UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Divided government in Canada Lockhart, Julia Kate 2004

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

Item Metadata

Download

Media
831-ubc_2004-0537.pdf [ 3.24MB ]
Metadata
JSON: 831-1.0091652.json
JSON-LD: 831-1.0091652-ld.json
RDF/XML (Pretty): 831-1.0091652-rdf.xml
RDF/JSON: 831-1.0091652-rdf.json
Turtle: 831-1.0091652-turtle.txt
N-Triples: 831-1.0091652-rdf-ntriples.txt
Original Record: 831-1.0091652-source.json
Full Text
831-1.0091652-fulltext.txt
Citation
831-1.0091652.ris

Full Text

DIVIDED GOVERNMENT IN CANADA J U L I A K A T E L O C K H A R T B.A. Hons., Un ivers i ty o f Br i t i sh Co lumbia , 2002 A T H E S I S S U B M I T T E D I N P A R T I A L F U L F I L M E N T O F T H E R E Q U I R E M E N T S F O R T H E D E G R E E O F M A S T E R O F A R T S in T H E F A C U L T Y O F G R A D U A T E S T U D I E S (Department o f Pol i t ica l Science, Facul ty o f A r ts ) W e accept this thesis as con fo rming to the required standard T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A August 2004 © Jul ia Kate Lockhar t , 2004 Library Authorization In presenting this thesis in partial fulfillment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Title of Thesis: ply/lct^d foveSf\1Klvfc M (jWXxdQ Name of Author (please print) Date (dd/mm/yyyy) Degree: Mooter o[M& Year: Department of fQ\\ The University of British Columbia Vancouver, BC Canada ABSTRACT D i v i d e d government i n Canada refers to the c o m m o n situat ion when the federal and prov inc ia l governments are held by di f ferent po l i t ica l parties. The study o f d iv ided government can aid i n the understanding o f voter behaviour. The thesis reviews the relevant l i terature on d iv ided government, spl i t - t icket vo t ing and party ident i f icat ion in Canada and the Un i ted States. F r o m the l i terature several voter strategies are extracted that describe the possible ind iv idua l level processes that result in the aggregate outcome o f d iv ided government. Th is l inkage, between ind iv idua l decisions and col lect ive outcomes, is crucia l to understanding d iv ided elect ion outcomes and i t is to the explorat ion o f this concept that the thesis contributes. Us ing a dataset o f party vote shares i n p rov inc ia l and federal elections f r o m 1904 to 2003, the thesis looks fo r aggregate effects o f the ind iv idua l level strategies that i t ident i f ies. The thesis argues that d iv ided government i n Canada is a result o f staggered elect ion t i m i n g and po l icy learning across levels w h i c h combine to produce a cyc l ica l effect i n elect ion results. i i T A B L E O F C O N T E N T S Abstract i i Table o f Contents i i i L is t o f Tables i v L is t o f Figures v Acknowledgements v i Dedicat ion v i i C H A P T E R 1 In t roduct ion 1 C H A P T E R 2 Li terature Rev iew 7 C H A P T E R 3 Voter Strategies 24 C H A P T E R 4 Data Analys is 41 C H A P T E R 5 Conclus ion 57 Append ix I Figures 60 Append ix I I Tables 62 B ib l iography 67 i i i LIST OF TABLES Table 1 - Summary o f M o d e l Predict ions 62 Table 2 - Federal Swings 63 Table 3 - Prov inc ia l Swings 64 Table 4 - Federal Combined M o d e l 65 Table 5 - Prov inc ia l Combined M o d e l 66 iv L I S T O F F I G U R E S Figure 1 - Conservat ive Federal Government (17=0.5) 60 Figure 2 - L ibera l Federal Government (q=0.5) 60 Figure 3 - Conservat ive Federal Government (t7=0.8) 61 Figure 4 - L ibera l Federal Government (q=0.S) 61 v ACKNOWLEDGEMENTS M y t ime at U B C has been greatly enriched by a couple o f professors to w h o m I owe a deep debt o f grati tude fo r their help, inspirat ion and insight: R ichard Johnston and Fred Cutler. Dr . Johnston and Dr . Cut ler suppl ied the dataset that this thesis is bu i l t on and in do ing so turned over many o f their ideas and their research area to me (at least temporar i ly ! ) . Together, they fo rmed an enviable supervision team, bo th o f their inf luences being clearly v is ib le throughout the text. Dr . Johnston f i rst int roduced me to the concept o f d iv ided government late in m y undergraduate program. Dr. Johnston has an amazing capacity to d r i l l d o w n an argument to its most essential points by posing very thought fu l research questions. Dr . Cut ler went out o f his way to help me w i t h bo th the data analysis and voter strategies sections. He of fered me a directed readings course in w h i c h much o f the pre l iminary model ing was done as we l l as p rov id ing the in i t ia l inspirat ion for the theoretical f ramework o f the thesis. A l o n g the way they both taught me a lot about the discipl ine o f po l i t ica l science and I am very grateful to them both. v i D E D I C A T I O N T o Dad , w h o taught me to reach fo r the stars. T o M o m , w h o was there to catch me when I fe l l . CHAPTER 1 - INTRODUCTION D i v i d e d government occurs i n po l i t ica l systems w h i c h permi t voters to express preferences by casting ballots fo r more than one inst i tut ion. The most famous example o f d iv ided government is Amer i can , when the Presidency and Congress are held by di f ferent parties. D i v i d e d government can also occur when the t w o chambers in the legislat ive branch are o f d i f ferent part isanship. I n essence d iv ided government refers to those t imes when t w o governments w i t h i n one country are held by di f ferent (and usual ly compet ing) po l i t i ca l parties. Thus wh i le much o f the d iv ided government l i terature focuses on d iv is ion w i t h i n the US federal government, federal consti tut ions in general prov ide opportuni t ies fo r d iv ided government. The theories wh ich are used to expla in d iv ided government i n the U S can be mod i f i ed to f i t the di f ferent, but structural ly s imi lar , inst i tut ional context o f federal ism. I n do ing so we can hope to exp la in more general ly how voters respond when faced w i t h more than one choice in a po l i t i ca l system. This paper considers the electoral causes o f d iv ided government w i t h i n the Canadian federat ion. H o w and w h y do Canadian voters select di f ferent parties fo r their p rov inc ia l and federal governments? The Canadian case is perhaps best understood as one o f d iv ided elect ion outcomes. Because the voters i n any g iven prov ince cannot uni lateral ly decide the federal elect ion outcome, d iv ided government outcomes are not consistent across the country nor are they necessarily the product o f a conscious choice. Nevertheless d iv ided elect ion outcomes (or d iv ided government as the results w i l l be k n o w n in this paper) occur f requent ly. There are t w o reasons w h y the study o f d iv ided outcomes is important . First, d iv ided government may impact po l icy outcomes and governmental strategies. W h e n 1 div is ion occurs w i t h i n one inst i tu t ion i t is more l i ke ly that po l i cy impacts w i l l be visible. Therefore, i t is f r o m this perspective that much o f the Amer i can research began. The second reason is that by studying d iv ided government we may learn more about h o w voters make decisions and use in fo rmat ion when going to the pol ls . I t is this reason that motivates much o f the Canadian research. D i v i d e d government represents the aggregation o f a set o f choices by voters. W h e n d iv is ion occurs i t is a product o f voter choice. Th is , more behavioural ist , interest is what provokes the research conducted in this paper. D i v i d e d government is an electoral puzzle. I n the process o f expla in ing i t we may learn about more than jus t the specif ic calculations at hand. I t is fo r this reason that the study o f d iv ided government is part icular ly useful w i t h i n the Canadian context. Anecdota l ly , many Canadians have experienced d iv ided government when their p rov inc ia l government was he ld by a di f ferent party than the federal government. Consider ing the federal elections held since 1904 1 i n each prov ince separately, voters chose the same party as their p rov inc ia l government, on aggregate, fo r the federal government on ly about 4 4 % o f the t ime. That is to say that less than ha l f o f the t ime voters chose the same party at a subsequent federal elections as the party they chose prov inc ia l l y . The prov inc ia l elections held since 1904 2 produced un i f ied results on ly 3 5 % o f the t ime. Overa l l on ly about 4 0 % o f elections i n Canada produced un i f ied government w i t h the same party ho ld ing the prov inc ia l and federal governments. B roken d o w n by prov ince, N o v a Scotia elections produced the highest percentage o f un i f ied governments, w i t h 6 1 % o f al l elections (federal or prov inc ia l ) resul t ing in un i f ied government. A lber ta elections, unsurpr is ingly g iven that the federal vo t ing patterns o f 1 Exc lud ing the June 28, 2004 federal elect ion. 2 U p to the June 9, 2003 N e w Brunsw ick elect ion. 2 Albertans are usual ly at odds w i t h the outcomes o f federal elections, produced the fewest un i f ied governments at 15%, f o l l owed by Ontar io at 2 6 % and B C w i t h 27%. Clear ly then, this effect is not s imply a product o f geographic and regional tensions on the federal level . W h i l e provinces that are more alienated f r o m the federal government do display more o f a tendency towards d iv ided government, even i n Ontar io (the prov ince that contains a th i rd o f the federal seats) un i f ied government occurs at a rate be low that o f the nat ional average. There are any number o f possible theoretical causes fo r these results. The results presented in this paper do not conclusively reject or support any o f the explanations presented be low. Rather, g iven the l imi tat ions o f the data available, the paper attempts to out l ine the possible voter strategies at w o r k and to ident i fy any aggregate trends that are vis ible. The paper w i l l begin by rev iew ing the relevant d iv ided government l i terature. A large por t ion o f the l i terature is devoted to balancing and moderat ing explanations o f d iv ided government. W h i l e balancing theories or ig inated in Canada, they have been r igorous ly tested i n the Amer i can context. A n attempt w i l l be made to l i nk our understanding o f d iv ided government and party ident i f icat ion. W h i l e d iv ided government i n Canada is not a part icular ly w e l l studied area, those w h o study party ident i f icat ion have long been interested in spl i t - ident i f iers w h o are probably l inked in some way w i t h d iv ided government outcomes. I t seems natural to assume that i f one carries one's federal and prov inc ia l party ident i f icat ions separately to the extent that they are in many cases di f ferent, one migh t also vote d i f ferent ly in federal and prov inc ia l elections such that d iv ided government ensues. F r o m the party- focused l i terature a cyc l ica l or po l icy learning perspective on d iv ided electoral outcomes has also emerged. 3 The paper w i l l also consider incumbency, issue ownership, and t i m i n g effects i n the US context. W h i l e some o f the specifics o f U S theories do not translate we l l to the Canadian case many o f their assumptions about voter strategies can be manipulated to f i t the federal context. The paper w i l l then present f ive possible strategies that Canadian voters migh t be using. Th is section w i l l out l ine the strategies and what the effect o f those strategies w o u l d be on an aggregate level i f every voter used the same strategy in the same condit ions. I n this way the paper w i l l attempt to overcome the d i f f icu l t ies presented by hav ing access on ly to aggregate level data. The key insight o f this section o f the paper is that d iv ided government may be the result o f d i f ferent voters responding in d i f ferent ways to their electoral choices. Obv ious ly al l Canadians do not respond to events and choices in exact ly the same fashion. D i f ferent voters w i l l employ the ident i f ied strategies depending on their level o f po l i t i ca l in fo rmat ion and partisanship. A l l o f the strategies ident i f ied be low may w e l l be at w o r k though i t may prove d i f f i cu l t to ident i fy wh ich are dominant f r o m aggregate level data alone. These strategies fa l l into t w o broadly conceived categories, intent ional models and unintent ional models, based on the strategies' understandings o f the thought process being used by the voter. The two balancing models are intent ional strategies: • a po l icy balancing strategy in w h i c h voters consciously choose to balance the parties i n power based on their preferences fo r more moderate po l icy outcomes; • an interest balancing strategy in w h i c h voters express their preference fo r part icular ist ic regional or prov inc ia l benefits i n the interest balancing model . 4 There are three remain ing strategies that al l produce d iv ided government unintent ional ly as a result o f the lack o f coordinat ion in the t i m i n g o f federal and prov inc ia l elections: • a general evaluative mode l w h i c h assumes that voters cannot dist inguish between federal and prov inc ia l po l icy areas or outcomes and that they s imply evaluate the total po l icy envi ronment before casting their vote; • a level-specif ic mode l w h i c h is s imi lar except that voters using this strategy evaluate the federal and prov inc ia l po l icy arenas separately; • a po l i cy learning model w h i c h suggests that voters use in fo rmat ion gathered at both levels to update their party preferences based on how those parties pe r fo rm i n government. Voters using intent ional strategies (pol icy and interest balancing) choose d iv ided government. Voters using unintent ional strategies (general evaluat ive, level-specif ic evaluat ive and po l icy learning) produce d iv ided government by accident. These people w o u l d l ike to be consistent across levels but because they only ever get to vote on one level at a t ime and because prov inc ia l electorates do not have complete contro l over federal outcomes they cannot always succeed in produc ing consistent results. The empi r ica l section o f the paper w i l l focus on aggregate elect ion outcomes in Canadian federal and prov inc ia l elections since 1904. The dataset, the most complete bu i l t so far, incorporates al l parties i n al l provinces at both levels. The paper suggests a mode l to capture both intent ional and unintent ional d iv ided government effects. E ight models are estimated. For both arenas, each major po l i t ica l party (Conservat ive, L ibera l and N e w Democrat ic Party) is modeled separately. A f ina l mode l fo r both the federal and prov inc ia l arenas is f i t ted w i t h al l three parties combined. W h i l e the models cannot 5 f u l l y test the theoretical posit ions out l ined w i t h i n the paper, they do prov ide evidence to suggest that d iv ided government is not an intent ional outcome o f voter choice. 6 C H A P T E R 2 - L I T E R A T U R E R E V I E W A large number o f d i f ferent theories have been used to expla in d iv ided electoral outcomes in both Canada and the Un i ted States. The most prominent explanations in both countries rest on a not ion o f balancing or moderat ion. Ba lanc ing theories suggest that i n a polar ized party system voters w i l l choose to d iv ide elect ion outcomes to achieve more moderate po l i cy results. Us ing both aggregate and ind iv idua l level data these balancing explanations have been empi r ica l ly tested in Canada, the US and Germany. Recent ly, i n response to the g row ing U S l i terature, balance theory has been reappl ied to the Canadian case w i t h m i x e d results suggesting that d iv ided government may be the result o f either t im ing or po l icy learning. Other theories have also emerged in bo th countries. US l i terature emphasizes a variety o f other explanations inc lud ing incumbency effects and gerrymander ing. The partisan "ownersh ip " o f various issues and as a result various branches o f government is another strand explored in the U S l i terature. F ina l ly , some research has suggested that d iv ided government is the product o f voters learning about part ies' po l i cy posit ions between staggered elections. Th is chapter w i l l explore these di f ferent conceptual explanations o f d iv ided government i n turn. Balancing W h i l e the not ion o f po l icy balancing is typ ica l ly understood to be an Amer i can concept, i t was " in t roduced by students o f Canadian pol i t ics long before its contemporary appl icat ion to A m e r i c a n po l i t i cs . " 3 Frank Underh i l l f i rst art iculated the balance theory stating that " b y some inst inct ive sub-conscious mental process the Canadian people have apparently decided that, since f reedom depends upon a balance o f power, they w i l l 3 Robert Er ikson and M i k h a i l F i l i ppov , "E lectora l Balanc ing i n Federal and Sub-nat ional Elect ions: The Case o f Canada" Constitutional Political Economy 12 (2001): 315. 7 balance the monopol is t ic power o f the L ibera l government at Ot tawa by setting up the effect ive countervai l ing power not i n Ot tawa but i n the prov inc ia l capi ta ls ." 4 Interest ingly, Underh i l l v iewed the vo t ing decisions that produced this pattern w i t h disdain, suggesting that i t was a fa i lure on the part o f federal opposi t ion parties to prov ide a credible alternative to voters. Dennis W r o n g ' s study o f party vo t ing i n Canada also describes a balancing effect. He argues that Canadian voters have chosen " to counter the power o f the nat ional administrat ion not by elect ing a strong federal opposi t ion but by vo t ing against the L ibera l Party i n p rov inc ia l e lect ions." 5 B i l l Reeves and Roger Gibbins w o r k e d to estimate the effect o f balancing on the Canadian electorate. They wr i te that "at least i n the short-run, we encountered no support for a balance hypothesis . " 6 Reeves and Gibbins make an interesting point not ing that bandwagon effects i m p l y the integrat ion o f federal and prov inc ia l elections wh i le balancing suggests that the relat ionship between the t w o levels is p r imar i l y conf l ic tua l . 7 I n contrast to these negative results, Robert Er ikson and M i k h a i l F i l i ppov f i n d evidence to support the balance model . Er ikson and F i l i ppov use a l im i ted dataset consist ing o f those prov inc ia l elections that f o l l o w federal elections f r o m 1949 to 1997. They also restr ict their analysis to the L ibera l party because i t was a both a major federal and prov inc ia l par ty . 8 E r ikson and F i l i ppov regress vote swings f r o m one prov inc ia l 4 Frank Underh i l l , "Canadian L ibera l Democracy i n 1955," i n Press and Party in Canada, eds. G.F. Ferguson and F.H. Underh i l l , (Toronto: Ryerson, 1955), 39-40. 5 Dennis W r o n g , " T h e Pattern o f Party V o t i n g in Canada" Public Opinion Quarterly 21 (1957): 255. 6 B i l l Reeves and Roger Gibb ins, "The Balance Theory: A n Empi r i ca l L o o k at the Interdependency o f Federal-Provincial Electoral Behav iour , " presented to the 1976 A n n u a l Mee t ing o f the Canadian Pol i t ica l Science Associat ion, Quebec C i ty , 23. 7 Reeves and Gibb ins, 4. 8 E r ikson and F i l ippov , 318. 8 elect ion to the next as w e l l as swings f r o m a federal elect ion to the next prov inc ia l one on a d u m m y indicat ing whether the L ibera l party was in power federal ly. The coeff ic ient fo r the d u m m y is negative leading them to conclude that the L ibera l party suffers when in power federal ly, consistent w i t h the log ic o f balancing. Expand ing the regression to inc lude pr io r federal elect ion outcomes as an independent variable produces s imi lar results. W h i l e the results contained w i t h i n this paper w i l l challenge those presented by Er ikson and F i l ippov , i t is important to note that the balancing explanat ion in the Canadian context is not w i thout its supporters. Balanc ing is not a "s t raw m a n " set up to be knocked d o w n by academics look ing fo r something interest ing to wr i te articles about. Balance theory was also proposed i n the U S as a possible explanat ion fo r the regular loss o f seats that the President's party experiences at m i d t e r m elections. Robert Er ikson, in his study o f m id te rm loss, proposes four potent ial explanations o f the phenomenon: regression to the mean, a surge and decl ine effect, referendum on presidential performance, and a presidential penalty. 9 Regression to the mean states that m id te rm loss is a func t ion o f the loss o f the presidential coattails. Surge and decl ine is a turnout based argument wh ich suggests that on-year elections energize large segments o f the electorate wh i le m i d t e r m elections are decided by "core partisan vo ters . " 1 0 The referendum model may at t imes be d i f f i cu l t to dist inguish f r o m the presidential penalty, but the key is that the referendum theory rests on evaluations o f presidential performance; the mere fact o f ho ld ing the presidency, i n this model , is not enough to cause the m id te rm loss. The presidential penalty mode l incorporates both negative vo t ing and po l i cy 9 Robert Er ikson, " T h e Puzzle o f M i d t e r m Loss " Journal of Politics 50 (November 1988): 1012-3. 1 0 Er ikson, 1013. 9 balancing. Er ikson draws a connect ion between this mode l and the early Canadian l i terature inc lud ing Underh i l l and W r o n g . 1 1 Er ikson conducts a study o f the relat ionship between Democrat ic vote in m id te rm and on-year elections. He discovers that data support the presidential penalty explanat ion to the exclusion o f the other possible models. Er ikson states that " the presidential penalty at m id te rm appears to ref lect more than voters ' unhappiness w i t h their president ." 1 2 However he declines to conclude whether this represents balancing or negative vo t ing because o f the l imi tat ions o f aggregate data. A lber to A les ina and H o w a r d Rosenthal develop and expand on the concept o f the presidential penalty, specif ical ly the po l icy balancing strand. D e f y i n g the log ic o f the median voter theorem, A les ina and Rosenthal assert that the Democrats and Republ icans do not have the same po l i cy preferences. 1 3 The i r claims about the polar izat ion o f parties are key to the logic o f balancing. Because po l icy outcomes are a product o f both the po l i cy preferences o f the executive and the legislature, those voters located between the part ies' ideal po l i cy points face a choice: they can either vote a straight party t icket fo r the party closest to them or they can spli t their t icket i n the hope o f produc ing more moderate outcomes w i t h both parties w o r k i n g together (or against each other). A les ina and Rosenthal argue that "midd le-o f - the- road voters . . . take advantage o f [the] legislat ive-execut ive interact ion in po l icy fo rmat ion to b r ing about moderate po l icy ou tcomes. " 1 4 They support their theory by stating that there is a relat ionship between the size o f the m i d t e r m loss for the President's party and voter surprise over the outcome o f 1 1 Er ikson, 1014. 1 2 Er ikson, 1027. 1 3 A lbe r to A les ina and H o w a r d Rosenthal, Partisan Politics, Divided Government, and the Economy, (Cambr idge: Cambr idge Univers i ty Press, 1995), 16. 1 4 A les ina and Rosenthal, 44. 10 the Presidential elect ion. W h e n voters are surprised by an elect ion outcome i t of ten means that they incorrect ly forecast the result o f the elect ion and therefore balanced (or d idn ' t balance) inappropriately. M o r r i s F io r ina categorizes possible explanations o f d iv ided government into t w o groups: accidental and intent ional models . 1 5 W h i l e F ior ina does not come to a f i r m conclusion as to the causes o f d iv ided government, he makes a useful cont r ibut ion to the balancing l i terature by descr ibing a l o w in fo rmat ion appl icat ion o f the balancing logic. He argues that voters do not have to choose consciously to balance in order to actual ly do so. He states that "hav ing made a decision to support Bush and fee l ing less than enthusiastic about i t , [voters] may be predisposed to l isten to Democrat ic appeals fo r other o f f i ces . " 1 6 I n do ing this, voters show a vague appreciat ion fo r the larger decision mak ing apparatus. Obv ious ly the predict ive power o f this theory is l im i ted in a party system w i t h more than t w o parties or i f d i f ferent parties compete at the t w o levels. F ior ina argues that "people cou ld be vo t ing as i f they are mak ing conscious choices to d iv ide government even i f their ind iv idua l decisions are w e l l be low the conscious l e v e l . " 1 7 W h a t is part icular ly interesting about this c la im is that i t suggests that we may never be fu l l y able to determine the reasons that people spl i t their t ickets. For i f voters use the balancing strategy subconsciously, i t w o u l d be d i f f i cu l t to prove or measure even w i t h access to sophisticated ind iv idua l level data. M a n y more authors have taken up the theoretical f ramework presented by Ales ina and Rosenthal and have undertaken empir ica l tests o f the various balancing assumptions 1 5 Th is d is t inct ion that is carr ied through this thesis. 1 6 M o r r i s F ior ina , Divided Government, 2 n d ed., (Boston: A l l y n and Bacon, 1996), 64. 1 7 I b i d . 11 and predict ions. Some have found direct evidence o f ind iv idua l level support fo r po l icy ba lanc ing. 1 8 Others assert that because po l icy balancing is a re lat ively sophisticated vo t ing strategy, i t is typ ica l ly on ly used by those w i t h h igh levels o f po l i t i ca l knowledge. James Garand and M a r c i Glascock L i c h t l state that, wh i le the basic models prov ide no support fo r po l i cy balancing, " w h e n the effect o f preferences fo r d iv ided government on spl i t - t icket vo t ing is mediated by po l i t i ca l knowledge, [ their ] results prov ide reasonably strong conf i rmat ion for [ba lanc ing ] . " 1 9 I n contrast, Wal te r Mebane and Jasjeet Sekhon assert that wh i le po l i cy balancing or moderat ion is the end result o f people 's electoral choices this is not a ref lect ion o f a preference for d iv ided government per se. I t is s imply a product o f an inst i tut ional structure that channels "each elector 's self ish efforts i n such a way that co l lect ive ly there is a moderated resul t . " 2 0 Electors, they argue consider each others' choices, recognize that the elect ion w i l l l i ke ly produce a moderated result and vote accordingly. Those that f i n d evidence o f po l icy balancing (sometimes cal led cogni t ive Mad ison ian ism 2 1 ) usual ly deal w i t h ind iv idua l level data. There are some d i f f icu l t ies using ind iv idua l level survey data because po l icy balancing is a relat ive compl icated concept to test w i t h i n the confines o f a telephone survey. Moreover i f one accepts that preferences fo r d iv ided government cannot be separated f r o m partisan preferences i t can 1 8 Charles Smi th , Robert B r o w n , John Bruce and L. M . Overby, "Par ty Balanc ing and V o t i n g fo r Congress i n the 1996 Nat iona l E lec t ion" American Journal of Political Science 43 (1999): 737-64. 1 9 James Garand and M a r c i Glascock L i c h t l , "Exp la in ing D i v i d e d Government i n the Un i ted States: Test ing an Intent ional M o d e l o f Spl i t - t icket V o t i n g " British Journal of Political Science 30 (2000): 186. 2 0 Wa l te r Mebane and Jasjeet Sekhon, "Coord ina t ion and Pol icy Modera t ion at M i d t e r m " American Political Science Review 96 (2002): 141. 2 1 Everett Car l l L a d d , "Pub l ic Op in ion and the 'Congress P r o b l e m ' " The Public Interest 100 (1990): 66-67. be d i f f i cu l t to design a survey question that permits voters to be true to their thoughts. Th is is an interesting argument because i t suggests that there is no perfect dataset fo r the study o f d iv ided government. Aggregate data is l im i ted in its explanatory power because i t cannot expla in why voters vote the way they do. Ind iv idua l level data is f lawed because the " w h y " is too hard to probe ef fect ively in a telephone survey. I t is hard to adapt the balancing model to a po l i t ica l system w i t h more than t w o parties. The presence o f a th i rd party suggests that voters already have a moderate choice in the f o r m o f the midd le party. A les ina and Rosenthal o f fer no insight as to h o w a midd le party m igh t affect the strategic calculus undertaken by voters. Nonetheless, the balancing mode l has been exported to the mul t i -par ty German federat ion. Susanne L o h m a n n , D a v i d Brady and Douglas Rivers conducted a study o f the roles o f party ident i f icat ion, retrospective vo t ing and moderat ion i n West Germany. They argue that the "modera t ing elections hypothesis provides an alternative explanat ion fo r the puzzle w h y m i d t e r m vote losses in by-elect ions or Land elections do not necessarily translate in to vote losses i n national e lect ions." 2 3 A f t e r demonstrat ing that po l i cy outcomes depend on both the nat ional and L a n d elections, they argue that midd le o f the road voters may use the federal system to achieve more moderate outcomes than w o u l d be possible by vo t ing a straight t icket. The German case is part icular ly interest ing because o f the presence o f a th i rd party though i n the German case the th i rd party is usual ly part o f a coal i t ion government. Lohmann , Brady and Rivers conduct a regression o f the swing i n 2 2 M ichae l Lew is -Beck and Richard Nadeau, "Sp l i t -T icke t V o t i n g : The Effects o f Cogn i t i ve Mad ison ian i sm" Journal of Politics 66 (February 2004) : 99. 2 3 Susanne Lohmann , D a v i d Brady, and Douglas Rivers, "Par ty Ident i f icat ion, retrospective V o t i n g , and Modera t ing Elect ions in a Federal System: West Germany 1961-1989," Comparative Political Studies 30 (August 1997), 424. 13 party vote share f r o m national to L a n d elections on G N P g rowth and incumbency. They f i n d that the nat ional incumbency d u m m y is s igni f icant in most variat ions o f the model leading them to conclude that there is some element o f moderat ion contained w i t h i n the German elect ion results. The models show that retrospective vo t ing and party ident i f icat ion also dr ive vote choice. Incumbency W h i l e balancing theory dominated m u c h o f the early Canadian research, early US studies o f d iv ided government focused on the role o f gerrymander ing and incumbency in perpetuating d iv ided contro l . W h i l e Canadian pol i t ic ians at both levels o f government do not enjoy the personal incumbency benefits o f members o f Congress, there is some evidence to suggest that p rov inc ia l governments are often dominated by one party fo r a long per iod o f t ime creating a type o f incumbency advantage. M a n y Republ ican activists focused on the Democrat ic cont ro l o f many state legislatures (the bodies responsible for redistr ic t ing) as the cause o f the Democrat ic strength in the House. They argued that because Democrats contro l led the redistr ic t ing process, they had the capacity to t ransform marg ina l distr icts into Democrat ic strongholds. 2 4 Because the t radi t ional pattern o f d iv ided government was a Republ ican president combined w i t h a Democrat ic Congress, this seemed to be a plausible, i f somewhat part isan, explanat ion fo r the phenomenon. H o w e v e r i t seems that most po l i t ica l scientists reject this l ine o f reasoning because i t impl ies that d iv ided government is caused by Republ ican House votes being wasted rather than by spl i t - t icket vo t ing . 2 5 2 4 F ior ina , 16. 2 5 I b id . 14 Another possible explanat ion for Democrat ic strength in the House is the power o f the incumbency effect. Because most incumbents are re-elected and because once a party takes contro l o f the House by def in i t ion a major i ty o f the incumbents are f r o m that party, i t seems that there is l i ke ly to be a strong tendency fo r that party to retain contro l . The second part o f the incumbency argument is that " o n the basis o f performance and issues, the contemporary electorate favors Republ ican presidential candidates." 2 6 W h i l e House races are dominated by incumbency wh ich favours the Democrats, the presidential contest is dominated by Republ icans. W h i l e this logic is certainly compel l ing , i t does not expla in the swi tch in partisan roles in the 1990s nor the level o f t icket spl i t t ing in open seat races. Bar ry Burden and D a v i d K i m b a l l also approach the p rob lem f r o m an incumbency oriented perspective that can be translated into the Canadian context. They study d iv ided government at the aggregate level arguing that i t is an aggregate result and is therefore best studied at that l eve l . 2 7 They assert that t icket-spl i t t ing is best understood at a distr ict level . I n the Canadian case, the equivalent o f this approach (and the one undertaken in this thesis) is to study the p rob lem on a prov inc ia l aggregate level . Burden and K i m b a l l note than in many districts the House race is not compet i t ive either because one party declines to run a candidate or because one o f the major party candidates is substantial ly better funded or more experienced. I n many prov inc ia l elections, the same cou ld be said fo r one party or another. Th is is, i n a way, an expansion o f the incumbency argument presented earlier. Burden and K i m b a l l argue that "d i v ided government is largely an 2 6 I b i d , 2 1 . 2 7 Bar ry Burden and D a v i d K i m b a l l , Why Americans Split Their Tickets: Campaigns, Competition and Divided Government, ( A n n Arbor : Un ivers i ty o f M i c h i g a n Press, 2002), 33-4. 15 accidental creat ion, a by-product o f lopsided congressional races around the country that foster spl i t - t icket v o t i n g . " 2 8 I n contrast to the po l icy balancing perspective, w h i c h sees t icket-spl i t t ing as a f u l l expression o f voters' preferences, Burden and K i m b a l l see d iv ided vo t ing as a ref lect ion o f the poor choices of fered by uncompet i t ive House races. B y extension, wh i le one-party dominance in p rov inc ia l elections may be the result o f voter preferences, i t cou ld also be a result o f s imply hav ing no credible alternative to the govern ing party. D i v i d e d government occurs i n this model because voters are on ly able express their true preferences on the federal level , the level on w h i c h they are presented w i t h a complete array o f choices. O n the prov inc ia l level they are " f o r c e d " to select the dominant party because the other options do not appear to have the capacity to f o r m government. Arena Separation and Issue Control Another theory o f d iv ided government i n the Un i ted States was advanced by Gary Jacobson. Jacobson's study o f the causes o f d iv ided government, publ ished in 1990, focused on the t radi t ional pattern o f d iv ided government: a Republ ican President fac ing a Democrat ic Congress. He argues that the causes o f d iv ided government are po l i t i ca l : " the Democrats ' cont inued dominance o f the House. . . despite Republ ican presidential victor ies is a consequence o f electoral pol i t ics: o f candidates, issues, electoral coal i t ions and voters ' reactions to t h e m . " 2 9 Th is challenged the convent ional w i s d o m o f many Republ icans w h o asserted that d iv ided government resulted f r o m structural constraints that prevented them f r o m w i n n i n g their " f a i r " share o f House seats. Jacobson asserts that 2 8 Burden and K i m b a l l , 40. 2 9 Gary Jacobson, The Electoral Origins of Divided Government, (Boulder, Co lo : Westv iew Press, 1990), 105. 16 voters understand that the President and Congress have di f ferent po l i cy funct ions. Congress is responsible fo r local d ist r ibut ive concerns wh i le the President handles fore ign po l icy , defense and other nat ional issues. M a n y voters want both l o w taxes, balanced budgets and responsible f iscal management as we l l as a strong mi l i ta ry , good publ ic educat ion and qual i ty social services. The dual elect ion opportunit ies o f presidential and house elections permi t people to express bo th sets o f preferences by vo t ing for Republ ican presidential candidates and Democrat ic House candidates. Th is is because "perce ived differences between parties coincide w i t h differences i n what people expect o f presidents and members o f Congress . " 3 0 A f te r the Democrats took contro l o f the Presidency i n 1992 and Republicans the House i n 1994, Jacobson had to update his theory o f d iv ided government. He argued that d iv ided government provides po l i t ica l cover fo r both parties to enact unpopular pol icies. I n this way d iv ided government perpetuates itself. Once the partisan af f i l iat ions o f the branches changed i t became un l ike ly that i t w o u l d change back because o f the po l i t ica l cover af forded by d iv ided government . 3 1 B u i l d i n g on the w o r k o f Jacobson, other US scholars have focused on the part isan cont ro l and "ownersh ip " o f issues and as a result branches o f government. John Petrocik and Joseph Doher ty advocate a theory o f issue ownership. Issue ownership asserts that " t icket -sp l i t t ing occurs when the issues dominat ing the presidential elect ion d i f fer f r o m Jacobson, Electoral Origins, 112. 3 1 Gary Jacobson, " D i v i d e d Government i n the 1994 Elect ions" i n Divided Government: Change, Uncertainty and the Constitutional Order, ed. Peter Galder is i , (Lanham, M D : R o w m a n and L i t t l e f ie ld Publishers, 1996), 8 1 . 17 those attracting attention in the congressional contest.' Th is argument is s imi lar to that presented by Jacobson except that i t does not t ie specif ic issues to either branch o f government a l low ing the theory to accommodate the swi tch i n part isan af f i l ia t ions that occurred i n the early 1990s. Petrocik and Doher ty use ind iv idua l level data to test fo r correlat ions between professed support fo r d iv ided government and t icket-spl i t t ing. They f i n d instead that there "are very few correlates o f expressed feel ings about d iv ided government . " 3 3 I n Canada i t is easy to see that p rov inc ia l and federal governments migh t o w n di f ferent issues. I n a federal system, such as Canada, issue ownership is def ined quite speci f ical ly i n the const i tut ion: federal and prov inc ia l governments have di f ferent areas o f responsibi l i ty . Thus the more f lex ib le not ion o f issue ownership, as presented by Petrocik and Doher ty , does not translate w e l l to the Canadian case as i t ignores the const i tut ional imperat ive. Issue contro l i n the Canadian context must become a type o f arena contro l w i t h specif ic parties showing part icular strength at one level or the other. W i t h i n the Canadian context, the argument fo r arena-separation is best presented by Dona ld B lake i n his book Two Political Worlds. I n i t he studies the relat ionship between federal and prov inc ia l vo t ing and partisanship i n Br i t i sh Co lumbia . B lake notes that in the t w o elections held w i t h i n 12 days o f each other in M a y o f 1979 Br i t i sh Columbians voted in dramat ical ly d i f ferent ways. He states that " the prov inc ia l and federal elections John Petrocik and Joseph Doher ty , " T h e Road to D i v i d e d Government : Paved w i thout In ten t ion" i n Divided Government: Change, Uncertainty and the Constitutional Order, ed. Peter Galder is i , (Lanham, M D : R o w m a n and L i t t l e f ie ld Publishers, 1996), 89. 3 3 Petrocik and Doher ty , 105. 18 i n M a y 1979 were close i n t ime but distant i n psychological space." The returns fo r the t w o elections were very di f ferent: federal ly Br i t i sh Columbians voted Conservat ive wh i le p rov inc ia l l y Social Credi t w o n . B lake argues that this was because the issues were very di f ferent in the t w o arenas and the voters were able to make those dist inct ions and act accordingly. He asserts that this type o f separation is not jus t a result o f the B C party system in wh ich the compet i t ive parties at one level are minor parties on the other leve l . 3 5 Richard Johnston's early w o r k on d iv ided government also approaches the prob lem f r o m the perspective o f parties and party cohesion. First, he highl ights the challenge o f determin ing the elect ion pairs that are o f interest in the Canadian case when elect ion t im ing is not as regular ized as i n the Un i ted States. 3 6 A f t e r determin ing that the most impor tant considerat ion is to select those elections w h i c h are closest i n t ime, he proceeds to examine the possible effect o f voter abstentions on d iv ided elect ion outcomes. Johnston concludes that the changes in partisan outcomes across levels are not the result o f selective abstentions based on his comparison o f the net change in vote share inc lud ing those voters w h o reported not vot ing. A second explanat ion fo r ind iv idua l level choices fo r d iv is ion explored by Johnston is the possible correlat ion between po l i t i ca l invo lvement and casting di f ferent federal and prov inc ia l votes. W h i l e there are good theoretical reasons fo r this correlat ion to exist, he f inds no conclusive support fo r the relat ionship. F ina l ly , Johnston tests to see i f part icular social or rel ig ious groups are more l i ke ly to swi tch their votes across levels. A g a i n wh i le some patterns emerge, they 3 4 Dona ld B lake w i t h D a v i d E lk ins and Richard Johnston, Two Political Worlds: Parties and Voting in British Columbia, (Vancouver: U B C Press, 1985), 135. 3 5 B lake, 168. 3 6 R ichard Johnston, "Federal and Prov inc ia l V o t i n g : Contemporary Patterns and Histor ica l E v o l u t i o n , " i n Small Worlds: Provinces and Parties in Canadian Political Life eds. D a v i d E lk ins and Richard Simeon, (Toronto: Methuen, 1980), 154. 19 are t rumped by the differences across provinces. Johnston's study suggests then, i n contrast to the argument advanced i n the in t roduct ion, that d iv ided electoral outcomes may w e l l be the result o f the di f ferent prov inc ia l party systems and arena separation. Cycling A f ina l perspective on d iv ided government, w h i c h or iginated i n Canada, suggests that these outcomes are the result o f hav ing staggered elections w h i c h permi t voters to learn about part ies' po l icy preferences and posit ions. Th is more unintent ional model was described by R. MacGregor Dawson. He suggested that the pattern is that " f i rs t , the great major i ty o f the D o m i n i o n and prov inc ia l governments w i l l be long to the same pol i t ica l party; second, the prov inc ia l governments w i l l begin to fa l l away to the opposi t ion party or parties un t i l these are i n a major i ty ; th i rd , there is an overturn in the D o m i n i o n Par l iament . " 3 7 H o w a r d Scarrow notes that wh i le elect ion results seem to indicate alternation and therefore balancing, he believes that what is more impressive is the " f requency o f instances where the result o f a prov inc ia l elect ion has correct ly forecast the result o f the succeeding federal elect ion, and vice versa. " 3 8 Th is suggests that d iv ided government is actual ly a product o f cyc l ica l victor ies w i t h staggered elections rather than o f balancing. He fur ther argues that the Canadian experience o f d iv ided government and alternating elections is not inconsistent w i t h the experience o f other federations. He argues that perhaps alternating vo t ing patterns are the result o f turnout di f ferentials between federal and prov inc ia l elections. Scarrow rejects the not ion o f intent ional balancing on the part o f the voter as the under ly ing cause o f alternation and d iv is ion 3 7 R. MacGregor Dawson, The Government of Canada, 5 t h ed, revised by N o r m a n W a r d , (Toronto: Un ivers i ty o f Toronto Press, 1970), 486. 3 8 H o w a r d Scarrow, "Federal -Provinc ia l V o t i n g Patterns i n Canada" Canadian Journal of Economics and Political Science 26 (1960): 2 9 1 . 20 stating that "Canadian pol i t ics are far too complex to permi t a simple interpretat ion such as the 'balance' theory to expla in alternating party cho ice . " 3 9 Canadian pol i t ica l scientists have not always clearly d rawn the dist inct ion between this cyc l ica l mode l and the balancing mode l described earlier. For example George Per l in and Patt i Peppin consider Dawson 's statement to be a var iat ion on the ideas ident i f ied by U n d e r h i l l . 4 0 Th is type o f confus ion is concerning because i t fai ls to recognize the fundamental ly d i f ferent mot ivat ions that underl ie balancing and cyc l ing . Balancing is an intent ional act; cyc l ing is a unintent ional by-product o f the po l i t ica l system. Party ident i f icat ion obviously has l inks to vo t ing behaviour. G iven that Canadians have the opportuni ty to vote on more than one level , they often ident i fy w i t h po l i t i ca l parties d i f ferent ly across levels. Mar ianne Stewart and Haro ld Clarke model a relat ionship w h i c h takes these connections in to account. They include i n their model a retrospective evaluat ion o f party performance w h i c h they use to l i nk the federal and prov inc ia l arenas. Stewart and Clarke assert that " federal systems do not erect f i rewal ls that hermet ical ly isolate federal and prov inc ia l pol i t ics in people's m i n d s . " 4 1 I n keeping w i t h the publ ic choice l i terature, they argue that "a l though voters may apply contextual ly condi t ioned discount rates, they w i l l not ignore a subset o f in fo rmat ion they have gathered because i t concerns party performance i n an arena other than one i n wh ich part isan updat ing is occur r ing . " 4 2 Stewart and Clarke suggest that the dual party 3 9 Scarrow, 297. 4 0 George Per l in and Patt i Peppin, "Va r ia t i on i n Party Support i n Federal and Prov inc ia l E lec t ions" Canadian Journal of Political Science 4 (1971): 282. 4 1 Mar ianne Stewart and Haro ld Clarke, "The Dynamics o f Party Ident i f icat ion i n Federal Systems: the Canadian Case" American Journal of Political Science 42 (January, 1998): 99. 4 2 Stewart and Clarke, 106. 21 ident i f icat ion, that B lake argues is part o f what produces d iv ided elect ion results, is actual ly part o f the same instabi l i ty that characterizes the party systems over t ime. They see party ident i f icat ion as on ly one part o f the in fo rmat ion that voters take w i t h them to the pol ls. For Stewart and Clarke, negative retrospective evaluations o f a party at the other level can w o r k to produce d iv ided government as we l l as dual partisanships. I t is the combinat ion o f retrospective evaluations and staggered elections that produces the cyc l ica l relat ionship between the t w o levels. R ichard Johnston and Fred Cut ler also apply a cyc l ing mode l to d iv ided elect ion outcomes i n Canada, arguing that wh i le federal and prov inc ia l vo t ing decisions appear to be connected, the relat ionship is not one o f ba lancing. 4 3 They f i rst note that d iv ided government is a frequent occurrence in Canada and, as was noted in the in t roduct ion, not jus t a product o f regional po l i t i ca l differences. They ident i fy three possible theoretical explanations o f d iv ided government: by-product o f federal ism, balancing, r ise-and-decl ine. Us ing their dataset o f p rov inc ia l elections that f o l l o w federal elections since 1908, Johnston and Cut ler note that the swing f r o m a federal to a prov inc ia l elect ion w i t h i n a g iven prov ince is usual ly quite large. This suggests that there are some fundamental differences between the parties that run i n federal elections and those that run prov inc ia l l y . Such an insight is not inconsistent w i t h the party ident i f icat ion l i terature described above. Johnston and Cut ler el iminate much o f the d i f f i cu l t y presented by these separated party systems by using swings rather than straight vote shares. Th is technique is carr ied fo rward i n this paper fo r reasons described be low. W i t h 4 3 R ichard Johnston and Fred Cut ler, "Popular Foundations o f D i v i d e d Government i n Canada," presented to the 2003 Annua l Mee t ing o f the Canadian Pol i t ica l Science Associat ion, Dalhousie Univers i ty , Ha l i fax , 13. 22 the l im i ted dataset that they constructed consist ing on ly o f p rov inc ia l and federal swings for nat ional winners i n the past federal elections they test fo r evidence o f the aggregate effects o f balancing and cyc l ing . They conclude that "movemen t i n one arena is shadowed by l i ke movement i n the o ther . " 4 4 Furthermore, Johnston and Cut ler suggest understanding o f the causal reasoning behind this l inkage may be aided by i nd i v i dua l 1 level data. However as was noted above w i t h reference to the US d iv ided government l i terature, i t is not clear that such data could ever g ive researchers access to ind iv idua l reasoning that leads to the aggregate results o f d iv ided elect ion outcomes. Conclusion The above l i terature rev iew highl ights the variety o f possible explanations that have been advanced to expla in d iv ided government i n Canada, the Un i ted States and Germany. W h i l e the specif ic tests contained w i t h i n many o f these works m igh t not be easily repl icable i n the Canadian case, the next chapter w i l l attempt to draw out h o w each o f these theories can be interpreted on the ind iv idua l level because, at the end o f the day, vote decisions are made on the ind iv idua l level . I n mak ing the connect ion f r o m the theory to the ind iv idua l level decisions and then back up to the aggregate results, the paper w i l l f i l l i n some o f the gaps in the l i terature so far. The balancing mode l needs to be adapted to a three party, federal context. The party effects experienced by voters must be specif ied. Retrospective concerns and the general separation o f the t w o levels o f government must be taken in to account. I t is to these issues that the paper w i l l now turn. Johnston and Cutler, 13. 23 CHAPTER 3 - VOTER STRATEGIES Based on the l i terature rev iewed above, the paper w i l l n o w attempt to describe h o w ind iv idua l voters migh t react to the strategic envi ronment prov ided by d iv ided electoral opportuni t ies. W h i l e d iv ided government is an aggregate level phenomenon, vo t ing decisions that take place at the ind iv idua l level are its cause. Th is means that every theory o f d iv ided government must rest on some k i n d o f ind iv idua l - leve l process. There are a variety o f d i f ferent processes that can be d rawn out f r o m the theories presented above. These di f ferent strategies, once aggregated, predict very di f ferent patterns o f d iv ided and un i f ied government. Some o f these strategies (po l icy and interest balancing) i m p l y an intent ional choice consistent w i t h some port ions o f the l i terature wh i le others (evaluative and po l icy learning models) of fer a more unintent ional perspective. The po l icy balancing strategy is d rawn out f r o m much o f the balancing l i terature explored above. The interest balancing strategy represents an attempt to apply balancing to the three-party federal system in Canada. The t w o evaluat ive models (general and level-speci f ic) i m p l y that d iv ided government is the result o f staggered elections as voters change their retrospective evaluations over t ime. The po l i cy learning mode l also suggests that d iv ided outcomes are the result o f t im ing as voters use the gaps between elections to learn more about the po l icy preferences o f the compet ing parties. D i f fe rent voters use di f ferent strategies and as a result i t is d i f f i cu l t to determine f r o m overal l elect ion outcomes what dominant strategies ( i f any) exist. I n fact assuming that al l voters employ the same strategic calculus is a s impl i f i ca t ion that undermines much o f the previous w o r k in this area. This paper w i l l attempt to describe the various possible voter strategies and the thought processes o f the voters that employ those 24 strategies. The paper w i l l also describe the predicted aggregate results i f every voter used that strategy and the voters were normal ly distr ibuted across the po l i cy space. The goal is to determine whether any o f the patterns found w i t h i n the aggregate data are more consistent w i t h some strategies than w i t h others. Policy Balancing The f i rst strategy is der ived f r o m the Amer i can d iv ided government l i terature. I t is most common ly referred to as the po l icy balancing theory o f d iv ided government. This mode l at its most basic level consists o f two parties (cal l them D and R ) compet ing in a one-dimensional (usually Le f t -R igh t ) po l icy space. The po l i t i ca l parties, despite the logic o f the median voter theory, are assumed to be more extreme than the preferences o f moderate, centrist voters. 4 5 Speci f ical ly , party D is to the lef t o f the median voter wh i le party R is to the r ight. Behaviour o f voters to the left o f party D 's ideal po in t and voters to the r ight o f party R's ideal point is easy to predict. These voters w i l l obv iously select the party that is closest to them at every elect ion as they prefer po l i cy outcomes that are more extreme than the po l i t ica l parties espouse. I t is the behaviour o f voters located between the t w o parties that is considered important under balancing theory.' These voters prefer po l icy options that are more centrist than those presented by either party. I n any one elect ion these voters are forced to choose a party that is more extreme than they w o u l d actual ly prefer. I n any system, however, that permits people to vote more than once (such as a presidential system, a two-chamber system or a federal system) these moderate voters have the opt ion o f casting contradictory ballots. W i t h t w o elections, voters may choose opposite parties to produce outcomes w h i c h are closer to their 4 5 A les ina and Rosenthal, 16. 25 preferred locat ion. Th is theory rests on the idea that the t w o levels o f government being elected interact i n some way to produce po l icy outcomes so that voters can reasonably expect the winners at one level to check the winners at the other level . I n a three party system the simple logic o f this argument is more d i f f i cu l t to apply. Imagine again a one dimensional po l icy space w i t h three parties arranged r ight to left: party C, party L and party N. There are obviously st i l l voters whose po l i cy preferences are more extreme than those o f either party C or party N . A s in the t w o party model the votes o f these people are easy to predict ; they w i l l select the party that is closest to them. The voters located between the extremist parties C and N are st i l l the focus o f the balance theory. I t is not clear, however, how one w o u l d go about balancing po l i cy preferences. Obv ious ly the party i n the midd le , party L, is l i ke ly to benef i t f r o m the existence o f more moderate voters w h o w o u l d presumably choose them over either o f the extremist parties. Ye t the very existence o f this centrist, moderate party challenges the log ic o f the balancing argument. A les ina and Rosenthal 's claims rest on the idea that the part ies' ideal po l icy posit ions do not converge to those o f the median voter. A centrist party confounds that log ic . 4 6 However , i t is st i l l possible to imagine some need to balance fo r those voters whose ideal po l i cy point lies between party C and party L or between party L and party N . Assuming that al l voters use this po l icy balancing strategy and that voters ' preferences are distr ibuted over the range o f po l icy choices in a cont inuous, single-peaked fo rmat ion , the aggregate results w o u l d depend on a variety o f factors. First, the 4 6 M o r r i s F ior ina suggests that th i rd parties are in many ways the par l iamentary equivalent to t icket spl i t t ing, thus, i n a way, agreeing that balancing in a three party system is not as compel l ing an explanat ion. F ior ina, 121. 26 results w o u l d be affected by the relat ive posit ions o f the parties over the po l i cy space. For the purpose o f this example, we w i l l assume that the L ibera l par ty 's ideal po l i cy point is located at the median voter 's ideal point wh i le the Conservatives and N e w Democrats are located an equal distance to the r ight and lef t o f the median, respectively. Second, the results depend on the relat ive contr ibut ions o f each government to po l icy fo rmat ion . The relat ionship between the po l i cy capacities o f the t w o governments can be expressed in a simple equation: po l icy = ^( federal po l icy) + (1 -q)(provincial p o l i c y ) . 4 7 The easiest way to mode l po l icy outcomes wh i le a l low ing for both governments to be invo lved in the process is to assume that they both contr ibute equal ly, that is to say that q equals 0.5. Obv ious ly this is s imply an assumption; as w i l l be examined be low, the cont r ibut ion o f the di f ferent levels to po l i cy fo rmat ion varies by issue. Moreover the size o f one's prov ince affects the extent to w h i c h the prov inc ia l government can alter the national po l icy outcomes. W h i l e there are any number o f possible variat ions to be considered, the paper w i l l focus on t w o examples: one where an extreme party is i n power at the other level and one where the centrist party is i n power at the other level . For the purpose o f the example we w i l l consider the case o f a federal Conservat ive government f o l l o w e d by the case o f a federal L ibera l government. I n either case those voters located either to the r ight o f the Conservatives (point C) or the lef t o f the N D P (point N ) w i l l never vote for d iv ided government as they prefer a more extreme po l icy . W h e n the federal government is Conservat ive (Figure l ) 4 8 a l l those voters w h o are located to the lef t o f the Liberals (point 4 7 Equat ion adapted f r o m F ior ina , 74. 4 8 W h i l e bo th Figure 1 and 2 are modeled fo r when the federal and prov inc ia l governments share equal ly i n the overal l po l icy outcome, i t is easy to adjust the model to 27 L ) or are closer to the Liberals than the midpo in t between the Liberals and the Conservatives (to the lef t o f po in t Y ) w i l l vote N D P . Those voters w h o are closer to the mid -po in t between the Liberals and the Conservatives w i l l vote L ibera l (voters between Y and Z ) , wh i le those closer to the Conservatives than to the midpo in t (to the r ight o f po in t Z ) w i l l vote To ry . The voters that are engaged in balancing are those voters w h o vote N D P despite hav ing true preferences that are closer to the Liberals (voters located between points N L and Y ) and those voters w h o vote L ibera l despite hav ing true preferences that are closer to the Conservatives (voters located between points C L and Z ) . The aggregate outcome o f this elect ion w i l l be a N e w Democrat ic government because the bu l k o f the voters w i l l have selected the N D P either as part o f a straight t icket or a balancing calculat ion. W h e n the Liberals ho ld the federal government (Figure 2) , the results are less conclusive. A n equal number o f voters, located to the outside o f points X and Y , w i l l select either the N D P and the Conservatives depending on w h i c h side they are on. Those voters located inside points X and Y w i l l vote L ibera l . The overal l aggregate results w i l l therefore be hard to determine because wh i le the L ibera l party w i l l occupy less po l icy space, more voters w i l l be clustered in that space. The exact elect ion results w i l l depend on h o w t igh t ly clustered the voters are around the median voter and the L ibera l par ty 's ideal point . I f the voters are very t ight ly packed i t is possible fo r the Liberals to w i n despite ho ld ing very l i t t le po l icy space. The assumption that voters are distr ibuted symmetr ica l ly around the median is also part icular ly impor tant to determining the accommodate other possible q values. A l l one has to do is adjust pos i t ion ing o f the l ine labeled " t o t a l " to the appropriate rat io and the lines w i l l fa l l at the correct locat ion for interpretat ion. For examples see Figures 3 and 4 where q equals 0.8. 28 outcome in this case. I f either the left w i n g or the r ight w i n g is larger, they may we l l w i n the prov inc ia l elect ion when a centrist party holds the federal government. There is some c i rcular i ty to the patterns that emerge f r o m the balancing model . Once an extreme party takes ho ld , the balancing theory predicts that the other level w i l l adopt the other extremist party. Th is arrangement is an equ i l i b r ium; balancing theory does not predict any var iat ion f r o m the above pattern. Depending on the exact d is t r ibut ion o f voters, L ibera l governments at either level under this strategy cou ld either be quite stable or cou ld lead to m inor i t y government. Obv ious ly because not every voter employs this strategy the equ i l i b r ium i t produces is not necessarily what occurs nor is i t necessarily stable. Those voters using other strategies may swamp the po l icy balancers, but the basic balancing strategy remains the same. Interest Balancing A possible var iat ion on the po l icy balancing strategy that m igh t make more in tu i t ive sense w i t h i n the Canadian context takes into account the regional tensions inherent i n the federat ion. A s was noted above fo r the po l icy balancing v iewpo in t to have value, the overal l nat ional po l icy must be some combinat ion o f the po l icy preferences o f both the federal and prov inc ia l governments. I n some po l i cy areas, such as health care, h igher education or the envi ronment , this is l i ke ly to be true. However i n other areas the boundaries between federal and prov inc ia l ju r i sd ic t ion are clearer and more def ined, fo r example the c r im ina l code, wel fare or fore ign po l icy . Thus to balance po l i cy posit ions in the Canadian context requires even higher levels o f in fo rmat ion fo r such a strategy to be effect ive. One must k n o w w h i c h level o f government is responsible fo r what aspects o f the po l icy as w e l l as the po l i t i ca l part ies' posit ions on the po l i cy o f 29 interest. A less demanding version o f balancing can be d rawn out f r o m some o f the Canadian l i terature on executive federal ism. I n this strategy voters choose di f ferent po l i t ica l parties s imply because they are di f ferent. The idea is that i f the t w o governments are o f the same party then the prov inc ia l government m igh t be more w i l l i n g to compromise at the expense o f the prov ince 's nar rowly def ined interests. Voters using this strategy bel ieve that by selecting a di f ferent party at each level they ensure that no government w i l l f o l d early i n an ef fort to assist their part isan fr iends at the other level. D r a w i n g this argument out even further, Steven M u l l e r sees that there migh t be some benef i t i n elect ing a prov inc ia l party that has no federal w i n g . He asserts that "such an autonomous party, i n essential cont ro l over the province, is free to bargain w i t h both o f the contending parties at Ot tawa each t ime there is a D o m i n i o n e lec t ion . " 4 9 The fundamental tension in this strategy has very l i t t le to do w i t h le f t - r ight po l icy concerns and much more to do w i t h part icular ist ic us-versus-them issues. Voters w h o want to ensure that their prov ince always comes out ahead dur ing federal -provinc ia l negotiat ions w i l l always vote fo r a party that is not i n power at the other level regardless o f the po l icy posit ions taken by any o f the parties invo lved . That disregard fo r po l i cy orientations makes this strategy substantially less sophisticated than the po l i cy balancing strategy despite their apparent similar i t ies. Th is log ic , w h i c h encourages d iv is ion, can be turned on its head to encourage the match ing o f federal and prov inc ia l governments. Ian Stewart 's study o f consistency in Prince Edward Is land is part icular ly interesting. He notes that p rov inc ia l po l i t ica l parties have campaigned both on p lat forms that encourage d iv is ion and on p la t forms that 4 9 Steven M u l l e r , "Federa l ism and the Party System in Canada" i n Canadian Federalism: Myth or Reality, ed. J. Peter Meek ison , (Toronto: Metheun, 1968), 127. 30 encourage matching depending on whether their party is in power federal ly. Stewart f inds considerable histor ical evidence o f parties campaigning based on the importance o f hav ing match ing federal and prov inc ia l governments to ensure that the prov ince benefits f r o m the federal government . 5 1 The idea behind this matching log ic is that pol i t ic ians are more l i ke ly to go out o f their way to assist other pol i t ic ians f r o m their o w n party. Th is strategy o f match ing governments is not any more sophisticated than the strategy o f d iv id ing them. Nei ther strategy has anyth ing to do w i t h po l icy preferences or party ident i f icat ion. Stewart notes that wh i le i t m igh t be tempt ing to suggest that un i f ied government i n a g iven prov ince is the product o f consistent party ident i f icat ion and vo t ing , this is not always the case as not al l provinces have a large enough impact in the federat ion to sway the outcome. 5 2 This po in t w i l l be elaborated on i n the data analysis. I f every voter adopted this strategy the aggregate results w o u l d be quite easy to predict. I f everyone voted to d iv ide governments then y o u w o u l d never f i n d un i f ied government except i n smal l provinces and then on ly fo r the t ime between a federal e lect ion that produced match ing government and the subsequent p rov inc ia l elect ion. N o prov inc ia l elect ion w o u l d ever produce un i f ied government because the federal government is f i xed and prov inc ia l electorates have complete contro l over the elect ion outcomes i n their province. Federal elections w i l l on ly produce un i f ied government i n those smal l provinces w h i c h previously had parties i n government that were i n opposi t ion bo th i n the federal government and the larger provinces. Th is is because voters i n large provinces w i l l never choose, under this strategy, to swi tch their federal votes to a party 5 0 Ian Stewart, "Fr iends at Court : Federal ism and Prov inc ia l Elect ions on Prince Edward I s land" Canadian Journal of Political Science 19 (March 1986): 133. 5 1 Stewart, 135. 5 2 Stewart, 142. 31 that is in government prov inc ia l ly . I n a three party system there is always the possib i l i ty that d i f ferent provinces may choose di f ferent federal opposi t ion parties fo r their p rov inc ia l governments. Obv ious ly the possib i l i ty o f un i f ied government occurr ing this way w o u l d prov ide incentives for prov inc ia l electorates to choose parties that exist on ly on a prov inc ia l level . This theory therefore may go some way to expla in ing the dominance o f p rov inc ia l -on ly parties i n provinces that are part icular ly alienated f r o m the federal government. I n part icular this strategy may describe the dominance o f the Social Credi t party i n B C and A lber ta and the U n i o n Nat ionale and Parti Quebecois i n Quebec. I n contrast i f every voter pursued the strategy o f match ing parties to produce un i f ied government we w o u l d expect the exact opposite. N o prov inc ia l elect ion w o u l d ever produce d iv ided government fo r the same reasons as used above under the strategy o f deliberate d iv is ion. S imi la r ly , federal elections w o u l d on ly produce d iv ided government i n smal l provinces f o l l o w i n g the reverse logic o f the argument previously presented. Moreover the react ion to the other level fo r voters using this strategy should be immediate; as soon as government is un i f ied the voter should set out to d iv ide it. Th is impl ies that the gap between the elections is i rrelevant to these voters. W h i l e i t w o u l d appear that this strategy cannot possibly be at w o r k because prov inc ia l elections regular ly produce results opposite to those predicted, i t is important to remember that each o f these strategies cou ld to some extent be present i n the populat ion. Inconsistent results may just be the product o f those voters using this strategy being swamped by those voters pursuing other more popular strategies. 32-Evaluative Strategies M o v i n g away f r o m intent ional balancing strategies, the paper w i l l n o w explore the effects o f retrospective par ty /pol icy vot ing. Bo th strategies are variants o f the incumbent or iented retrospective vot ing strategies ident i f ied i n the U S l i terature. Voters using these strategies have adopted the simple decision rule that " i f something good occurs g ive the incumbents credit for i t ; i f something bad occurs b lame them for i t . " 5 3 There are t w o possible variants o f this strategy, one more complex than the other. The f i rst strategy is a general retrospective evaluat ion, ignor ing the differences between the federal and prov inc ia l governments. The second strategy requires more po l i t i ca l knowledge to dist inguish those aspects o f the po l i t ica l c l imate that are the responsibi l i ty o f the government up fo r elect ion but st i l l rests on voters' ind iv idua l evaluations o f " h o w things are" po l i t ica l ly . I n either case d iv is ion is caused by staggered elections. General evaluative strategy Us ing the f i rst s imple retrospective strategy each person judges h o w things are at the moment they go to the pol ls. The evaluat ion is general and broad. The voter using this strategy does not need to k n o w what pol ic ies they prefer or w h i c h parties w i l l del iver those preferred outcomes or w h i c h level o f government is responsible for the issues in quest ion; he or she must s imply decide whether they l ike the current po l i t i ca l c l imate or not. I f the voter is unsatisf ied w i t h the current si tuation he w i l l vote against the incumbent party. I f , on the other hand, the voter is satisfied she w i l l reward the incumbent party and vote fo r them. W h i l e voters i n this strategy show l i t t le awareness o f 5 3 D. Roder ick K iew ie t , Macroeconomics and Micropolitics: The Electoral Effects of Economic Issues, (Chicago: Un ivers i ty o f Chicago Press, 1983), 7. 33 the di f ferent responsibi l i t ies o f the t w o levels o f government, they are aware o f wh ich parties are in power i n either arena. They are jus t as l i ke ly to punish the federal L iberals because the prov inc ia l L iberals are in power as they are to punish the federal L iberals fo r their o w n actions. Predict ing the aggregate results when al l voters use this strategy is more d i f f i cu l t because o f the possible permutat ions. Consider ing four di f ferent scenarios, however, should prov ide enough detai l to understand the impl icat ions o f the strategy. The f i rst type o f si tuation occurs when the voters l ike the current c l imate and the federal and prov inc ia l governments are held by the same pol i t ica l parties. The choice fo r these voters is s imple; they reward the incumbent party (who jus t so happens to be the same as the incumbent on the other level) . The second scenario is ident ical to the f i rst , except the voters are dissatisf ied w i t h the current situation. I n a three party system they can choose either o f the t w o opposi t ion parties that are not i n government. Th is strategy provides no predict ions about wh ich o f the t w o parties the voters w i l l select because for voters emp loy ing this strategy i t does not matter. However , g iven their distaste fo r the current government such voters may vote fo r the stronger o f the t w o opposi t ion parties in an attempt to coordinate w i t h other voters to oust the government party. I n the th i rd scenario, the voter remains satisfied w i t h the po l i t ica l outcomes, but the federal and prov inc ia l governments are held by two di f ferent parties. W h i l e the strategy described above gives no hints as to wh ich party w i l l be rewarded, i t is l i ke ly that despite the general lack o f po l i t i ca l knowledge among these voters they w o u l d reward the correct incumbent at each elect ion rather than reward ing the party i n government at the other level . F ina l ly , when the governments are held by di f ferent parties and voters are 34 dissatisf ied, this theory predicts that the voters w i l l choose whatever party is not i n government at either level . The aggregate impl icat ions o f these four scenarios are di f ferent. Moreover because the differences between many o f the scenarios are merely the perceptions o f voters, i t w i l l be d i f f i cu l t to apply this theory to aggregate level data. W h a t is clear, however, is that d iv ided government i n this mode l is a func t ion o f the staggered t i m i n g o f elections permi t t ing voters' evaluations o f the po l i t i ca l c l imate to change between elections. Level specific evaluations A more sophisticated version o f this evaluat ive strategy permits voters to analyze w h i c h aspects o f the po l i cy envi ronment at elect ion t ime are the responsibi l i ty o f the government that is up fo r elect ion. That is to say that wh i le the voters have no specif ic po l i cy or party preferences, they can dist inguish between w h i c h governments are up fo r elect ion and w h i c h are not. These voters evaluate the po l i cy performance o f the government i n quest ion, ignor ing the effects or responsibi l i t ies o f the other level . I n the minds o f voters emp loy ing this strategy the t w o levels are separate and their electoral decisions are just as separate. D i v i d e d government, then, occurs when these voters ' analyses o f the po l i t i ca l cl imates o f each level are di f ferent. I n contrast to al l the strategies explored so far voters i n this theory do not draw a connect ion between the t w o arenas. Some o f the l i terature in Canada on party systems suggests that such a strategy m igh t not be unheard of. Dona ld B lake makes the case that because di f ferent issues dominate the t w o arenas voters make their decisions separately. W h i l e his argument is not necessarily retrospective i n nature, the voters he studied s imply drew a d is t inct ion between the concerns that were relevant to the federal elect ion and those that were 35 relevant prov inc ia l ly . The logic o f his argument is not d issimi lar to that i n the issue ownership l i terature in the Un i ted States wh ich suggests that some issues are best dealt w i t h by some parties, though Blake 's c la im migh t be more consistent w i t h earl ier formulat ions o f the issue ownership l i terature that imputed ownership to inst i tut ions rather than parties. The aggregate results o f this strategy are not as d i f f i cu l t to get a handle on as the results f r o m the overal l evaluative model yet they st i l l depend on voters ' decisions about h o w happy they are at the t ime o f the elect ion. Regardless o f the level o f the elect ion, voters using this strategy w i l l decide whether or not they are happy and then reward or punish the party in government accordingly. I n a three party system, when voters decide to punish the government they are st i l l lef t w i t h a choice between the t w o parties i n opposi t ion. Th is strategy does not suggest h o w voters w i l l choose between those t w o parties other than stating that the voters w i l l not refer to any in fo rmat ion gathered at the other level . I f , however, the voters using this strategy are happy w i t h the current government they w i l l s imply vote fo r the party i n power. Because o f the l imi ts o f the dataset, neither evaluative theory can be appropriately tested in this paper. Understanding the use o f both o f the theories requires some in fo rmat ion about voters ' levels o f satisfaction w i t h the government, and, fo r the second mode l , separate in fo rmat ion fo r both the federal and prov inc ia l governments. Nevertheless, i t is clear that these evaluat ive strategies combined w i t h the staggered nature o f Canadian federal and prov inc ia l elections may w e l l help create d iv ided government. 36 Cycling and Policy Learning The f ina l strategy that voters may use is the most complex yet at the same t ime the most in tu i t ive ly sat isfying. I n this strategy voters use in fo rmat ion f r o m both levels o f government to update their in fo rmat ion about po l i t ica l part ies' po l icy preferences and competencies. Voters also rely on their party ident i f icat ion to help them when in fo rmat ion about the parties is less readi ly available ( for example when the party is i n opposi t ion on both levels o f government) . This strategy is consistent w i t h the w o r k o f Stewart and Clarke. They suggest that i t is "sensible fo r voters w h o have paid in format ion-acquis i t ion and del iberat ion costs to evaluate party performance at a g iven level to use those evaluations when updat ing their party ident i f icat ions at the other l e v e l . " 5 4 W h e n faced w i t h a choice at elect ion t ime, these voters evaluate the pol ic ies o f bo th the federal and prov inc ia l governments to determine what the part ies' true po l icy orientations are. They also use this in fo rmat ion to determine whether or not the parties i n government are "competent " to govern. This strategy is more consistent w i t h the po l i cy -oriented retrospective vo t ing l i terature as these voters are concerned w i t h wh ich pol ic ies are enacted and not jus t what the incumbent is responsible for. These voters are more sophisticated because " instead o f s imply b laming the incumbent fo r any and a l l forms o f economic d i f f i cu l t y , po l icy-or iented voters support the party w h i c h places a higher p r io r i t y on attacking the part icular . . . p rob lem they are concerned w i t h . " 5 5 The in fo rmat ion that is available to these voters then, depends on w h i c h parties are in government and whether government is d iv ided or un i f ied at the t ime o f the elect ion. I f Stewart and Clarke, 113. K iew ie t , 8. 37 government is already divided then these voters have access to more information with which to make their decision than if government is unified. Examining a few possible scenarios should help explain how these voters make decisions. Consider the situation of a voter in a province with a Liberal government while the federal government is also Liberal. During the upcoming provincial election this voter has the choice of either voting Liberal, Conservative or NDP. Unfortunately because government is unified this voter does not benefit from any additional information acquired at the federal level. Nonetheless the voter has learned something about the policies supported by the Liberal party and their capacities to implement those policies. If this voter was using a simple evaluative strategy she would then decide whether or not she was happy and vote accordingly. However, this voter is interested in more than a mere referendum on government performance. So instead she uses the information she has gathered to update her understanding of the Liberal party's preferred policy. She then decides based on her own policy preferences (and assisted by her party identification) which party is the best fit. If the Liberal party in government has proven itself to be more leftist and the voter in question is a left leaning Liberal she will likely support them; however if she is a Conservative or right-wing Liberal she probably will not. Consider now the same voter in the same situation except that the federal government is now Conservative. The voter now has access to information about the policies and competencies of both the Conservatives and Liberals. She can now use this information to determine which parties are most consistent with her own preferences. Perhaps the provincial Liberals are left leaning, while the Conservatives are more 38 centrist. I f the voter prefers s l ight ly r ight o f centre pol ic ies she may choose to vote Conservat ive rather than L ibera l because she knows that the Conservat ive party is a better f i t for her preferences. The aggregate impl icat ions i f every voter used this strategy are l i ke ly less precise than those presented fo r the other voter strategies. Because every voter has di f ferent preferences, i t is d i f f i cu l t to understand how they w o u l d aggregate over a who le prov ince or country. However i t is l i ke ly that because voters using this strategy update their preferences based both on federal and prov inc ia l in fo rmat ion , the federal and prov inc ia l elect ion results w i l l to some extent track each other. Thus i t is these voters w h o learn f r o m both levels o f government that structure the aggregate cyc l ica l effects that many academics have noted. Cyc l i ng or the rise and decline model is explored in m u c h o f the Canadian l i terature, inc lud ing Dawson, Scarrow and Johnston and Cutler. The argument presented by these authors is that " c o m m o n forces. . . may pervade both arenas such that pressures for convergence exist, but the t im ing o f elections produces leads or lags i n the expression o f these forces and, at least temporar i ly , divergence between leve ls . " 5 6 I f voters use in fo rmat ion f r o m both arenas in mak ing their choices at elect ion t ime, the results o f the elections at both levels should be l inked in some way, un l ike in the four th mode l discussed where voters use in fo rmat ion on ly fo r the level f r o m w h i c h i t was gathered. Therefore when analyzing aggregate data we w o u l d expect to see some connections between the trends on the federal level and the trends on the prov inc ia l level . Because o f the irregular t im ing o f Canadian elections, i t w i l l never be clear wh ich level is Johnston and Cut ler, 5. 39 leading and wh ich level is f o l l o w i n g , nonetheless some cyc l ica l connect ion ought to be detected. Conclusion The f i ve strategies presented in this chapter out l ine the various ways that voters' choices can produce d iv ided government. The f i rst t w o models are strategies that in tent ional ly produce d iv ided government. Voters using those strategies prefer d iv ided government to un i f ied government fo r either po l icy or regional reasons. The other three models suggest that d iv ided government is more a product o f unintent ional t im ing effects than deliberate voter choice. 5 7 T w o o f these t i m i n g based models consider voters that see l inkages between the t w o levels o f government and choose parties accordingly. I n the other evaluat ive model , voters treat the t w o arenas as ent i rely separate. A s was noted earlier the effects o f these strategies w i l l be d i f f i cu l t to measure because o f the lack o f ind iv idua l level data. I t is possible that there are voters out there pursuing al l o f these di f ferent decision mak ing systems; however i f any one system is dominant i t is l i ke ly that the effects o f that dominant system w i l l swamp the results produced by the less popular strategies. Th is tension between intent ional and unintent ional models pervades much o f the scholarship i n this area. F ior ina makes a s imi lar d ist inct ion between the l i terature that suggests that d iv ided government is accidental and that wh ich suggests that i t is intent ional . F ior ina , 143. 40 C H A P T E R 4 - D A T A ANALYSIS The dataset analyzed in this thesis consists o f data on party vote shares and elect ion winners for the provinces and the federal government f r o m 1904 to June 2003 exc lud ing those elections for w h i c h no records could be found. The cases in the dataset consist o f e lect ion years w i t h i n provinces. Each prov inc ia l elect ion is a separate case and each federal elect ion produces 10 separate cases. The dependent variable is a measure o f vote sw ing for a g iven party f r o m one elect ion to the next on a g iven level . For example, one o f the cases o f the dependent variable is vote swing fo r the federal L iberals i n B C between 1997 and 2000. The independent variables are: • a measure o f the previous vote share (not swing) for the party o f interest at the level o f interest (PV) . Car ry ing on w i t h the previous example, this variable w o u l d be, fo r this case, the federal L ibera ls ' percentage o f vote at the f i rst elect ion i n the swing pair, that is, the 1997 elect ion, i n B C . • a measure o f the most recent vote swing fo r the party o f interest at the opposite level p r io r to the second elect ion in the sw ing pair (OPS). I n the example, this w o u l d be the vote sw ing fo r the L ibera l party in B C f r o m 1991 to 1996, the most recent pair o f elections that fal ls complete ly before the second elect ion in the federal swing pair (that is, the 2000 elect ion). • a d u m m y variable indicat ing whether at the t ime o f the second elect ion in the dependent variable swing pair, the party o f interest was in power at the opposite level (G) . F r o m the example, this variable w o u l d indicate whether the Liberals were in power i n B C in 2000. 41 • the gap since the last elect ion on the other level measured in months ( G A P ) • an interact ion between the swing on the other level and the gap since the last elect ion on the other level ( O P S * G A P ) • an interact ion between the other level swing and the d u m m y indicat ing whether the party o f interest was in power at the other level ( O P S * G ) • an interact ion between the gap since the last elect ion at the other level and the d u m m y indicat ing whether the party o f interest was in power at the other level ( G A P * G ) • an interact ion between the gap since the last elect ion at the other level , the d u m m y indicat ing whether the party o f interest was in power at the other level and the other level swing ( G A P * G * O P S ) . 5 9 Based on the research o f Johnston and Cutler, the data analysis contained w i th in this paper w i l l focus on vote swings rather than straight vote shares. Us ing vote swings for a part icular party on a g iven level controls for some o f the var iat ion in party strength across provinces. B y using vote swings rather than vote shares we ef fect ively el iminate much o f the regional ism in Canadian party systems. Vo te swings, however, do not cont ro l fo r the differences between federal and prov inc ia l party systems w i t h i n a g iven province. The mode l used be low assumes that party names and labels have some 5 8 The federal mode l was also estimated inc lud ing a quadratic measure o f t ime (gap since the last e lect ion on the other level squared). Th is was an attempt to account fo r the possib i l i ty that the relat ionship between t ime and the vote swing was not l inear due to the existence o f a " h o n e y m o o n " fo r new governments dur ing w h i c h voters learn posi t ive things about the party i n power. The coeff ic ient on this term was ins igni f icant fo r al l three parties, suggesting that the relat ionship w i t h t ime is not quadratic. 5 9 W h e n the mode l produced no signi f icant results, i t was re-estimated w i thou t the last three interact ion terms ( O P S * G , G A P * G and G A P * G * O P S ) in an attempt to determine whether the interactions were masking the main effects. 42 impor tant degree o f signif icance fo r voters even when parties o f the same name have very di f ferent po l icy preferences. That is to say, voters in B C treat the L ibera l Party o f Canada and the Br i t i sh Co lumb ia L ibera l Party as t w o parts o f the same pol i t ica l phenomenon. There are good reasons w h y voters w i t h very l o w levels o f po l i t ica l in fo rmat ion m igh t make this assumption. However w i t h even a smal l amount o f awareness about party pol ic ies i t becomes evident that the two parties have very di f ferent preferences. Whether or not treating parties o f the same name as the same invalidates the results o f this study is unclear. The processes captured in the models be low suggest that there is a relat ionship between nomina l ly s imi lar parties despite any po l i cy differences. I f the assumption is unfa i r then the net result w o u l d be to undermine the causal l inks established by the various theories studied rather than to inval idate the empir ica l conclusions about the causes o f d iv ided government. Thus the va l id i ty o f this assumption, wh i le impor tant in understanding h o w people go about vo t ing , is not cr i t ica l to understanding the aggregate level electoral causes o f d iv ided government. The data seem to indicate a connect ion in voters ' minds between parties o f the same name; how that connect ion got to be there is to a certain extent i rrelevant to understanding its effects on vo t ing patterns. The variables above were selected because they represent an ef for t to capture both intent ional and unintent ional d iv ided government effects. Clear ly the challenge here is to ident i fy possible aggregate trends that migh t indicate ind iv idua l voters ' thought processes. The d i f f i cu l t y is that aggregate outcomes can mask the variety o f d i f ferent patterns o f behaviour. The overal l vote swing in any g iven elect ion is the result o f several m i l l i o n ind iv idua l decisions. I t is possible that a s igni f icant amount o f intent ional 43 div is ion occurs but that these effects are swamped by w o r l d events or even larger numbers o f unintent ional decisions. T o a certain extent the single member p lura l i ty electoral system also distorts aggregate elect ion results as people's vo t ing decisions are altered by the landscape o f the race in their r id ings in a variety o f ways that are not necessarily connected to notions o f d iv ided government. Thus the testable impl icat ions presented be low represent generalizations about trends rather than specif ic in fo rmat ion about causal processes. Despite these aggregation problems, predict ions for the signs and signif icance o f each variable are contained w i t h i n Table 1. F r o m the table i t becomes clear that w i t h the data we have there is no useful way to dist inguish between po l icy and interest balancing as those theories predict the same results. These intent ional balancing strategies, at an aggregate level , l ook the same. A s a result both strategies w i l l be discussed together under the label o f intent ional balancing. The previous vote variable (PV) , though impor tant f r o m an empi r ica l perspective because i t a l lows parties to experience bo t toming and topping out (when a party has done part icular ly w e l l or part icular ly poor ly i t is l i ke ly that i n the next elect ion that t rend w i l l reverse), is not very he lp fu l i n determin ing w h i c h strategy is at w o r k fo r the predicted sign (negative) is the same for al l f i ve theories. Regardless o f the strategy employed by voters, parties are expected to experience a bo t toming or topping out effect over t ime. Therefore i t has been omi t ted f r o m both the table and the discussion. S imi la r ly the main effect o f the gap ( G A P ) since the last elect ion variable and the swing, gap, w i n interact ion ( G A P * G * O P S ) have no substantive importance and are therefore not discussed in the text. They are expected to be zero. The level specif ic model predicts no relat ionship for al l variables except the 44 previous vote on the level o f interest. Th is is because the voters emp loy ing that strategy should make no connections between the t w o levels. Because al l the regressions contain some signi f icant results, this model is also not discussed fur ther be low. The general evaluat ive mode l is very s imi lar to the po l icy learning model . W i t h the current dataset i t is impossible to dist inguish effects that are unique to i t as i t depends on voter evaluations o f the general we l l -be ing o f the Canadian po l i ty . A s a result discussion o f the general evaluat ive model w i l l be l im i ted . The cr i t ica l variables are those incorporat ing the other level sw ing variable (OPS) and the d u m m y variable ind icat ing whether the party was i n power at the other level (G) . The intent ional balance theory predicts that the coeff ic ient fo r the d u m m y variable G w i l l be negative suggesting that a party is punished more harshly i f i t is i n power at the other level . I n contrast, the po l i cy learning hypothesis suggests that the coeff ic ient w i l l be ins igni f icant or posi t ive ind icat ing a movement towards that party (a ' r ise ' ) when i t holds power at the other level . Moreover i t is possible that ho ld ing a prov inc ia l government w i l l confer benefits on that party i n terms o f organizat ion, fundrais ing power and pol i t ica l experience. The coeff ic ient fo r OPS using the po l icy learning model is expected to be posi t ive as i t is argued that a swing on the opposite level is l i ke ly to be matched by a s imi lar sw ing on the level o f interest as voters learn about the pol ic ies o f the various parties and update their preferences. The intent ional balancing theories prov ide no predic t ion as to the sign or signif icance o f OPS as the relat ionship between the t w o levels is cont ingent upon the other level winner. Thus the intent ional balancing theory predicts that the O P S * G interact ion w i l l be negative wh i le the po l i cy learning mode l predicts insigni f icance. 45 The effect o f the interact ion O P S * G A P is important to mode l ing the po l i cy learning hypothesis. The rise and decline model predicts that the coeff ic ient w i l l be negative because the larger the t ime dif ference between the two elections the more l ike ly the party w i l l have suffered a decline in the intervening months as voters update their in fo rmat ion about the party o f interest. I n contrast, the intent ional balancing theories predict that these t i m i n g measures w i l l be insigni f icant as the react ion to the other level should be immediate, as po l icy balancers and interest balancers do not need to learn about the preferences o f the various parties before they have a mot iva t ion to balance. A n y type o f lag or learning impl ies that voters are using the cyc l ica l strategy and not a strategy o f intent ional d iv is ion. Re in fo rc ing this relat ionship is the interact ion between the other level w inner d u m m y and the gap term ( G A P * G ) . The balancing theory predicts that this variable w i l l be ins igni f icant because the response should be immediate wh i le the learning mode l suggests that i t w i l l be either negative or posi t ive as voters learn more about the party when i t is i n government at the other level . Us ing these variables, six basic models must be constructed: one fo r each o f three di f ferent po l i t i ca l parties on t w o di f ferent levels. A s the l i terature rev iew suggests, the effects on the federal and prov inc ia l levels may we l l be di f ferent as the process o f updat ing in fo rmat ion on po l i t i ca l parties may not be symmetr ic over the two levels. I n part icular Ian Stewart notes that most provinces ' voters cannot real ly expect to contro l the outcome o f the federal elect ion. W h i l e his study focuses on the most extreme example o f this (Prince Edward Island w i t h on ly 4 seats), many Canadian voters in small provinces must realize that an attempt to balance their p rov inc ia l governments w i t h the federal government may not succeed as their vote is d i lu ted by the votes o f Canadians in 46 larger provinces w i t h potent ia l ly di f ferent balancing imperat ives. Voters therefore cannot cond i t ion al l their votes based on the expectat ion that the party they vote fo r w i l l acquire power over a l l the po l icy space. A t the same t ime, in fo rmat ion acquired by voters about their p rov inc ia l governments is l i ke ly to be more direct ly relevant and apparent to them as that government is closer to them, increasing the potent ial fo r balancing as they become more aware o f the prov inc ia l govern ing party 's preferences. S imi la r l y the relat ionships fo r the various po l i t i ca l parties might not be ident ical fo r a who le host o f reasons. A t the most obvious level because the federal N D P has never held government i t is impossible to ascribe balancing mot ivat ions to that par ty 's p rov inc ia l vote swings. The party separated models were estimated using Ordinary Least Squares (OLS) regression. The federal combined model was estimated using O L S w i t h dummies fo r each party and year interacted. The prov inc ia l combined mode l was estimated using O L S w i t h dummies fo r each party (year dummies are not necessary as each prov inc ia l elect ion is independent and therefore does not need to be contro l led fo r ) . There are a number o f possible assumption violat ions that result f r o m this decision. First, the dataset contains cases def ined by units o f t ime. Th is makes i t a f o r m o f t ime series. Moreover the model uses a variety o f t ime series techniques inc lud ing lagged variables and differences. A t the same t ime the cases are not measured at predetermined t ime intervals. The gap between elections can vary dramat ical ly and the gap since the last other level elect ion is s imi lar ly poor ly def ined. W h i l e one can s imply treat the elections as i f the exact t i m i n g does not matter, this ignores the fact that the decision to cal l an elect ion is a po l i t ica l choice based on the previous government 's percept ion o f its abi l i ty to w i n . Ye t this endogeneity is un l i ke ly to be a serious p rob lem because incumbent governments of ten 47 lose and therefore the ability to control the timing of an election is not an insurmountable advantage. Certain types of time series techniques will not work with this data but as long as we are conscious of the incomplete nature of the time series it should not be too difficult to deal with. The second concern relates to the independence of the cases. Properly understood the divided government dataset is a time series cross section with the set of provinces representing the panel measured at different times (i.e. the elections). By treating each province separately we are better able to model the relationship between provincial and federal elections. If we aggregated the provinces not only would we lose many cases but we would also lose the ability to test what we are interested in understanding. Unfortunately when studying the federal level this means treating the outcomes of the same election in different provinces as if they were independent. Even without referring to the data, it is clear that this assumption is probably not correct. This makes sense because federal elections are fought across the entire country on the same issues with the same leaders while provincial elections are internal affairs that do not tend to make news or effect politics outside the provincial borders. A scatterplot of vote swings across time seems to indicate that election results across the country are not fully independent. In fact the results for all the provinces tend to be closely grouped. In years when a particular party does well, they tend to do well in all the provinces and vice versa. Interestingly the grouping is most pronounced for the Progressive Conservative party and least evident for the Liberals. This correlation will again be visible in the regression residuals. It seems, however, that despite this violation the major results of the paper are still valid because the inclusion of dummy variables for each election year (controlling 48 fo r the overal l trends experienced across the country by a g iven party) does not alter the sign or statistical s ignif icance o f the cr i t ica l variables. The prov inc ia l level mode l faces no such d i f f i cu l t y as prov inc ia l elections are easily understood to be independent. The models that combine the results fo r al l three parties suffer some o f the same independence problems. W h e n al l the parties are combined together the O L S assumption is that their results are independent. Th is is clearly not the case because an increase in one par ty 's vote share necessitates a drop for another party. However this correlat ion is taken in to account by using a series o f d u m m y variables: one fo r each party i n the p rov inc ia l mode l (w i th the Conservatives as the base) and one fo r each party, year and party and year interacted in the federal mode l (w i th the Conservatives and 1904 as the bases). A d d i n g the elect ion year dummies to the federal mode l captures any part o f the relat ionship between vote swings in the ten provinces that migh t be part icular ly associated w i t h that elect ion year. Nevertheless, the assumption o f independence represents one o f the more fundamental challenges to mode l ing d iv ided government i n Canada. The paper w i l l now examine in detai l each o f the six basic models beginning w i t h the federal parties (L ibera l , Conservat ive and N D P ) f o l l o w e d by their p rov inc ia l counterparts. Then the paper w i l l examine t w o models that combine the effects fo r al l three parties fo r each level . These models w i l l re inforce the conclusions d rawn f r o m the party-separated models. Federal Models The mode l fo r the federal L iberals is easily the most robust o f al l six. The coeff ic ients are consistent w i t h an unintent ional po l icy learning mode l o f d iv ided 49 government. H o l d i n g a prov inc ia l government has a posi t ive effect on federal L ibera l results. I n those provinces where the Liberals were in power they cou ld expect posi t ive increase in their elect ion to elect ion swing o f around 3% i f the t w o elections were close enough in t ime. A s the gap between the elections increased that effect decl ined as voters learned about the pol ic ies o f their p rov inc ia l L ibera l governments. The prov inc ia l swings are also posi t ive ly related to federal swings, a key l i nk i n the cyc l ica l model . A n upward trend prov inc ia l l y tends to translate in to an upward trend federal ly w i t h a one percent b u m p at the lower level predict ing a 0.61 percent b u m p at the nat ional level . Unsurpr is ing ly , this relat ionship grows weaker as the distance in t ime between the two elections increases. Th is is v is ib le f r o m the negative effect o f the interact ion term: the larger the gap i n months between the elections the less relevant the in fo rmat ion voters gathered about that early choice is to the decision at hand. Moreover , the gap al lows voters to gain some in fo rmat ion about the po l icy preferences o f the L ibera l party. Thus i f their votes are based on programmat ic concerns a larger gap cou ld produce greater changes in party po l icy or greater voter knowledge o f party po l icy . Fur thermore, after about twen ty - two months the effect o f p rov inc ia l sw ing reverses, suggesting that voters swi tch their understanding o f the party the longer the gap between elections. The relat ionship between prov inc ia l swing cont ingent on ho ld ing the prov inc ia l government and the federal sw ing was ins igni f icant and therefore not consistent w i t h the intent ional balancing strategy. The results fo r the federal Conservatives are very s imi lar to those fo r the Liberals. Interest ingly, results f r o m the previous elect ion are more impor tant to predict ing the current sw ing , suggesting that Conservatives experience s l ight ly stronger topping out 50 effects than the other parties. A s w i t h the Liberals, Conservat ive prov inc ia l governments are associated w i t h posi t ive swings in federal vote share. Conservat ive federal governments experience twice the b u m p at the pol ls f r o m ho ld ing prov inc ia l governments as Liberals (7 .4%). This result is therefore fundamental ly inconsistent w i t h an intent ional or balance explanat ion o f d iv ided government patterns. Prov inc ia l swing is again posi t ive ly related to federal swing. W h e n prov inc ia l Conservatives receive a one percent b u m p at the pol ls , their federal counterparts can expect an increase o f about a th i rd as m u c h (0 .33%). U n l i k e fo r the Liberals, this relat ionship does not appear to vary over t ime as the interact ion term ( though negative) is insigni f icant . Th is suggests that voters are not as f i ck le i n their support fo r the Conservatives as distance between elections seems to have a smaller effect on their vo t ing choices. The interact ion between ho ld ing prov inc ia l governments and prov inc ia l swing is again inconsistent w i t h balancing explanations as i t is not negative ( i t is ins igni f icant) . O n the other hand the gap-w in interact ion is also ins igni f icant , a result wh ich is consistent w i t h balancing explanations. Voters do not seem to learn more about federal Conservatives the longer they ho ld prov inc ia l governments. The N D P exhibi ts some o f the most stable support out o f the three parties studied. The relat ionship between vote swings and previous elect ion totals is weakest fo r the N D P possibly as N e w Democrat ic support is re lat ively consistent over t ime. The N e w Democrats receive a b u m p f r o m ho ld ing prov inc ia l governments in i t ia l ly ( though the effect is not s igni f icant) but over t ime this effect reverses quite strongly as voters learn about N D P governments. Th is effect is captured by the interact ion between gap since the last other level elect ion and the.other level w inner d u m m y . Over the course o f t w o years 51 the N D P is predicted to go f r o m a 3% posi t ive bump to a 3% penalty fo r ho ld ing a p rov inc ia l government. Th is suggests that voters do not l i ke what they get f r o m N D P governments. They learn about the party and then reevaluate their choice at the other level . A g a i n this s igni f icant effect is inconsistent w i t h intent ional balancing. A s w i t h the Liberals and Conservatives, p rov inc ia l swing is posi t ive ly related to federal swing patterns. Prov inc ia l swing is w o r t h about 2 8 % on the federal level : a one percent shif t in p rov inc ia l swing w i l l produce a federal swing o f about 0 .28% in the same direct ion. The N D P does experience some decline in that relat ionship as the distance between the t w o elections grows. However the decl ine is not suff ic ient to produce a reversal i n the relat ionship, as the curve tops out after more than six years. The general pattern, then, appears to be that swings across levels are posi t ive ly related. H o l d i n g a prov inc ia l government provides a boost to the federal party immediate ly but this effect declines and even reverses over t ime. The effect o f other level swing is not cont ingent upon whether the party is i n power. The predict ive value o f other level sw ing seems to decline over t ime. A l l these conclusions are more consistent w i t h po l i cy learning strategies than w i t h balancing theories. Provincial Models Turn ing now to the prov inc ia l level , three further models were estimated. Beg inn ing again w i t h the Liberals, i t is clear that the prov inc ia l picture is quite di f ferent f r o m the federal version. First, al l the substantive variables are insigni f icant. W h a t this seems to suggest is that the prov inc ia l L iberals do in fact experience some level specif ic effects that are not captured in the model o f d iv ided outcomes. The previous vote share has the same sign as in the federal model but is smaller, suggesting that bo t toming out 52 effects are less dramatic on the prov inc ia l scene. Th is may ref lect the tendency fo r provinces to be dominated by one party to the exclusion o f others fo r a long stretch o f t ime. M o r e interesting is that federal vote swings appear to be unrelated to prov inc ia l vote swings. The coeff ic ient fo r the relevant variable is ins igni f icant on its o w n and interacted w i t h the gap since the last federal elect ion or the w i n d u m m y . I t is this relat ionship that Er ikson and F i l i ppov model . W h i l e they use di f ferent regressions (federal to p rov inc ia l swings as we l l as using only one independent variable, the d u m m y fo r l iberal federal incumbency) they also f i n d support fo r the intent ional balancing model . Est imat ing this mode l w i thou t the sw ing -w in , gap-win and gap-sw ing-w in interactions produces results that are rough ly consistent w i t h the results o f Er ikson and F i l ippov . W h e n the Liberals are in government federal ly, p rov inc ia l L iberals can expect to suffer a penalty o f around 3.5%. This effect is s igni f icant and i t contradicts the logic o f the cyc l ica l theory o f d iv ided government. Th is i n combinat ion w i t h insigni f icance o f the federal swing seems to suggest that the prov inc ia l level mode l is very d i f ferent f r o m the federal one. W h i l e the federal L ibera l party seems to prov ide robust support fo r an unintent ional mode l o f d iv ided government, p rov inc ia l L iberals seem to prov ide an excel lent example o f intent ional balancing effects. Whether these penalties accrue because o f interest balancing or po l icy balancing cannot be determined f r o m the aggregate results. I n the Conservat ive party p rov inc ia l model the previous vote, gap and gap-w in interact ion are s igni f icant at the 5% conf idence level . The effect o f previous vote share on current sw ing is s imi lar i n size and di rect ion to the prov inc ia l L ibera l model . A g a i n this variable has a much smaller effect than on the federal level . Previous federal 53 Conservat ive vote swing is posi t ive ly related to prov inc ia l vote sw ing ; when the interactions are dropped the effect is signif icant. A one point decrease in federal vote is predicted to produce a 0.24 po in t decrease in p rov inc ia l vote. Th is relat ionship is again consistent w i t h a cyc l ing model o f unintent ional ly d iv ided government. I n the l im i ted model the other level w inner d u m m y is negative and signi f icant however i n the expanded mode l this effect seems to be captured by the gap-win interact ion. This interact ion suggests that when the Conservatives ho ld the federal government voters learn that they do not l i ke them over t ime. A g a i n , this s igni f icant result is consistent w i t h the cyc l ing or po l i cy learning models and not the balancing explanations. F ina l ly , the prov inc ia l N e w Democrat ic mode l is quite d i f ferent f r o m the other parties and levels. A s the N D P has never held government federal ly, the d u m m y variable and its interactions are dropped f r o m the mode l as they contain no var iat ion. The federal swing and previous prov inc ia l swing terms are signi f icant at the 5% conf idence level . As a result the on ly in fo rmat ion gained f r o m the model is that N D P vote swings are inversely related to previous vote shares, the now- fami l ia r bo t toming out effect. These results are d i f f i cu l t to analyze using the models o f intent ional and unintent ional d iv ided government because o f the l imi tat ions o f the data. For the N D P at least i t seems there is a substantial disconnect between the federal elections and the prov inc ia l results, at least f r o m the prov inc ia l angle. Combined Models The combined mode l produces results that are rough ly consistent w i t h the party separated models. For the federal case the mode l was estimated using dummies for party, year and party interacted w i t h year to account fo r the lack o f independence ident i f ied 54 above. I n the combined model , on ly the coeff ic ients measuring the effect o f the previous federal vote, the prov inc ia l swing, the other level winner , the interact ion between the gap since the last p rov inc ia l elect ion and the prov inc ia l swing and the interact ion between the gap and the other level w inner d u m m y are signif icant. The return on prov inc ia l swing at the federal level is around 34%. The d u m m y indicat ing whether the party is i n power p rov inc ia l l y is s igni f icant and posit ive. However over t ime this effect reverses, consistent w i t h the idea that voters learn about parties when they ho ld government at the other level and then apply that knowledge to the level o f interest. Return ing to the strategies out l ined above, on ly the cyc l ing po l icy learning mode l and the general evaluat ive mode l predict vote swings to be related. Balancing theories require the d u m m y to be signi f icant and negative. Thus i t appears that, f r o m the federal angle at least, unintent ional d iv ided government models are more persuasive. The interact ion term, w i t h a coef f ic ient o f -0.009, suggests that the in fo rmat ion used to cast a bal lot in a p rov inc ia l elect ion, becomes irrelevant to the federal choice after about 3 years. Thus the combined mode l produces results that, wh i le not f u l l y consistent w i t h the separated versions, re inforce the conclusion that intent ional balancing models lack explanatory power . 6 0 The combined prov inc ia l model incorporated on ly dummies fo r each party because o f the inherent independence between prov inc ia l elections he ld i n di f ferent provinces. The combined mode l reinforces the connect ion between federal and prov inc ia l vote swings w i t h a return o f about 2 2 % on previous federal swing in a p rov inc ia l elect ion. The model also displays the usual bo t toming out effects. T o test 6 0 I t is impor tant to remember that because this model incorporates a large number o f dummies, the r-square is to a certain extent ar t i f ic ia l ly in f la ted. 55 f u l l y the idea that vote swings in the balancing model are cont ingent upon the other level w inner , we look to the interact ion between the other level swing and the other level winner. A s w i t h the federal party separated model , the results were not s igni f icant and as a result not consistent w i t h the balancing theory w h i c h predicts that they w o u l d be negat ively related. Moreover the posi t ive ( i f ins igni f icant) coef f ic ient fo r the other level w inner combined w i t h a negative coeff ic ient fo r the gap and winner interact ion term is on ly t ru ly consistent w i t h the po l icy learning model as voters change their minds over t ime rather than responding immediate ly by balancing. Conclusion Aggregate data, then, suggests that intent ional balancing models are not the best way to understand d iv ided outcomes in Canadian federal and prov inc ia l elections. I n most models estimated federal and prov inc ia l swings were related. Th is appears to e l iminate any explanat ion, such as the level specif ic evaluat ive model , that does not a l low fo r the levels to be connected. Moreover i n every case except the prov inc ia l L iberals, inc lud ing the ins igni f icant ones, the relat ionship was posi t ive ind icat ing that federal and prov inc ia l results track one another. The results produced by the d u m m y indicat ing when the party being analyzed was in power at the other level produced the least consistent results. I n some cases balancing explanations seemed to be supported but as has been previously described, balancing requires not on ly an aggregate penalty but also a inverse relat ionship between the swings once the party in power at the other level is taken in to account. W h i l e no aggregate results can fu l l y determine w h i c h strategies are at work , i t appears that the po l i cy learning model may be more dominant i n the publ ic to the extent that i t produces aggregate effects. 56 CHAPTER 5 - CONCLUSION D i v i d e d government i n Canada is not as w ide ly studied as d iv ided government i n the Un i ted States. W h i l e some o f the early US l i terature cites Canada as the source o f some o f the theoretical f rameworks under study, the l iteratures d iverged as d iv ided government became more and more c o m m o n in the US. Ye t part isan d iv is ion between federal and prov inc ia l governments continues to be quite c o m m o n in Canada. Voters of ten select di f ferent parties for their federal and prov inc ia l governments. This paper has presented a variety o f voter strategies that might produce this outcome. W h i l e not every voter employs the same strategy in the same way , these strategies can be combined to produce d iv ided government. Aggregate level elect ion results were analyzed to show that wh i le no f i r m conclusions can be reached, i t w o u l d appear that prov inc ia l and federal elect ion swings track each other over t ime. Th is impl ies that d iv ided government is more l i ke ly the result o f unintent ional voter decisions rather than intent ional balancing efforts. The paper advances the argument that f r o m the l i terature rev iewed there are several possible voter strategies at w o r k across the electorate. N o t every voter must respond to mu l t ip le elections in the same way. The paper specif ied f i ve possible strategies and the aggregate impl icat ions o f each. The f i rst set o f t w o strategies suggest that voters intent ional ly balance part ies' strength in one arena w i t h weakness in the other. Th is is done either to produce more moderate po l icy outcomes, as in the f i rst po l icy balancing strategy, or to produce d iv is ion at the federal -provinc ia l bargain ing table, as in the second interest balancing strategy. The next set o f t w o voter strategies i m p l y more s imple retrospective evaluations either evaluat ing the po l icy env i ronment as a whole , in the th i rd overal l evaluat ive strategy, or separating the responsibi l i t ies o f the t w o arenas, 57 i n the four th level specif ic evaluat ive strategy. F ina l ly , some voters l i ke ly use strategies that recognize the effects o f party ident i f icat ion on voter choice. Th is l inkage was recognized in the f i f t h po l icy learning based strategy. Perhaps most impor tant ly the thesis recognizes that al l the strategies may we l l be at w o r k i n the populat ion at large. The paper makes use o f the best available aggregate data set to determine whether some voter strategies are more clearly v is ib le (and therefore more dominant) than others. The data analysis incorporates both levels and the three nat ional po l i t i ca l parties: the Conservat ives, the Liberals and the N e w Democrats. Swings across levels appear to be posi t ive ly related, an effect that is not cont ingent upon wh ich party is i n power at the other level . That is to say, federal and prov inc ia l results track each other over t ime as the po l icy learning mode l predicts. N o party at either level ever suffers a statist ically s igni f icant penalty fo r ho ld ing government at the other level immediate ly upon elect ion. However , over t ime most parties begin to lose support as voters learn more about their pol ic ies. W h i l e the po l icy learning perspective does not predict or require that al l learning be negative, the data suggests that this is i n fact the case. Thus the overal l conclusions o f the paper suggest that the dominant strategy is one o f po l i cy learning. W h i l e the paper takes advantage o f much o f the in fo rmat ion contained w i t h i n the dataset, i t is possible that there is a better way to capture the relat ionship between federal and prov inc ia l elections w i thout los ing any o f the meaning contained w i t h i n these models. M o r e compl icated models that f u l l y recognize that a loss fo r one party necessarily requires a gain fo r another w o u l d do a better j o b o f understanding the relat ionships between the elections. A model wh ich permi t ted us to dist inguish between the general evaluat ive model and the po l icy learning model w o u l d also be valuable. Such 58 a mode l w o u l d l i ke ly require in fo rmat ion on the economy or some other measure o f general po l i t ica l performance o f the government. The mode l w o u l d also have to focus more exp l ic i t l y on parties i n government rather than al l parties. Nonetheless, i t seems we have exhausted the possible conclusions that can be reached f r o m aggregate data alone. I n order to understand more fu l l y how prov inc ia l and federal vot ing decisions are related (or not related), po l i t ica l scientists need to start asking voters w h y they behave the way they do. The strategies described in this paper represent an ef for t to begin this study. Ind iv idua l level survey data w i l l not be a panacea. Determin ing the dif ference between the various strategies w i l l be d i f f i cu l t because i t w i l l require that accurate in fo rmat ion be col lected f r o m voters on their past choices in both arenas. The differences between the strategies are sometimes subtle mak ing i t hard to design survey questions that w o u l d tap the ideas that mot ivate d iv ided government choices. Moreover there is l i t t le agreement among US scholars on what are the best questions to ask to determine reasons and propensity to vote for d i f ferent parties fo r d i f ferent posit ions. The study o f d iv ided government, however, remains an excel lent way to approach the puzzle o f vo t ing patterns across Canada. 59 APPENDIX 1 - FIGURES FIGURE 1: CONSERVATIVE FEDERAL GOVERNMENT Total policy outcome = ^ federal policy) + (1-^(provincial policy) Note: For Figures 1 and 2, q=0.5 Fed Total W NL ' Y , 'CL Z ' L < Prov NDP Liberal Conservat ive FIGURE 2 : LIBERAL FEDERAL GOVERNMENT Fed Total W , 'NL X Y C L \ Z Prov NDP Liberal Conservat ive F I G U R E 3: C O N S E R V A T I V E F E D E R A L G O V E R N M E N T Total policy outcome = ^ (federal policy) + (l-g)(provincial policy) Note: For Figures 3 and 4, g=0.8 Fed Total ' W NL X ' Y ,"CL -Z ^ N L / C Prov NDP Liberal Conservative F I G U R E 4: L I B E R A L F E D E R A L G O V E R N M E N T Fed Total W NL -X Y\ CL Z Prov NDP Liberal Conservative Z O H U Q w OH Q O o r * I ti < i i X £ to 13 03 > _J l i o l > on ai > ai i—I LH ai .fl o o 60 c 00 X •4-> 1/1 03 i—I o fl 13 > 1) O 60 cl 00 l i > Oi I-I I-I (U . f l o t-l l i N ^3 c 03 o cl 6p oo Cl ai > 60 ai ai N c 03 O c 60 •rn oo Cl Hi _> 03 60 ai Ci -fl Ci e o fl ai OH \ C 2, o 5 ai S TH P Hi a e Oi > o CH 60 00 c OH CQ l i LSI c Ci fl 60 9i » 60 1> O VH 4i SI c 03 O '8 60 •l—I 00 Cl Hi > 60 l i 2 C i o c S Ji ^ 60 * S SH fl * e Hi b od o Ci OH 60 C H £ .S 60 C *-» -i-i oo Ci <D Cl S3 £ 5 pq o i-H Hi e c 03 C i c 60 Oi _> '•4-» 03 60 l i GO O CH I-I o o IH GO O OH Cl o Hi CH O 2; c o •I o '-3 ai CH O l i _> '-4-» 03 60 l i c o •a o a CH O l i _> '•*-> 'GO O OH l i s ^ 8 ^ ai > o I-I o N ^3 Cl 03 Ci fl 60 GO fl O l i N c 03 O c 60 O I-I l i N "£> c 03 C i C fl O l i C 03 O c 60 • 1—I oo C O I-I fl fl 60 • i—i oo C Ci <D l i > CH ai X > J? ai > CN ai o 8 IH o I-I ai N ai X> o -a i i •*-» o -a ai t-l CH oo c _o *H—» Ci ai 03 ai XI O 2 ai ^2 fl T3 I-I 8 § ai > IH ai XI V-l . O ai fl 03 RT 60 (30 1) rH C - f l 03 ^ rS " 03 •> X3 S -S IH -rH l i •3 CH l i 03 ft M W VH oo o O O > ' a ^ o 00 • .2 ^ > CH £ 1 ft* s 73 u fa CM Q O Z ti Q ti ti l fN ti ti M H 64 B 73 >-CU ti u CM DJ C '% II t>0 73 u , & T J *> I ti 73 i . „ w * * * g N rr> vo <*5 V© (N fxl O fN CM fN o O © * cn ^ vo O ft Tf o © o o 5 o d d f; as © o © p © S * ^ t— oo * _ o ^ o © w a _^ O CM © P * ^ ON cn © o 1 1 © w CO © © © » tN — S — 2 » o 2 - O o o © o o o © S. ° S. © S-( T ) CM * _ * _ r> i - N » TT p- Id vo <_l a F~ £2 P K © w 0 w * ^ — fN © oo S o ea m = — © o S o S O © o © o 5 S- © S ° S © w w w i w TT -a-i s in oo i / l P _: o vo o © ON i l © w VO p 2 o © o © p ~i O VO VO t> © o © d VB £• S £ g-o © -o © o 9 d © d ^ d 1-H tS i-l o © " s © * ^ ft S 3 CM . o © w 3 s H -•> 3 ft 2 fN p s - ft n f^  S iti <--© cn JN vo A O fN S o tN o © o © o P o o ft !C? fN 2 S 3 ~ o © '-' * _ vo & VO 5 fN P P. S © 8 ° CM © P " O VO ~ © o p a 8 8 8 f? r n CM © a I 8 © o © p. 00 fN , •—. Cr f*> — OV #sJ — § © S3 d w © o d 4—> o > 00 s Ji pq 00 ° P oo c > s CL, W X w X > p oo —I J X w X w J J o o oo oo o « oo c3 00 •J J X J a O a O ca ca o a w oo § ^ c O o U •* P 00 tN "^ t IT) OO fN C N fN OO CN ^ CN ^ z a cn vo o d V o bp 'c^ 1 XI CL) O o CL) •4-1 O c <u T3 1/3 1^ •a C/3 O z I—* I—H u z o 0$ OH I fO W H 2 "3 e > O t-OH OH Q Z 6J C c/5 *4 CM § s. © S © $ £" © o © o o p « p. 1 S © o <u a a. o u a a, o >-T3 a. a, o i-S3 2 © — © q <-? p. ft* c '% I  M is "3 c o I  OH U e ft is c '> o 1 -13 I f NC m — !-H P © S ITi p 5 2 «i d © w § P © S | <F § « § 8 § S 5 o © o © S © o © S • . CM © N© ^ r~ © r~ ro u~i in CM «s © o © © © s r o IQ 00 -f. 1 s © o © p. ON ~ OS 2 © CM © P. * co © O 9 © P © © s © oo o r o CM © p * _ V> O co 5 p • o OO CM t~- CM £j CM 2 m pn o CO CM © O S •* CT) —i © o S o © o © o © o © s © s. ° s in ON * _ 00 S CM r-!_i o © so © p, 5 © o © OS OS © © o OS CM CO ND CM « © © OS ,-H * O NO i/) in © © 9 © SS r o rx o iri © q © o J*J © o © P . o > 3 O '> Ml C w w co oo co O CO O g w co O c CO r 8 OH W X w *J co « co cS CO c3 CO J J J J r 8 X •a PH BH 60 W X W CH O CO CO O « CO 3 co CO •a W u CO * CO J J X J CH O CH O ca O C3 NO in NO O 8 ^ NO NO 3 ^ CN r-< CM CM 2 OH1 so o o V C~H C3 C3 O fl SH C3 XI c 4) O o 1) o fl 1) T3 •c TABLE 4 - FEDERAL COMBINED MODEL Federal Swing Variable Previous Vote -0.1393* OLS SE (0.0227) Provincial Swing 0.3410* OLS SE (0.0696) Other Level Winner 0.0226* OLS SE (0.0109) Gap Last 0.0003 OLS SE (0.0002) Gap Last X Prov Swing -0.0095* OLS SE (0.0024) Prov Swing X Prov Win -0.0448 OLS SE (0.1116) Gap Last X Prov Win -0.0012* OLS SE (0.0003) Gap X Prov Swing X Prov Win 0.0018 OLS SE (0.0040) Liberal Dummy 0.0017 OLS SE (0.0818) NDP Dummy -0.0456 OLS SE (0.0272) Constant 0.0574 OLS SE (0.0589) N 668 R 2 1 .74 A l s o inc luded i n the mode l were dummies for each year except 1904 and each year and party interacted (except 1904 and the Conservatives). The F-statistic for these dummies was 17.27 w i t h 70 and 590 degrees o f f reedom, ind icat ing that the dummies were h igh ly signif icant. The inc lus ion o f these dummies also resulted i n a substantial increase in the R 2 . Aster isks denote coeff ic ients that are signi f icant at /?<0.05. TABLE 5 - PROVINCIAL COMBINED MODEL Provincial Swing Variable Previous Vote -0.1473* OLS SE (0.0218) Federal Swing 0.2239* OLS SE (0.0965) Other Level Winner 0.0104 OLS SE (0.0149) Gap Last 0.0008* OLS SE (0.0002) Gap Last X Fed Swing -0.0030 OLS SE (0.0032) Fed Swing X Fed Win -0.0505 OLS SE (0.1369) Gap Last X Fed Win -0.0018* OLS SE (0.0004) Gap X Fed Swing X Fed Win 0.0052 OLS SE (0.0045) Liberal Dummy 0.0090 OLS SE (0.0088) NDP Dummy -0.0331* OLS SE (0.0100) Constant 0.0425* O L S S E (0.0119) N 617 R 2 1 .18 Aster isks denote coeff ic ients that are signi f icant at p<0.05. 66 BIBLIOGRAPHY Ales ina, A lber to and H o w a r d Rosenthal. Partisan Politics, Divided Government and the Economy. Cambr idge: Cambr idge Univers i ty Press, 1995. B lake, Dona ld w i t h D a v i d E lk ins and Richard Johnston. Two Political Worlds: Parties and Voting in British Columbia. Vancouver: U B C Press, 1985. Burden, Barry and D a v i d K i m b a l l . Why Americans Split Their Tickets: Campaigns, Competition and Divided Government. A n n Arbor : Un ivers i ty o f M i c h i g a n Press, 2002. Dawson , R. MacGregor . The Government of Canada. Toronto : Un ivers i ty o f Toronto Press, 1947. Er ikson, Robert . " T h e Puzzle o f M i d t e r m Loss" Journal of Politics 50 (1988): 1011-29. Er ikson, Robert and M i k h a i l F i l ippov . "E lectora l Balanc ing i n Federal and Sub-nat ional Elect ions: The Case o f Canada" Constitutional Political Economy 12 (2001): 313-331 . F ior ina, Mor r i s . Divided Government. N e w Y o r k : M a c m i l l a n , 1992. Garand, James and M a r c i Glascock L i ch t l . "Exp la in ing D i v i d e d Government i n the Un i ted States: Test ing an Intent ional M o d e l o f Spl i t - t icket V o t i n g " British Journal of Political Science 30 (2000): 173-91. Jacobson, Gary. The Electoral Origins of Divided Government: Competition in US House Elections, 1946-1988. Boulder , Co lo : Westv iew Press, 1990. . " D i v i d e d Government in the 1994 Elect ions." I n Divided Government: Change, Uncertainty, and the Constitutional Order, ed. Peter Galder is i . Lanham, M D : R o w m a n and L i t t l e f ie ld , 1996. Johnston, Richard. "Federal and Prov inc ia l V o t i n g : Contemporary Patterns and His tor ica l E v o l u t i o n . " I n Small Worlds: Provinces and Parties in Canadian Political Life eds. D a v i d E lk ins and Richard Simeon, 131-178. Toronto: Methuen , 1980. Johnston, Richard and Fred Cutler. "Popular Foundations o f D i v i d e d Government in Canada." Presented to the 2003 Annua l Meet ing o f the Canadian Pol i t ical Science Associat ion, Dalhousie Univers i ty , Ha l i fax . K iew ie t , D. Roder ick. Macroeconomics and Micropolitics: The Electoral Effects of Economic Issues. Chicago: Un ivers i ty o f Chicago Press, 1983. 67 Ladd , Everett Car l l . "Pub l ic Op in ion and the 'Congress P r o b l e m ' " The Public Interest 100 (1990): 57-67. Lew is -Beck , M ichae l and Richard Nadeau. "Sp l i t -T icke t V o t i n g : The Effects o f Cogn i t i ve Mad ison ian ism" Journal of Politics 66 (February 2004) : 97-112. L o h m a n n , Susanne, D a v i d Brady, and Douglas Rivers. "Par ty Ident i f ica t ion, retrospective V o t i n g , and Modera t ing Elect ions i n a Federal System: West Germany 1961-1989" Comparative Political Studies 30 (August 1997): 420-449. Mebane, Wal te r and Jasjeet Sekhon. "Coord ina t ion and Pol icy Modera t ion at M i d t e r m " American Political Science Review 96 (2002): 141-57. Mu l l e r , Stephen. "Federa l ism and the Party System in Canada." I n Canadian Federalism: Myth or Reality ed. J. Peter Meek ison , 119-34. Toronto : Metheun, 1968. Per l in , George and Patt i Peppin. "Va r ia t i on i n Party Support in Federal and Prov inc ia l E lect ions" Canadian Journal of Political Science 4 (1971): 280-6. Petrocik, John and Joseph Doher ty . " T h e Road to D i v i d e d Government : Paved W i t h o u t In ten t ion" I n Divided Government: Change, Uncertainty, and the Constitutional Order, ed. Peter Galder is i . Lanham, M D : R o w m a n and L i t t l e f i e ld , 1996. Reeves, B i l l and Roger Gibbins. "The Balance Theory: A n Empi r i ca l L o o k at the Interdependency o f Federal-Provincial Electoral Behav iour . " Presented to the 1976 A n n u a l Mee t ing o f the Canadian Pol i t ica l Science Associat ion, Quebec Ci ty . Scarrow, Howard . "Federal -Provinc ia l V o t i n g Patterns in Canada." Canadian Journal of Economics and Political Science 26 (1960): 289-98. Smi th , Charles, Robert B r o w n , John Bruce and L. M . Overby. "Par ty Balancing and V o t i n g for Congress i n the 1996 Nat iona l E lec t ion" American Journal of Political Science 43 (1999): 737-64. Stewart, Ian. "Fr iends at Court : Federal ism and Prov inc ia l Elect ions on Prince Edward I s land" Canadian Journal of Political Science 19 (1986) : 127-50. Stewart, Mar ianne and Haro ld Clarke. " T h e Dynamics o f Party Ident i f icat ion in Federal Systems: the Canadian Case" American Journal of Political Science 42 (January, 1998): 97-116. 68 Underb i l l , Frank. "Canadian L ibera l Democracy i n 1955." I n Press and Party in Canada eds. G.F. Ferguson and F.H. Underh i l l , 27-46. Toronto : Ryerson, 1955. W r o n g , Dennis. " T h e Pattern o f Party V o t i n g in Canada" Public Opinion Quarterly 19 (1957): 252-64. 69 

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.831.1-0091652/manifest

Comment

Related Items