQUENCY: Monthly SCALING FACTOR: units DECIMALS: 1 GDP 1961-2000 CANSIM II SERIES V498943 T A B L E N U M B E R : 3800002 T A B L E TITLE: GROSS DOMESTIC PRODUCT (GDP), EXPENDITURE-BASED Data Sources: IMDB (Integrated Meta Data Base) Numbers: # 1901 - National Income and Expenditure Accounts SERIES TITLE: C A N A D A ; 1992 CONSTANT PRICES; UNADJUSTED; GROSS DOMESTIC PRODUCT (GDP) A T M A R K E T PRICES CANSIM I Series Number: D15721 SERIES FREQUENCY: Quarterly SCALING FACTOR: millions DECIMALS: 0 Also run with CANSIM II SERIES VI992259 T A B L E N U M B E R : 3800002 T A B L E TITLE: GROSS DOMESTIC PRODUCT (GDP), EXPENDITURE-BASED Data Sources: IMDB (Integrated Meta Data Base) Numbers: # 1901 - National Income and Expenditure Accounts SERIES TITLE: C A N A D A ; 1997 CONSTANT PRICES; S E A S O N A L L Y ADJUSTED; GROSS DOMESTIC PRODUCT (GDP) A T M A R K E T PRICES CANSIM I Series Number: D100525 SERIES FREQUENCY: Quarterly SCALING FACTOR: millions GDP 1926-1961 Label: D14606 Title: SELECTED PER PERSON SERIES IN C & K $ / G.D.P. A T M A R K E T PRICES IN CONSTANT (1986) D O L L A R S Subtitle: SELECTED PER PERSON INCOME A N D PRODUCT SERIES A T CURRENT PRICES A N D A T 1986 PRICES, A N N U A L L Y , F R O M 1926. Factor: U N S C A L E D Unit: K D O L L A R S Source: SDDS 2501 STC (13-531 & 13-201) Update : 25 June, 1996 Period : 1926 - 1995 Frequency : annual 286 Unemployment 1976-2000 CANSIM II SERIES VI59752 This series has been deleted by Statistics Canada. Use only for comparisons with your earlier retrievals. T A B L E N U M B E R : 2790001 T A B L E TITLE: L A B O U R FORCE S U R V E Y ESTIMATES (LFS), B Y A G E GROUP A N D SEX, C A N A D A Data Sources: IMDB (Integrated Meta Data Base) Numbers: # No sources available SERIES TITLE: C A N A D A ; U N E M P L O Y M E N T RATE; B O T H SEXES; 15 Y E A R S A N D OVER; S E A S O N A L L Y ADJUSTED CANSIM I Series Number: D980745 SERIES FREQUENCY: Monthly SCALING FACTOR: units DECIMALS: 1 Unemployment 1946-1975 Title: Historical statistics of Canada / F. H. Leacy, editor. — Author: Leacy, F. H. Statistics Canada-Social Science Federation of Canada. Published: Ottawa : Statistics Canada in joint sponsorship with Social Science Federation of Canada, cl983. Series: D491 Series Title: Unemployment Rates, Canada total, annual averages SERIES FREQUENCY: annually Edition: 2nd ed. Vote Share/Popularity 1956-2000 Title: Gallup poll [electronic resource!. Uniform title: Gallup poll (Canadian Institute of Public Opinion) Other titles: Canadian Gallup poll CIPO Gallup poll Author: Canadian Institute of Public Opinion. Canadian Gallup Poll Limited. Published: Toronto: Canadian Institute of Public Opinion, Party popularity and valid Ns were calculated from a dataset compiling individual level data from all Canadian Institute of Public Opinion surveys including the question "If a federal election were held today, which party would you favour?," from 1945 to 2000. This dataset builds upon the file compiled by Jean A. Laponce, David Fenn, Jeffrey Rustand, John R. Wright and Helen Ray (Laponce et al 1982). Their compilation was based on CIPO data for 1945-1983. Cameron Ortis and Mark Pickup, working under the direction of Richard Johnston, compiled standardized coding for CIPO polls between 1983 and 2000 to merge them with each other and finally with the Laponce data; creating one large dataset for the period 1945-2000. One of the variables gleaned from the CIPO polls was vote-intention. In 1992, Gallup began to consistently ask a follow-up question of those that could not name a party for the vote-intention question. The question asked which party the respondent may be leaning towards. To retain consistency 287 throughout the popularity series, this "leaning" question was not used in the calculation of government or party popularity at any time. The general strategy for the compilation was to follow a modular approach for each year. A file in each year was individually imported and transformed. Following that, files within each year were then merged together, concluding with a final merge of the 1983-2000 data with the Laponce file. The Laponce dataset was considered dominant in the final coding procedure. Each file between September 1983 and December 2000 underwent the following processing procedures: a) the original data file was checked to determine the location and coding of each variable. b) SPSS templates were then created in order to import the text files and capture the relevant variables. c) within each year common data files were then merged. d) using the Laponce coding each of the variables were then transformed and recoded. e) files were double checked from available codebooks and frequencies were run at each stage of the project to confirm the samples sizes in each file. It should be noted that any project that merges and compiles survey data that span long time periods inevitably faces several methodological issues. First, as noted in Laponce (1982), the CIPO data do not follow a consistent coding scheme across time. This was true for the original Laponce merge and is also a problem present in the 1983-2000 data; although the variation in coding schemes gets smaller through the 1990s. Second, a small number of CIPO files were not added to the final data set because the associated code book entries were incomplete, inaccurately documented, or the surveys were carried out with only a small number of socio-economic variables and no voting data (eg. sub-national surveys). Many of the CIPO data files are based on the now antiquated multi-punched card files. In some instances many of the code books only document a modest amount of the overall data contained in the CIPO files. In part this is due to privacy agreements with the data provider where certain questions asked were guaranteed confidential. In other cases the "extra" data contained in each file remains unaccounted for in the code books. The resulting data file - containing CIPO data spanning May 1945 to December 2000 - contains 514770 logical records. Vote-intention results published in The Gallup Poll were used for May 1995, November and December 1995, April 1996, April-August 1978, and June 1968 due to errors within the electronic files for these months. F L Q Media Count Data The FLQ media count variable was produced using the dataset "Globe and Mail : Canada's Heritage From 1844." This is an electronic full-page newspaper archive of all the editions and versions of The Globe from June 1844 to The Globe and Mail until December 2000. In recent years, it is the Metro edition of which the National Edition is a subset. The variable is a monthly count of the number of pages of The Globe and Mail on which the term F L Q or "Front de Liberation du Quebec" appear. 288 Canadian Newsdisk Canadian Newsdisk is a companion to the Canadian Business and Current Affairs (CBCA), except that it provides full- text articles from Canadian newspapers rather than from magazines from journals. C B C A is a database of full-text articles from Canadian magazines and journals. Canadian election study, 2000 Principal investigator: Andre Blais, Elisabeth Gidengil, Richard Nadeau, and Neil Nevitte Distributor: Universite de Montre. Faculte des artes et des sciences. Departement de science politique The 1997 Canadian election survey Principal investigator: Andre Blais, Elisabeth Gidengil, Richard Nadeau, and Neil Nevitte Distributor: York University. Institute for Social Research; ICPSR 2593; Universite de Montreal. Faculte des artes et des sciences. Departement de science politique 1993 Canadian election study incorporating the 1992 referendum survey on the Charlottetown Accord Principal investigator: Richard Johnston, Andre Blais, Henry E. Brady, Elisabeth Gidengil, and Neil Nevitte Distributor: York University. Institute for Social Research; ICPSR 6571 The 1974-1979-1980 Canadian national elections and Quebec referendum panel study Principal investigators: Harold Clarke, Jane Jenson, Lawrence Leduc, Jon Pammett Distributor: ICPSR 8079 Canadian national election study, 1965 Principal investigators: Philip Converse, John Meisel, Maurice Pinard, Peter Regenstrief, Mildred Schwartz Distributor: ICPSR 7225 289 Appendix C I N T E R P R E T I N G S I G N I F I C A N C E U S I N G T H E B A Y E S I A N E S T I M A T E D D I S T R I B U T I O N S O F T H E P A R A M E T E R S For each of the state-space model parameters estimated, a non-informative prior distribution is used. This means the parameters being estimated by Bayesian methods are equivalent to those that would be estimated using the principle of maximum likelihood. In each case, the parameters are estimated such that they that maximize the probability (likelihood) of observing the data that was sampled. However, a maximum likelihood estimation is a point estimation of the parameter (producing a single value), while Bayesian methods estimate the distribution of the parameter. If the distribution of the parameter is roughly normal, the maximum likelihood estimation is essentially the mode (and ideally the mean and median) of the distribution determined through Bayesian methods. In this case, interpreting the statistical significance of the parameter estimated by maximum likelihood methods and the mode of the parameter distribution estimated by Bayesian methods will produce similar conclusions. In the case of maximum likelihood, the standard errors and confidence intervals are calculated based on large sample properties and significance is determined accordingly. In the case of Bayesian estimation, the appropriate percentiles are calculated from the distribution - typically 2.5 and 97.5 percentiles - and significance is determined by whether the null hypothesis for the mode (e.g., equal to zero) falls between these percentiles. If the distribution is not normal - skewed, bimodal, etc. - interpretation is not as straightforward in the Bayesian case. With many such distributions, measures of central tendency such as mean, median and mode are poor descriptions of the parameter distribution. If the distribution is bimodal for example, the median may be very close to zero while each of the two modes are clearly different from zero. Interpreting the parameter as statistically insignificant because the median falls between the 2.5 and 97.5 percentiles would be erroneous. Fortunately, the distributions for almost all the parameters of the state-space model were estimated to be roughly normal. As long as this was the case, significance was determined by whether zero fell within the 2.5 and 97.5 percentiles. In a few instances, zero fell within 0.001 of 291 a percentage point from either the 2.5 or 97.5 percentile. These values were interpreted as significant. The only parameter distributions determined to not be roughly normal were for some but not all of the AR(1) terms. These distributions are identified in the text of the dissertation. For these parameters, decisions regarding their significance are somewhat subjective. Therefore, the justification for the decision either way is presented along with the estimated distribution. 292 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0099803/manifest

Comment

Related Items