UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Essays on political participation and the quality of democracy Held, Alexander 2019

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

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata


24-ubc_2019_september_held_alexander.pdf [ 2.9MB ]
JSON: 24-1.0380576.json
JSON-LD: 24-1.0380576-ld.json
RDF/XML (Pretty): 24-1.0380576-rdf.xml
RDF/JSON: 24-1.0380576-rdf.json
Turtle: 24-1.0380576-turtle.txt
N-Triples: 24-1.0380576-rdf-ntriples.txt
Original Record: 24-1.0380576-source.json
Full Text

Full Text

Essays on Political Participation and the Quality ofDemocracybyAlexander HeldA THESIS SUBMITTED IN PARTIAL FULFILLMENTOF THE REQUIREMENTS FOR THE DEGREE OFDoctor of PhilosophyinTHE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES(Political Science)The University of British Columbia(Vancouver)August 2019c© Alexander Held, 2019The following individuals certify that they have read, and recommend to the Faculty of Grad-uate and Postdoctoral Studies for acceptance, the thesis entitled:Essays on Political Participation and the Quality of Democracysubmitted by Alexander Held in partial fulfillment of the requirements for the degree of Doc-tor of Philosophy in Political Science.Examining Committee:Alan M. Jacobs, Political ScienceSupervisorRichard G. C. Johnston, Political ScienceSupervisory Committee MemberFrederick E. Cutler, Political ScienceUniversity ExaminerM. Marit Rehavi, EconomicsUniversity ExaminerAdditional Supervisory Committee Members:Christopher Kam, Political ScienceSupervisory Committee MemberiiAbstractIn advanced industrial democracies, a substantial number of citizens feel alienated from main-stream politics and political elites. This dissertation analyzes factors that help mitigate twocrucial aspects of citizen political alienation in these countries: political disengagement by anincreasing segment of the electorate, especially poor and young voters; and the turn to radicalalternatives such as far-right populist politicians and parties.Study 1 assesses the effect of procedural information costs - in particular, uncertainty aboutwhether one has to be 18 by the registration deadline or by Election Day - on youth voterturnout across U.S. states. Using a regression discontinuity design with official state voterrecords and leveraging a discontinuity in voter turnout around the registration deadline, thisstudy shows that uncertainty about the registration requirements for first-time voters depressesvoter turnout both in the immediate, and also in subsequent, elections among this group ofvoters, turning many of them into habitual non-voters.Study 2 takes a new look at the relationship between levels of political participation andsupport for left-wing parties and policies. It reanalyzes a critical case - Australia in theearly 20th century - frequently cited as a strong demonstration of such a relationship. Basedon an original and more fine-grained dataset of district electoral data in combination with adifference-in-differences design, this study tests the robustness of the previously found rela-tionship and investigates its mechanisms.Study 3 uses survey experimentation to test the responsiveness of populist voters to main-stream political messages. Based on a large-scale survey experiment with the polling firmiiiYouGov shortly before the 2017 German federal election, it finds that emphasizing the goodperformance of the German economy was the most effective strategy to increase support forthe incumbent Christian Democrats among likely rightwing populist voters.Overall, these findings speak to ongoing debates about the ability of politicians to shapecitizens’ political behavior. Improving on previous quantitative research in this area, this re-search highlights the limitations of institutional fixes and provides new insights into the role ofprocedural information and political framing for civic engagement.ivLay SummaryA increasing number of citizens feel alienated from mainstream politics and political elites inadvanced industrial democracies. This dissertation uses state-of-the-art quantitative techniquesof survey experimentation and quasi-experiments to analyze factors that help mitigate twocrucial aspects of citizen political alienation in these countries: political disengagement by anincreasing segment of the electorate, especially poor and young voters; and the turn to radicalalternatives such as far-right populist politicians and parties. Overall, this dissertation points tolimitations of institutional fixes such as compulsory voting laws to address issues of economicinequality and highlights the crucial role of procedural information and performance-basedcampaign messages to increase civic engagement in established democracies.vPrefaceThis dissertation is original, unpublished, independent work by the author, A. Held. The surveyreported in Chapter 4 was covered by UBC Ethics Certificate number H17-01708.viTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiLay Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiiiAcknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 Turning 18, but Not Quite: The Impact of Procedural Information Costs onTurnout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62.2 Procedural Information Costs and Political Participation . . . . . . . . . . . . . 102.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142.3.1 Visualizing Registration Deadlines and Turnout Patterns . . . . . . . . 162.4 Identification Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22vii2.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252.5.1 Persistence of Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . 272.5.2 Alternative Explanations: Formal Rules and Concurrent Administra-tive Acts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353 The Electoral Consequences of Increasing Voter Turnout: New Evidence froma Natural Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383.2 Compulsory Voting, Politically Disengaged Citizens, and Partisan Support . . . 433.3 Compulsory Voting in Australia . . . . . . . . . . . . . . . . . . . . . . . . . 483.4 Data and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513.4.1 Data and Dependent Variables . . . . . . . . . . . . . . . . . . . . . . 513.4.2 Research Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 533.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 593.5.1 State-Level Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 593.5.2 Federal-Level Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 673.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 724 Can Potential Populist Voters Be Framed? Results from a Survey Experimentin the 2017 German Election . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 764.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 764.2 Populist Voters and Mainstream Non-Immigration Messages . . . . . . . . . . 794.2.1 Performance, Policies, and Uncertainty . . . . . . . . . . . . . . . . . 844.3 Data and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 884.3.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88viii4.3.2 Dependent Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . 914.3.3 Independent Variables . . . . . . . . . . . . . . . . . . . . . . . . . . 914.3.4 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 944.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 964.4.1 Exploratory Analyses: The Role of Political Knowledge . . . . . . . . 1004.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1035 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111A Supporting Materials to Chapter 2 . . . . . . . . . . . . . . . . . . . . . . . . . . 124A.1 Smoothness of Density around Registration Deadline Cutoff . . . . . . . . . . 124A.2 Registration Deadline Discontinuities in Texas . . . . . . . . . . . . . . . . . . 125A.3 Separate Treatment Around School Entry Cutoff . . . . . . . . . . . . . . . . . 127A.4 Discontinuity in Daily Vote Totals for Florida . . . . . . . . . . . . . . . . . . 128A.5 Robustness of RD Estimates to Alternative Bandwidths . . . . . . . . . . . . . 129A.6 Robustness of RD Estimates to Alternative Model Specifications . . . . . . . . 130A.7 Robustness of Persistence of Effect to Varying Bandwidths . . . . . . . . . . . 134A.8 Information on Registration Requirements Across U.S. States . . . . . . . . . . 135A.9 Voter Registration Form from Florida . . . . . . . . . . . . . . . . . . . . . . . 137B Supporting Materials to Chapter 3 . . . . . . . . . . . . . . . . . . . . . . . . . . 139B.1 Within-State Analysis of Treatment Effects . . . . . . . . . . . . . . . . . . . 139B.2 Within-State Analysis of Changes in Turnout and Labor Seat Share . . . . . . . 141C Supporting Materials to Chapter 4 . . . . . . . . . . . . . . . . . . . . . . . . . . 145C.1 Question Wording of Treatments (English Translation) . . . . . . . . . . . . . 145ixC.2 Question Wording of Treatments in Survey (German Original) . . . . . . . . . 146C.3 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148C.4 Results with Alternative Dependent Variable: Party Favorability . . . . . . . . 149C.5 Results with Continuous Political Knowledge Variable . . . . . . . . . . . . . 149xList of TablesTable 2.1 Turnout in U.S. Presidential Elections for 1990 Birth Cohorts . . . . . . . . 28Table 2.2 CACE of 2008 Incorrect Procedural Beliefs on Downstream Voting . . . . . 30Table 3.1 The Adoption of Compulsory Voting in State and Commonwealth ElectionsAcross Australia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49Table 3.2 Effect of Compulsory Voting in State Assembly Elections . . . . . . . . . . 60Table 3.3 Effect of Compulsory Voting Using Within-State Variation in Treatment In-tensity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63Table 3.4 Effect of Compulsory Voting in Commonwealth House of RepresentativesElections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71Table 4.1 Expected Effect on Electoral Support . . . . . . . . . . . . . . . . . . . . . 95Table 4.2 Vote Intention in 2017 German Federal Election . . . . . . . . . . . . . . . 98Table 4.3 2017 Vote Intention by Level of Political Knowledge . . . . . . . . . . . . 102Table A.1 Turnout in U.S. Presidential Elections for 1990 Birth Cohorts . . . . . . . . 131Table A.2 Turnout in U.S. Presidential Elections . . . . . . . . . . . . . . . . . . . . 132Table A.3 Voter Registration Information Across Seven U.S. States . . . . . . . . . . 137Table B.1 Effects of Compulsory Voting Using Within-State Variation in TreatmentIntensity - Non-Labor Candidates . . . . . . . . . . . . . . . . . . . . . . . 140xiTable B.2 Effects of Compulsory Voting Using Within-State Variation in TreatmentIntensity - Independent Candidates . . . . . . . . . . . . . . . . . . . . . . 140Table C.1 Party Favorability Rating in 2017 German Federal Election . . . . . . . . . 150Table C.2 Models for 2017 Vote Intention with Continuous Political Knowledge Vari-able . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151xiiList of FiguresFigure 2.1 Turnout in Florida for People Born in 1986/87 and 1990/91 . . . . . . . . 18Figure 2.2 Registration Deadline Discontinuities Across U.S. States in 2008 Election . 21Figure 2.3 Registration Timing for Floridians born Feb 4-Aug 3, 1990 . . . . . . . . . 33Figure 3.1 Change in Turnout and ALP Vote Share in New South Wales, 1927-1930 . 64Figure 3.2 Change in Turnout and ALP Vote Share in Tasmania, 1928-1931 . . . . . . 66Figure 3.3 Voter Turnout in State and Federal State Assembly Elections, by State . . . 69Figure 4.1 Share of Respondents Intending to Vote for Mainstream Parties, by Treat-ment Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99Figure 4.2 Share of Respondents Intending to Vote for the AfD, by Political Knowl-edge and Treatment Condition . . . . . . . . . . . . . . . . . . . . . . . . 103Figure A.1 Distribution of Births in Florida between May 5, 1990 and May 4, 1991 . . 125Figure A.2 Turnout in Presidential Elections in Texas . . . . . . . . . . . . . . . . . . 126Figure A.3 Turnout in 2012 Presidential Election in Ohio . . . . . . . . . . . . . . . . 127Figure A.4 Votes in 2008 and 2012 Presidential Elections in Florida - Excluding Week-ends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129Figure A.5 Votes in 2008 and 2012 Presidential Elections in Florida - Including Week-ends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130xiiiFigure A.6 Varying Bandwidths Around Registration Deadline Cutoff . . . . . . . . . 134Figure A.7 Voter Registration Form from Florida . . . . . . . . . . . . . . . . . . . . 138Figure B.1 Change in Turnout and ALP Vote Share in Queensland, 1912-1915 . . . . 141Figure B.2 Change in Turnout and ALP Vote Share in South Australia, 1941-1944 . . 142Figure B.3 Change in Turnout and ALP Vote Share in Victoria, 1924-1927 . . . . . . 143Figure B.4 Change in Turnout and ALP Vote Share in Western Australia, 1936-1939 . 144Figure C.1 Effect of Economic Voting Treatment on AfD Support . . . . . . . . . . . 150Figure C.2 Effect of Social Inequality Treatment on AfD Support . . . . . . . . . . . 152xivAcknowledgmentsDuring my PhD studies I was fortunate to be surrounded by a fantastic group of individualsand exemplary scholars. My dissertation committee members Alan Jacobs, Dick Johnstonand Chris Kam provided superb guidance along the various stages of my PhD, were constantsources of inspiration and true role models. For all the incredible support that they provided tome over the years, I am forever grateful.Alan has accompanied my years in graduate school like no other. During all this time, hewas not only my research supervisor, but a true mentor to me. His generous advice, constantsupport and sharp comments on my research did not only improve my research, but they alsohelped me to grow as a scholar. We had many discussions about the current state and futuredirection of the discipline – ranging from issues such as research transparency and replication topreregistration and experimental research – that shaped my thinking as an empirical researcher.For all the support and encouragements throughout graduate school, I thank him with all myheart.I had my first extended discussion about voting behavior with Dick when I was at theend of my second year and studied for my comprehensive examinations. Without question,Dick is one of the most knowledgeable individuals on anything survey- and campaign-relatedthat I have ever met, but he is also one of the most good-humoured and supportive. Both hisintellectual and financial support were crucial when I set out to field my own survey experimentshortly before the 2017 German federal election. I am extremely lucky to have him on mycommittee and thank him for all his support.xvAt the end of my first year in graduate school, I took Chris’ course in advanced quantitativemethods. This was a transformative experience for me. It lay the groundwork for my keeninterest in quantitative methods and played a crucial part in shaping the trajectory of my futureresearch. Chris has been a constant source of advice about my research and the profession, hesupported parts of my research financially when no other sources were available and offeredme the opportunity of co-authorship. For all this, I am extremely grateful.I thank Greg Huber and Ben Cashore for being my mentors during my time as a visitingpre-doctoral fellow at Yale University. They gave me excellent advice on my research andhelped me make this stay a truly rewarding experience. Antje Ellermann and Yves Tiberghienprovided excellent research support and mentorship during the early stages of my PhD, forwhich I am grateful. Antje, as director of the Institute for European Studies, also providedme with study space, which allowed me to be part of the vibrant community of graduate andpostdoctoral students at the institute. I thank Rob Franzese and Andrew Owen for introducingme to the teaching of quantitative methods in Political Science. They were truly outstandingrole models and I learnt a great deal from them. A special thank you to Tom Cornwall andMatt Graham. Tom was my study companion and hiking buddy at UBC. There is no topic inmy dissertation that Tom has not commented on. Matt provided excellent feedback during thefinal stages of my dissertation writing. Their comments and critical questions challenged meto think harder about my research and were of tremendous help in improving it.I would like to thank the professors, graduate students and staff at UBC and Yale Universitywho have supported me in my academic endeavor. Several people have commented on orprovided advice on aspects of my research, among them particularly David Armstrong, PeterAronow, Andre Blais, Alexander Coppock, Cesi Cruz, Fred Cutler, Sonke Ehret, Cees vander Eijk, Jeremy Ferwerda, Alexander Fouirnaies, Alan Gerber, Joshua Gottlieb, Gyung-HoJeong, Josh Kalla, Thomas Lemieux, Winston Lin, Molly Offer-Westort, Marit Rehavi, DanielRubenson, Fredrik Savje, Campbell Sharman and Francesco Trebbi. Part of my research hasbeen presented to audiences at the University of British Columbia, Yale University, TrinityCollege Dublin, Oxford University, the University of Mannheim, the NYU-CESS Conference,xviand the Midwest Political Science Association and American Political Science AssociationAnnual Meetings, and I am grateful to all the comments that I received on my research and thathelped me sharpen my argument.I am lucky to have been surrounded by a wonderful group of fellow graduate students. Iespecially thank Yoel Kornreich, Pascale Massot, Miriam Matejova, Andrea Nuesser, BrentSutton, Dominik Stecula, Daniel Westlake, Sule Yaylaci, Salta Zhumatova and my fellow col-leagues at the IES (Chase Foster, Sara Pavan, Conrad King) at UBC and Scott Bokemper, Gau-tam Nair, Lilla Orr, Kyle Payton, Eli Rau, Joan Ricart-Huguet, Johannes Wiedemann, JenniferWu and the other Fox International Fellows at Yale University.I would like to thank Alison Fox, the Fox Fellowship staff and Yale University for givingme the opportunity to spend an unforgettable year as a Fox Fellow at Yale University duringwhich I could focus on finalizing my dissertation. I gratefully acknowledge financial supportfor my research from the Public Scholars Initiative (PSI), the Department of Political Science,the Faculty of Arts (all at UBC) and the Humboldt Foundation (PI: Richard Johnston). Gen-erous funding to present parts of this research at conferences was provided by the Institute forEuropean Studies (UBC), the Department of Political Science (UBC), the American PoliticalScience Association, and Yale University. I am grateful to Antony Green and Campbell Shar-man for sharing data and to staff in the Departments of State and Departments of Health inFlorida, New York, Ohio, Pennsylvania and Texas for providing access to voting records andbirth data. Any published findings and conclusions are those of the author and do not neces-sarily represent the official position of the Departments of State or Departments of Health inthese states. I thank Arian Zand and Bahja Alammari for excellent research assistance.A special thank you to Christian Breunig who played a key role in my decision to apply tograduate school.Last but certainly not least, I would like to thank my parents, Eva and Michael, for theirlove, support and patience.xviiChapter 1IntroductionOn June 23, 2016 a majority of 51.9% of British voters cast their ballot in favor of leavingthe European Union in the 2016 EU Membership referendum. The outcome of the referendumtook most politicians and political observers in Britain as well as abroad by surprise (Evansand Menon 2017, ch. 3). It sent shock waves across British politics, has led to a politicalstalemate in Westminister and is expected to have a fundamental impact on both future politicsand the economy in Britain. There is little doubt that a significant number of British votersused the referendum to vent their dissatisfaction with the political establishment. However, thephenomenon of citizen political alienation is not limited to the British Isles. Events such asthe election of Donald Trump in late 2016 or recent electoral successes of right-wing populistparties in several European countries have shown the extent to which this trend affects estab-lished democracies more generally. As a result, citizens, politicians and political scientistsalike have focused on similar questions: What explains the frustration of an increasing numberof citizens with their political elites that so suddenly erupted in Britain’s referendum vote onJune 23, 2016? Would election and referendum outcomes in these countries look different ifgroups with traditionally low levels of political participation – such as the young or the socioe-conomically worse off – voted in larger numbers? What strategies can politicians both on themoderate political left and right adopt to regain the trust of disaffected citizens?The goal of the research that is presented in the following pages is to help contribute an-1swers to at least some of these questions. In three separate, but related papers, this dissertationuses state-of-the-art econometric methods of causal inference and survey experimentation toadvance our understanding of two crucial aspects of citizen alienation in advanced industrialdemocracies: political disengagement by an increasing segment of the electorate, especiallypoor and young voters; and the turn to radical alternatives such as far-right politicians andparties.Study 1 focuses on youth voter turnout in the United States. It addresses a literature thatuses large administrative datasets in combination with regression discontinuity (RD) designsto draw inferences about the effects of policy reforms on voting rates. Such studies typicallycompare voter turnout of young Americans whose 18th birthday falls shortly before ElectionDay with those who turn 18 shortly after. Using data from official state voter files from 18U.S. states, this study shows that for many states there is a second consequential discontinuityin voter turnout that is not around the Election Day, but around the registration deadline 30days earlier. It also shows that this discontinuity has serious policy implications. Proceduralinformation costs that are associated with this deadline, and in particular the incorrect beliefthat one has to be 18 not by Election Day, but by the registration deadline, reduce turnoutamong certain young voters in some states by up to 25% (or 9 percentage points) in U.S.presidential elections.Study 2 leverages an arguably exogenous shock to voter turnout due to the introduction ofcompulsory voting laws in Australia in the early 20th century to study the relationship betweenlevels of political participation and support for left-wing parties and policies. Several studiesfind a direct link between higher levels of voter turnout and the electoral success of left-wingparties, suggesting that an institutional fix like the introduction of mandatory voting could haveserious political consequences. This study improves on previous quasi-experimental researchby using a unique dataset of more fine-grained electoral district election data in combinationwith a difference-in-differences design to provide a more direct empirical test of the assumedrelationship. Based on results from this analysis, it shows that empirical support for the previ-ously claimed association is weaker and more mixed than previously assumed. It also identifies2other factors that affect voter turnout – such as a party’s strategic choice in which districts itwill run candidates and the geographic size of a district – and whose analysis might give usa fuller picture of the conditions under which higher turnout is likely to shift power towardleft-wing parties.Study 3 presents a theoretical framework that outlines the conditions under which main-stream messages may overcome populist citizens’ feeling of resentment toward the estab-lished elite and shift their vote intentions. While populist voters might be attracted by theanti-immigrant rhetoric of right-wing populist politicians, there are other policy considerations(e.g. economic redistribution, performance of the economy) that are of interest to this groupof voters and where mainstream parties can credibly claim to better represent the interests ofpopulist voters. It tests hypotheses on the effect of messages about economic performance,welfare policies, strategic considerations, and uncertainty on these people’s vote intentionsthrough survey experimentation on a targeted sample of 1,800 citizens with a high propensityfor supporting anti-immigrant populist parties during the 2017 German election. The findingssuggest that populist voters are hardly swayed by these mainstream messages. However, someof them are responsive to messages that highlight the good performance of the German econ-omy, increasing support for the incumbent Christian Democrats among this group of votersby about 6 percentage points. Overall, these results provide new insights into the potential offraming strategies for shaping citizen political behavior in democracies.Beyond their substantive contribution to understanding citizen alienation, all three studiesshare two further features: they all rely on new and large data sources, and they apply soundcausal identification strategies to the research on politically disaffected citizens.Despite their recent growth across advanced industrial democracies, politically disaffectedcitizens still represent a small group of the population in most countries. In addition to their rel-atively small group size, disaffected citizens represent two additional challenges to researcherswho want to study them and who rely on traditional, nationally representative public opinionsurveys: 1) given their generally lower interest in politics and their lower socioeconomic sta-tus, they are less likely to be included in these surveys to begin with; and 2) even if they are3included, it is hard to identify them as a group of citizens because they do not share a clearlydistinctive feature that makes them easily identifiable as “politically alienated.” For these rea-sons, there are limitations to the questions about these voters than can be answered with datafrom existing, nationally representative surveys. The three studies in this dissertation broadenthe study of disaffected citizens by identifying new data sources that can be used to studytheir behavior quantitatively. These studies illustrate how large and unique data sources suchas administrative voter registration records, fine-grained historical election returns or an origi-nal survey experiment promise to shed new light on our understanding of disaffected citizens’engagement with politics.The study of voter turnout has recently seen a shift toward experimental and quasi-experimentalresearch designs that improve on traditional observational studies in this field by providing amore convincing empirical test of the effect of electoral reforms. Two papers of my dissertationdirectly speak to this literature by highlighting potential challenges to or limitations of some ofthese recent quasi-experimental studies themselves. Study 1 finds an additional discontinuityin voter turnout in the vicinity of the Election Day cutoff. This should caution researchers whouse regression discontinuity designs in combination with voter file data and a person’s dateof birth to take this discontinuity seriously and account for it in their study designs in orderto avoid confounding through multiple treatments. Study 2 illustrates how a seemingly minormodification to a difference-in-differences design – the move from a cross-state to a within-state analysis – leads to a very different outcome. However, these findings are neither meantas a challenge to these studies nor to the use of quasi-experimental research designs more gen-erally. Instead, they highlight the benefits of reflecting upon – and empirically testing – theassumptions that are used in these quasi-experimental research designs.Taken together, the findings from all three studies paint a picture of politicians’ ability toshape alienated citizens’ political behavior that neither warrants exaggerated optimism nor isit reason to despair. They show that carefully crafted electoral reforms and policies are able toreach at least some of the politically disaffected citizens. They also show that these reformsmake these voters more likely to go to the polls and, for some of them, to vote according to4their material self-interest on Election Day. However, the studies also point to the limitationsof such interventions, which are frequently less effective than commonly thought, and theyhighlight some unintended – and potentially unwanted – consequences of electoral reforms.5Chapter 2Turning 18, but Not Quite: The Impact ofProcedural Information Costs on Turnout2.1 IntroductionFormal institutional rules determine who can vote in elections and how. Extensive research hasshown that these rules - from voter ID laws (Hajnal, Lajevardi and Nielson 2017; Grimmer et al.2018) to voter registration requirements (Burden et al. 2014; Wolfinger and Rosenstone 1980)- have a direct impact on voter turnout. In contrast, we know relatively little about the beliefsthat citizens hold about these rules and how these beliefs affect their likelihood to participatein elections. Yet, beliefs about procedural aspects of elections play an important role. Citi-zens in the U.S. are increasingly concerned about so-called disinformation campaigns. Thesecampaigns, which are frequently attributed to radical right or foreign sources (Bennett and Liv-ingston 2018), intentionally spread incorrect or misleading information among people from theopposing side in order to harm moderate politicians or the Democratic party (Bennett and Liv-ingston 2018; Kim et al. 2018). While most of the recent academic debate has focused on fakenews and misleading reporting about issues such as immigration or crime in the context of the2016 presidential election (Guess, Nyhan and Reifler 2018; Kim et al. 2018), disinformationcampaigns also include targeted attempts to use incorrect procedural information to deter vot-6ers from the opponent to cast a vote (Stringer 2008). In the context of the 2018 congressionalelection, this has even led major national newspapers to explicitly warn voters about proceduralmisinformation that is used by certain groups to deter citizens from voting.1Thus, procedural information seems to play a crucial role for voter turnout. Voters whoknow how to vote and where to find their polling place are more likely to participate in elec-tions. As a result, providing citizens with information about the location of their polling placehas been shown to increase turnout (Brady and McNulty 2011). Similarly, turnout of Hispaniccitizens goes up if their procedural information costs are reduced through language supportin Spanish (Hopkins 2011). Yet, beyond a mere focus on procedural information about theelection as a means to primarily make the voting process easier or more convenient for voters,there are many situations where citizens may hold outright incorrect beliefs about the electoralprocess. These incorrect procedural beliefs usually do not affect all citizens, but only a sub-group of people. Yet, for those who are affected by them, the effect may be extremely strong.Especially in cases where these incorrect procedural beliefs let citizens doubt their electioneligibility or where they make them erroneously think that voting costs are substantially higherthan they really are, these beliefs are likely to strongly depress voter turnout among those whohold them. For example, in the absence of correcting information, voters who erroneouslybelieve they are ineligible to vote are unlikely to cast a ballot in an election (cf. Gerber et al.2015). Similarly, doubting the secrecy of the ballot in a context where ballot secrecy is strictlyenforced may plausibly lead people who hold such beliefs to abstain from voting due to privacyconcerns (Gerber et al. 2013).However, not every person is equally likely to be susceptible to this kind of misinforma-tion. We would expect that high-information citizens, that is, those with higher education,that are more socially embedded and that have more experience voting are less likely to beswayed by new and incorrect procedural voting or registration information. In contrast, peoplefrom low-information groups often lack the resources and the motivation to question dubious1See, for example, Kevin Rose, “6 Types of Misinformation to Beware of on Election Day.(And What to Do if You Spot Them.)” New York Times, November 5, 2018, available at:https://www.nytimes.com/2018/11/05/us/politics/misinformation-election-day.html.7procedural information about elections and thus, are more likely to hold incorrect proceduralbeliefs. Recent research (Meredith and Morse 2014, 2015) supports this view by showing thatex-felons – a group that is likely to have a rather low level of procedural information – arefrequently poorly informed about their eligibility status in elections. A considerable number ofthem assumes that they are ineligible to vote even though their voting rights had been restoredpreviously and, as a result, does not vote in elections. This study focuses on another largegroup of low-information citizens that has received little attention so far in this literature (butsee Corvalan and Cox 2018): first-time voters who have no or little experience with the votingsystem and thus, who are more likely to hold incorrect procedural beliefs on the administrationof elections than older and more experienced voters.In order to assess the effect of procedural information costs among young people, this studyleverages a natural experiment from the U.S. context. In particular, it uses a large dataset thatis based on voter file data from 18 U.S. states in combination with a regression discontinuitydesign to causally identify the effect of incorrect beliefs on voter turnout. In contrast to mostprevious research that use a similar design (Coppock and Green 2016; Holbein and Hillygus2016; Meredith 2009; Nyhan, Skovron and Titiunik 2017), the present study does not focus on adiscontinuity in turnout around the Election Day, but instead shifts the focus on a discontinuityaround the registration deadline, which, for most states, happens to be roughly four weeksearlier. Doing so, I am able to show that there is a sharp drop in turnout right at the registrationdeadline in many U.S. states. This suggests that the problem of incorrect beliefs among youngAmericans about the registration requirements is not an isolated phenomenon, but instead quitecommon across U.S. states. More specifically, I present evidence that shows that people whoturn 18 after the registration deadline, but before Election Day are significantly less likely tovote than those who turn 18 a few days earlier. This drop in turnout can be substantively large(up to 9 percentage points or 25%) and is consistent with an account that attributes the dropin turnout right after the registration deadline to incorrect beliefs about the voter registrationrequirements. Moreover, in line with research on the habitual nature of voting (Coppock andGreen 2016; Dinas 2012; Fujiwara, Meng and Vogl 2016; Gerber, Green and Shachar 2003;8Meredith 2009) and on the effect of voter registration drives on voter turnout (Nickerson 2015),the decision not to register and vote in their first election has long-term consequences for theseyoung people and makes them significantly less likely to vote in future elections. For example,37% of young Floridians who were deterred from voting in 2008 as a result of erroneouslybelieving they are ineligible due to their 18th birthday being after the registration deadline didnot vote in the 2016 election 8 years later for the simple reason of having held these incorrectbeliefs previously. In addition, I address two plausible alternative explanations that attributethis drop in turnout to formal rules or concomitant administrative acts that may plausibly reduceregistration costs for young people are not able to account for the sharp drop in turnout right atthe registration deadline.This study makes three contributions. First, it sheds new light on the crucial role of beliefsabout procedural aspects of elections in shaping voter turnout. Based on a large amount ofnew evidence, it shows both their widespread existence and their long-term effects on votingbehavior for a low-information group of citizens: first-time voters. Second, it makes a method-ological contribution by highlighting the crucial role that temporal discontinuities can play instudying procedural information effects. While this study highlights the relevance of a tem-poral discontinuity around the registration deadline for youth voter turnout, similar age-baseddiscontinuities exist in other politically important areas and determine a citizen’s eligibility forsocial welfare benefits or citizenship and immigration status. Thus, the approach presented inthis paper can be applied to other policy areas where procedural information effects are likelyto be highly consequential, but have not been a focus of previous research. Third, my findingshave direct policy implications. They suggest that complex rules add additional costs to citi-zens, making program uptake less likely. Thus, in order to increase voter turnout or the propen-sity of citizens to make use of specific government programs, governments do not necessarilyhave to provide additional (financial) incentives or run expensive mobilization campaigns, butcould start by simplifying the rules.92.2 Procedural Information Costs and Political ParticipationIn line with a large literature on voter turnout that goes back to Downs (1957)’s foundational“An Economic Theory of Democracy,” I assume that the act of voting entails costs for voters.These costs that are associated with casting a vote include, among other things, acquiringinformation about the procedural aspects of an election. At the very least, voters need to knowhow and when to register for an election and where and how to cast their vote on Election Day.Thus, it is not surprising that interventions that reduce these procedural information costs havebeen shown to have a positive effect on turnout. Brady and McNulty (2011), for example, findthat informing citizens about the exact location of their polling place and about how to get thereon Election Day makes them more likely to vote. Hopkins (2011) find a similar positive effectfor an intervention that lowers the language barriers to Hispanic citizens by offering them theoption to use a Spanish-language ballot.However, there are different kinds of procedural information and not all of them simplyhelp citizens cast a vote more conveniently. Another type of such procedural information (orlack thereof) are incorrect procedural beliefs. Citizens may hold beliefs that make them thinkerroneously that certain rules do not apply to them or that these rules work differently fromhow they actually work. Problematic are these incorrect beliefs especially in situations wherethey may make the act of voting appear substantially more costly in these people’s eyes thanit actually is, thereby strongly reducing the likelihood of these people to turn out and vote.For example, with respect to the institution of the secret ballot, Gerber et al. (2013) find thateven in a long-standing democracy such as the United States where the secret ballot has beenin place for more than a 100 years, a significant share of the population doubts its secrecy.Furthermore, in a field experiment, the authors show that informing citizens about the secrecyof the ballot by mail helps to correct these incorrect beliefs and, as a result, increases voterturnout. Another area where incorrect procedural beliefs may have a strong negative effecton turnout is when they are about a person’s eligibility status in an election. Citizens whooutright – but erroneously – believe that they are ineligible to register or vote in an election are10not very likely to participate in that election. Therefore, I would expect a strong reduction inturnout among this group of people. Consistent with this argument, Gerber et al. (2015) findthat informing misinformed citizens about their eligibility status and correcting their wrongprocedural beliefs can significantly increase political participation.Incorrect procedural beliefs are not likely to affect all citizens equally, though. Especiallylow-information citizens – those with little prior personal exposure to elections and those mostdisadvantaged socially and economically (Brady, Verba and Schlozman 1995; Delli Carpiniand Keeter 1996) – are most likely to lack the resources to effectively counter incorrect proce-dural beliefs. Several recent studies have started to analyze political participation among mem-bers of one such low-information group: ex-felons whose voting rights have been reinstated(Gerber et al. 2015; Meredith and Morse 2014, 2015). In line with the argument presentedhere, these studies find that ex-felons often incorrectly assume that they are still ineligible tovote even after their voting rights have been reinstated and as a result, are less likely to votein elections (Gerber et al. 2015; Meredith and Morse 2015). These authors also show that ad-ministering an informational treatment that corrects these wrong beliefs significantly increasedturnout among this group of people (Gerber et al. 2015; Meredith and Morse 2015).2 However,while most of the existing research on incorrect procedural beliefs is limited to ex-felons, thephenomenon is likely to extend to people of other low-information groups, too.Young people are another plausible low-information group. While certainly not everyyoung person is politically uninformed, young people tend to be – almost by necessity – politi-cally less experienced than many older people simply due to their lower age and as a result, thefact that they had fewer opportunities to engage in political activities such as voting. This lackof political experience combined with their generally lower political knowledge (Delli Carpiniand Keeter 1996) will make them less likely to detect and question incorrect procedural infor-mation than older and politically more experienced citizens. This is particularly likely to be the2The two studies discussed in the text focus on the states of Iowa and Connecticut. Meredith and Morse(2014)’s failure to replicate the effect for two other U.S. states suggests that the effectiveness of informationaltreatments is potentially dependent on additional factors, such as the type of delivery or the specific content of thetreatment message, which were not systematically studied in these papers.11case for first-time voters who, by definition, never had the chance to participate in an electionpreviously.3 For example, it is possible that first-time voters whose 18th birthday is after theregistration deadline, but before Election Day erroneously assume that one has to be 18 by theregistration deadline already in order to vote in an upcoming election and as a result, do notregister and vote in this election. While this belief is wrong and contrary to existing law in allU.S. states, such a misperception seems to be common enough among young Americans thatstates explicitly address this issue in their voter guides or on their state election office web-sites.4 In addition to that, not only rules about when and how people can register, but also theclarity with which this information is conveyed to citizens varies across states, thus potentiallycreating explaining the confusion in states with less clear instructions on websites and voterregistration forms.5Some evidence about procedural information costs among young voters comes from Cor-valan and Cox (2018) who investigate this phenomenon in the context of Chile. Using a re-gression discontinuity design around the registration deadline cutoff in Chilean elections, theyfind that Chileans who turn 18 after the registration deadline, but before Election Day have asignificantly lower voter turnout than their peers whose birthday is right before that deadline.However, as the authors point out, Chile is particular in several respects: First, it is a veryyoung democracy where incorrect beliefs about the population might be more prevalent thanin an established democracy like the U.S. Second, the temporal gap between the registrationdeadline and the Election Day is much larger than in any U.S. state, being between 3 and 6months compared to only a few days or weeks in the U.S. Third, up until a change to theelection law, young people in Chile had to be 18 not be Election Day, but by the registrationdeadline already if thy wanted to vote in an election (Corvalan and Cox 2018, 8). Thus, the3In addition, the national context is likely to matter, too. The likelihood of first-time voters holding incorrectprocedural beliefs is more plausible for countries like the U.S. where the electoral rules are more complex becausevoters have to register themselves (as opposed to automatic enrolment by the government) and where rules abouthow and when to register and vote may vary by state.4See, for example, the Arizona voter guide at https://www.azvoterguide.com/faqs/. Corvalan and Cox (2018,fn7) find similar questions on the websites of 10 other U.S. states.5For some illustrative evidence of these differences, see Table A.3 in the supporting information.12lower turnout for people turning 18 after the registration deadline in Chile might be the resultof path dependence where some people are simply not aware of the changed rules to registerin Chile. For these reasons, it is not clear to what extent the empirical evidence from Chileapplies to other countries where this is different, such as the U.S.Finally, not voting due to incorrect procedural beliefs in one election is likely to affecta person’s propensity to vote in future elections. There are good reasons to assume that notvoting in the present election also makes people less likely to vote in future elections. First,we know that voting is habit forming. People are more likely to vote in future elections for themere fact of having voted previously, a phenomenon that is sometimes called “habitual voting”in the literature (Coppock and Green 2016; Fujiwara, Meng and Vogl 2016; Gerber, Green andShachar 2003; Meredith 2009). Thus, a young person who fails to vote in the first election inwhich they are eligible to vote fails to make the first and crucial step toward establishing such ahabit and is less likely to vote in future elections. Second, simply the fact of registering to voteearly on and being added to a voter file may increase a person’s likelihood to vote in futureelections. Especially in countries such as the U.S. where people have to actively register tovote, registering produces some – albeit small – cost to voters. Thus, having already registeredin the past could make it more attractive to people to vote in future elections. Related to that,the simple fact of being added to a voter file increases a person’s likelihood to be contactedby future political election campaigns and thus, to be mobilized to cast a vote (cf. Nickerson2015). For these reasons, any effect of incorrect beliefs that depresses voter turnout in theimmediate election is likely to have long-run effects on the likelihood of a person to cast a votein future elections.This theory of procedural information costs leads me to formulate the following two hy-potheses:Incorrect procedural beliefs hypothesis: Young citizens who turn 18 after the registrationdeadline, but before Election Day are less likely to vote than otherwise identical young peoplewho turn 18 before or on the registration deadline.Persistence hypothesis: Young people who do not register or vote in their first election13due to incorrect procedural beliefs are less likely to vote in future elections.2.3 DataThe empirical analysis in this study uses state voter records to calculate youth voter turnoutin U.S. presidential elections. The advantage of these administrative data is that they avoidshortcomings of traditional survey-based analyses of voter turnout such as social desirabilitybias and sample sizes that are often too small to precisely estimate effects for societal sub-groups. The voter files include registration information and voting histories for each singlevoter and were obtained directly from the Secretary of State offices for five states (Florida,New York, Ohio, Pennsylvania, Texas) and are taken from (Coppock and Green 2016) for 11others (Arkansas, Colorado, Connecticut, Iowa, Illinois, Kentucky, Missouri, Montana, NewJersey, Nevada, Oklahoma, Oregon, Rhode Island). The statistical analyses focus on the largerof these states. Due to their larger population sizes, I can restrict my analysis to people whoturned 18 shortly before or after the registration deadline for the 2008 U.S. presidential electionand still get precise estimates of voter turnout and treatment effects for each individual state.Despite its advantages, using voter file data for the study of voter turnout poses certain em-pirical challenges, above all, the risk of “differential registration bias” (DRB) (Nyhan, Skovronand Titiunik 2017). Such bias occurs if we estimate voter turnout based on the number of peo-ple registered in the voter file, but registration rates differ between treatment and control group.For example, as part of my analysis I am comparing citizens whose 18th birthday falls shortlybefore or after the registration deadline for the 2008 presidential election. If those marginallyolder were able to vote in one more presidential election, their registration rate is likely to behigher. Ultimately, DRB is a denominator problem because we usually lack the exact numberof eligible voters needed to calculate voter turnout and as a consequence, have to rely on aproxy measure or alternative estimation strategies that do not require knowledge of the votingeligible population.To avoid the risk of DRB, I adopt two strategies that have been successfully used in previous14research with voter file data. The first strategy uses birth data as a proxy for the unknownvoting-eligible population (Meredith 2009; Nyhan, Skovron and Titiunik 2017). Birth datawith the exact number of people born in a given state on a given day were accessed throughthe U.S. Center for Disease Control (CDC)’s National Vital Statistics website for people bornbefore 1989. Due to more stringent confidentiality standards starting in 1989, data on dailybirth counts for later years had to be requested from the respective Departments of Health in agiven state. Birth data has two big advantages over alternative measures. First, it is temporallyvery fine-grained. In contrast to census data, we can collect daily birth counts and use them tocalculate voter turnout by birthdate cohort, that is, for all people born on the same day. Second,birth counts are plausibly exogenous to the treatment of election eligibility. It is unlikely that aperson’s date of birth could possibly have been affected by whether or not that person is eligibleto vote in an election 18 years later. For these reasons political scientists have used it before tostudy socialization effects and procedural information costs with voter file data (Corvalan andCox 2018; Meredith 2009; Nyhan, Skovron and Titiunik 2017).However, one potential concern with birth data is that people move in and out of statesbefore they turn 18. While in and out migration of states is likely to be more or less randomlydistributed on both sides of the registration deadline discontinuity6 and thus, mere noise inour statistical analysis, I adopt a second strategy to retrieve the complier average causal effect(CACE) of procedural information costs on future voter turnout whose estimation is solelybased on vote totals (Coppock and Green 2016). As a consequence, the estimation of theCACE, which I discuss below, is unaffected by differential registration rates between treatmentand control group.6To make sure that this assumption is not violated by young people graduating from high school and movingto other states for college, the analyses are almost exclusively based on states where the official school entry cutoffdate is outside the RD estimation window that I use.152.3.1 Visualizing Registration Deadlines and Turnout PatternsThis section provides a first, informal look at the data. In a first step, I will provide a morein-depth look at one particular state, Florida, over several election cycles while in a secondstep, I provide a cross-sectional overview of all 18 U.S. states for which I have voter file datain the 2008 presidential election. This serves to give us a better understanding of the nature ofthe discontinuity in turnout right at the registration deadline and of its prevalence both acrossU.S. states.Figure 3.3 illustrates the effect of incorrect procedural beliefs on youth voter turnout forFlorida for different birth cohorts and across different elections. Each graph plots voter turnoutby birthdate cohorts for young Floridians in three U.S. presidential elections. The first rowshows turnout for people who were born between May 4, 1990 and May 4, 1991, the secondrow does the same for those born four years earlier, between May 4, 1986 and May 4, 1987.In Figure 3.3a we can see a clear and stark drop in 2008 turnout directly at the registrationdeadline cutoff of October 7, 2008, with turnout for those turning 18 right after (i.e. to theright of) the registration deadline significantly lower than for those whose 18th birthday wasa few days earlier. Interestingly, we find this discontinuity in turnout rates at the registrationdeadline cutoff not only for these people’s first presidential election, but also in their subse-quent presidential elections four and even eight years later (see Figures 3.3b-c). Comparingthis with people who turned 18 four years earlier and who could vote for the first time in the2004 presidential election, we find the same pattern: four, eight and even twelve years afterthe first election in which these citizens were eligible to vote, those who turned 18 right afterthe registration deadline of October 5, 2004 are significantly less likely to vote in presidentialelections than their slightly older peers (Figures 3.3d-f). In contrast, the initially much largerdiscontinuity at the Election Day cutoff between eligible and ineligible voters in the 2004 and2008 elections (see Figure 3.3a) produces a discontinuity in subsequent elections that, fromgraphical inspection, seems substantively smaller in size (Figures 3.3b-f). This suggests thatin states where we find a clear discontinuity in turnout at the registration deadline, this dis-16continuity is temporally persistent over multiple elections and does not seem to be an isolatedphenomenon restricted to a birth cohort or specific electoral context. Figure A.2 in the sup-porting information provides a graphical illustration of a similar discontinuity for presidentialelections in Texas. It is also worth pointing out that this discontinuity is not a result of the waywe measure turnout and persists if we replace the turnout measure with the daily vote totals(see Figure 4.2 below and Figures A.4 and A.5 in the supporting information).To get an idea of the prevalence of these incorrect procedural beliefs and how they affectvoter turnout across U.S. states, Figure 4.2 plots the total number of votes cast by birthdatecohort for people born May 4, 1990-May 4, 1991 for all 18 states for which I have voterfile data. The advantage of using total votes over a measure that computes turnout amongthose registered in the voter file is that it avoids the potential pitfall of differential registrationbias (Nyhan, Skovron and Titiunik 2017). DRB would be particularly problematic in thiscase because turning 18 right after the registration deadline is likely to reduce both a person’spropensity to register and to vote, thus biasing a turnout measure based on the number ofregistered people. The total number of votes, on the other hand, is unproblematic because,despite some seasonal fluctuations in total eligible voters, there is no reason to expect a suddendrop in vote totals right at the registration deadline: Parents are unlikely to anticipate theregistration deadline for their children 18 years later and to make the day of their child’s birthdependent on that deadline. In order to empirically test these expectations, Figure A.1 in theAppendix plots the number of people who were born on each single day in Florida for the May5, 1990 to May 4, 1991 period. We can see that in line with our expectations, the density ofthat variable is smooth around the registration deadline. At the same time, there is seasonalfluctuation in the number of births, with births being highest in September and lowest in May.77Given the typical day-of-the-week pattern of eligible voters per day, there is a much lower total number ofvotes for weekend days. For easier visual interpretation of the general trends around the registration deadline,graphs exclude data for people born on Saturdays and Sundays. Including data for these people does not affectthe interpretation of our results at the registration deadline.17Figure 2.1: Turnout in Florida for People Born in 1986/87 and 1990/91Eligible in 2008 Ineligible in 2008EligibilityRegistration0%10%20%30%40%50%60%Turnout5/4/90 8/4/90 11/4/90 2/4/91 5/4/91Date of Birth(a) 2008 Presidential ElectionEligible in 2008 Ineligible in 2008EligibilityRegistration35%40%45%50%55%60%Turnout5/4/90 8/4/90 11/4/90 2/4/91 5/4/91Date of Birth(b) 2012 Presidential ElectionEligible in 2008 Ineligible in 2008EligibilityRegistration35%40%45%50%55%60%Turnout5/4/90 8/4/90 11/4/90 2/4/91 5/4/91Date of Birth(c) 2016 Presidential ElectionEligible in 2004 Ineligible in 2004EligibilityRegistration35%40%45%50%55%60%Turnout5/2/86 8/2/86 11/2/86 2/2/87 5/2/87Date of Birth(d) 2008 Presidential ElectionEligible in 2004 Ineligible in 2004EligibilityRegistration40%45%50%55%60%65%Turnout5/2/86 8/2/86 11/2/86 2/2/87 5/2/87Date of Birth(e) 2012 Presidential ElectionEligible in 2004 Ineligible in 2004EligibilityRegistration40%45%50%55%60%65%Turnout5/2/86 8/2/86 11/2/86 2/2/87 5/2/87Date of Birth(f) 2016 Presidential Election18Each graph in Figure 4.2 fits three lines, one for each of the three groups we are interestedin: pre-registration deadline eligibles to the left of the registration deadline, post-deadline eli-gibles in between the registration deadline and Election Day, and ineligibles to the very right.Fitted lines use local mean-smoothing, with 95% confidence intervals.8 Results are for the2008 presidential election.Looking at the graphs for all 18 states, a clear pattern emerges. People turning 18 right afterthe registration deadline in their respective state are less likely to vote in the 2008 presidentialelection than those who turn 18 shortly before that deadline. For the majority of states there is aclear and statistically significant drop in the vote totals right after the registration deadline (e.g.Florida, Illnois, Texas). For four states (Kentucky, New York, Ohio, Pennsylvania) the drop invote totals for post-deadline eligibles is relatively small and the confidence intervals overlap.Thus, we cannot be sure of a statistically significant difference between pre- and post-deadlineeligibles in these states. There are only four states (Iowa, Montana, Nevada, Oregon) wherethere is no evidence for a procedural information effect on turnout at all, with daily vote totalsin these states being roughly the same on both sides of the registration deadline. Note thattwo of these states, Iowa and Montana, enacted Election Day Registration (EDR) before the2008 election.9 These laws allowed voters to both register and vote on Election Day itself, thusstrongly reducing any registration costs to first-time voters in these two states. Overall, thispattern suggests that the discontinuity at the registration deadline is not limited to a few states,but indeed a common phenomenon across U.S. states.Importantly for the plausibility of my causal story about the role of incorrect proceduralbeliefs, we can see that across states the drop in turnout is intimately linked with the registrationdeadline. In fact, this drop occurs right at the registration deadline and, very crucially, variesacross states in line with the specific registration deadline in each state. Most states have a8I chose local mean-smoothing over fitted polynomial regression lines to allow for more flexibility. As a result,seasonal birth trends, possible confounding factors in a given state such as the school entry cutoff date or the shortwindow between registration deadline and Election Day are less likely to affect the discontinuity at the registrationdeadline.9Montana and Iowa enacted EDR in 2005 and 2007 respectively and are the only states in my sample withEDR in 2008.19registration deadline of 28 days and, as we would expect, turnout drops exactly for those whohave their 18th birthday 28 days before Election Day (e.g. Florida, Illinois, Texas). However,the drop occurs after only 25 (e.g. Oklahoma), 21 (e.g. New Jersey) or even 14 days (e.g.Connecticut) in some states where the registration deadline is 25, 21 or 14 days before ElectionDay, respectively. In contrast, there is no drop in turnout before Election Day in Iowa andMontana, the only two states with Election Day Registration in 2008 in my sample. Thisfurther bolsters our confidence that it is the registration deadline - and not something else - thathas to be responsible for the drop in voter turnout.Interestingly, there is a second drop in the total number of votes in some states (e.g. Texas,Illinois) to the left of September 1. This date coincides with these states’ school entry cutoffdate in the 1990s. It suggests that those who turned 18 before September 1, 2008 and thus,who are more likely to have finished high school by the time of the 2008 election are lesslikely to vote in that election (see also Figure A.2 in the supporting information). To avoidconfounding due to this additional discontinuity, the statistical analyses exclude people bornbefore September 1, 1990 and focus exclusively on people whose 18th birthday is within aone-month window around the registration deadline.20Figure 2.2: Registration Deadline Discontinuities Across U.S. States in 2008 Election0204060Votes6/4/90 11/4/90 4/4/91Date of BirthAK050100150Votes6/4/90 11/4/90 4/4/91Date of BirthCO020406080100Votes6/4/90 11/4/90 4/4/91Date of BirthCT0100200300400Votes6/4/90 11/4/90 4/4/91Date of BirthFL0102030Votes6/4/90 11/4/90 4/4/91Date of BirthIA0100200300Votes6/4/90 11/4/90 4/4/91Date of BirthIL020406080100Votes6/4/90 11/4/90 4/4/91Date of BirthKY050100150Votes6/4/90 11/4/90 4/4/91Date of BirthMO010203040Votes6/4/90 11/4/90 4/4/91Date of BirthMT050100150200250Votes6/4/90 11/4/90 4/4/91Date of BirthNJ0204060Votes6/4/90 11/4/90 4/4/91Date of BirthNV0100200300Votes6/4/90 11/4/90 4/4/91Date of BirthNY0100200300Votes6/4/90 11/4/90 4/4/91Date of BirthOH020406080Votes6/4/90 11/4/90 4/4/91Date of BirthOK020406080Votes6/4/90 11/4/90 4/4/91Date of BirthOR0100200300400500Votes6/4/90 11/4/90 4/4/91Date of BirthPA010203040Votes6/4/90 11/4/90 4/4/91Date of BirthRI0100200300400Votes6/4/90 11/4/90 4/4/91Date of BirthTXNote: Figures plot total number of votes by birthdate cohorts.212.4 Identification StrategyTo test the previously specified hypotheses about procedural information costs, I will estimatethe effect of beliefs about registration requirements, in particular the effect of erroneouslythinking that one is ineligible to vote because one’s birthday is after the registration deadline,on current and future rates of youth voter turnout. The dependent variable in the empiricalanalyses is voter turnout, measured either at the individual level as whether or not a person inthe dataset cast a vote or aggregated by birthdate cohort, that is, for people born on the sameday, as described below. Individuals within a narrow window on either side of the registrationdeadline are as-if randomly assigned as to whether or not they face higher costs to voting dueto uncertainty about whether they have to be 18 by the registration deadline or by ElectionDay. This allows me to use a regression discontinuity design to estimate the treatment effectof these procedural information costs on voter turnout. Given the large size of the dataset, Iam able to treat the RD as a “local randomized experiment” (Cattaneo, Titiunik and Vazquez-Bare 2017; Cattaneo, Idrobo and Titiunik 2018; Dunning 2012; Nyhan, Skovron and Titiunik2017), that is, I restrict the estimation to observations within a narrow window around theRD cutoff and use difference-in-proportions and a Wald estimator to retrieve the treatmenteffect. Thus, the empirical analysis primarily focuses on young Americans born September10-November 4, 1990, that is, people who turned 18 within a one-month window on eitherside of the registration deadline for the 2008 presidential election in their respective state.10Additional age cohorts are constructed analogously. The one-month (or, more precisely, 25 or28-day window) is the maximum temporal distance between the registration deadline and theElection Day and thus, maximizes efficiency while avoiding bias due to confounding.11In a first step, I compute differences in proportions in voter turnout between those whoturned 18 right after the 2008 registration deadline (i.e. October 7, 2008) and those who turned10For comparison, Coppock and Green (2016)’s study on habitual voting uses a much larger 12-month windowon either side of the RD cutoff.11For example, our estimates would potentially be confounded if the window was larger and included theElection Day or the school entry cutoff date, which might have an independent effect on voter turnout.2218 right before it for both the 2008 and 2012 U.S. presidential elections. The difference be-tween these two groups captures the average treatment effect (ATE) of procedural informationcosts on turnout in each state. To avoid differential registration bias (Nyhan, Skovron andTitiunik 2017), this part of the empirical analysis calculates turnout by using total daily birthcounts as a proxy for the voting eligible population in a given state.In a second step, I estimate the complier average causal effect (CACE) that being exposedto these procedural information costs in a person’s first election has on their propensity to votein future elections. While we cannot randomly assign the non-voting as a result of proceduralinformation costs, we can get at it by leveraging a natural experiment which as-if randomlyincreases the procedural information costs and thus, discourages some young people from vot-ing. Both among those who turn 18 right before and among those who turn 18 right after theregistration deadline in 2008, some are likely to think that a person has to be 18 not by ElectionDay, but by the registration deadline in order to vote in that election. However, this incorrectbelief only materializes and affects the behavior of those who turn 18 after the registrationdeadline while it is inconsequential for those who turn 18 before that date. As a result, onlythe former experience increased procedural information costs: if they are unclear about theireligibility status in that election, they would have to take additional steps to find out, which arenot necessary for those who turn 18 a few days earlier.However, not all eligible voters who turn 18 after the 2008 registration deadline hold theseincorrect beliefs and thus, are exposed to the treatment, that is, these higher procedural infor-mation costs. In fact, the treatment is restricted to those who hold these incorrect beliefs andwho, as a result of holding them, do not vote in 2008. In other words, they are the compliersin this study. They are the ones who, due to higher procedural information costs, are preventedfrom registering and voting in 2008 and thus, are less likely to vote in future elections. To beclear, the claim here is that some of them do not register and vote and thus, are less likely tovote in future elections only because they erroneously believe that they are not eligible to votein their first election.To estimate the complier average causal effect (CACE), I use a Wald estimator (Angrist23and Pischke 2009). Following (Coppock and Green 2016) I call the procedural informationcosts Z ∈ {0, 1}, with 1 indicating that a person is exposed to a discouragement to vote inan election due to higher procedural voting costs and 0 otherwise. The estimand of interestis “voting in an upstream election V1 on voting in a downstream election V2” (Coppock andGreen 2016, 1046).My approach rests on four key assumptions (Angrist and Pischke 2009). First, Z is as-if ran-domly assigned at the registration discontinuity and thus, independent of potential outcomes.Second, to satisfy the exclusion restriction, Z has to affect V2 only through its effect on V1. Inother words, I assume that the only reason a person who was discouraged from voting in 2008due to higher procedural information costs is less likely to vote in 2012 is because of her notvoting in 2008. Third, Z impacts the endogenous regressor V1, an assumption that can be di-rectly tested through the first stage of the Wald estimator. Fourth, the monotonicity assumptionrequires that those who are responsive to the treatment respond to it in the same way.Defining the Average Upstream Treatment Effect (AUTE) as E[V1i(1)] − E[V1i(0)] andthe Average Downstream Treatment Effect (ADTE) as E[V2i(1)] − E[V2i(0)],12 the followingestimator gives us the complier average causal effect (CACE) (Coppock and Green 2016, 1046-1047):ˆCACE =Eˆ[V2i|Zi = 1]− Eˆ[V2i|Zi = 0]Eˆ[V1i|Zi = 1]− Eˆ[V1i|Zi = 0]=ˆADTEˆAUTE(2.1)Given that voter files contain only those who are registered, I avoid differential registrationbias in computing the CACE by following Coppock and Green (2016)’s strategy of using thetotal number of votes cast by birthdate cohort as dependent variable. I include the daily votetotals lagged by a year on the right-hand side of the equation to account for seasonal or day-of-the-week fluctuations in the number of eligible voters. This gives us a Wald estimator with the12V1i(1) and V2i(1) are unit i’s treated, V1i(0) and V2i(0) untreated potential outcomes.24following two-stage form:UpstreamV otesCastj = γ0 + γ1Zj + γ2LaggedDownstreamj + ρj (2.2)DownstreamV otesCastj = β0 + β1 ˆUpstreamV otesCastj + γ2LaggedDownstreamj + j(2.3)Equation 3.2 displays the first stage. The total number of votes by birthdate cohort in theupstream election (Vj) are regressed on a treatment indicator Zj and LaggedDownstreamj ,the dependent variable lagged by one year. In the second stage shown in equation 3.4, Yj is thetotal number of votes by birthdate cohort in the downstream election and Vˆj are the predictedvalues from the first stage. j indexes birthdate cohorts.2.5 ResultsTable 2.1 provides statistical estimates of the effect of erroneously believing one has to be 18to register to vote on voter turnout for five large U.S. states. I focus my analysis on peopleborn September 10-November 4, 1990, that is, people turning 18 within a 28-day window oneither side of the registration deadline. The results in the top part of the table are from the 2008U.S. presidential election. We can see that in four of the five states there is a significant dropin turnout for those turning 18 after the registration deadline compared to those having theirbirthdays before October 7, 2008. This drop is largest for Florida, Illinois and Texas, threestates for which the graphical analysis above (see Figure 4.2) revealed a clear discontinuity inturnout at the registration deadline. The estimated treatment effect is -6.72 percentage pointsfor Florida and -8.89 percentage points for Texas, showing that post-registration deadline el-igibles who might erroneously think they are not eligible to vote in 2008 have a much lowerturnout in 2008 (45.24% for Florida and 25.39% for Texas) than pre-deadline eligibles (51.69%25and 34.27% respectively). We find a similar, but smaller effect of -2.27 percentage points forNew York and -3.48 percentage points for Pennsylvania, two states where our graphical anal-ysis above already suggested that the effect is smaller. Finally, Ohio is the only state amongthese six where voter turnout between pre- and post-registration deadline eligibles is roughlythe same, at about 46.85%, suggesting that post-registration deadline eligibles in this state aregenerally better informed about the age requirement for registering to vote and voting.The middle section of Table 2.1 presents estimates for the same group of people (i.e. 2008eligibles), but for the 2012 presidential election four years later, that is, the downstream effectof 2008 erroneous beliefs on 2012 turnout. We can see that for Florida and Texas, the two stateswith the largest effect in the 2008 election among the five states included in Table 2.1, turnoutfor post-registration deadline eligibles is still significantly lower than for pre-registration eli-gibles. In Florida turnout for the former is 48.10% while it is -2.59 percentage points lower(45.52%) for those who have their 18th birthday right after the registration deadline. In Texasthe difference in turnout between these two groups is -1.74 percentage points. For Illinois, NewYork and Pennsylvania, 2012 turnout for those who turned 18 right after the 2008 registrationdeadline is now only marginally lower to that of pre-registration deadline eligibles, at -1.05,-0.30 and -0.91 percentage points respectively, a difference that is no longer statistically signif-icant at the 95% level. As for Ohio, the difference in 2012 turnout between these two groupsis a positive 1.50 percentage points, suggesting that post-registration deadline eligibles do nothave a lower, but in fact a higher turnout rate (40.11%) than pre-deadline eligibles (38.61%)in this state.13 Furthermore, all five states have similar registration rules - voters have to regis-ter roughly a month before Election Day and same-day-registration14 is not possible - and wefind large effects both in “battleground” (Florida, Pennsylvania) and “non-battleground” states(Texas). This further bolsters our confidence that the effect can be attributed to erroneous be-13A possible explanation for this surprising finding is that results for Ohio might be affected by that state’slate school entry cutoff date of September 30, which falls clearly within the one-month window to the left ofthe registration deadline that I used for the analysis (see Figure A.3 in the supporting information for furtherdiscussion).14Same-day-registration allows voters to register and vote in one single step on Election Day.26liefs about the age requirement to register and vote in these states and not to differences inregistration rules or the competitiveness of the election across states.Finally, the voter file for Florida also shows whether a person voted in the 2016 presidentialelection. As we can see in the bottom section of Table 2.1, 2008 eligibles who turned 18 afterthe 2008 registration deadline are still significantly less likely to vote eight years after theywere first eligible to vote in a presidential election than otherwise identical citizens who turned18 a few days earlier. Based on the results in Table 2.1, their probability to cast a vote is -2.45 percentage points lower. Interestingly, the size of the effect is practically identical to theeffect four years earlier, suggesting that not voting in 2008 due to misperceptions about theregistration requirements in that election has long-term effects, deterring some of these peoplefrom participating in future elections. Importantly, these results are generally robust both tovarying bandwidths around the registration deadline and to different model specifications (seeTables A.1 and A.2 in the supporting information).152.5.1 Persistence of EffectsThe results from Table 2.1 above illustrate the existence and size of the effect that incorrectbeliefs about the registration requirements have on voter turnout. However, they do not giveus a direct estimate of the effect that being discouraged from voting in 2008 due to theseinformational hurdles has on a person’s propensity to abstain from voting in future elections.This complier causal average effect (CACE) of being deterred from voting in the upstreamelection (i.e. 2008) on turnout in a downstream election (i.e. 2010 or 2012) is reported in Table2.2. As explained above, the coefficient on the registration beliefs variable is the ratio of the2010 or 2012 average downstream treatment effect over the 2008 average upstream treatmenteffect. Multiplying this ratio by 100 gives us the percentage point difference between treated15For most states, the size of the effect is largely unaffected by using the shorter 10 day window on either side ofthe registration deadline (see Table A.1 in the supporting information). Note, however, that for New York, resultsare sensitive to bandwidth selection and further research is required to unambiguously establish the existence ofthe effect identified in Table 2.1 for this state.27Table 2.1: Turnout in U.S. Presidential Elections for 1990 Birth Cohorts(FL) (IL) (TX) (NY) (OH) (PA)Registration Deadline Discontinuity2008Eligibles post-deadline 45.24 35.41 25.39 29.47 46.87 39.75Eligibles pre-deadline 51.96 40.87 34.27 31.73 46.83 43.23Difference -6.72 -5.46 -8.89 -2.27 0.04 -3.48[-7.81, -5.62] [-7.23, -3.68] [-9.68, -8.09] [-3.12, -1.41] [-1.18, 1.26] [-4.66, -2.29]N 32,084 28,859 50,727 45,540 25,627 26,6952012Eligibles post-deadline 45.52 30.16 24.91 29.62 40.11 35.82Eligibles pre-deadline 48.10 31.21 26.65 29.92 38.61 36.73Difference -2.59 -1.05 -1.74 -0.30 1.50 -0.91[-3.68, -1.49] [-2.86, 0.76] [-2.50, -0.97] [-1.15, 0.55] [0.30, 2.70] [-2.07, 0.24]N 32,084 28,859 50,727 45,540 25,627 26,6952016Eligibles post-deadline 43.93Eligibles pre-deadline 46.38Difference -2.45[-3.54, -1.36]N 32,084Note: Table presents difference-in-proportions with 95% confidence intervals in brackets for young Americans born in 1990. Eachestimation uses a 28-day-window (25 for NY) on either side of the registration deadline in 2008.and untreated compliers.16 The table presents results for the five states from Table 2.1 above forwhich erroneous beliefs about the registration requirements matter. It does no longer includeOhio, for which I failed to find any effect of incorrect procedural beliefs on 2008 voter turnoutamong post-registration deadline eligibles.Based on the top part of Table 2.2, we can see that independent of the initial size of theeffect in 2008 that I identified above, not voting in 2008 increases a person’s likelihood to notvote in the 2012 election four years later. For all five states the effect is statistically significantat the 95% confidence level and substantively large, ranging from 0.174 in New York to 0.514in Pennsylvania, with the values for the other three states falling in between this range. AˆCACE of 0.425 in Florida, for example, indicates that being discouraged from voting in 2008because a person erroneously thinks she is not eligible to vote in that election increases her16The dependent variable in Table 2.2 is the number of people for each bithdate cohort (i.e. people who areborn on the same day) who voted in either 2010 or 2012. Therefore, the constant gives the number of people whoare born on the same day and who voted in 2010 or 2012 when all other variables are set to 0.28probability of not voting in 2012 by a full 42.5 percentage points.17Luckily, the data in the voter file for the state of Florida allows me to analyze an even moreextended period of time and observe the voting behavior of the same group of people over morethan two presidential elections. Based on the results in the last column of the table, we can seethat not voting in the 2008 election due to incorrectly believing one is ineligible increases theprobability to not vote in the 2016 presidential election 8 years later by 37.1 percentage points.This pattern is consistent with our findings in Figure 3.3 above which showed that both the 2004and the 2008 cohorts of post-registration deadline eligibles are less likely to vote in presidentialelections eight (i.e. in 2012 and 2016, respectively) or even twelve years later (i.e. in 2016 forthe 2004 eligibles). After an initial drop between the first and the second presidential election,the effect seems to stabilize and persist over time.Results from the bottom part of Table 2.2 report the downstream effect of being deterredfrom voting in the 2008 presidential election on voting in the 2010 midterm election. Whileeffects are somewhat smaller, the pattern is largely the same as before. Across all five statesthe effect of being discouraged from voting in the 2008 election strongly and consistentlyincreases the probability of not voting in the 2010 midterm election. The smallest effect isstill a remarkable 10.1 percentage points in Texas and the largest effect is 22.7 percentagepoints in New York. These ˆCACEs are statistically significant at p<0.05 for all states exceptPennsylvania, whose estimate has a p-values of 0.09.18Overall, these results suggest that young people who incorrectly think they are not eligibleto vote in their first presidential election due to their birthday falling after the registrationdeadline and who, as a consequence, fail to vote in that election are significantly less likely to17Consistent with the assumption that post-deadline eligibles are less likely to vote in 2008 than pre-deadlineeligibles, the instrument of procedural voting costs is negatively correlated with voting in the upstream election inthe first-stage results (not reported). This association is statistically significant for all Wald estimators reported inthis study.18Further analyses (results not reported) fail to produce a similar long-term effect for upstream midterm elec-tions on downstream midterm or presidential elections. For example, incorrect beliefs in the 2006 midterm elec-tion only depress turnout in 2006, but do not produce a significant drop in turnout in the 2008 and 2010 elections.This suggests that incorrect procedural beliefs are more consequential for first-time eligible voters in presidentialthan in midterm elections.29Table 2.2: CACE of 2008 Incorrect Procedural Beliefs on Downstream Voting(FL) (IL) (NY) (PA) (TX) (FL)Presidential on Presidential2008 - 2012 2008 -2016Registration Beliefs .425* .280* .174* .514* .265* .371*(.114) (.121) (.082) (.114) (.050) (.134)Lagged DV .424* .411* .667* .317* .454* .471*(.143) (.153) (.088) (.127) (.089) (.158)Intercept 37.365* 35.472* 35.601* 17.333* 55.554* 34.111*(15.756) (12.606) (16.776) (10.901) (12.805) (17.146)N (Days) 56 56 56 56 56 56Presidential on Midterm2008 - 2010Registration Beliefs .167* .227* .191* .160 .101*(.038) (.044) (.067) (.094) (.027)Lagged DV .221 .255 -.174 .151 .240*(.120) (.144) (.154) (.178) (.086)Intercept 15.782* 4.979 5.022 22.445 29.005*(5.298) (6.434) (10.142) (19.161) (6.049)N (Days) 56 56 56 56 56Note: Table presents results from a Wald estimator (second stage). Estimations use a 25 (NY) or 28-day-window (FL, IL, PA, TX) on eitherside of the cutoff. *p<.05.vote in future elections. As the ˆCACEs for various states show, these longer term effects aresubstantively large. They are significantly larger than similar effects of previous non-voting onfuture voter turnout that have been identified in the literature for people who turn 18 aroundthe Election Day (see, for example, Coppock and Green 2016). However, we have to keepin mind that the effect works on a relatively narrow group of people: those who are deterredfrom voting because they turn 18 between the registration deadline and Election Day, and whoerroneously think they are not eligible to vote in their first presidential election. The evidencein this study suggests that this specific group of people benefits a lot from being added to thevoter rolls early on and also possibly from casting a vote in their first election. Preventing thesepeople from doing so turns many of them into non-voters in future general elections. We canassume that many of these very same people would have voted in their first and subsequentelections, had they known that they were eligible to register while still 17.302.5.2 Alternative Explanations: Formal Rules and Concurrent Adminis-trative ActsHow certain can we be that the drop in voter turnout that we observe right at the registrationdeadline is really due to young people’s incorrect beliefs about the age requirement to register?Given that this study uses observational data, the credibility of the causal story about incor-rect beliefs crucially depends on me being able to rule out alternative explanations that couldplausibly account for the sudden drop in voter turnout among those who turn 18 right after theregistration deadline. Key among them are the following two: (1) formal rules that prevent17-year-olds from registering and (2) concurrent administrative acts that make young peoplemore likely to register on their 18th birthday due to lower registration costs.First, the drop in turnout for post-registration deadline eligibles could be due not so muchto their beliefs that they are ineligible to register and vote, but rather to actual rules or adminis-trative practices that prevent this group of people from or make it much more difficult for themto register and vote in their first election in some states. However, such an explanation wouldgo against the stipulations of the 26th Amendment of the U.S. Constitution, which grants U.S.citizens “18 years of age or older” the right to vote. Accordingly, most state election codes orstate voter registration forms explicitly state that citizens have to be 18 by Election Day, notby the registration deadline to register to vote. Even in states where the wording in officialdocuments may be less clear about the specific age that is required to register (e.g. Florida,Idaho, Oklahoma, Texas – see Table A.3 in the supporting information), election officials arereadily available to provide clarification. For example, employees at Secretary of State andCounty Supervisor of Election offices in Florida and Texas consistently confirmed to me thatan otherwise eligible person is able to register in these states even before she turns 18 as long asshe will be 18 by Election Day.19 Thus, legal rules and their enforcement by election officialsseem unlikely to be able to explain the sharp drop in voter turnout at the registration deadline.19Personal communication by phone with the Florida and Texas Secretary of State and several Florida CountySupervisor of Election offices (Alachua, Baker, Bay, Bradford, Brevard, Broward, Miami-Dade, Leon, Orange,Palm Beach) on December 12, 2018.31Second, it is possible that another administrative act that young people tend to perform righton their 18th birthday significantly reduces registration costs, makes them much more likely toregister on that day and thus, increases their likelihood to vote. For example, if a large numberof young people were to get their driver’s license exactly on the day of their 18th birthday and,at the same time, registered to vote at the DMV office, this could explain the abrupt drop inturnout for those who turn 18 right after the registration deadline. Fortunately, we can use thevoter file data to test the plausibility of this alternative hypothesis empirically. First, I create acounterfactual of the number of young people who would have registered on the day of their18th birthday if their 18th birthday had not been affected at all by the registration deadline.Second, I compare this counterfactual number to the number of people who would have hadto register on their 18th birthday in order to account for the stark drop in turnout right at theregistration deadline.To estimate the number of people who register on their 18th birthday due to a concomi-tant administrative act, I analyze the registration behavior of young Floridians who turn 18between February 8 and August 7, 2008. The advantage of focusing on this group of peoplefor the counterfactual is that their decision to register on their 18th birthday is less likely tobe influenced by other factors such as incorrect beliefs about the registration age. In contrast,those who turn 18 in September or early October 2008 are not suited for this purpose becausetheir decision to register could primarily be a result of their incorrect beliefs that they have towait until their 18th birthday to register, but have only few days left to do so before the reg-istration deadline. Thus, finding a high number of people who register exactly on their 18thbirthday among this group of citizens does not allow us to differentiate between the role ofincorrect beliefs and administrative acts that citizens perform on the same day because bothmake them more likely to register on the day they turn 18. In contrast, for people whose 18thbirthday is several months before the registration deadline, there is less reason to assume thatincorrect beliefs are the main reason why many of them register exactly on the day they turn18. Registering a few days or even weeks later has no apparent disadvantage to them.The black line in Figure 2.3(a) plots the average number of young Floridians for a single32Figure 2.3: Registration Timing for Floridians born Feb 4-Aug 3, 199001020304050Number of People-30 days -15 days 18th Birthday +15 days +30 daysAge at Registration(a) Voter Registrations010203040Number of People-30 days -15 days 18th Birthday +15 days +30 daysAge at Registration(b) Voting in 2008birthdate cohort who register 30 days before and after their 18th birthday. Of roughly 393people who are born on the same day and who registered by the 2008 election, relativelyfew people registered on each single day around their 18th birthday. The average numberis between 1 and 2 registrants per day, with the number of registrations increasing to 3 onthe day of the 18th birthday itself. Thus, while people are somewhat more likely to registeron the day they turn 18, this increase affects only a very small number of people. Can thisdifference in 1 to 2 more people who register on their 18th birthday account for the drop ofbetween 6 and 7 percentage points in turnout20 for those who turn 18 right after the registrationdeadline? The answer is no. It cannot explain the drop of 6 to 7 percentage points, whichcorresponds to a drop of roughly 36 votes – from an absolute number of 278 to 242 votes –for the average birthdate cohort after October 7, 2008 (see Figure 2.3(b)). The grey line inFigure 2.3(a) illustrates this graphically: it shows the number of people who would have hadto register on each single day in order to account for the drop in the registration rate at theregistration deadline between pre- and post-deadline eligibles.21 As we can see, the spike onthe day of the 18th birthday for the grey line is much larger (at 48 required registrants) than for20See Table 2.1 above and Table A.1 in the supporting information.21The registration rate is calculated by dividing the number of people registered by the 2008 election overthose eligible to register, by birthdate cohort. Given that we do not know the voting-eligible population, I usethe number of people born on the same day as a proxy for those eligible to register. The drop in the registrationrate between pre- and post-deadline eligibles is 8.7 percentage points or, in absolute numbers, roughly 47 fewerregistrations for each post-registration deadline birthdate cohort.33the black line (at only 3 estimated registrants), showing that the registration behavior of youngFloridians (black line) is inconsistent with a causal story that attributes the drop in turnoutto a concomitant administrative act that young people perform on their 18th birthday. In otherwords, neither formal rules nor concurrent administrative acts can explain the pattern in turnoutthat I identified above.What role do social mechanisms play in this context? Most of those who turn 18 shortlybefore or after the registration deadline will still live at home. Yet, children at home mightexperience a significant social shock when they turn 18 that might include comments from theirrelatives about their eligibility and possibly even encouragements to vote in the impendingelection. This might make them more likely to register and vote in that election. Can thissocial mechanism possibly explain most of the difference in turnout between pre- and post-registration deadline eligibles? Like with the other alternative explanations before, this socialmechanism can only account for the strong drop in turnout right at the registration deadline ifit leads young people who turn 18 before the registration deadline to register exactly on theday of their 18th birthday.22 However, this is not the case. A look at the number of peoplein Florida who turn 18 shortly before the registration deadline reveals that they are not morelikely to register on their 18th birthday than on any other day shortly preceding or followingtheir birthday.23 Therefore, this social mechanism is unable to account for the drop in turnoutright at the registration deadline.24Finally, it is possible that voter mobilization campaigns might be responsible for the differ-22Instead, if it only made them more likely to register a few days before or after their birthday, we would notexpect to see a discontinuity right at the registration deadline, but rather a smooth decline in turnout around thatdate.23For example, for those who turned 18 one day before the registration deadline, two people registered on theirbirthday and 3, 7, 7 and 2 people on each of the four days directly preceding their birthday. For those who turn18 two days before the deadline, two people registered on their birthday and 12, 10, 6 and 12 people on each ofthe previous four days. Note that these numbers are higher than those in Figure 2.3 above for the simple reasonthat, independent of the date they turn 18, citizens are more likely to register to vote shortly before a registrationdeadline.24Note that this does not preclude the possibility that incorrect beliefs about the registration deadline eligibilitymay be amplified or dampened by this social mechanism. To the extent that relatives might provide correct (orincorrect) information about the registration requirements for the upcoming election, this might lower (or increase)procedural information costs for young voters.34ence in turnout between pre- and post-registration deadline eligibles. This would be the caseif campaigns that are aimed at increasing turnout among first-time voters specifically targetyoung citizens who turn 18 before the registration deadline. These mobilization campaignswork under financial constraints and might focus their efforts primarily on people who theyknow are already 18 and thus, eligible to vote in order to make the most efficient use of theirresources. However, it seems highly unlikely that these campaigns would all – independentof the state they are run in – specifically target people who happen to turn 18 right before theregistration deadline.For all these reasons, incorrect beliefs about the procedural registration requirements arethe most likely explanation for the sharp drop in turnout among Americans who turn 18 rightafter the registration deadline.2.6 ConclusionNot only formal institutional rules, but also beliefs about these rules matter for voter turnout.Combining RD designs with voter file data can be a powerful empirical strategy to convincinglyevaluate the impact of these procedural beliefs on youth voter turnout. This study leverages adiscontinuity in voter turnout among young Americans depending on whether they turn 18 afew days before or after the voter registration deadline in their respective state for the first pres-idential election in which they are eligible to vote. Doing so, I am able to show that incorrectprocedural beliefs about whether one has to be 18 by the registration deadline or by ElectionDay can decrease turnout by up to a very noticeable 9 percentage points (or 25%). Crucially,incorrect procedural beliefs do not only depress voter turnout in the immediate election, buthave long-lasting effects on those young people who are affected by them. In Florida 37% ofyoung Americans who were deterred by these beliefs from registering and voting in 2008 didnot vote in the 2016 presidential election 8 years later although they would have done so, hadthey not incurred these higher procedural information costs in 2008. What is more, alternativeexplanations - in particular, concurrent administrative acts such as getting one’s driver’s license35and registering at the same time at the DMV office - are not able to account for this drop inturnout, bolstering our confidence that it is actually beliefs and not something else that mattersmost in this case.However, one caveat applies to the quasi-experimental research design used in this study.While it provides compelling evidence for the effect of procedural beliefs on voter turnout, itis based on observational data and as such, does not allow for a direct manipulation of youngvoters by the researcher. In other words, the strength of my empirical approach relies onme being able to plausibly rule out potential alternative explanations. In particular, I haveshown that two competing hypotheses – formal rules requiring young people in some statesto be 18 to register and administrative acts that Americans may perform when they turn 18and that make them more likely to register – are unable to explain the sharp drop in voterturnout right at the registration deadline. Yet, a direct test of the effect of procedural beliefson voter turnout would require an experimental (or field experimental) research design wherethe researcher is able to assign young citizens to different degrees of procedural information.Given the increased concern about the degree of misinformation in U.S. election campaigns,this represents a promising route for future research.Two implications flow from my findings. First, a methodological implication suggests thatfuture studies should more explicitly exploit temporal discontinuities in order to study pro-cedural information costs. In many areas that political scientists care about – from welfarestate benefits to a person’s citizenship or immigration status, just to name a few – eligibilitydepends on a temporal discontinuity: a person’s age, time of residence in a country, durationof employment, etc. Given that incorrect beliefs about these other deadlines are not less likelythan they are for voting behavior, exploiting these temporal discontinuities may provide impor-tant insights into the extent to which erroneous beliefs limit program uptake of welfare stateprograms or citizenship rights.A second implication is more policy-oriented. Citizens are often faced with a complex setof rules that adds an additional burden on them. This may deter especially low-informationcitizens – that is, in many cases, the very group of people that may be the primary target of a36government intervention – from enrolling into a program or registering for benefits that theyare eligible for. A common strategy in these cases is to either ignore the problem or to addressit by increasing the financial incentives for program uptake or to start a costly promotion cam-paign. Another possible solution that the findings from this paper suggest is to simplify therules, a strategy that has the combined advantaged of likely increasing program uptake withoutrequiring governments to spend additional substantial amounts of funding.37Chapter 3The Electoral Consequences of IncreasingVoter Turnout: New Evidence from aNatural Experiment3.1 IntroductionDoes increasing voter turnout help shift the balance of power from right-wing parties to thoseof the political left? A large literature in political science has produced consistent evidence thatshows substantial inequalities in political participation across societal subgroups in advancedindustrial democracies: the poor, the less educated, the young, and ethnic minorities are gen-erally significantly less likely to vote in elections than the average citizen (Brady, Verba andSchlozman 1995; Kasara and Suryanarayan 2015). Yet, whoever does not vote in electionsis also less likely to be represented politically. However, if the voices of poor people are notheard by politicians, this is likely to perpetuate or even increase existing economic inequali-ties. It is this direct relationship between political and economic inequality that Lijphart (1997)had in mind when he talked about persistently low voter turnout among the poor in advancedindustrial democracies as “democracy’s unresolved dilemma” in his 1996 American Political38Science Association (APSA) presidential address.One of the most straightforward ways to solve this dilemma – and one that was proposedby Lijphart (1997) himself – is to increase voter turnout among the poor and other politicallydisadvantaged groups by making voting compulsory for every voting eligible citizen. In par-ticular with respect to socioeconomically worse off citizens, the logic runs as follows: 1) Poorpeople are generally less likely to vote. 2) Making voting compulsory will disproportionatelyincrease their share among those who cast a vote. 3) Given poor people’s policy preferencesfor more redistribution, they are likely to vote for left-wing parties, thus increasing the voteand seat shares of these parties. 4) Once in power, these left-wing parties will adopt redis-tributive policies that will disproportionately benefit the poor (cf. Lijphart 1997; Meltzer andRichard 1981).1 While initial empirical analyses produced somewhat mixed results about thepolitical consequences of compulsory voting, especially recent empirical evidence that usessound causal identification strategies seems to corroborate the assumed causal link betweenhigher voter turnout and the electoral support for left-wing parties and policies (Fowler 2013;Bechtel, Hangartner and Schmid 2016; Carey and Horiuchi 2017).However, despite this compelling account of how increasing turnout through compulsoryvoting might lead to more electoral support for left-wing parties, several concerns are war-ranted. First, the assumed causal mechanism that links voter turnout and left-wing party sup-port fundamentally rests on the assumption that voting patterns of poor people who did notvote previously will be similar to the average poor person once they are compelled to cast avote. Yet, Selb and Lachat (2009) and Singh (2019b) show that non-voters who are compelledto vote tend to be more politically disaffected, less informed about politics and less politicallyinterested. Thus, the extent to which these people are similar to other poor people who used tovote in previous elections and really will cast their vote for left-wing parties is unclear. Sec-ond, if we assume that parties behave strategically and want to maximize their electoral gains1Given decreasing levels of voter turnout (Franklin 2004) and increasing economic inequality in many ad-vanced industrial democracies (e.g. Atkinson and Piketty 2007; Bartels 2008), it is not surprising that we haverecently seen a renewed interest in this electoral reform among both scholars of voting behavior (Bechtel, Hangart-ner and Schmid 2016; Singh 2019a,b) and political theorists (Chapman 2019; Umbers 2018).39(e.g. Downs 1957), the introduction of compulsory voting is likely to affect partisan strate-gies. Right-wing parties might respond to an electoral reform that disproportionately increasesturnout among poorer voters by shifting their policy positions to the left. In addition, the intro-duction of compulsory voting is likely to directly affect partisan competition within electoraldistricts. Candidates who previously thought they did not stand a good chance to win might re-consider their decision and run a campaign if they know that, despite possibly smaller chancesof winning, their own supporters are compelled to cast a vote. These general concerns aresupported by new research that finds that the empirical evidence is more mixed. Two recentquasi-experimental studies have used convincing, sound causal identification strategies and ap-plied them to new contexts, but failed to find empirical support for the assumed relationshipbetween turnout and left-wing support (Ferwerda 2014; Miller and Dassonneville 2016).Given these mixed findings, this paper presents a new perspective on the political participation-left vote relationship. It takes into account the greater political disaffection of the group of poorvoters who have been compelled to vote as well as parties’ strategic response to a reform thatis likely to directly affect electoral competition between parties. This theoretical account nolonger assumes that power shifts from the political right to the left, but rather expects to seeonly minor changes to each parties’ electoral fortunes and an overall balance of power betweenthe main parties that is largely unchanged by the reform. Possibly small increases for bothleft- and right-wing parties are assumed to be associated with small decreases in seat and voteshares for independent candidates and smaller parties.These theoretical expectations are tested empirically in a reanalysis of a crucial case thatis often seen as the most compelling empirical evidence for the assumed positive relationshipbetween voter turnout and left-wing parties’ vote and seat shares: the adoption of compulsoryvoting laws at different points in time across Australian states in the first half of the 20th cen-tury (Fowler 2013). Yet, the Australian case represents a major puzzle: If compulsory votingstrongly shifted the balance of political power in favor of the Australian Labor Party (ALP),why did the major non-Labor parties consistently support this electoral reform? Ignorance ormisperceptions about the political consequences of the reform might explain initial support40among non-Labor parties in the 1910s. However, they can hardly account for continued sup-port among these parties for a reform that threatened to seriously undermine their chances ofwinning elections twenty or thirty years later, that is, at a time by which they would have hadample evidence of this policy’s effects from other Australian states.The present quasi-experimental study builds on a similar study by Fowler (2013) and im-proves on it in three ways: First, the analysis uses more fine-grained, district-level electoraldata. This allows to test the extent to which compulsory voting affected parties’ strategic cam-paign behavior, and in particular the non-Labor parties’ decision to run their own candidates inpreviously unchallenged Labor districts after the adoption of compulsory voting. While Fowler(2013)’s analysis is based on state-level aggregate data that assumes that compulsory voting didnot affect candidate entry in electoral districts, the more fine-grained district-level data in thisstudy allows for a direct empirical test of this assumption. Second, this study adds to Fowler(2013)’s cross-state analysis a within-state difference-in-differences design. It leverages varia-tion in the treatment intensity, that is, in the extent to which compulsory voting boosted turnoutacross electoral districts within each state, to estimate the effect of this policy on a party’s voteshare. The advantage of this within-state analysis is that all voters are exposed to the sameelection campaign, parties and policy platforms and cast their vote on the same day in a givenstate election. Thus, such a within-state design provides a context where the parallel trendsassumption that is required for this kind of analysis is highly plausible. Third, the presentstudy extends the original study which was exclusively focused on the state level to federal (orCommonwealth) elections and tests the effect of compulsory voting in Commonwealth Houseof Representatives elections in Australia.An initial replication exercise with the district-level data that are used for analysis in thisstudy shows that they produce – once they are aggregated to the state level – the exact sameresults as Fowler (2013). Yet, the complete set of analyses in this paper gives a much morenuanced view on the political consequences of increased voter turnout as a result of compulsoryvoting. Two key findings emanate from this study: First, results both from the state-level andfederal-level analyses are more mixed for Australia than previous research suggested. There is41empirical support for a class voting view that suggests that compulsory voting helps left-wingparties and hurts those on the right. However, a more thorough look at within-state and within-Commonwealth variation reveals that there is more variation, with some states aligning moreclosely to a class voting view and others with a minimal effects view. As a graphical analysis ofwithin-state variation shows, additional factors – such as parties’ strategic decisions to competein a district and geographic size of a district – seem to matter too and decisively shape therelationship between turnout and left-wing party support. Second, parties respond strategicallyto compulsory voting by adjusting their campaign strategies, such as the number of candidatesthey run in previously unchallenged seats that are held by the opposing party. Accountingfor parties’ strategic behavior, the positive relationship between voter turnout and a left-wingparty’s vote share becomes weaker. For example, Fowler (2013)’s finding that compulsoryvoting increased the ALP’s vote share by more than 9 percentage points in Australia is likelyto be a very optimist estimate, with the true effect likely to be a couple of percentage pointssmaller.This study makes two main contributions. First, it directly speaks to the debate in politicalbehavior about the political consequences of higher voter turnout. Based on a sound causalidentification strategy and new, more fine-grained district level data, this study advances thisdebate by providing a thorough empirical analysis of a crucial case for this literature. Doingso, the study raises questions about the empirical evidence that is used to support the dominantclass voting approach and presents an alternative approach that focuses more explicitly on thepsychological differences between poor voters who already used to vote before the adoption ofcompulsory voting and those who did not. Second, it speaks to the literatures on the effective-ness of electoral reforms and on economic inequality. In particular, it highlights the limitationsof institutional fixes such as compulsory voting laws in helping solve the problem of increasingeconomic inequality across advanced industrial democracies.423.2 Compulsory Voting, Politically Disengaged Citizens, andPartisan SupportEver since Lijphart (1997) advocated compulsory voting as the most effective means to ad-dress “Democracy’s unresolved dilemma” of unequal political participation, several studieshave analyzed the consequences of compulsory voting on political participation and electionoutcomes. Yet, while a large literature has shown that compulsory voting substantially in-creases voter turnout (Hirczy 1994; Birch 2008), the findings about the political consequencesof this electoral reform are less clear.In this context, I present a theory of the minimal effects of compulsory voting on politicsand contrast this approach with the dominant view in this literature: the class voting approach,which is also called the “bias correction argument” (Bechtel, Hangartner and Schmid 2016).Both approaches start from the assumption that an exogenous shock2 to voter turnout suchas compulsory voting targets citizens who are different from those who already vote: in ad-vanced industrialized countries they are generally socioeconomically worse off (Brady, Verbaand Schlozman 1995; Kasara and Suryanarayan 2015), less educated, and less interested inpolitics. By exogenously increasing the cost of voting (usually by fining those who do not casta ballot), this group of citizens who would otherwise have abstained from voting is compelledto participate in elections. Based on the dominant class voting approach, these previous non-voters’ lower socioeconomic status makes them more likely to favor redistributive policies andthus, to have policy preferences that align with the policy positions of left-wing parties. As aconsequence, increasing the share of poor voters among the electorate as a result of compulsoryvoting will mainly help left-wing (or Labor) parties and boost their overall electoral support atthe expense of right-wing parties or candidates (Bechtel, Hangartner and Schmid 2016; Fowler2013; Lijphart 1997; Meltzer and Richard 1981).2While the decision to adopt compulsory voting is likely to not be exogenous, the shock to voter turnout thatit produces can be considered exogenous to the extent that the policy was not primarily adopted to improve aparty’s electoral fortunes at the cost of other political parties, but rather for other, more practical (e.g. efficiency,legitimacy of the election outcome) reasons.43The empirical evidence for the class voting view is mixed, with some researchers being ableto find evidence that higher voter turnout benefits left-wing parties (Herron 1998; Mackerrasand McAllister 1999) and others not (Highton and Wolfinger 2001; Rubenson et al. 2007). Sev-eral recent studies have attempted to advance this debate by using natural experiments in orderto more convincingly isolate the causal effect in question. For example, Bechtel, Hangartnerand Schmid (2016) leverage cross-cantonal variation in compulsory voting laws in Switzer-land to estimate the effect of increasing voter turnout on referendum outcomes, finding thatin cantons with compulsory voting support for left-wing policies increased substantially. Ina study most similar to my own analysis, Fowler (2013) uses variation in the timing of theadoption of compulsory voting across Australian states to gauge the effect of an exogenousshock to voter turnout on the Australian Labor Party’s vote and seat shares. Overall, that studyfinds a strong effect, with the Labor Party’s vote and seat shares increasing by between 7-10percentage points after voting was made mandatory.3 Thus, these quasi-experimental studiesseem to have settled the debate in favor of the class voting approach. However, other studieswith similarly convincing quasi-experimental research designs fail to find evidence for a rela-tionship between voter turnout and left-wing party support in other countries (Ferwerda 2014;Miller and Dassonneville 2016; van der Eijk and van Egmond 2007). One interpretation ofthese conflicting findings that is consistent with the class voting approach is that increasingturnout through compulsory voting is more likely to tanslate into a significant boost of theleft-wing vote share if the institutional context is conducive to it, but not in others. For exam-ple, Ferwerda (2014) suggests that finding a class-based effect of compulsory voting is ratherunlikely in electoral contexts that use proportional instead of majoritarian electoral formulas.Given the generally larger number of effective parties and an overall higher rate of turnout un-der proportional representation, it is less likely to see under such a system the combination of astark change in voter turnout with a single left-wing party that benefits from this change, whichare both required for detecting any meaningful effect (Ferwerda 2014, 249). Instead of focus-3These two studies leverage within country variation in the timing of the adoption of compulsory voting acrossgeographic subunits. Other quasi-experimental studies have tried to estimate the political consequences of com-pulsory voting through a cross-country comparison (e.g. Carey and Horiuchi 2017).44ing on contextual factors that might moderate class voting, this paper suggests an alternativeinterpretation to the class voting view that questions the plausibility of the assumed positiverelationship between voter turnout and left-wing partisan support.This study proposes a second, alternative perspective on the relationship between turnoutand left-wing partisan support, which I call the minimal effects view. It assumes a fundamentaldifference between traditional voters who are poor, but who would have voted even in theabsence of compulsory voting and those poor voters who only vote because they are compelledto do so. People in this latter group are expected to be generally more politically disaffected,less interested in and less informed about politics (Selb and Lachat 2009; Singh 2019b). Justlike other poor voters, they might have policy preferences for economic redistribution.4 Yet,due to their limited political knowledge, they are less likely to know which party is most likelyto represent their policy preferences and thus, also less likely to have clearly formed partisanpreferences and partisan attachments than socioeconomically similar citizens who used to votein past elections (Selb and Lachat 2009).5What does this lower level of political sophistication mean for their party choice on ElectionDay? These new voters are likely to resort to strategies that are known to be used by voters inorder to compensate for a lack of knowledge. One possible strategy is to avoid taking a politicaldecision by simply casting a blank vote. This strategy seems especially appealing to thosevoters who do not know what they want, are unaware of the candidates’ policy positions and4Recent research in Economics has started to question Meltzer and Richard (1981)’s assumption that poor peo-ple generally share preferences for economic redistribution. Alesina and Ferrara (2005) and Alesina, Stantchevaand Teso (2017) empirically show that a person’s desire for redistribution is highly context dependent, with indi-viduals believing in intergenerational mobility holding weaker preferences for redistribution. In addition, Fisman,Kuziemko and Vanutelli (2017) find that redistributive preferences among socioeconomically worse off citizensare usually weaker than often assumed, with many of them preferring a more targeted taxation of the rich thanMeltzer and Richard (1981)’s theory would suggest. Finally, Kuziemko et al. (2015) show that even if poor peo-ple hold preferences for redistribution, their possible lack of trust into the government might make them not verysupportive of redistributive policies. Thus, this literature provides an alternative account of why compelling thepoor to vote might not increase support for left-wing parties that is consistent with the minimal effects view thatis discussed in this paper.5The situation would be different if voters could directly vote on policies themselves, as is the case in referen-dums (Bechtel, Hangartner and Schmid 2016). However, most democracies are representative democracies wherecitizens choose parties or politicians who represent their interests in the political arena.45do only care about the election outcome to a limited extent. Previous research that has found anincrease in invalid votes after the introduction of compulsory voting provides evidence that atleast some voters seem to resort to this strategy (Mackerras and McAllister 1999; Singh 2019b).Another strategy for voters is to rely on the pieces of information that are most easily availableto them when making their vote choice. Voters who, for example, lack details about the policypositions of the candidates in their electoral districts are likely to lean towards candidates thathave been more visible to them and for which they can assume that they have more politicalexperience than alternative candidates who are completely unknown to them (Luskin 1990;Zaller 1992). In most cases, this will be a candidate from one of the major parties because sheis more likely to be seen as the default candidate than an independent candidate or a candidatefrom a minor party where it will often be less clear to an uninformed voter what policies thatcandidate stands for. Name recognition may reinforce this tendency. Candidates from largerparties tend to have been around for longer, as is the case for the party’s name. Thus, for lowinformation citizens, the default vote choice is likely to be for such a candidate or party. In theabsence of strong partisan preferences for either a left-wing or right-wing candidate (cf. Selband Lachat 2009), uninformed citizens who are compelled to vote are expected to support mainparties’ candidates over independent candidates or candidates from minor parties. Unless poorvoters are embedded in the Labor movement by, for example, being members of a trade union,this lack of political knowledge and interest could make it difficult for them to clearly aligntheir material interests with the party system. Presented with right-wing candidates that arevying for the support of poorer middle class and rural voters, it might be hard for many of themto clearly identify the party that they think is mostly likely to improve their own economicsituation in the medium term. Faced with the task of choosing a candidate from the list ofcandidates on election day, these voters’ choice is likely to resemble a coin flip between themajor party alternatives, mostly driven by idiosyncratic factors. As a result, compulsory votingis not expected to produce a general shift in the relative power (i.e. vote shares) between left-wing and right-wing parties, but rather to modestly increase support for both parties or haveno effect at all. At the same time, there could be a small decrease in support for independent46candidates or candidates from minor parties.In addition, parties tend to behave strategically in order to maximize their electoral returns(Downs 1957). As a result, they might react to the introduction of compulsory voting by adjust-ing their behavior accordingly. For example, if we assume that in a two-party context partiescompete for the median voter (Downs 1957), right-wing parties may respond to a leftwardshift among the median voter as a result of compulsory voting (cf. Meltzer and Richard 1981)by shifting their own policy position leftward too. However, if they do, their behavior wouldminimize any negative effect of compulsory voting on their vote shares. Beyond adjustingtheir policy position, right-wing and left-wing parties might also react to the introduction ofcompulsory voting by changing their campaign strategies. Campaigning is costly and a party’sresources are limited. As a result, parties might strategically choose not to waste money onrunning candidates in districts where they do not stand a serious chance to win and rather fo-cus their attention and resources on more competitive district. Considerations such as these areusually made responsible for the comparatively high number of uncontested seats in Australianelections in the first half of the 20th century (Sharman 2003). However, a party’s calculus ofwhether to run their own candidate in a district is likely to be directly affected by the adop-tion of compulsory voting. After compulsory voting, every voting-eligible citizens has to vote.Even in districts where, for example, right-wing parties do not stand a good chance of winningand where they previously did not run a candidate, their supporters are now compelled to poll.So all else equal, the chances of a right-wing candidate to win this district increase, as does thelikelihood of the right-wing party to enter into the electoral race in this district. This in turnwould reduce the positive effect that compulsory voting, and higher voter turnout, are expectedto have on a left-wing party’s vote share, thus potentially contributing to a minimal effect ofthis reform on the overall partisan balance of power.To test this minimal effects view, the empirical section reanalyzes a crucial case that isoften seen as the most compelling evidence for the class voting argument: the gradual adoptionof compulsory voting at the state and federal levels in Australia in the first half of the 20thcentury. The present study improves on Fowler (2013)’s analysis by leveraging more fine-47grained district electoral data for the analysis. The key advantage of these data is that theyallow to more thoroughly test the previous findings that found a positive relationship betweencompulsory voting and left-wing partisan support in Australia. First, these data also includeinformation about unchallenged districts in these elections and thus, allow for a direct empiricaltest of how the introduction of compulsory voting affected the main parties’ decision to put uptheir own candidates in less competitive districts in state elections. Second, these data are thebasis for additional, arguably more internally valid analyses. These analyses do no longer relyon cross-state variation, but instead leverage variation in changes in turnout across electoraldistricts within a single state to estimate the effect of compulsory voting on a party’s voteshare.3.3 Compulsory Voting in AustraliaIn the late 19th and early 20th century, Australia was a pioneer as far as electoral rules areconcerned and it was among the first countries to adopt universal suffrage, the Australian - orsecret - ballot, instant run-off voting (also known as alternative vote) and compulsory voterregistration and voting. Compulsory voting was introduced at different points in time acrossAustralia. Queensland was the first of the six Australian states to use it for its 1915 stateassembly election. The federal government passed legislation in 1924 to make voting in thefederal House of Representatives elections mandatory across Australia. After that, all fiveremaining states followed suit and adopted compulsory voting in 1926 (Victoria), 1928 (NewSouth Wales and Tasmania), 1936 (Western Australia) and 1941 (South Australia; see Table3.1).To what extent does compulsory voting in the different Australian states fulfill the require-ment that its adoption was as-if randomly assigned? This requirement would be violated if theadoption of this policy was endogenous to strategic considerations by the political parties thatpushed for its adoption. For example, it is possible that the Australian Labor Party supportedthe adoption of compulsory voting because it anticipated an electoral advantage from such a48Table 3.1: The Adoption of Compulsory Voting in State and Commonwealth Elections AcrossAustraliaState First Election under CV Parties Supporting CVQueensland 1915 Liberal (non-Labor) PartyCommonwealth 1925 Both Labor and non-Labor PartiesVictoria 1927 Both Labor and non-Labor PartiesNew South Wales 1930 Both Labor and non-Labor PartiesTasmania 1931 Both Labor and non-Labor PartiesWestern Australia 1939 Both Labor and non-Labor PartiesSouth Australia 1944 Both Labor and non-Labor PartiesNote: With the exception of South Australia which had a Labor government at the time of the adoption ofcompulsory voting, all other states and the Commonwealth made voting mandatory while non-Laborgovernments were in power. Except for Queensland, compulsory voting laws received unanimous support fromall MPs in the respective state and Commonwealth legislature.reform.The literature cites several reasons for the adoption of compulsory voting in Australia.While partisan motivations for the adoption of compulsory voting have mattered to some ex-tent, they usually played a secondary role and were largely limited to early considerations ofthis policy in the 1910s (Brett 2019). For example, the incumbent Liberal (i.e. non-Labor)government under Premier Digby Denham in Queensland adopted compulsory voting in 1914largely because it hoped that it would hurt the Australian Labor Party and thus, help the Lib-erals avoid or at least reduce the size of the anticipated defeat in the 1915 state election(Brett2019; Evans 2006). This is in line with Helmke and Meguid (2010)’s claim that compulsoryvoting was sometimes supported by right-wing parties as a means to counter the electoral threatfrom the left. However, as it turned out, mandatory voting did not disadvantage the Labor Partyand they clearly won the 1915 election in Queensland. Possibly as a result of these early expe-riences with compulsory voting, partisan considerations played less of a role in the followingdebates that led to the adoption of compulsory voting after 1924 in the Australian states andin Commonwealth elections. Instead, rather technical concerns are frequently mentioned asthe decisive factors for the broad cross-partisan support for this electoral reform (Brett 2019;Evans 2006; Mackerras and McAllister 1999). First, this reform was aimed at reducing the49costs of campaigning and mobilizing voters for candidates in a context where an increasingnumber of candidates complained about the burden of having to drive voters in cars to thepolling places on election day (Brett 2019, 132). Second, an immediate impetus for the adop-tion of compulsory voting was an election with a particularly low voter turnout and the concernamong politicians that this might undermine their political legitimacy (Brett 2019, 134, 137).Thus, in contrast to other electoral reforms such as the adoption of the alternative vote or theredistricting of electoral boundaries, compulsory voting was a relatively uncontentious elec-toral reform after the mid-1920s in Australia and rather seen as a technical solution to verytangible problems that candidates faced in their districts. This rather technical perception ofthe reform is supported by the fact that, with the exception of Queensland in 1914, it enjoyedbroad cross-partisan consensus in state assemblies and in the House of Representatives. Theadoption of compulsory voting can therefore largely be considered exogenous to the Labor orthe main non-Labor parties’ strategic electoral considerations.Once compulsory voting was implemented, every voter was supposed to go to the pollingplace and cast a ballot on election day. Eligible citizens who did not vote and could not providea valid excuse for not voting had to pay a fine of up to two pounds, a substantial amountof money that corresponds to A$160 or US$120 in today’s money and that is likely to haveprovided a strong financial incentive – especially for poorer people – to participate in elections.6Most crucially for the test of the minimal effects account, previous research has shown thatit is especially poor voters who were less likely to vote in Australia in the early 20th centuryunder a system where voting was not compulsory (Fowler 2013). As a consequence, makingvoting mandatory for these people increases the share of poor voters among the electorate,which is a key prerequisite for the test of the proposed theory.6The value of two pounds in 1925 is converted into 2018 dollars based on the Reserve Bank of Australia’spre-decimal inflation calculator, https://www.rba.gov.au/calculator/annualPreDecimal.html (last accessed on May11, 2019).503.4 Data and Methods3.4.1 Data and Dependent VariablesThe empirical analysis in this study is based on a dataset that includes electoral district datafor state and federal (Commonwealth) Lower House elections in Australia for the period 1910-1950. To create this dataset, I digitized historical voting records from state assembly electionsin all six states and for the Commonwealth House of Representatives (see Carr 2019; Hughesand Graham 1974a,b, 1976). These district-level data have two key advantages over data that isaggregated at the state level, which has been used in previous studies (e.g. Fowler 2013). First,they allow for a direct empirical test of the extent to which previous estimates of the effect ofcompulsory voting on the Australian Labor Party (ALP)’s vote share have been confoundedby that party’s changing share of unchallenged seats. The state-level aggregate vote share forthe ALP in a given state election is commonly calculated by adding all votes that were cast forthis party in this particular election over all voting-eligible citizens in this state who reside inelectoral districts that were challenged.7 In other words, unchallenged districts with only onecandidate who is not opposed by any other candidate within her district are commonly treatedas missing data and simply excluded from this calculation. This procedure is a reasonableapproach to estimate a valid vote share measure for a given state for a particular election.However, it may undermine the validity of analyses that rely on a before-after comparison ofa party’s vote share over time (for example, in a difference-in-differences design) if not onlythe vote share, but also the number or the kind of districts which are unchallenged are affectedby over-time variation. Unless the exact same districts are unchallenged in all elections for agiven state and thus excluded from analysis, this is a potential problem. Having the more fine-grained district-level vote shares for each political party allows me to directly test to whichextent adjusting for these unchallenged seats for a given party affects that party’s electoralfortunes.7See, for example, the state-level aggregate election data from the Australian Politics and Elections Database,which is available at: http://elections.uwa.edu.au/.51The second advantage is that more fine-grained district-level data provide additional op-portunities for analysis which may give new insights into how compulsory voting affects aleft-wing party’s electoral support. In this specific study, they allow to add to the cross-statecomparison of the effect of compulsory voting an analysis that leverages within-state variationacross electoral districts. This within-state analysis can use variation in turnout changes acrosselectoral districts and link them, for example, to changes in left-wing party support to providea more direct test of the claim that increases in voter turnout help left-wing parties. The mod-elling choices that are associated with this kind of within-state analysis and the advantages ofthis approach are discussed in more detail below.The key dependent variable is a party’s vote share over all valid votes that were cast in agiven election. Further dependent variables are the overall voter turnout and a party’s share ofuncontested seats among all the uncontested seats in a given election. The coding of the partyvariables that is required to determine a party’s vote and seat shares needs some explanationdue to Australia’s relatively fluid party labels before World War II. The Australian Labor Party(ALP) as well as the Lang Labor Party both targeted working-class voters and are coded asone single Labor party. The latter temporarily split from the Australian Labor Party after the1929 economic crisis and competed elections in the early 1930s as a separate party, but bothparties merged back together into a single party in the mid-1930s. This Labor party was usuallyopposed by a coalition of two non-Labor parties: the main non-Labor party, which, dependingon the election and the state, is either called the Liberal, National or United Australia Party, anda smaller Country Party. Given that both parties usually formed electoral coalitions in whichthey supported each other’s candidates against Labor, they are coded as a single party which Icall ”non-Labor coalition”. Candidates from smaller parties (including the Social Credit Partyand the Communist Party) and all independent candidates did usually not officially cooperatewith either of the two main parties and are all grouped into a third category for which the terms”other parties” or ”independent candidates” are used interchangeably in this paper.523.4.2 Research DesignThe challenge to empirically estimate the political consequences of increased voter turnout isthat changes in voter turnout are usually not exogenous, but directly influenced by the specificelectoral context of any given election such as, for example, the competitiveness of the electionor the attractiveness of the candidates and their programs. This makes it difficult to assess howhigher voter turnout would impact a given party’s electoral success. To limit the risk of con-founding, this study leverages a natural experiment: the gradual adoption of compulsory votingat different points in time across Australian states as well as within-state variation in treatmentintensity across electoral districts. A possible concern with this research design might be thatthe timing – which state switched when – could have been shaped by strategic considerations.However, a look at the relevant literature suggests that this concern is likely to be unfoundedin the Australian context. The timing – except for the early adopter Queensland in 1914 – wasgenerally less influenced by strategic considerations and rather the result of politicians tryingto reduce voter mobilization costs and potential threats to the legitimacy of election outcomes(Brett 2019; Evans 2006; Mackerras and McAllister 1999). More specifically, in some cases(e.g. for the federal level and in South Australia) a strong motivation for reform was providedby an immediately preceding election with a particularly low level of voter turnout (Brett 2019,134, 137). In other cases, the adoption of compulsory voting seems to have followed a patternof policy diffusion, such as when Victoria, New South Wales and Tasmania – and with a delayof a few years, possibly Western Australia – all made voting mandatory for their lower-houseelections shortly after it was introduced at the federal level in 1924 (Brett 2019, 137). For thisreason as well as those that were discussed in detail above, the adoption of compulsory votingproduced a shock to voter turnout that can arguably be considered as exogenous. In particu-lar, compulsory voting received cross-partisan support in parliament and seemed to have beenprimarily adopted for practical, as opposed to strategic, considerations in order to make theadministration of elections less costly and more efficient.For the empirical analyses in this study, two different kinds of difference-in-differences53designs are used. One design leverages cross-state variation in the timing of when a particularstate adopted compulsory voting to estimate the effect of this electoral reform, and the boost toturnout that followed, on each parties’ vote share. A second design uses within-state variationin treatment intensity, that is, changes to voter turnout as a result of compulsory voting. Thisanalysis compares changes in voter turnout across electoral districts from before and after theintroduction of this reform to changes in partisan support in these districts within a single state.Both analyses use the same dataset of district electoral data that was described above. Forthe cross-state analysis, these data are aggregated into state-level measures of voter turnoutand vote shares for each party. This design follows the modelling choices in Fowler (2013) inorder to allow for a direct replication of the findings from that study. The within-state analysisrequires the use of district-level electoral data. It follows similarly designed studies on thewomen’s suffrage (Morgan-Collins and Teele 2016) and only uses data from two electionsin each state, that is, from the election that directly preceded and the one that followed theintroduction of compulsory voting in a state.For the cross-state analysis, the following model is used for state elections between 1910and 1950:Yst = αs + γt + δDst + st (3.1)where Yst is a party’s vote share in state s in year t, αs are state fixed effects, γt yeardummies, Dst an indicator showing whether state s had compulsory voting in year t and  theerror term. To adjust for dependence across elections from the same state, standard errors areclustered at the state level (Angrist and Pischke 2009). The dataset includes data for stateassembly elections in all six Australian states and states had on average 14 elections in thistime period.In this cross-state difference-in-differences design, the estimation of the effect of compul-sory voting crucially hinges on the plausibility of the parallel trends assumption. Variation inthe profiles and policy platforms of the state Labor and non-Labor parties across states, state-54specific election campaigns and the fact that dates for state elections usually lie months apartfrom each other are all reasons to question the validity of this assumption. In addition, the as-sumption of parallel trends across states for the 1910-1950 period might be further challengedby these states’ electoral reforms such as the changes in electoral formulas (for an overview,see Farrell and McAllister 2006, 50-51) or the redrawing of electoral district boundaries.8Therefore, this study complements the cross-state analysis with an arguably more soundand internally valid within-state difference-in-differences design that leverages variation intreatment intensity across electoral districts within a single state (cf. Morgan-Collins and Teele2016). In particular, this approach uses the change in voter turnout in a given district betweenthe pre- and the post-reform elections as a measure of that district’s treatment intensity. Dis-tricts with a larger share of poor citizens are, all else equal, expected to have a lower voterturnout pre-reform and thus to see a stronger increase in voter turnout from the pre- to the post-reform election than districts where the poor represent a smaller share of the electorate. If theclass voting view is correct, we should therefore see the Labor vote share to increase more inthose districts where the increase in voter turnout was larger. The fact that all electoral districtswithin a state were exposed to the exact same state Labor and non-Labor parties, experiencedthe same election campaign and had the election on the same day provides strong support tothe plausibility of the parallel trends assumption in this context. The following model is usedfor the empirical analysis of the within-state variation:∆Yd = α + β∆Turnoutd + ∆d (3.2)where ∆Yd is the change in a party’s vote share in each electoral district from the elec-tion preceding to the election following the adoption of compulsory voting in a given state,8For example, New South Wales switched from a majority-runoff system with single-member districts to thesingle transferable vote in multi-member districts in 1918, and then to alternative vote in single-member districtsin 1926 (Farrell and McAllister 2006, 50). The only two states that did not change their electoral formula between1910 and 1950 are Tasmania, which used the single-transferable vote in multi-member districts, and WesternAustralia with the alternative vote (and compulsory preferences).55∆Turnoutd the change in turnout between the two elections and d the error term.9 ∆Turnoutdis a continuous variable that measures the intensity with which each electoral district was ex-posed to the compulsory voting treatment. Electoral districts with a larger number of poorvoters who did not vote previously are expected to see a stronger increase in turnout as a resultof the introduction of compulsory voting than districts with fewer such citizens.The within-state analysis only includes electoral districts in the analysis where more thana single candidate competed in both elections. This means that districts that are unchallengedeither in the election that precedes the adoption of compulsory voting or in the election thatimmediately follows this reform are completely dropped from the analysis. This approachmakes sure that it is possible to calculate a meaning change in turnout and a party’s voteshare between the pre- and the post-reform elections. Despite several changes to their electoralformulas in the 1910-1950, no state made changes to their electoral formula around the timethat they adopted compulsory voting.10Furthermore, the redistricting (or redistribution) of electoral boundaries in between thetwo elections is a potential threat to the validity of this within-state design. For example, ifright-wing politicians strategically redraw the electoral boundaries in order to shield them-selves from the negative electoral effects of compulsory voting, the above within-state modelspecification would produce downwardly biased estimates for the relationship between voterturnout and a left-wing party’s vote share. For most of the states in the analysis – Queensland,South Australia, Western Australia, Tasmania as well as for the federal level (see further be-low) – electoral boundaries remained unchanged in between the elections in question (Bennettand Bennett 1986; Hughes and Graham 1968). Furthermore, the major redistricting in New9With observations from only two time points, this is equivalent to an equation with district and year fixedeffects (see Angrist and Pischke 2009).Ydt = βd + γt + δ(Turnoutd ∗ dt) + dt (3.3)In this equation βd are district fixed effects, γt a dummy for the election year with compulsory voting, the vari-able Turnoutd the share of people who voted in each district and dt a time dummy for observations after theintroduction of compulsory voting.10New South Wales modified its alternative vote in single-member districts in 1928 by replacing the contingentvote with compulsory preferences (Farrell and McAllister 2006, 50).56South Wales that turned its previously multi-member districts into single-member districts andthat accompanied this state’s transition from single transferable vote to the alternative votehappened before 1927. Thus these changes did not affect the 1927 and 1930 state elections.However, in New South Wales electoral boundaries were changed again by the non-Labor gov-ernment in 1929 (Clifford, Clune and Green 2006) as were electoral boundaries in Victoria in1926 (Hughes and Graham 1968, 464), which each redistricting falling in between the 1927and 1930 elections in New South Wales and those in 1924 and 1927 in Victoria respectively.While it is not clear to what extent incumbent right-wing governments in these two states wereable to use redistricting to shield their party from the effects of compulsory voting, this redis-tricting might undermine the validity of the effect estimates from these two states. As I havenot been able to locate material yet that would allow me to identify the districts which wereaffected by this redistricting, empirical results from these two states have to be interpreted withcaution.11For the estimation of the effect of compulsory voting on parties’ vote share at the federallevel, two empirical strategies are chosen that to some extent mirror the analyses at the statelevel that were explained above. One is a direct application of the within-state difference-in-differences design to the federal level. The model that is used for this analysis is the same as inequation 3.1 above, with the only difference that state districts are replaced by federal districtsand state lower-house elections by the 1922 and 1925 elections to the Commonwealth Houseof Representatives. Importantly for the validity of this approach, there were no changes to theelectoral formula or to the electoral district boundaries between 1922 and 1925.12Analogous to the cross-state analysis, the second approach uses variation in the timing ofwhen the reform was adopted across different levels of government. However, in contrast tothe state-level analysis, the estimation of the effect of compulsory voting for Commonwealth11A name change for an electoral district is frequently indicative of a major redrawing of its electoral boudaries.Thus, districts which changed their name as a result of the redistricting were dropped from the analysis for thesetwo states.12The transition from single-member plurality to alternative vote for House of Representatives elections hadalready happened several years prior, in 1918.57(federal) elections requires cannot simply adopt the cross-state difference-in-differences designthat was explained above. Due to the simultaneous introduction of mandatory voting across allsix Australian states for the 1925 federal election, there is no geographic variation in the tim-ing of when a given state adopted compulsory voting in federal elections. Thus, a standarddifference-in-differences design is not applicable in this setting. Yet, it is possible to compareelections within each state by leveraging information from elections at different levels of gov-ernment. More specifically, we can compare voting behavior right before and after the adoptionof compulsory voting at the federal level with voting behavior at the state level that was notdirectly affected by this electoral reform.13 Such an analysis would zoom in a specific periodin Australian politics: the adoption of compulsory voting at the federal level between the 1922and 1925 Commonwealth elections and state elections in all six states that happened around thesame time, that is, in the mid-1920s right before and after the adoption of the reform in 1924.Luckily for our purpose, both in the states and at the federal level a party system had emergedat the beginning of the 19th that, despite some regional differences,14 pitted everywhere theLabor Party against non-Labor coalition parties (Jaensch 1994, 20). Importantly, while issuessuch as conscription in World War I or policy responses to the 1929-1930 economic crisis ledto divisions and splits within the Labor Party (e.g. Jaensch 1994; Weller and Fleming 2003, 21,25), the early to mid-1920s were a time of little frictions between the state branches themselvesor the state branches and the federal branch of the Labor Party. During this time, a primaryfocus of the Australian Labor Party is on promoting social welfare for the poor and improvingthe working conditions of those in the labor force (McKinlay 1981, 55-57). Therefore, there isno obvious reason to assume that voting patterns are any less likely to move in tandem acrossthe federal and the state level within each state in this specific period of time (i.e. between 1922and 1926) than they do across states (see Jaensch 1994; Moon and Sharman 2003; Sharman13Such an empirical strategy is possible because the adoption of compulsory voting at the state level doesnot temporally coincide with the introduction of this policy at the federal level, with most states making votingcompulsory a few years after the federal parliament did so.14For example, it is regularly emphasized that the New South Wales state branch of the Australian Labor Partywas more conservative than other state branches of this party (e.g. Smith 2003, 60).58and Sayers 1998). Yet, the results section below provides an empirical test of this assumptionto directly address concerns in this respect.The model that is used for this specific analysis has similarities with the cross-state analysisthat is discussed above. This model is run on data from both the state and the federal level, butonly for elections directly preceding and following the adoption of compulsory voting at thefederal level in 1924:Ydse = αs + θe + δDse + dse (3.4)where Ydse is a party’s vote share in district d within state s in election e, αs are fixed effectsfor each Australian state, θe a dummy indicating whether the election occurred right after or inthe same year as the adoption of compulsory voting as opposed to occurring right before 1924,Dse the treatment of compulsory voting and  the error term. As with equation 3.1, standarderrors are clustered at the state level.3.5 ResultsThe results section starts with difference-in-differences analyses of the adoption of compulsoryvoting in Australian states in the early 20th century. Then, the section proceeds to extendthe view by looking at the federal level, that is, compulsory voting in the elections for theAustralian House of Representatives.3.5.1 State-Level ResultsTable 3.2 presents the results from a difference-in-differences design that leverages the gradualadoption of compulsory voting in Australian states between 1914 and 1943. Results in columns1 and 2 are a direct replication of Fowler (2013)’s table 2, columns 1 and 3. The adoption ofcompulsory voting led to a strong increase in voter turnout in these states by an average 24.7percentage points (see column 1). In other words, voter turnout that tended to be around 68%59Table 3.2: Effect of Compulsory Voting in State Assembly ElectionsTurnout Vote Share without Unch. Seats ALP Unch. Vote Share with Unch. SeatsALP NLC Ind Seat Share ALP NLC Ind(1) (2) (3) (4) (5) (6) (7) (8)Compulsory Voting .247 .093 -.019 -.080 -.209 .066 -.001 -.072(.044) (.028) (.027) (.046) (.096) (.018) (.027) (.040)State fixed effects X X X X X X X XYear fixed effects X X X X X X X XN 86 86 86 86 601 86 86 86R2 .930 .814 .832 .765 .220 .779 .762 .760Note: Table presents difference-in-differences estimates. Unch.: Unchallenged. Turnout data are aggregated tothe state level. Standard errors in parentheses are clustered at the state level. ALP: Australian Labor Party; NLC:Non-Labor Coalition; Ind: Independents and minor parties.went up to around 92% as a result of this institutional electoral reform. Columns 2 to 4 reportthe effect of this institutional change on a party’s vote share. Column 2 shows that the voteshare of the Australian Labor Party increased by 9.3 percentage points as a result of compulsoryvoting. This effect is substantially large and statistically significant at p<0.05. Interestingly,both the effect of compulsory voting on turnout and on the Australian Labor Party’s vote shareare practically identical to those from Fowler (2013), who reports an increase in turnout of24.3 and in Labor’s vote share of 9.2 percentage points.15 This shows that using district-levelelection returns from the dataset that is the basis of this study and aggregating them to thestate level gives us the exact same results as those that are reported in Fowler (2013). Theresults in columns 3 and 4 extend the analysis to non-Labor coalition parties and independentcandidates. They suggest that the gain for Labor did not come at the cost of support for thenon-Labor coalition parties, whose vote share is largely unaffected by this electoral reform,but instead is primarily the result of weakened support for independent candidates, whosevote share decreased by 8 percentage points (p=.146). Overall, these results support a class-based account according to which a strong increase in voter turnout primarily benefits left-wingparties and speak against a minimal effects view.However, it is possible that political parties respond strategically to the introduction of com-15The very small difference of 0.1 percentage points in Labor vote shares between Fowler (2013)’s study andthe present analysis could be due to marginally differing coding decisions about which candidates belong to theAustralian Labor Party in a given election and which do not.60pulsory voting, with non-Labor coalition parties more likely to run a candidate in an electoraldistrict after voting is mandatory. It is plausible that these new non-Labor candidates receive alower and Labor candidates in these districts a higher vote share than the respective non-Laborand Labor candidates in other districts with a more established competition between the twoparties. To the extent that compulsory voting directly affects the share of unchallenged seatsthat are held by each party and that results in columns 2-4 simply exclude all unchallengedelectoral districts from the analysis, the results in columns 2-4 might overestimate the effect ofcompulsory voting on the Australian Labor Party. To test this possibility, column 5 reports theeffect of compulsory voting on the share of unchallenged seats that are held by the ALP. Thecoefficient of -0.21 shows that the share of unchallenged Labor seats among all unchallengedseats declined by 21 percentage points after the introduction of compulsory voting. This effectis substantively large and, with a p-value of p=0.096, it is statistically significant at p<0.10.This suggests that the positive effect of compulsory voting on the Labor vote share in column2 would probably become smaller if the strategic behavior of political parties and the changingnumber of unchallenged seats are taken into account.Columns 6 to 8 present results that illustrate how estimates change once the lower shareof unchallenged Labor seats after the introduction of compulsory voting is accounted for. Themodel specification is identical to the corresponding models in columns 2-4. However, formodels in columns 6-8, unchallenged districts without any competition are no longer dropped,but included into the dataset and the missing vote shares for the main parties imputed. A seat isusually unchallenged because one party’s candidate holds such a strong electoral advantage inthis district that other parties assume that they do not stand a reasonable chance of winning thisseat. As a result, these other parties choose not to spend their campaign resources on a race in anelectoral district they are unlikely to win and instead invest them into electoral districts wherethe competition between their own candidate and those of the other parties is more balanced.How do parties know whether they stand a chance to win in a given electoral district? I assumethat in most cases a party’s best guess is to look back to previous election outcomes in thatdistrict. In particular, the outcome in the previous election is likely to serve as an indicator61for a party’s chance to win in the next election. The imputation follows this reasoning andtherefore replaces all parties’ vote shares in an unchallenged district with the vote share thateach party obtained in the most recent election in which that district was challenged. This valueis likely to be the best approximation of the relative electoral power that each party holds inthat district.Following this adjustment that accounts for the parties’ strategic response to the reform,the effect of compulsory voting on a party’s vote share is marginally affected and decreasessomewhat in size for all parties (see columns 6-8). Most importantly, compulsory voting stillleads to an increase in the Labor Party’s vote share that is statistically significant at p<0.05, butthe effect size shrinks from 9.3 to 6.6 percentage points. Thus, strategic dynamics matter forestimating the effect. While there clearly is a positive relationship between compulsory voting(and the boost in turnout it produces) and the electoral support for the Australian Labor Party,the true size of the shift in Labor’s vote share is likely to be somewhat smaller than previouslythought, at around 6.6 percentage points.A possible concern with the results from Table 3.2 is that states might have been affected byfactors that undermine the plausibility of the parallel trends assumption for the 1910-1950 pe-riod. Several Australian states switched to a different electoral formula or changed the districtmagnitude during the first half of the 20th century. In addition, parties’ responses to challengessuch as the two World Wars or the economic crisis of 1929-1930 varied to some extent acrossstates. Table 3.3 therefore reports results from a within-state difference-in-differences designthat are less likely to be affected by these threats to internal validity. Columns 1-6 are for eachof the six Australian states while column 7 pools the data from the districts in all six states.This within-state analysis only includes districts where seats were challenged both in the elec-tion preceding the adoption of compulsory voting and in the election following it. Note alsothat in two of the six states, New South Wales and Tasmania, voter turnout in the state electiondirectly preceding the adoption of compulsory voting was already relatively high, at 83% and82% respectively. This may suggest that non-voters in these states are more likely to be ’occa-sional’ or ’accidental’ non-voters and therefore we might expect to find less of a class effect of62Table 3.3: Effect of Compulsory Voting Using Within-State Variation in Treatment Intensity∆ Vote Share of Australian Labor PartyNSW QLD SA TAS VIC WA All States1927- 1912- 1941- 1928- 1924- 1936-1930 1915 1944 1931 1927 1939(1) (2) (3) (4) (5) (6) (7)∆ Turnout .549 -.165 .018 -3.060 .269 -.758 -.057(.359) (.251) (.360) (.373) (.409) (.254) (.165)N 77 61 25 5 25 29 222R2 .051 .017 .000 .806 .016 .384 .270Note: Table presents difference-in-differences estimates of treatment (change in turnout as a result ofcompulsory voting) on change in ALP vote share. The pooled model in column 7 includes state fixed effects.Robust standard errors in parentheses. Commonw.: Commonwealth; Elect.: Elections.compulsory voting on turnout in these two states.There is a positive relationship between voter turnout and the vote share for the Labor Partyin only two of the six states, New South Wales and Victoria. The coefficient of 0.549 in column1 suggests that an electoral district in New South Wales with a 10 percentage point increase invoter turnout between the 1927 and the 1930 elections saw on average a 5.5 percentage pointincrease in the vote share for the Labor candidate. This effect is substantively large, but notstatistically significant at the p<0.05 level, something that might partly be attributable to therelatively small sample size of N=77. The coefficient for Victoria is even smaller in size andwell below the p=0.05 threshold of statistical significance. As it turns out, the only two stateswith positive coefficients, New South Wales and Victoria, are precisely the two states that saw aredistricting in between the election that immediately preceded the introduction of compulsoryvoting and the election that followed it. We therefore cannot exclude the possibility that thenon-Labor government’s redistricting might have influenced the results for these two states,possibly giving us too small an estimate of the true effect of the reform on the Labor voteshare.However, more worryingly for the class voting view is that in three of the six states therelationship is a negative one that is substantively large and statistically significant at p<0.05for two of these states, Tasmania and Western Australia. In a fourth states, South Australia, the63coefficient is substantively close to 0, suggesting that there is no relationship between turnoutchange and Labor vote share in this state at all. Despite the extremely small sample size ofN=5 for Tasmania which cautions us not to put too much weight into the statistical findingfrom this state, the consistency of this pattern across states and the fact that results in noneof these four states were possibly confounded by redistricting leave us with a pattern that ishardly reconcilable with the class voting view. This interpretation is corroborated by the resultfrom the analysis that pools the data from all six states. The effect is small and statisticallyindistinguishable from 0 (see column 7). Thus, the overall pattern across the six states as wellas the effect from the pooled analysis of this within-state analysis are generally more in linewith a minimal effects view. This interpretation is largely supported by similar analyses for thenon-Labor parties and independent candidates in all six states, which are reported in AppendixTables B.1 and B.2.Figure 3.1: Change in Turnout and ALP Vote Share in New South Wales, 1927-1930WoollahraBotanyWaverleyMarrickvilleDulwich HillMudgeeGranvilleBondiCanterburyAshfieldBalmainIllawarraGoulburnWilloughbyRydeCastlereaghRandwickPaddingtonBathurstDrummoyneLakembaCroydonBankstownGordonHamiltonLiverpool PlainsHurstvilleLeichhardtBarwonAnn n aleAuburnYoungParramattaHornsbyMaitlandNewcastleCootamundraAshburnhamGlebeNeutral BayHartleyManlyMonaroAlburyWagga WaggaLane CoveGloucesterRedfernOrangeWollondillyNorth SydneyNewtownVaucluseKurri KurriNepeanCoogeeNamoiBurwoodArmidalePhillipTemoraUpper HunterMurrayLismoreCorowaSouth CoastTamworthRaleighMurrumbidgeeOxley2HawkesburySturtCessnockLachlanMosmanKingClarence-2002040% Change in ALP Vote Share5 10 15 20 25% Change in Turnout64Finally, Figures 3.1 and 3.2 illustrate the relationship between turnout and Labor vote sharefrom the within-state analysis for two of the states graphically.16 The goal of the graphicalexercise is to point out patterns in the data that may provide the basis for a more in depthanalysis of individual electoral districts that could include their socioeconomic composition,geographic location and redistricting history. By looking at the distribution of the districts fromNew South Wales in Figure 3.1, the state with the potentially strongest positive relationshipbetween turnout and Labor Party support, the following pattern appears: the positive coefficientin Table 3.3 above is mostly driven by a handful of districts in the top right hand corner of thegraph that saw a strong increase in both turnout and Labor vote share between the 1927 andthe 1930 elections and one district at the bottom left hand corner, Woollahra, where turnoutpractically stayed the same while Labor support dropped by 20 percentage points. For the restof the districts, voter turnout and the Labor Party vote share seem mostly unrelated to eachother. In other words, some of the districts in New South Wales exhibit a pattern that seemsto align well with the class voting view while other districts do not. A closer analysis of thefive districts in the top right hand corner of the graph shows that none of them had a Laborcandidate compete against the other candidates in that district in 1927, but in all of them aLabor candidate ran three years later in 1930. As for Woollahra, the seemingly strong drop insupport for the ALP is the result of redistricting and of the candidate’s strategic response to it.In 1929, the district boundaries for both Woollahra and the adjacent district of Paddington wereredrawn in such a way that the 1930 district of Woollahra included large parts of the formerdistrict of Paddington and the 1930 district of Paddington large parts of the former Woollahra(Clifford, Clune and Green 2006, 61,69). What is more, this 1929 redistricting is likely to havefundamentally changed the electoral composition of each of these two districts, leading boththe incumbent Labor candidate from Woollahra and the incumbent Nationalist candidate fromPaddington to swap districts and to successfully run for re-election in each other’s district in16The analysis in the main body of the text focuses on the state with the largest increase and the one with thelargest drop in Labor vote share respectively. For graphs of the other Australian states, see Figures B.1, B.2, B.3and B.4 in the Appendix.65the 1930 election.17 In other words, instead of being support for the class voting approach, theseemingly positive association between turnout and the ALP vote share in New South Wales israther evidence for the candidates’ strategic response to the introduction of compulsory votingor to the redrawing of electoral district boundaries.Figure 3.2: Change in Turnout and ALP Vote Share in Tasmania, 1928-1931DenisonLyon/WilmotFranklinBassBraddon/Darwin-20-15-10-50% Change in ALP Vote Share10 12 14 16% Change in TurnoutFigure 3.2 plots the data for Tasmania, the state with the strongest negative relationshipbetween voter turnout and Labor Party support. This graph provides evidence that clearlysuggests that geography – and especially district size – matters for the turnout-Labor vote sharerelationship. As the graph shows, the district of Denison saw the smallest increase in turnoutbetween 1928 and 1931, but it was the district where support for the Labor Party stayed roughlythe same whereas it declined strongly in all other districts. However, Denison is different from17If these five districts in the top left-hand corner and the problematic districts of Woollahra and Paddingtonare dropped from the graph, the slope of the regression line in Figure 3.1 changes from 0.549 to -0.236.66the other districts in many respects. Most importantly, it is much smaller in size than all otherfour districts and much more urban, largely built around the Tasmanian capital of Hobart.Given that people in Denison have to travel much shorter distances to vote on election day,voting is likely to be easier and voting costs significantly lower for people who live in thisstate.18 As a consequence, making voting compulsory more strongly increases turnout in thelarger, more rural districts of Tasmania where turnout tended to be lower before the adoptionof the reform. As it turns out, these voters from the more rural areas seem less likely to vote forthe Labor Party in Tasmania than people who live in the more urban capital region. However,neither the class voting nor the minimal effects view properly accounts for this finding fromTasmania that geography matters.3.5.2 Federal-Level ResultsGiven these somewhat mixed results from the state-level analysis, this study expands the anal-ysis by looking at the effect of compulsory voting at the federal level. Like before, the analysisat the Commonwealth level proceeds in two steps: First, it uses variation in the timing of whencompulsory voting was adopted at the federal (or Commonwealth) and the state level. Sec-ond, it leverages variation in the intensity of the compulsory voting treatment across electoraldistricts within federal elections, that is, Commonwealth elections to the House of Representa-tives.At the Commonwealth level, voting was made compulsory for all states starting in the 1925House of Representatives election. Given this simultaneous introduction of compulsory votingfor all Australian states in federal elections, there is no geographic variation in the timing ofwhen compulsory voting was adopted across Australia. Thus, a direct application of a tradi-tional difference-in-differences design that leverages cross-state variation is not possible. Yet,the different adoption dates of compulsory voting in state and Commonwealth elections allowus to compare voting patterns in federal elections before and after the adoption of compulsory18The next larger district of Braddon is roughly 27 times larger than Denison while the largest district ofLyons/Wilmot is about 127 times larger.67voting in a given state with the voting behavior in the same state in similarly timed electionsone level below, that is, in state elections. Like the difference-in-differences design, such ananalysis crucially hinges on the validity of the parallel trends assumption. While it is plausibleto assume that turnout and voting patterns in federal and state elections move at least as muchin tandem as they do across states, Figure 3.3 provides an empirical test of this assumption.Across all six states, Figure 3.3 shows that with a few exceptions such as Victoria between1910 and 1920, turnout at the state and federal level tend to move in parallel. As we wouldexpect, the only time these trends do clearly not move in tandem is after one level adoptedcompulsory voting, but the other did not. In these instances, which are demarcated by twovertical lines for each state, turnout in the elections under compulsory voting is significantlyhigher than in the elections without.68Figure 3.3: Voter Turnout in State and Federal State Assembly Elections, by State. 1920 1930 1940 1950YearNSW. 1920 1930 1940 1950YearQLD. 1920 1930 1940 1950YearSA.4.6.81Turnout1910 1920 1930 1940 1950YearTAS. 1920 1930 1940 1950YearVIC. 1920 1930 1940 1950YearWANote: Red line shows voter turnout in federal and blue line in state assembly elections for each state. The twovertical lines indicate the adoption of compulsory voting. With the exception of Queensland where compulsoryvoting was adopted at the state level in 1914, the first vertical line is for the federal and the second line for thestate level. The area in between the two vertical lines indicates the period in which voting was only compulsoryat one level, but not the other.Based on the results in Table 3.4, the adoption of compulsory voting increased voter turnoutin Commonwealth elections by 30.4 percentage points (column 1), boosting turnout fromaround 60% in the 1922 federal election to just above 90% in 1925. Results in columns 2-4 suggest that this strong shift in voter turnout primarily helped the Australian Labor Party,increasing its vote share by 2.8 percentage points (p<0.05), but did not affect the vote shareof the non-Labor coalition parties. However, as in columns 2-4 in Table 3.2 above, results incolumns 2-4 in Table 3.4 are based on an analysis that set observations for unchallenged elec-69toral districts to missing values, but included these very same districts for an election in whichmore than one candidate ran in them. As before, it is possible that parties respond strategicallyto the introduction of compulsory voting. Therefore, these results may be affected by parties’strategic decisions about whether to run their own candidate in a given district and might possi-bly overestimate the positive electoral effect for the ALP. To remedy this problem, columns 5-7present results from an analysis that uses the same models as in columns 2-4, but only keepselectoral districts for which we have vote share data for both the pre- and the post-compulsoryvoting election. In other words, it excludes unchallenged districts not only from the electionin which they were unchallenged, but completely drops them from the analysis. This gives usa balanced panel that includes the exact same electoral districts for both the pre- and the post-mandatory voting election. Based on this balanced panel, the substantively small coefficient of0.007 in column 5 suggests that compulsory voting did not have any effect on the ALP’s voteshare. As for the non-Labor parties’ electoral fortunes, their vote share might have increaseda little (2.8 percentage points), but due to a large standard error this effect is not statisticallysignificant at p<0.05. The vote share for independent candidates slightly decreased after vot-ing became compulsory, but the effect is small and statistically not significant (see columns 4and 7). Thus, the analysis of the introduction of compulsory voting at the federal level thatleverages variation in the timing of the reform between the federal and the state levels does notsupport the class-based view and is more consistent with a minimal effects perspective.70Table 3.4: Effect of Compulsory Voting in Commonwealth House of Representatives ElectionsComparison Across Levels: Federal – State Comparison Within Federal LevelTurnout Vote Share Vote Share ∆ Vote ShareWithout Missing Data With Balanced PanelALP NLC Ind ALP NLC Ind ALP NLC Ind(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Compulsory Voting .304 .028 .009 -.019 .007 .028 -.015(.057) (.007) (.036) (.028) (.014) (.036) (.030)∆ Turnout .164 -.285 .120(.140) (.140) (.121)State fixed effects X X X X X X X1924-1926 election dummy X X X X X X XN 543 543 543 543 492 492 492 69 69 69R2 .714 .089 .069 .027 .113 .080 .032 .024 .043 .010Note: Table presents estimates from two analyses. Columns 1-7 are analogous to a difference-in-differences design that contrasts the before-after difference inturnout and vote shares at the federal level with a before-after difference at the state level. Before-after refers to the introduction of compulsory voting at thefederal level in 1924. Columns 8-10 report results from variation in treatment intensity (change in turnout between 1922 and 1925 elections) across electoraldistricts on the change of a party’s vote share. Standard errors in parentheses are clustered at the state level for columns 1-7; for columns 8-10, heteroskedasticityrobust standard errors are reported in parentheses. The unit of observation is at the district level. ALP: Australian Labor Party; NLC: Non-Labor Coalition; Ind:Independents and minor parties.71Columns 8-10 shift the focus from the cross-level analysis to a within-Commonwealth one.Results are based on a within-Commonwealth difference-in-differences analysis that tests theextent to which changes in turnout between 1922 and 1925 are related to changes in supportfor different political parties across federal electoral districts. As before, this research designis likely to produce internally valid estimates of the effect of compulsory voting for the federallevel.As the coefficient of 0.164 and 0.120 in columns 8 and 10 show, there is a small positiveeffect from an increase in voter turnout on support for the Labor Party and for independent can-didates, respectively. If turnout increased by 10 percentage points between 1922 and 1925 in afederal district, the Labor vote share went up by 1.6 percentage points. This effect is moderatein size, but fails to be statistically significant at p<0.05. However, things look different for thenon-Labor coalition parties. The coefficient of -0.285 means that districts with a 10 percentagepoint increase in turnout saw support for non-Labor parties drop by 2.9 percentage points, aneffect that is statistically significant at p<0.05. Thus, while the within-Commonwealth analy-sis does not find direct support for the positive relationship between turnout and Labor support,the negative relationship between turnout and support for non-Labor parties is consistent withthis view and clearly at odds with the minimal effects perspective.3.6 ConclusionPolitical scientists have debated for quite some time whether increasing voter turnout shifts thebalance of political power by increasing support for left-wing parties at the expense of right-wing parties. Several recent articles that use natural experiments and leverage an exogenousshock to voter turnout due to the introduction of compulsory voting seem to provide compellingevidence that supports such a class voting perspective: compulsory voting disproportionatelyincreases voter turnout among poorer citizens, thereby shifting the political power in favor ofleft-wing parties vis-a-vis their right-wing counterparts.This paper advances a theory of minimal effects that challenges this account. This study ar-72gues that compelling less politically interested and informed voters who would not have votedotherwise to cast a vote on Election Day will at most marginally affect the overall partisanbalance. Instead, we are likely to see a movement of voters away from independent candidatesand minor parties toward the more visible and well known main parties on both the politicalleft and right. This paper evaluates this theoretical account empirically by reanalyzing a crucialcase: the adoption of compulsory voting in Australia in the early 20th century. The empiricalfindings are rather mixed. At first view, a state-level analysis seems to support the class votingview while a federal-level analysis aligns better with the minimal effects view. However, closeranalysis that leverages within-state and within-Commonwealth variation gives a more nuancedpicture, showing that class voting and minimal effects view are both supported in some in-stances and not in others. This suggests that additional factors which have only been hinted atin this study need further analysis. Key among them are parties’ strategic responses to the in-troduction of compulsory voting as well as electoral geography – in particular, the geographicalsize of electoral districts. With respect to the class voting view, findings in this study suggestthat the Australian Labor Party did not consistently benefit electorally from compulsory vot-ing and if it did, benefits were often somewhat smaller than those that had been attributed tocompulsory voting in previous research.However, a word of caution is warranted. The interpretation of the effects from the analysesoften face a small N problem. As a result, even substantively meaningful effects frequently failto be statistically significant at conventional levels. Unfortunately, with observational historicaldata there is no direct solution to how we could improve on such an analysis because it isimpossible for the researcher to rerun history in order to increase the sample size. One possibleway forward that has been hinted at in the results section is a more thorough and in depthanalysis of specific electoral districts. A second possibility is to match the district-level datain this study with additional district-level covariates to increase efficiency. While certainlypossibly, this is a daunting task due to the fact that census districts frequently do not align withelectoral districts for the period under investigation.What are the implications from these findings? First, this study contributes to the debate on73the relationship between political participation and left-wing party support. The findings fromthis study caution against too optimistic a view on this relationship, suggesting that in someinstances higher turnout is unlikely to increase a left-wing party’s electoral support. How canwe reconcile this finding with studies that have directly linked increasing voter turnout with theadoption of policies that benefit the poor? For example, Fowler (2013) directly attributes anincrease in pension spending in Australia to the adoption of compulsory voting in this country.Yet, given that pension spending in Australia was largely a matter for the federal governmentand that compulsory voting did not shift power towards Labor at the federal level, increas-ing support for the ALP is unlikely to be the mechanism through which higher voter turnoutaffected pensions. A possible explanation that could account for the policy shift in pensionspending in Australia in the absence of a significant shift in political power from the non-Laborcoalition parties to the ALP could emphasize the strategic behavior of the right-wing parties. Itis possible that these non-Labor parties adjusted their own policy positions and shifted towardsthe left in an attempt to appeal to these new, socioeconomically worse off voters. However,while quite plausible, we cannot totally rule out the alternative possibility that there was nodirect effect of higher voter turnout on pension spending and that we erroneously attributed thepolicy shift to the adoption of compulsory voting. Given that most of the empirical leverage inthe pension spending analysis in Fowler (2013) comes from a comparison of pension spendingin Australia with that of New Zealand and Denmark, a qualitative, in depth analysis of thesethree cases could provide additional insights into the mechanism that was leading to increasedpension spending in Australia in the early 20th century (cf. Castles and Uhr 2007).Second, this study casts new light on the role of electoral reforms and their ability to helpalleviate the problem of increasing economic inequality. Overall, the findings from this studysuggest that compulsory voting is an effective strategy to address the issue of political inequal-ity – after all, voter turnout increased strongly after the adoption of compulsory voting –, butof limited usefulness as a tool to remedy economic inequality. Thus, the present study illus-trates both the potential and the limitations of an institutional fix such as compulsory voting.Therefore, compulsory voting is only likely to solve part of the democratic dilemma that was74identified by Lijphart (1997) in his 1996 APSA presidential address: it mechanically increasesvoter turnout among the poor, but it might not necessarily and automatically guarantee a betterpolitical representation of their policy preferences.75Chapter 4Can Potential Populist Voters Be Framed?Results from a Survey Experiment in the2017 German Election4.1 IntroductionThe last few years have seen a strong increase in the number of people across establisheddemocracies who support right-wing populist parties and politicians. These populist voters’suspicion of elites and establishment actors (Cramer 2016; Mudde 2007) would seem to makethem a “lost cause” to mainstream parties. At the same time, mainstream parties have po-tentially powerful considerations working in their favor, at least some of the time: e.g., goodeconomic performance (Duch and Stevenson 2008), a policy agenda of economic redistribu-tion (Boix 1998), or government experience (Meguid 2005, 349). These include issues thatshould be of importance even to voters who tend to be susceptible to populist appeals. Whatare the possibilities and limits of influencing populist voters and having them consider issuedimensions favorable to mainstream parties?Previous research has shown that the media, parties and politicians are able to prime citi-76zens, that is, to influence their political choices by talking about and thus, raising the salienceof some topics while ignoring others (Iyengar and Kinder 1987; Johnston et al. 1992; Krosnickand Kinder 1990; Lenz 2012). To the extent that populist voters represent “frustrated issuepublics” (Chou et al. 2018) who care at least to some extent about policies, we would expectthem to generally be receptive to priming. After all, it is these people’s feeling that issuesthat matter most to them, such as restricting immigration (Dinas et al. 2019; Mutz 2018) andalleviating economic hardship that is a result of globalization (Colantone and Stanig 2018a,b;Dal Bo et al. 2018), have been ignored by the established elites that at least partially drivestheir decision to support radical right parties.What messages are likely to be most effective for mainstream parties? One possible strat-egy for mainstream parties is to try to copy the right-wing populist party’s policy position byadopting a tougher stance on issues such as immigration, a strategy that Meguid (2005) calls“accommodative”. However, such a strategy is risky. First, it might be ineffective with populistvoters who consider the populist right-wing party as the more credible proponent of radical pol-icy change on issues such as immigration (Meguid 2008, chapter 6). Second, even if such astrategy was effective with some populist voters, it might risk alienating more moderate votersand thus, lead to a net vote loss for the mainstream party (Chou et al. 2018; Krause, Cohenand Abou-Chadi 2019). An alternative and more promising strategy for mainstream parties istherefore to emphasize mainstream issue dimensions where they arguably have a comparativeadvantage over right-wing populist parties, a strategy that Meguid (2005) has dubbed “dismis-sive”. Doing so, they might be able to appeal to populist voters without alienating their owncore electorate.For this reason, this study explores to which extent non-populist political messages areable to shift the vote intention of voters who are susceptible to right-wing populist appeals.It proposes a theoretical framework that builds on existing theories of voting behavior thatconceives of voters as performance-oriented (Duch and Stevenson 2008), strategic (Downs1957) and self-interested actors (Meltzer and Richard 1981), but that puts the psychology ofpotential populist voters centre-stage. In particular, it takes into account that any blunt attempt77to try to influence these voters with openly anti-populist messages is likely not to work ormight even backfire by making them not less, but even more supportive of populist candidates(cf. Chong and Druckman 2007, 2010). This framework suggests that by taking these voters’concerns – such as their resentment toward established political elites, their feeling of insecurityand of not getting their fair share of the economic pie (Cramer 2016; Mudde 2007) – seriously,existing theories of voting behavior may help us design mainstream political messages thatappeal to this group of voters.This study tests these expectations through survey experimentation during a real-worldelection campaign in which populist messages were highly salient on a large pool of roughly1,800 voting-age German citizens who were specifically selected for this study due to their highpropensity to vote for a right-wing populist party. The key finding of this study is that noneof the four messages that were tested were able to reduce support for the right-wing populistAlternative for Germany (AfD) party. However, we need to take into account that these nullfindings are the result of a conservative test where it was attempted to shift potential populistvoters’ vote intentions only two weeks before a real world election, that is, at a time wheremost people’s vote intentions are likely to have been firmed up. In contrast to these null find-ings, highlighting the good performance of the Germany economy proved to be a potentiallypowerful strategy for the incumbent Christian Democrats, increasing their vote share amongthis group of voters by 6 percentage points from 14% to 20%. Given that the increased supportfor the Christian Democrats did not hurt the populist AfD party, this suggests that the economicmessage that was tested was most effective with those dissatisfied and anti-immigrant voterswho had abandoned the Christian Democrats not for the populist challenger, but for mainstreamopposition parties.The findings of this study have broad implications for our understanding of populist vot-ers, framing, and party strategies. First, it contributes to debates about strategies that are mostsuccessful for mainstream parties in the face of niche party competition. Previous research hasprimarily focused on strategies that directly challenge the radical right party on immigrationpolicy by either adopting a more liberal or more restrictive stance on this issue. Instead, this78study explores the effectiveness of dismissive strategies that have received less attention so farand that aim at reducing the salience of the immigration issue by focusing the voters’ attentionon alternative policy considerations. Second, this study provides new insights into the issuesbeyond immigration that populist voters care about. In a context where immigration is a highlysalient issue, this study finds only very limited evidence that these voters are still receptiveto mainstream messages. Third, and in line with previous research, this study presents someexploratory analyses that hint at the crucial role of a populist voter’s level of political knowl-edge in moderating their resistance to mainstream messages. It seems that particularly thosepotential populist voters with less political knowledge might be more receptive to mainstreammessages that convey new and relevant information to them.4.2 Populist Voters and Mainstream Non-Immigration Mes-sagesThere is considerable evidence for the ability to shape citizens’ political evaluations. For exam-ple, the literature on priming has shown that by talking about certain issues and ignoring others,the media, parties and politicians can influence the issues to which citizens pay attention whenmaking political choices (Iyengar and Kinder 1987; Krosnick and Kinder 1990; Johnston et al.1992). As a result, priming makes some considerations more accessible to citizens than othersand doing so, changes the “standards that people use to make political evaluations” (Iyengarand Kinder 1987, 63). Priming is a particularly powerful strategy if it is used with so-called“valence” judgements that focus citizen attention on the performance of the economy or on can-didates’ personal characteristics (Lenz 2012; Vavreck 2009). In addition, the closely relatedliterature on framing has shown that even in a context of competing frames, stronger framesand those received later in time tend to have a stronger effect on people’s opinions (Chongand Druckman 2007, 2010). Recent research has refined this debate by pointing out that it isoften not so much the mere priming of an issue, but rather information-based persuasion that79is able to shift citizens’ opinions (Leeper and Slothuus 2018). In other words, citizens do notso much reconsider their vote choice simply because an issue has been made more salient tothem, but rather because they – advertently or inadvertently – received new information thatallowed them to re-evaluate and potentially change their existing opinions.However, while previous studies have extensively tested the effectiveness of a variety of dif-ferent frames, these studies have largely been restricted to either the average voter or to small,non-representative samples of college students (Chong and Druckman 2007, 2010; Huber andLapinski 2006; Lenz 2012; Mendelberg 1997). Yet the effectiveness of framing strategies, es-pecially when combined with mainstream political messages, may be more limited with thegroup of voters who are least susceptible to these sorts of political appeals: the increasinglylarge number of citizens in established democracies who tend to distrust and feel abandoned bycurrent political elites. These citizens frequently share a feeling of resentment towards main-stream politicians and perceive them as either unable or unwilling to solve the most urgentproblems that they themselves and other people like them face today (Cramer 2016; Mudde2007). As a consequence, they are likely to be more responsive to the anti-establishmentrhetoric and appeals of (often right-wing) populist parties and politicians and rather scepticaltoward messages from mainstream parties.However, things may look different if we acknowledge that these voters are not only protestvoters, but citizens who care about policies (Chou et al. 2018) and policy change. It is in thisarea where mainstream parties may present themselves as the better and more credible alter-native to right-wing populist parties, both due to their greater experience of running a country(Meguid 2005, 349), their successful management of the economy (Duch and Stevenson 2008)or a more credible agenda for redistributive economic policies (Boix 1998). To what extent canmainstream parties use their strengths in these areas to connect with populist voters and havethem consider issue dimensions that are favorable to mainstream parties?Following Meguid (2005, 2008)’s seminal research on mainstream parties’ strategies in theface of niche party competition, we can distinguish two strategies that mainstream parties canchoose from if they want to lure voters away from a right-wing populist party. A mainstream80party can follow an “accommodative” strategy and compete with the radical right party onthe latter’s main policy issue. By trying to copy the niche party and adopting, for example, atougher stance on immigration, the mainstream party can hope to win back some of the votersit has lost to the niche party competitor. However, the empirical evidence about the effective-ness of this strategy is somewhat mixed. While Meguid (2005, 2008) finds that such a strategyreduces electoral support for niche parties, recent research has questioned these previous find-ings. Krause, Cohen and Abou-Chadi (2019)’s replication of Meguid (2005, 2008) extendsthe time frame by more then ten years to include the most recent elections and combines bothmacro-level election data and micro-level survey data, but fails to find any negative effect ofthis strategy on radical right party support. In addition, recent survey experimental researchinto right-wing populist party support has cast doubt about the usefulness of this strategy formainstream parties. For example, Chou et al. (2018) show that adopting a right-wing populistparty’s anti-immigrant policy position is likely to lead to a net loss for mainstream parties be-cause any increase in support from previous radical right voters is more than offset by coremainstream party voters who would abandon the mainstream party. Thus, even if an accom-modative strategy might reduce electoral support for a right-wing populist party, it might notbe a viable strategy for a mainstream party if that strategy weakens its own electoral positionand undermines its chances of being reelected.For these reasons, the “dismissive” strategy might be a better alternative for many main-stream parties. A mainstream party that follows this strategy reduces the salience of the nicheparty’s core policy issue by primarily campaigning on alternative programmatic issues wherethe mainstream party has a natural comparative advantage over the niche party competitor. Soinstead of trying to replace the niche party as the better anti-immigrant party, a mainstreamparty could appeal to populist voters by highlighting its strength in alternative areas such as81income redistribution, health care or other social policies.1 Assuming that some right-wingpopulist voters do not only or even primarily care about immigration, but also about other pol-icy issues that directly affect them – especially welfare (Chou et al. 2018) and economic issues(Colantone and Stanig 2018a,b) –, this mainstream party strategy may be able to lure themaway from the right-wing populist party. Previous research that empirically tests the effective-ness of the dismissive strategy is limited (but see Meguid 2005, 2008). In particular, there isvery little research that systematically tests which non-immigration issue might be most ap-pealing to right-wing populist voters. One exception is Chou et al. (2018) who, in a conjointanalysis, find suggestive evidence that right-wing populist voters prefer “candidates who areattentive to pensions and who propose increasing taxes on the rich” (Chou et al. 2018, 29).The choice of the strategy – accommodative versus dismissive – affects the issue that main-stream parties may use to appeal to voters, but it does not tell us much about the type ofmessages, that is, whether a message is mainstream or populist. I call a message populist if itfulfills two criteria:2 First, it proposes a radical policy shift from the current policy status quo.Second, this radical policy shift is presented as an effective solution to a perceived policy prob-lem, but it is an ineffective or suboptimal way of addressing the problem.3 For example, mostright-wing populist parties support a drastic reduction of any future immigration or even a totalban of certain immigrant groups to their country. Such a policy position represents a markedshift from the status quo in all liberal democracies. Right-wing populist parties frequently jus-tify their tough stance on immigration as the best way to address an existing policy problem(e.g. a supposedly high crime rate or labor market repercussions from technological change).1Meguid (2005, 2008) identifies a third mainstream party strategy, which she calls adversarial. In this case, amainstream party competes with the niche party on the latter party’s core issue, but instead of adopting a positionmore similar to the niche party, the mainstream party emphasizes how its policy stance is different from the nicheparty’s. This strategy will generally increase electoral support for the niche party (Ibid.) – possibly partiallyfuelled by a backlash effect (cf. Chong and Druckman 2007, 2010) among voters who are susceptible to populistappeals – and therefore, is not a viable strategy to reduce niche party support.2I thank Milan Svolik for shaping my thinking on these two criteria.3In line with standard definitions of populism in the literature, this definition does not link populism to any par-ticular policy issue, but instead highlights its anti-establishment character (Mudde 2007; Mudde and Kaltwasser2017; Norris and Inglehart 2019). It is different from these other definitions in more explicitly spelling out howthis anti-establishment character translates into policies that are supported by populist parties.82However, while the proposed populist “solution” might seem appealing to some people as aneasy fix to a serious policy problem, it is either unlikely to solve the problem or, if it does,comes at a higher cost for certain societal groups (e.g. people in economic sectors that benefitfrom immigration) than alternative, less radical policy solutions. In other words, it is not somuch the fact of wanting to reduce immigration levels that makes a policy position populist,but rather the extremeness with which populist radical right parties want to achieve this goal.In contrast, I define as “mainstream” a message that is not populist, that is, that does not fulfillthese two criteria.In line with this study’s analytic goals, its focus is on mainstream messages that can be usedas dismissive strategies by mainstream parties. Given that immigration is the core issue formost right-wing populist parties in Europe, a dismissive strategy has to focus non-immigrationissues. Thus, this study analyzes mainstream messages that are not on immigration. For sim-plicity, it uses mainstream message as a short form for “mainstream non-immigration mes-sage”.Finally, the context in which citizens are exposed to new messages is likely to matter, too.All else equal, we would expect that shaping potential populist voters’ vote intention with main-stream messages during and especially toward the end of a real-world election campaign willprove more difficult than early in a campaign. Over the course of an election campaign thesevoters will have been exposed to many frames and counterframes – including both mainstreamand populist ones – and party cues leaving little room for any new mainstream message to altertheir vote choice late in the campaign (Chong and Druckman 2007; Druckman 2004; Kalla andBroockman 2018). This is particularly likely to be true for many established democracies thesedays where major events, such as the economic repercussions of the 2008 financial crisis andthe influx of Syrian and other refugees since 2015, have helped raise the salience of issues thatbenefit especially right-wing populist parties and their anti-immigrant policy agendas.834.2.1 Performance, Policies, and UncertaintyDespite the challenges to shape the vote choice of citizens who are susceptible to right-wingpopulist appeals with mainstream messages, carefully crafted non-populist messages may shiftthese people’s vote away from populist and toward mainstream parties. Grounded in existingtheories of voting behavior, this study presents four alternative theoretical logics of framingthat specifically focus on this group of voters and that outline the conditions under which wewould expect these citizens to be swayed by non-populist messages. Based on these theoreticallogics, a series of hypotheses are specified in this section. All hypotheses were registered in apre-analysis plan at the EGAP website before data access by the researcher.4The first of these theoretical logics assumes that, in line with a long literature on economicvoting (e.g. Duch and Stevenson 2008; Powell and Whitten 1993), citizens who are inclined tovote for populist parties are rational retrospective voters who care about the performance of theeconomy. Just like any other voter, they want their country’s economy to do well and evalu-ate incumbent politicians accordingly. In particular, they use information about past economicperformance to assess the economic competence of the incumbent government. This assess-ment influences their selection of future government parties that they want to be economicallycompetent (Duch and Stevenson 2008, 28-31).5 As a consequence and despite their fundamen-tal dissatisfaction with some of the government’s policies (e.g. on immigration), these voterswill be responsive to messages that highlight the current state of the national economy (cf.Alt, Lassen and Marshall 2016; Lenz 2012; Vavreck 2009). They are expected to support theincumbent party or, in the case of coalition governments, the party of the incumbent primeminister if the economy is doing well and to prefer other parties if this is not the case. Thus,emphasizing the good economic performance is likely to increase support for the incumbentwhereas highlighting economic difficulties will decrease it. Applying this to a national contextwhere the economy is doing well, this leads to:4The EGAP preregistration ID number is 20170920AA.5In contrast to such a selection model, a sanctioning model emphasizes much more voters’ desire to punishbad incumbents for past economic performance at the ballot box (Duch and Stevenson 2008, 10-15).84Hypothesis 1 (Economic Voting): Emphasizing and providing information about the econ-omy when the economy is doing well in a country makes voters who are susceptible to populistappealsa) less likely to vote for the right-wing populist party andb) more likely to vote for the party of the incumbent prime minister or chancellor.Just like for other citizens, populist voters’ vote choice is likely to be shaped by both strate-gic (Downs 1957; Cox 1997) and expressive (Pons and Tricaud 2018) motivations. Given theirgeneral level of dissatisfaction with the established elite, some populist voters may use theirvote strategically and vote for an “outsider” party to throw the incumbent parties or politiciansout of office. In contrast, those who are more strongly driven by expressive motivations mightcare less about who will be in power after the election and vote for populist parties primarilyto signal their protest to incumbents (Mudde 2007, 226-229). Yet in some situations, thesetwo motivations may be in conflict with each other. For example, many of these voters sharea feeling that the current political elite builds a sort of cartel in which politicians are primarilyinterested in clinging to political power and no longer offer true policy alternatives (cf. Katzand Mair 1995), something that has been found to drive up support for populist challengerparties (Kriesi 1999, 419-420). However, in order to disturb this political cartel, it might bemore effective not to vote for a populist party, but rather for a mainstream opposition party.This is especially true in the context of a grand coalition government where a strategic votermight want to strengthen mainstream opposition parties so that this opens up viable coalitionalternatives to the existing grand coalition government. These strategic versus expressive mo-tivations will be most starkly in contrast in situations where the major mainstream parties havedeclared one or more populist parties as uncoalitionable. In such a context, casting a protestvote for a populist party to signal one’s opposition to the grand coalition government will makeit more difficult to build majority coalition governments that do not need the support of both ofthe two major mainstream parties. Therefore, telling populist voters about their option to casttheir vote strategically not for their preferred (i.e. populist) party, but against the continuationof the grand coalition might sway the more strategically minded among them.85Hypothesis 2 (Strategic Voting): Priming the strategic aspect of their vote and informingthem about how to use it to reduce the electoral chances of a future grand coalition will makevoters who are susceptible to populist appealsa) less likely to vote for the right-wing populist party andb) more likely to vote for the mainstream opposition parties who are potential future coalitionpartners outside of a grand coalition.Voters may support politicians or political parties not so much for their past economic per-formance in office or because they are attracted by their policy promises (Lenz 2012), but alsobecause they expect from them a reduction in political and economic uncertainty (cf. Jacobsand Matthews 2017). This might be particularly true for voters who lean toward (right-wing)populist parties. These people often share the feeling that due to macro-level developmentssuch as globalization, the recent surge in mass immigration or economic and political crises,things are getting worse for them (Colantone and Stanig 2018a,b; Cramer 2016; Mudde 2007;Mutz 2018; Norris and Inglehart 2019) and they feel insecure and vulnerable. As a conse-quence, they may be less opposed to mainstream politicians who they think may credibly re-duce uncertainty from these international economic and political risks. This will especially betrue for incumbent mainstream politicians who have proven their leadership skills in previousinternational negotiations with other world leaders. Thus, explicitly mentioning uncertaintyand the importance of leadership characteristics might make populist voters more likely to votefor the incumbent and her party, especially if the incumbent is generally perceived to be acapable leader with strong leadership skills.Hypothesis 3 (Uncertainty): Emphasizing political uncertainty and the importance of lead-ership characteristics makes voters who are susceptible to populist appealsa) less likely to vote for the right-wing populist party andb) more likely to vote for the party of the incumbent prime minister or chancellor.Finally, voters who are dissatisfied with mainstream politicians are likely to care not onlyabout performance, but also about policies. In many cases it is these citizens’ very disagree-ment with how politicians have handled certain issues – from immigration to Eurozone finan-86cial bailouts and trade policy – that has pushed them to consider voting for populist alternativesin the first place (Chou et al. 2018; Colantone and Stanig 2018a,b; Mutz 2018). A long lineof research in political science that goes back to Meltzer and Richard (1981)’s seminal articlestarts from the assumption that citizens’ self-interest shapes their attitudes toward economicredistribution and their vote choice. Voters who are leaning toward populist parties are likelyto be no different in this respect. What is more, many of them are “economic losers” (Dal Boet al. 2018) who have the feeling of not getting the share of the economic pie that they deserveand of being ignored by mainstream politicians (Cramer 2016). Therefore, explicitly address-ing the issue of social inequality and supporting policies that aim at effectively reducing thegap between rich and poor will allow left-wing parties to win back some of these alienated vot-ers. This will particularly be true if left-wing parties manage to convince these voters that theycan be trusted to adopt policies that benefit these voters once they are in power (cf. Kuziemkoet al. 2015). In a context where extreme left parties are considered uncoalitionable by themainstream left or where voting for the more moderate left-wing party increases the chancesof a future left-wing government, support will mostly go to the mainstream left-wing party.In contrast, in situations where this is not the case, voters who care about redistribution maychoose the far-left party with its more extreme redistributive policy agenda instead in orderto achieve a more substantial left-wing policy shift on redistribution (Kedar 2005). For theGerman context, this leads to:Hypothesis 4 (Social Inequality): Emphasizing social inequality combined with informa-tion about the main left-wing party’s policy positions and their main competitors’ positions onthis issue makes voters who are susceptible to right-wing populist appealsa) less likely to vote for the right-wing populist party andb) more likely to vote for the main left-wing party strongly supporting redistributive policies.The next section discusses a survey experiment that was conducted during the 2017 German87federal election to test these hypotheses.64.3 Data and Methods4.3.1 DataThe survey experiment was specifically designed to sample a large group of voters who aresusceptible to populist appeals in the context of a real-world election campaign where populistmessages were prevalent. It is on this group of voters that is least likely to be swayed bynon-populist messages that the above hypotheses will be tested.Campaign Context. The survey experiment was conducted shortly before the 2017 Ger-man federal election. This was the first federal election after Germany had seen a large influxof mostly Syrian refugees in 2015 and 2016, with a peak of 890,000 asylum seekers registeringwith German authorities in 2015 alone.7 This influx of refugees in combination with reports inGerman media about crimes committed by groups of young men from mostly North Africancountries in several German cities led to a sharp increase in support for the anti-immigrant pop-ulist Alternative for Germany (AfD) party in opinion polls and state elections across Germanyin 2016 and 2017. In this context the survey was fielded between September 7 and Septem-ber 15, 2017, that is, shortly before the German federal election held on September 24, 2017and thus, in the peak of the election campaign.8 Respondents were surveyed between Septem-ber 7 and September 15, 2017, that is, roughly two weeks before the election on September6The registered pre-analysis plan also specifies a series of hypotheses that test for heterogeneous treatmenteffects among strongly and weakly populist voters. However, the separation of survey respondents into stronglyand weakly populist voters based on demographic characteristics proved more difficult than initially assumed,making any results highly sensitive to the researcher’s judgement about how these two groups are built. For thisreason, the study focuses only on the pre-registered hypotheses of the treatment effects for the whole sample ofpopulist voters.7Numbers are from the Office of Migrants and Refugees (BAMF), Germany.8YouGov Deutschland administered the survey and I did not have access to any of the survey data before itsofficial pre-registration on EGAP.8824, 2017.9 As a result, most respondents in the survey will have been extensively exposedto real-world campaign messages of the 2017 German federal election by the time they takethe survey. Most importantly, the AfD party heavily criticized the incumbent government andGerman chancellor Angela Merkel for their immigration policies during the campaign, thuspriming the issue of immigration in this election.Sampling procedure. This study is interested in the voting behavior of people with ahigh propensity to vote for populist parties. To identify these people, it uses respondents’anti-immigration attitudes as a proxy for their susceptibility to vote for the German AfD party.Doing so has several advantages over alternative approaches. Especially in the German con-text where the recent surge in support for the populist AfD is directly linked to the salienceof the immigration issue, focusing on immigration attitudes is probably the best way to tap re-spondents’ potential to vote for populist parties. Consistent with this assumption, data from theGerman Longitudinal Election Study (GLES) show that in the fall of 2017 anti-immigration at-titudes among German voters are highly predictive of AfD vote in that election.10 On the otherhand, asking respondents in the pre-screening question directly about their attitude toward theAfD would likely have restricted the pool to the narrow group of hard-core AfD partisans.Furthermore, asking respondents about their partisanship shortly before the measurement ofthe outcome variable, i.e. vote intention, might have suppressed any treatment effects fromnon-populist priming messages.Participants were recruited by the survey firm YouGov. In order to gauge respondents’anti-immigration attitudes, this study used a pre-screening question that directly asked respon-dents about their immigration attitudes and retained only those respondents from the YouGovonline panel with the most extreme anti-immigration attitudes for the survey experiment. Thepre-screening question is a slightly adapted standard item from the GLES tapping respondents’position on the libertarian-authoritarian value dimension. It reads as follows (English transla-9The large majority of respondents (77%) filled in the survey after September 11, 2017.10For example, AfD vote intention in the GLES pre-election survey is only 2.59 percent for those who score 8and below on the 1 to 11 anti-immigration scale in the GLES while it is much higher, at 20.16 percent, for thosewho score 11 on this scale.89tion): “Do you think that immigration to Germany should be made rather easier or rather moredifficult?”11 Respondents indicate their support or opposition to this statement on an 11-pointscale where higher values correspond to a more anti-immigration attitude. In order to adminis-ter the survey experiment only to the most anti-immigrant third of respondents, YouGov startedby retaining only respondents from categories 8 to 11 on this pre-screening question and, afteran initial soft launch, narrowed it further down to only include respondents from the most anti-immigrant category 11. As a result, the vast majority of respondents in the sample (1,664 outof 1,786) fall into the highest (i.e. most anti-immigration) category on this variable.Sample characteristics. The final sample consists of 1,786 German voting-age citizenswith a high propensity to vote for the anti-immigrant populist AfD party. This sample islargely representative of the overall German population with respect to gender, age and ed-ucation.12 However, as intended by the pre-screening, the sample is not representative withrespect to respondents’ partisan affiliations. There is a clear over-representation of partisansfor the right-wing populist AfD party, with most respondents (21%) in our sample identifyingwith this party.13 Similarly, AfD vote intention among people in the sample is at 27%14 andthus, substantially higher than the average support for this party in the German population,which was at around 10% at the time of the survey.The survey itself is administered online. In contrast to other surveys like the GLES whichare conducted by phone or personal interview, administering the survey online has the ad-vantage that results are less affected by social desirability bias (Gooch and Vavreck 2019). Inparticular, respondents are expected to more truthfully answer the pre-screening question about11“Sind Sie der Meinung, die Zuzugsmoeglichkeiten fuer Auslaender sollten eher erleichtert oder ehereingeschraenkt werden?” To the extent that the pre-screening question primes the issue of immigration in respon-dents’ minds and thus, raises the salience of this issue, the results in this study will represent a rather conservativeestimate of the effect of non-populist messages on these voters.12There is a small under-representation of less educated people and, quite surprisingly, of very young peoplein the sample. The share of people aged 18-29 in the sample is 9% whereas their share is 14% in the overallpopulation. Data on the German population are from the website of the German statistical office.13Support for the two major mainstream parties, the CDU/CSU and the SPD, follows at 18% and 14% respec-tively. See summary statistics in Appendix for details.14This number is calculated based solely on people in the control group.90their anti-immigration attitudes and express their vote intention for the right-wing populist AfDparty when they fill in the survey at their own computer at home than if they were asked thesequestions in a direct conversation with an interviewer.4.3.2 Dependent VariablesFour dichotomous variables measure respondents’ vote intention for each of the following fourparties or combinations of parties: the right-wing populist Alternative for Germany (AfD)party, the center-right Christian Democratic Party (CDU/CSU), the center-left Social Demo-cratic Party (SPD), and for the two mainstream opposition parties, the Liberals and the Greens(FDP/Greens), combined. For each of these variables, respondents are coded as 1 if they pickedthe respective party as their answer to the survey question “Which party would you vote for ifthere was a federal election this Sunday?”15 and as 0 otherwise.164.3.3 Independent VariablesTreatments. The survey experiment has four treatments and all 1,786 participants were ran-domly assigned to either the control or one of the four treatment groups. The treatments cor-respond to hypotheses 1 to 4 that are specified in the theory section above. In each case, atreatment consists of roughly two sentences that directly precede the question that taps a re-spondent’s vote intention in the upcoming election and are displayed on the same screen as thevote intention question. The four treatments read as follows (English translation):17Economic performance: “The German economy is doing very well, with unemploymentat a historic low. The upcoming election will shape Germany’s economic prospects for the nextyears.”15This question was asked in German and represents the standard survey item tapping respondents’ vote inten-tion that is used election surveys in Germany. For the original German translation, see the Appendix.16The focus of this study is on respondents’ vote intention. Results with an alternative dependent variable ofparty favorability, which was part of the registered pre-analayis plan, are reported in Table C.1 in the Appendix.17The German text of the treatments can be found in the Appendix.91Strategic voting: “The CDU/CSU is very likely to win the federal election and form thenew government. Whoever wants political change has only one option: to make the FDP or theGreens as strong as possible in order to avoid another four years of a Grand Coalition.”Uncertainty: “These are uncertain times and the world order is unstable. In times like thesemany think it is important for Germany to have an experienced leader.”Social inequality: “In the upcoming election the SPD wants to make Germany fairer, pro-vide more support to people who are struggling, and have the rich pay more in taxes. TheCDU/CSU and the FDP oppose those changes. Thus, the outcome of the election is likely tohave a direct impact on the extent of social inequality in Germany.”Based on assignment status, four dichotomous variables indicate whether a respondent re-ceived either of the four treatments.Finally, it is worth emphasizing that two of the treatments, the economic voting and thesocial inequality message, do not only prime respondents, but also contain informational com-ponents about the different political parties in the election. The former provides informationabout the current economic situation of stable economic growth and low unemployment inGermany and, at least implicitly, links this to the performance of the parties in government.The latter explicitly informs respondents about the policy positions of the German mainstreamparties on the issue of social inequality, telling them which parties are willing to raise taxesin order to redistribute income from the rich to the poor and which parties do not.18 As a re-sult, any effects that this study will uncover from these two treatments could be a combinationof pure priming (or “emphasis framing”) and information-based persuasion (see Leeper andSlothuus 2018). To the extent that these two treatments prove more effective than the other twotreatments, this would be consistent with recent empirical findings in the framing literature thatshow that information-based persuasion tends to be more effective than pure emphasis framing(Leeper and Slothuus 2018).Political Knowledge. The survey includes five questions that tap different aspects of re-18The strategic voting message also contains an informational component by telling people that the CDU/CSUis likely to win the election and by explicitly naming potential parties that might help avoid a new grand coalition.92spondents’ political knowledge.19 Two of the questions are about procedural aspects of theGerman electoral system: One of them is about the precise election threshold of valid votesthat parties need to win in order to be allocated any seats in the national legislature (answer:5 percent); the other asks respondents which of the two votes each voter has in the election ismore important for the allocation of seats in parliament (answer: the second vote). The otherthree questions gauge the extent to which respondents are informed about the election cam-paign itself by asking them to name the party each of the lead candidates from three smallerparties (CSU, FDP, Greens) belongs to.20 Based on the additive index of political knowledgefrom these questions ranging from 0 to 5 (alpha = .72), a knowledge dummy variable was cre-ated where respondents below the mean are coded as less politically knowledgeable and thoseabove it as more politically knowledgeable.Control Variables. To increase efficiency, the analysis that was pre-registered online in-cludes a series of standard control variables that have been shown to influence citizens’ votingbehavior. The most important of these variables is party identification. To avoid any primingof partisanship and thus, confounding of the treatment effects, this item was not included in thesurvey itself. Instead, it is based on party identification information that is routinely collectedand regularly updated by YouGov for all members of their online panel. Using this informa-tion, six partisanship indicators are created, one each for AfD, CDU/CSU, SPD, FDP/Greens,and mainstream parties’ (i.e. CDU/CSU, SPD, FDP, Greens, the Left) combined partisanship,as well as one for those with no partisanship at all.As socio-demographic control variables, this study includes indicators for gender, differentage groups, region (East versus West Germany), levels of education (bottom tier, medium tier,top tier), income groups (including a dummy for respondents with missing income values),trade union membership, and most important topic (immigration, terrorism, crime). Finally,a dichotomous variable indicates if a respondent does not fall into the most extreme (i.e. re-19The section on political knowledge is not part of the pre-analysis plan.20While the knowledge questions were asked after randomization, these questions are about factual knowledge.As a result, the knowledge variable is likely to be “a measure of pre-randomization conditions, and treatmentassignment had no effect on measurement error” (Lin, Green and Coppock 2016, 16).93strictive) category of the pre-screening question about immigration attitudes, which is the casefor 122 of our 1,786 respondents. With the exception of union membership and income, allcontrol variables were measured before respondents were exposed to a treatment. In the caseof union membership and income, it is assumed that responses to these factual questions areunaffected by the treatments. Following common practice in the literature, missing values fortrade union membership, party identification and most important topic are treated as absenceof these things and are coded as MethodsTo test the hypotheses that were specified above, I estimate the following OLS model21Yi = β0 + β1T1i + β2T2i + β3T3i + β4T4i + γXi + i (4.1)where Yi is a respondent’s vote intention toward one of the parties (1) AfD, (2) CDU/CSU,(3) SPD or (4) FDP/Greens,22 T1i to T4i are indicators for the four campaign message treat-ments of economic performance, strategic voting, uncertainty and social inequality primes re-spectively, Xi is a vector of covariates and i indexes individual respondents. One-tailed signif-icance tests are used if the direction of a hypothesis was explicitly specified in the pre-analysisplan (see Table 4.1). In all other cases, two-tailed significance tests are adopted. In all cases,Huber-White robust standard errors are calculated. Based on the hypotheses, we would expectnegative coefficients on all four treatments with AfD vote intention and positive coefficients21Unless explicitly stated otherwise, the analysis follows the study design that was pre-registered on EGAP,with two additions. First, the models in the pre-analysis plan included a dummy variable to identify the 122respondents that are less anti-immigrant than the rest of the sample. Initial analyses suggest that these 122 citizensreact differently to the four treatments than the other 1,664 respondents. Therefore, interaction terms between thisdummy and the treatment variables should have been included in the pre-registered model and were added in theanalysis. Second, the pre-analysis plan explicitly breaks down the party identification variable only for the modelon AfD vote intention, but fails to do so for the models for the other political parties. Models of the vote intentionfor the other parties are therefore specified analogously by: (1) including a party identification dummy for theparty whose vote share is analyzed by a given model, (2) an AfD partisanship dummy and (3) a dummy for nopartisan affiliation (reference category: partisanship for all other mainstream parties).22FDP and Greens are treated here as a single party so as to capture support for mainstream opposition partiesmore generally.94on T1i and T3i with CDU/CSU vote intention, T4i with the SPD and T2i for the FDP andGreens. For additional analyses on how political knowledge conditions the effect of the fourtreatments, regressions additionally include a political knowledge dummy and its interactionwith each of the four treatments to the model.23All models control for a respondent’s answer to the pre-screening question about their im-migration attitudes by including an indicator for those less opposed to immigration and inter-actions between this indicator and the four treatments.24 This serves to avoid any confoundingof our estimates by the 122 respondents who are less hostile to immigration and thus, are likelyto respond differently to the treatments.Table 4.1 summarizes the four treatments and the expected effects from each of them onthe electoral support for different German political parties: the Christian Democratic Union(CDU/CSU) of incumbent chancellor Angela Merkel, the main left-wing Social DemocraticParty (SPD), the right-wing populist Alternative for Germany (AfD) party and the two main-stream opposition parties, the Liberals (FDP) and the Greens.Table 4.1: Expected Effect on Electoral Support(1) (2) (3) (4)AfD CDU/CSU SPD FDP/GreensPopulist Incumbent Left-Wing OppositionEconomic Performance (H1) – +Strategic Voting (H2) – +Uncertainty (H3) – +Social Inequality (H4) – +– : decreased support; + : increased support. AfD: far-right populist party; CDU/CSU: centre-right ChristianDemocrats/Christian Social Union (party of incumbent chancellor); SPD: centre-left Social Democrats (coalitionpartner in grand coalition); FDP: Liberals; empty cells: no predictions from theoretical model.Multiple Comparisons Adjustment. Due to the large number of hypotheses that are testedin this study, the p-value threshold to determine whether a treatment is statistically significant23The model with the knowledge dummy and its interactions was not specified in the pre-analysis plan.24These interactions are not included in the pre-analysis plan.95needs to be adjusted. In classical hypothesis testing, the “alpha value” indicates a researcher’swillingness to commit a Type I error. With the commonly accepted alpha level of 0.05, wewould expect to incorrectly reject the null hypothesis about 5% of the time. However, the riskto incorrectly reject a true null hypothesis increases with the number of hypotheses that aretested. A standard way to counteract this risk is to adjust the target p-value downwards so thatwhen this new threshold is used for hypothesis testing, the chance of making a mistake stays atthe alpha level of 0.05. Based on the frequently used Bonferroni correction, we would obtaina new target p-value of 0.00625 (= 0.05/8) by dividing the p-value of 0.05 by the number ofhypotheses (in this case, eight) that are tested. However, the Bonferroni correction is increas-ingly considered as too restrictive because it does not account for the fact that the multiplecomparisons tend to be correlated with each other, commonly leading researchers to not re-ject null hypotheses that should be rejected, that is, to commit a Type 2 error (Coppock 2015;Gelman, Hill and Yajima 2012). Therefore, an approach that provides a better compromise be-tween trading off Type 1 against Type 2 error is to simulate the error rate and use results fromthe simulation to determine the appropriate p-value for hypothesis testing (Ibid.). FollowingCoppock (2015), this procedure gives us a target p-value of p=0.00841 that is slightly largerthan the p-value from the Bonferroni correction. Based on this new threshold, a treatment ef-fect from any of the 8 statistical tests in this study with a p-value below p=0.00841 can beconsidered statistically significant at the 95% level.4.4 ResultsTable 4.2 presents the results of the survey experiment. It shows the effect of each of the fourmainstream messages on the vote intention in the 2017 German federal election for people whoare very strongly opposed to any further immigration to Germany and thus, susceptible to ap-peals from the anti-immigrant populist Alternative for Germany (AfD) party. The coefficientsthat are displayed in bold directly test the hypotheses that were specified above. Column 1shows the effect of the four treatments on support for the Alternative for Germany (AfD) party.96Based on these results and contrary to what was expected, none of the four messages leads toa reduction in overall support for the Alternative for Germany (AfD) party among this groupof voters. All four coefficients are substantively small and indistinguishable from 0. Thus, themain hypotheses are not confirmed. None of the four messages seems to have an impact onsupport for the populist party.However, the situation looks a little bit different for the treatment effects that are antici-pated to be positive, which are displayed in columns 2 to 4. Focusing exclusively on the fourcoefficients in bold which test the specified hypotheses, two of the four coefficients of primaryinterest are substantively large and in the expected direction: The coefficients of 0.059 and0.038 in column 2 suggest that providing messages to respondents about the good performanceof the German economy and appealing to their desire for less uncertainty both benefit chan-cellor Angela Merkel and her incumbent CDU/CSU and increase support for this party amonganti-immigrant citizens by 5.9 and 3.8 percentage points respectively.25 However, accountingfor the multiple comparisons in this study, only the former, that is, the economy frame, is sta-tistically significant at the 95% level with a p-value of 0.007, which is below the target p-valueof 0.00841. The p-value of 0.047 of the uncertainty frame is well above this threshold andthus, not statistically significant. Finally, there is no evidence that either the strategic votingor the social inequality frame work on these voters. Both coefficients are close to 0 and notstatistically significant at conventional levels.Based on these regression estimates, Figure 4.1 illustrates the effects of the treatments onmainstream parties graphically. For each of the three mainstream parties, it plots the predictedvote intention for the control group and to the left of it, the predicted vote share after exposureto one of the treatments. The CDU/CSU vote share, which is relatively low in this samplewith only about 14%, climbs to roughly 20% and 18% after exposure to the economic per-formance and the uncertainty frames. The two treatments for the FDP/Greens and the SPDleave the overall vote intention for either party more or less unaffected at around 11% and 17%25Models without control variables, which are not part of the pre-analysis plan, produce coefficients of 0.035(instead of 0.059) and 0.005 (instead of 0.038). The difference in effect sizes is likely to be due to imbalances inrespondents’ partisanship between the treatment and control groups.97respectively.Table 4.2: Vote Intention in 2017 German Federal Election(1) (2) (3) (4)AfD CDU/CSU SPD FDP/GreensEconomic Voting -0.0085 0.059 -0.015 -0.046(0.027) (0.024) (0.022) (0.019)Strategic Voting 0.0014 0.044 -0.043 -0.0073(0.027) (0.023) (0.022) (0.021)Uncertainty 0.0038 0.038 -0.014 -0.037(0.027) (0.023) (0.022) (0.019)Social Inequality -0.000087 0.0052 0.0047 -0.010(0.027) (0.021) (0.022) (0.020)Constant 0.17 0.018 0.066 0.090(0.044) (0.037) (0.039) (0.034)Controls Yes Yes Yes YesObservations 1786 1786 1786 1786R2 0.45 0.41 0.42 0.27Note: Dependent variable is vote intention in 2017 German federal election for the party indicated at the top ofeach column. OLS regression with control variables: indicators for region (East vs. West), age group, gender,income group, education level, party identification, and most important topic (immigration, crime, terrorism).Models control for 122 respondents who are less anti-immigrant (indicator and interaction with treatments).Coefficients of primary interest for hypothesis tests are in bold.All in all, the evidence provides very limited support that citizens susceptible to AfD ap-peals are responsive to non-populist campaign messages. Most importantly, exposure to thesemessages does not lead to an overall reduction in electoral support for the German right-wingpopulist party. Results in Table 4.2 do not provide support for any of the (a) variants of the fourhypotheses that predict a decrease in the vote share for populist parties (H1a, H2a, H3a, H4a).However, what we see is some shift in vote intentions between mainstream parties: Increasedpopularity of the incumbent CDU/CSU primarily comes at the expense of small mainstreamopposition parties like the liberal FDP and the Greens (see column 4 in Table 4.2). Evidencefrom Table 4.2 and Figure 4.1 support the (b) variants of the economic voting hypothesis (H1b)whereas there is little support for the uncertainty, strategic voting and the inequality hypotheses98Figure 4.1: Share of Respondents Intending to Vote for Mainstream Parties, by TreatmentCondition0%5%10%15%20%25%Vote IntentionDV: Incumbent CDU/CSU DV: Liberals/Greens DV: Social DemocratsControlEconomyAnti-Establ.InsecurityInequalityNote: DV = dependent variable. Graphs include 95% confidence intervals.(H2b, H3b, H4b).26 This may indicate that rather than reaching those hard-core populist voterswho have already turned to the populist AfD party, the economic voting message primarilymanages to win back dissatisfied voters who have defected to other mainstream parties.A second observation from the results in Table 4.2 is that they seem to align well with Lenz(2012)’s previous findings about the superiority of performance primes over policy primes:Even with strongly anti-immigrant citizens, priming the economy – a valence issue – proves amore effective means to win votes by the incumbent party than priming redistributive policy – aposition issue –, which does apparently not manage to shift votes. However, as the next sectionshows, this conclusion may need some qualification once the group of strongly anti-immigrantvoters is split into different levels of political knowledge and people with low and high politicalknowledge are analyzed separately.26There is some suggestive evidence in Table 4.2 that the strategic voting message might have shifted somevotes from the SPD to the CDU/CSU. While such a shift is not part of the hypotheses that were specified above,it would be consistent with the overall goal of this treatment, which is to increase the vote share for a possiblegoverning coalition among mainstream parties that would not include the SPD. Coefficients are slightly above theconventional threshold of p=0.05 and thus, well above the target p-value of 0.00841 that is used in this study.994.4.1 Exploratory Analyses: The Role of Political KnowledgeA possible reason why the previous analysis failed to find any framing effects on the supportfor the anti-immigrant populist AfD party could be that voters respond differently to thesemessages. For example, extensive previous research has shown that a person’s level of politicalknowledge affects how she processes new information and how likely this new informationis to change her beliefs (e.g. Zaller 1992). Thus, this section explores the extent to whicha respondent’s responsiveness to the various treatments may be moderated by her level ofpolitical knowledge.27How would we expect a populist voter’s level of political knowledge to affect the previouslydiscussed treatments? In order for a new message to shift a person’s vote intention, it bothneeds to be received and accepted by this person. In other words, new information needsboth to be available and accessible to the voter in order for it to shift her opinions (Zaller1992). While availability of new information is less of a concern in the context of a surveyexperimental intervention where both less and more knowledgeable respondents receive thesame pieces of information, there will be differences between these two groups with respectto the accessibility of this information, that is, with how they process this information (seealso Chong and Druckman 2007, 2010). Following Zaller (1992), more knowledgeable voterswith an inclination toward right-wing populism would be expected to be more likely to resistmainstream primes that try to shift their vote away from these parties and toward non-populistparties. Due to their higher level of political attentiveness, these voters will already be awareof arguments both in favor and against different political parties and thus, their vote intentionis less likely to be moved by a single message in a survey. Less knowledgeable voters, on theother hand, will lack the contextual knowledge necessary to evaluate and resist the informationconveyed in the treatment messages and as a result, be more likely to change their vote. Inother words, less knowledgeable populist voters are expected to be more likely to respond tothe treatments in this study than more politically knowledgeable ones.27This exploratory analysis on the moderating effect of political knowledge was not pre-registered.100To examine the moderating effect of political knowledge, Table 4.3 shows results frommodels that interact each of the treatment variables with a political knowledge indicator. Over-all, there is some evidence that is consistent with moderation by political knowledge, in par-ticular for the economic voting and the uncertainty framing in column 1. Each of these twotreatments is associated with a strong decrease in support for the right-wing populist AfD partyamong less knowledgeable respondents, by 8.9 and 9.9 percentage points respectively. In linewith the theoretical expectation about the greater resistance to these messages by more knowl-edgeable voters, these two treatments do not reduce AfD support among more knowledgeablevoters. At 4.1 (= 13.0 - 8.9) and 6.1 (= 16.0 - 9.9) percentage points respectively, the effects aremuch smaller and have also changed in direction, suggesting that these two treatments wouldmake more knowledgeable voters rather more – and not less – likely to vote for the populistAfD party in an election.28 Figure 4.2 plots the predicted vote intention for the populist AfDparty for these two treatments for both knowledge groups. AfD support drops from 29% toroughly 20% and 19% after exposure to the economy and inequality treatments among the lessknowledgeable while support among the more knowledgeable stays the same or possibly evenslightly increases.Given the exploratory nature of the analysis on the moderating effect of knowledge, thisstudy refrains from reporting p-values and tests of statistical significance for this part of theanalysis. Not only do post hoc analyses not lend themselves for classical hypothesis testing,but we also lack any way of determining how to adjust for multiple hypothesis testing. Asthe heterogeneous treatment effects were not part of the pre-analysis plan, it is impossible toapply the previously discussed Bonferroni correction or, alternatively, simulation to determinethe appropriate level of statistical significance in this case. This means that without furtherempirical testing, we should treat the evidence that is consistent with a moderating effect ofpolitical knowledge on populist voters’ vote choice very cautiously.As for the other treatments, evidence for moderation by political knowledge is even less28This small increase in support for the AfD among more knowleageable voters is consistent with Chong andDruckman (2007) and Chong and Druckman (2010)’s finding that frames frequently backfire with motivatedrespondents and may lead to an effect that is the opposite of what was intended with the initial frame.101Table 4.3: 2017 Vote Intention by Level of Political Knowledge(1) (2) (3) (4)AfD CDU SPD FDP/GreensEconomic Voting -0.089 0.023 0.023 -0.059(0.044) (0.035) (0.038) (0.028)Strategic Voting -0.042 0.067 -0.037 -0.013(0.041) (0.034) (0.036) (0.032)Uncertainty -0.032 0.056 -0.0019 -0.033(0.045) (0.037) (0.038) (0.029)Social Inequality -0.099 0.0093 0.050 0.0082(0.040) (0.032) (0.039) (0.033)Knowledge 0.0010 -0.0024 0.033 0.0056(0.039) (0.034) (0.032) (0.030)Economy*Knowledge 0.13 0.057 -0.061 0.021(0.054) (0.047) (0.045) (0.038)Strategy*Knowledge 0.075 -0.041 -0.0068 0.0100(0.053) (0.045) (0.044) (0.041)Uncertainty*Knowledge 0.057 -0.027 -0.019 -0.0057(0.053) (0.045) (0.045) (0.038)Inequality*Knowledge 0.16 -0.0065 -0.072 -0.029(0.050) (0.041) (0.045) (0.041)Constant 0.20 0.016 0.048 0.087(0.048) (0.039) (0.045) (0.037)Controls Yes Yes Yes YesObservations 1786 1786 1786 1786R2 0.46 0.41 0.42 0.27Note: Dependent variable is vote intention in 2017 German federal election for the party indicated at the top ofeach column. 688 respondents are in the less knowledgeable group and 1098 in the more knowledgeable one.OLS regression with control variables: indicators for region (East vs. West), age group, gender, income group,education level, party identification, and most important topic (immigration, crime, terrorism). Models controlfor 122 of the 1,786 respondents who are less anti-immigrant (indicator and interaction with treatments).Coefficients of primary interest for this study are in bold.102Figure 4.2: Share of Respondents Intending to Vote for the AfD, by Political Knowledge andTreatment Condition0%10%20%30%40%AfD Vote IntentionLow Knowledge High KnowledgeControlEconomyInequalityclear. A cursory look at the coefficients in Table 4.3 seems to suggest that treatment effects forlow knowledge respondents in the sample tend to be larger in absolute terms, but these effectsare substatively smaller than for the two previously discussed treatments.29All in all, the results from Table 4.3 may suggest that further empirical tests of hypothe-ses H1a, H2a, H3a and H4a would be desirable. At least for the less knowledgeable votersin the sample, there is some suggestive evidence that both economic performance (H1a) andsocial inequality (H4a) framing may substantively reduce support for the populist AfD party,practically cutting support for this party by a third among those who are inclined to vote for it.4.5 ConclusionThis study examines the power of framing effects in a context where we would expect them theleast: among voters who are susceptible to populist appeals in a context where they are likelyto have previously been heavily exposed to populist messages due to a real-world election29A possible exception is the economic voting message and CDU/CSU vote intention in column 2.103campaign. Evidence from a survey experiment suggests that this group of voters is hardlyresponsive to any of the messages that were tested. The main finding is that overall supportfor right-wing populist parties is unaffected by these messages. In other words, there is littleevidence that would support hypotheses H1a, H2a, H3a and H4a.There is only very limited evidence that supports the (b) variants of the hypotheses. Aspredicted by one of the theoretical logics that were presented above, being able to shift thesevoters’ attention to performance-related issues – in particular, a well-performing economy –benefits the incumbent party and may help them win elections (Hypothesis H1b). However,contrary to what was expected, this increased support for the incumbent does not primarilycome at the expense of the right-wing populist challenger, but instead seems to shift votesaway from minor mainstream opposition parties toward the party of the incumbent chancellor.There is no empirical support for hypotheses H2b, H3b or H4b.I hypothesize and find some evidence that suggests that these null-findings may partiallybe due to the moderating effect of political knowledge.30 While the economic and the socialinequality frames strongly reduce support for a populist party among the less politically knowl-edgeable, they have no effect – or possibly even backfire – on more knowledgeable voters.In uncovering these effects, the study extends our understanding of the scope of framingeffects to a group of voters that has hitherto received rather scant attention: voters who aresusceptible to support (right-wing) populist parties. Overall, the ability of mainstream partiesto use mainstream non-immigration messages to successfully lure these voters away from anti-immigrant right-wing parties is very limited. To the extent that it seems possible to potentiallysway the vote of at least some of them and if we take into account the exploratory analyseson the moderating effect of political knowledge, it seems that there is not one single strategythat is clearly superior to all other framing strategies and that works with all potential populistvoters. For mainstream parties, this means that they have to make trade-offs when selectinga specific campaign strategy, deciding to focus more on certain groups of populist voters than30This aspect of the theoretical framework is not part of the registered pre-analysis plan.104on others. While the results of this study suggest that economic performance messages maybe the most promising electoral strategy, politicians trying to target specific groups of voters(e.g. potential populist voters with little political knowledge) might be better off highlightingtheir policy positions on issues such as economic redistribution. Thus, the choice of the idealframing strategy for potential populist voters depends both on the specific goals of politiciansand on the target audience of their messages. Preliminary evidence suggests that badly chosenstrategies are not only ineffective, but potentially may even backfire and harm a party’s electoralsuccess.A few caveats apply. The empirical test in this study represents a very conservative test ofmainstream parties’ ability to sway potential populist voters. Most respondents were surveyedless than two weeks before the 2017 German federal election, that is, at a time where they hadalready been extensively exposed to both populist and mainstream political messages. As aresult, most of them are likely to have made up their mind about the upcoming election. There-fore, future research might test the effectiveness of mainstream non-immigration messages atan earlier point in time during or possibly even completely outside of a real-world electioncampaign, that is, at a time where vote intentions are not firmed up yet.It is also worth pointing out that the null findings might partly be the result not so much ofthe substantive content of the messages that were used in this study, but rather of the way theywere formulated. In designing the various mainstream non-immigration messages, priority wasgiven to the realism of the frames by trying to mimic political messages that could plausiblyhave been used by mainstream parties in the 2017 German election. However, there might be atrade off between the greater realism of a frame and its persuasive strength, especially duringthe final stages of an election campaign. Thus, future research might consider broadening thescope of this study and putting more weight on the argumentative strength of the frames thanon their realism.Finally, it is worth emphasizing that two of the treatments – the economic voting and thesocial inequality messages – provide information both about the current state of the econ-omy and about mainstream parties’ policy positions. Any treatment effects that suggest that105populist voters may be swayed by mainstream messages were concentrated among these twoinformational frames. This study design does not allow to distinguish between information-based persuasion and pure emphasis framing. Yet, the fact that exploratory analyses suggestedthat both treatments had a disproportionate effect among the less knowledgeable respondentswould be consistent with an account that highlights the crucial role of information-based per-suasion. Populist voters who are least politically informed and thus, most difficult to reach bymainstream parties are the ones that are usually considered as the most staunch supporters ofright-wing populist parties. Thus, future research might want to explore more fully the extentthat information may play in shaping these citizens’ vote choice.Other questions remain for future research to consider as well. First, the findings fromthis study are based on findings from a multi-party context. This particular context meansthat dissatisfied voters can choose between smaller mainstream opposition parties and moreextreme right-wing populist parties to express their dissatisfaction with the incumbent govern-ment. Given that practically all of the shift in votes that this study uncovered happened be-tween mainstream opposition and mainstream incumbent parties, it is not clear to what extenteconomic performance messages (and, to a lesser extent, uncertainty messages) is an equallypowerful strategy in a two-party context. Second, future research could also more thoroughlyinvestigate the conditions under which some of the frames that are included in this study maybackfire and produce counterproductive effects. This issue has only been tangentially touchedupon in this study, but I find some evidence that suggests that especially among the more po-litically knowledgeable voters, treatments that might be perceived as too directly opposed toright-wing populist parties might produce backlash effects.106Chapter 5ConclusionThis dissertation aimed at identifying electoral reforms and campaign strategies that may in-crease political participation and reduce support for populism among politically alienated citi-zens. To do so, it used econometric methods for causal inference – a regression discontinuitydesign in Study 1 and a difference-in-differences design in Study 2 – and survey experimen-tation in combination with large original datasets to test the effectiveness of government inter-ventions and campaign messages. Improving on previous quantitative research in this area, thisresearch highlights the limitations of institutional fixes and provides new insights into the roleof information and political framing for civic engagement: Study 1 shows that informationalambiguity due to institutional complexity may confuse young voters and lead some of themto question their eligibility status, thus depressing voter turnout among this group of voters.Study 2 finds only limited support for the assumed positive relationship between voter turnoutand support for left-wing parties. The mixed findings from Study 2 raise doubts about manda-tory voting laws as an effective tool to address economic inequality in established democracies.Study 3 identifies mainstream economic messages as a potentially promising strategy for main-stream parties to win back moderately dissatisfied voters.The findings from these studies have direct practical implications for electoral reform de-bates, political campaign strategies and policy making in advanced industrial democracies.For example, voter turnout is particularly low among young, and especially first-time, voters.107Some people attribute the blame for this primarily to young voters themselves and to their lackof interest in politics and elections. However, findings from Study 1 suggest that for some ofthem, it is not so much that they do not want to vote, but that the procedural information costsrepresent particularly high hurdles to them. They may be confused by the specificities of theirstate voter registration rules and as a result, erroneously think that they are ineligible to vote.Thus, a direct implication from Study 1 is that in order to increase voter turnout particularlyamong those who turn 18 shortly before Election Day, reforms should aim at making the act ofregistering to vote easier for these citizens. Some U.S. states have recently switched to ElectionDay registration where citizens can both register and vote at the same time on Election Day.A few U.S. states have followed Oregon and now automatically add their citizens to the voterroles. In line with the findings on procedural information costs, both of these reforms will sig-nificantly reduce the risk that first-time voters are misled by registration rules in these states.Thus, both reforms are likely to lead to higher voter turnout among this group of citizens.Findings from Studies 2 and 3 have the potential to inform debates about the appropriatestrategies of how to address concerns about increasing economic inequality and the rise ofpopulist parties across advanced industrial democracies. To help tackle the former problem,some people – possibly most prominently former U.S. president Barack Obama – have musedabout the benefits of making voting compulsory. The idea is that if poor people, who are gen-erally less likely to vote, are compelled to cast a vote on Election Day, this will help left of thecenter political parties, potentially bringing them to power and as a consequence, leading tomore redistributive policies. However, findings from Study 2 caution against putting too muchhope into mandatory voting laws as a means to reduce economic inequality. While these lawsdefinitely prove very effective in bringing poor voters to the polls and thus, help to addressthe problem of political inequality, the political consequences of more equal political partici-pation on economic redistribution seem rather modest. Only a slightly more satisfying insightfor mainstream politicians who are electorally challenged by right-wing populist parties comesfrom Study 3. Mainstream parties, especially those that are right of the center, are increasinglyseen as trapped in a situation where they lose voters to anti-immigrant right-wing parties, but108refrain from adopting too anti-immigrant a political stance in order to avoid alienating moremoderate core voters. In this situation, Study 3 primarily highlights the constraints that main-stream parties face in luring voters away from anti-immigrant populist parties. However, italso shows that by downplaying the immigration issue and focusing instead its efforts on theeconomy, an incumbent moderate right-wing party might be able to increase its net electoralsupport. This happens not so much through drawing voters away from the more populist al-ternative, but rather through winning back dissatisfied voters that they have previously lost tomore moderate opposition parties.A few caveats apply, however. While a major strength of these three studies is that theirfindings are based on three large and unique datasets, these datasets have their own limitations.For example, in an ideal world Study 1 would not have to rely on birth data as a proxy forthe voting eligible population to calculate its youth voter turnout measure, but could directlyuse the number of voting eligible citizens by birthdate cohort. While using birth data does notlead to any bias in the treatment effect estimates – the difference in turnout between treatedand untreated young citizens is robust across a variety of specifications –, one needs to beaware that the turnout rates in this study are likely to be somewhat different from those in thereal world. Another shortcoming is not so much with the existing data, but rather with thedata that we do not have. For example, making inferences based on the data from Australianelections in Study 2 is limited by the small number of Australian states and, within some ofthese states, the relatively small number of electoral districts. Given these data limitations andthe mixed findings of this study, this research cannot claim to offer the final answer on thequestion whether higher voter turnout among the poor leads to higher support for left-wingparties. Yet, as a result of the fine-grained nature of the data and the research design, we haveinternally valid effect estimates that give us reason to question aspects of the received wisdomon this topic.Several implications for future research directly flow from these findings. First, futureresearch could use the insights from Study 2 to more thoroughly investigate the contextualfactors that seem to shape the relationship between voter turnout and support for left-wing109parties. Preliminary findings from this dissertation suggest that geographic factors – especiallywhether an electoral district is urban or rural and its geographic size – are powerful factors inshaping both voter turnout among the poor and partisan support. Second, the finding in Study 3that even potential populist voters are receptive to economic frames invites additional work onthese people’s policy preferences and, above all, how they trade off different policy preferencesif they are in conflict to one another.Finally, to the extent that rule complexity is not limited to elections, but extends to manyareas in which government agencies interact with citizens, the findings from Study 1 demandfor a broader analysis of procedural information costs and their impact on program uptake.Many social welfare programs require that citizens actively enrol into them. Previous researchlargely attributes non-enrolment into such programs either to discrimination by bureaucrats orto a lack of interest among the poor. However, the findings from Study 1 suggest a possiblealternative explanation. It could be that complex rules about program eligibility lead some ofthose who are most in need of these programs – the less educated and the socioeconomicallyless well off – to erroneously think they are ineligible and as a result, they fail to enrol. Giventhe widespread use of social welfare programs to alleviate extreme poverty and to help thosewho are most vulnerable in society, figuring out whether non-enrolment into beneficial gov-ernment programs by poor people is an active choice or rather the result of unclear eligibilityrules warrants further investigation.110BibliographyAlesina, Alberto and Eliana La Ferrara. 2005. “Preferences for Redistribution in the Land ofOpportunities.” Journal of Public Economics 89:897–931.Alesina, Alberto, Stefanie Stantcheva and Edoardo Teso. 2017. “Intergenerational Mobilityand Preferences for Redistribution.” NBER Working Paper 23027 .Alt, James E., David D. Lassen and John Marshall. 2016. “Credible Sources and SophisticatedVoters: When Does New Information Induce Economic Voting?” The Journal of Politics78(2):327–342.Angrist, Joshua D. and Jorn-Steffen Pischke. 2009. Mostly Harmless Econometrics: An Em-piricist’s Companion. Princeton, NJ: Princeton University Press.Atkinson, A. B. and Thomas Piketty, eds. 2007. Top Incomes over the Twentieth Century:A Contrast Between Continental European and English-Speaking Countries. Oxford/NewYork: Oxford University Press.Bartels, Larry M. 2008. Unequal Democracy: The Political Economy of the New Gilded Age.Princeton: Princeton University Press.Bechtel, Michael M., Dominik Hangartner and Lukas Schmid. 2016. “Does Compulsory Vot-ing Increase Support for Leftist Policy?” American Journal of Political Science 60(3):752–767.111Bennett, Scott and Barbara Bennett. 1986. Tasmanian Electoral Handbook 1851-1982. Kens-ington: History Project Incorporated.Bennett, W Lance and Steven Livingston. 2018. “The Disinformation Order: Disruptive Com-munication and the Decline of Democratic Institutions.” European Journal of Communica-tion 33(2):122–139.Birch, Sarah. 2008. Full Participation: A Comparative Study of Compulsory Voting. Manch-ester: Manchester University Press.Boix, Carles. 1998. Political Parties, Growth, and Equality: Conservative and Social Demo-cratic Economic Strategies in the World Economy. Cambridge/New York: Cambridge Uni-versity Press.Brady, Henry A., Sidney Verba and Kay Lehman Schlozman. 1995. “Beyond SES: A ResourceModel of Political Participation.” American Political Science Review 89(2):271–294.Brady, Henry E. and John E. McNulty. 2011. “Turning Out to Vote: The Costs of Finding andGetting to the Polling Place.” The American Political Science Review 105(1):115–134.Brett, Judith. 2019. From Secret Ballot to Democracy Sausage: How Australia Got CompulsoryVoting. Melbourne: The Text Publishing Company.Burden, Barry C., David T. Canon, Kenneth R. Mayer and Donald P. Moynihan. 2014. “Elec-tion Laws, Mobilization, and Turnout: The Unanticipated Consequences of Election Re-form.” American Journal of Political Science 58(1):95–108.Carey, John M. and Yusaku Horiuchi. 2017. “Compulsory Voting and Income Inequality:Evidence for Lijphart’s Proposition from Venezuela.” Latin American Politics and Society59(2):122–144.Carr, Adam. 2019. Psephos: Adam Carr’s Election Archive. available at: http://psephos.adam-carr.net/ (last accessed March 31, 2019).112Castles, Francis G. and John Uhr. 2007. “The Australian Welfare State: Has Federalism Madea Difference?” Australian Journal of Politics and History 53(1):96–117.Cattaneo, Matias D., Nicolas Idrobo and Rocio Titiunik. 2018. A Practical Introductionto Regression Discontinuity Designs: Volume II. New York: Cambridge UniversityPress. Draft. http://www-personal.umich.edu/ titiunik/books/CattaneoIdroboTitiunik2018-Cambridge-Vol2.pdf.Cattaneo, Matias D., Rocio Titiunik and Gonzalo Vazquez-Bare. 2017. “Comparing Infer-ence Approaches for RD Designs: A Reexamination of the Effect of Head Start on ChildMortality.” Journal of Policy Analysis and Management 36(3):643–681.Chapman, Emily Booth. 2019. “The Distinctive Value of Elections and the Case for Compul-sory Voting.” American Journal of Political Science 63(1):101–112.Chong, Dennis and James N. Druckman. 2007. “Framing Public Opinion in CompetitiveDemocracies.” American Political Science Review 101(4):637–655.Chong, Dennis and James N. Druckman. 2010. “Dynamic Public Opinion: CommunicationEffects over Time.” American Political Science Review 104(4):663–680.Chou, Winston, Rafaela Dancygier, Naoki Egami and Amaney Jamal. 2018.“The Illusion of Radical Right Partisan Stability: How Party PositioningAffects Radical Right Voting in Germany.” Working Paper, available at:https://scholar.princeton.edu/negami/publications/illusion-far-right-partisan-stability-how-party-positioning-affects-far-right (last accessed April 23, 2019) .Clifford, Eamonn, David Clune and Antony Green, eds. 2006. The Electoral Atlas of NewSouth Wales 1856-2006. Bathurst: Department of Lands.Colantone, Italo and Piero Stanig. 2018a. “Global Competition and Brexit.” American PoliticalScience Review 112(2):201–218.113Colantone, Italo and Piero Stanig. 2018b. “The Trade Origins of Economic Nationalism: Im-port Competition and Voting Behavior in Western Europe.” American Journal of PoliticalScience 62(4):936–953.Coppock, Alexander. 2015. “10 Things You Need to Know about Multiple Comparisons.”EGAP Methods Guides, available at: https://egap.org/methods-guides/10-things-you-need-know-about-multiple-comparisons (last accessed April 20, 2019) .Coppock, Alexander and Donald P. Green. 2016. “Is Voting Habit Forming? New Evidencefrom Experiments and Regression Discontinuities.” American Journal of Political Science60(4):1044–1062.Corvalan, Alejandro and Paulo Cox. 2018. “The Impact of Procedural Information Costs onVoting: Evidence from a Natural Experiment in Chile.” Political Behavior 40:3–19.Cox, Gary W. 1997. Making Votes Count: Strategic Coordination in the World’s ElectoralSystems. Cambridge/New York: Cambridge University Press.Cramer, Katherine J. 2016. The Politics of Resentment: Rural Consciousness in Wisconsin andthe Rise of Scott Walker. Chicago, IL: University of Chicago Press.Dal Bo, Ernesto, Frederico Finan, Olle Folke, Torsten Persson and Johanna Rickne. 2018.“Economic Losers and Political Winners: Sweden’s Radical Right.” Working Paper, avail-able at: perseus.iies.su.se/ tpers/papers/Draft180902.pdf (last accessed April 23, 2019) .Delli Carpini, Michael X. and Scott Keeter. 1996. What Americans Know about Politics andWhy It Matters. New Haven, CT: Yale University Press.Dinas, Elias. 2012. “The Formation of Voting Habits.” Journal of Elections, Public Opinionand Parties 22(4):431–456.Dinas, Elias, Konstantinos Matakos, Dimitrios Xefteris and Dominik Hangartner. 2019. “Wak-ing Up the Golden Dawn: Does Exposure to the Refugee Crisis Increase Support forExtreme-Right Parties?” Political Analysis 27(2):244–254.114Downs, Anthony. 1957. An Economic Theory of Democracy. New York: Harper and Row.Druckman, James N. 2004. “Political Preference Formation: Competition, Deliberation, andthe (Ir)relevance of Framing Effects.” The American Political Science Review 98(4):671–686.Duch, Raymond M. and Randolph T. Stevenson. 2008. The Economic Vote: How Political andEconomic Institutions Condition Election Results. New York: Cambridge University Press.Dunning, Thad. 2012. Natural Experiments in the Social Sciences: A Design-Based Approach.New York: Cambridge University Press.Evans, Geoffrey and Anand Menon. 2017. Brexit and British Politics. Cambridge: Polity Press.Evans, Tim. 2006. “Compulsory Voting in Australia.” Australian Electoral Commission Reportpp. available at: https://www.aec.gov.au (last accessed April 2, 2019).Farrell, David M. and Ian McAllister. 2006. The Australian Electoral System: Origins, Varia-tions and Consequences. Sydney: University of New South Wales Press.Ferwerda, Jeremy. 2014. “Electoral Consequences of Declining Participation: A Natural Ex-periment in Austria.” Electoral Studies 35:242 – 252.Fisman, Raymond, Ilyana Kuziemko and Silvia Vanutelli. 2017. “Distributional Preferencesin Larger Groups: Keeping up with the Joneses and Keeping Track of the Tails.” WorkingPaper .Fowler, Anthony. 2013. “Electoral and Policy Consequences of Voter Turnout: Evidence fromCompulsory Voting in Australia.” Quarterly Journal of Political Science 8(2):159–182.Franklin, Mark N. 2004. Voter Turnout and the Dynamics of Electoral Competition in Estab-lished Democracies since 1945. Cambridge/New York: Cambridge University Press.Fujiwara, Thomas, Kyle Meng and Tom Vogl. 2016. “Habit Formation in Voting: Evidencefrom Rainy Elections.” American Economic Journal: Applied Economics 8(4):160–188.115Gelman, Andrew, Jennifer Hill and Masanao Yajima. 2012. “Why We (Usually) Don’t Haveto Worry About Multiple Comparisons.” Journal of Research on Educational Effectiveness5:189–211.Gerber, Alan S., Donald P. Green and Ron Shachar. 2003. “Voting May Be Habit-Forming:Evidence from a Randomized Field Experiment.” American Journal of Political Science47(3):540–550.Gerber, Alan S., Gregory A. Huber, David Doherty, Conor M. Dowling and Seth J. Hill. 2013.“Do Perceptions of Ballot Secrecy Influence Turnout? Results from a Field Experiment.”American Journal of Political Science 57(3):537–551.Gerber, Alan S., Gregory A. Huber, Marc Meredith, Daniel R. Biggers and David J. Hendry.2015. “Can Incarcerated Felons Be (Re)integrated into the Political System? Results from aField Experiment.” American Journal of Political Science 59(4):912–926.Gooch, Andrew and Lynn Vavreck. 2019. “How Face-to-Face Interviews and Cognitive SkillAffect Item Non-Response: A Randomized Experiment Assigning Mode of Interview.” Po-litical Science Research and Methods 7(1):143162.Grimmer, Justin, Eitan Hersh, Marc Meredith, Jonathan Mummolo and Clayton Nall. 2018.“Obstacles to Estimating Voter ID Laws’ Effect on Turnout.” Journal of Politics 80(3):1045–1051.Guess, Andrew, Brendan Nyhan and Jason Reifler. 2018. “Selective Exposure to Misinforma-tion: Evidence from the Consumption of Fake News during the 2016 U.S. Presidential Cam-paign.” Working Paper . http://www.ask-force.org/web/Fundamentalists/Guess-Selective-Exposure-to-Misinformation-Evidence-Presidential-Campaign-2018.pdf.Hajnal, Zoltan, Nazita Lajevardi and Lindsay Nielson. 2017. “Voter Identification Laws andthe Suppression of Minority Votes.” Journal of Politics 79(2):363–379.116Helmke, Gretchen and Bonnie Meguid. 2010. “Endogenous Institutions: The Origins of Com-pulsory Voting Laws.” Working Paper, available at: https://www.gretchenhelmke.com (lastaccessed May 14, 2019) .Herron, Michael. 1998. “The Presidential Election of 1988: Low Voter Turnout and the Defeatof Michael Dukakis.” Working Paper .Highton, Benjamin and Raymond E. Wolfinger. 2001. “The Political Implications of HigherTurnout.” British Journal of Political Science 31:179–223.Hirczy, Wolfgang. 1994. “The Impact of Mandatory Voting Laws on Turnout: A Quasi-Experimental Approach.” Electoral Studies 13:64–76.Holbein, John B. and D. Sunshine Hillygus. 2016. “Making Young Voters: The Impact ofPreregistration on Youth Turnout.” American Journal of Political Science 60(2):364–382.Hopkins, Daniel J. 2011. “Translating into Votes: The Electoral Impacts of Spanish-LanguageBallots.” American Journal of Political Science 55(4):814–830.Huber, Gregory A. and John S. Lapinski. 2006. “The ”Race Card” Revisited: Assessing RacialPriming in Policy Contests.” American Journal of Political Science 50(2):421–440.Hughes, Colin A. and B. D. Graham. 1968. A Handbook of Australian Government and Politics1890-1964. Canberra: Australian National University Press.Hughes, Colin A. and B.D. Graham. 1974a. Voting for the Australian House of Representatives,1901-1964. Canberra: Australian National University Press.Hughes, Colin A. and B.D. Graham. 1974b. Voting for the Queensland Legislative Assembly,1890-1964. Canberra: Australian National University Press.Hughes, Colin A. and B.D. Graham. 1976. Voting for the South Australian, Western Australianand Tasmanian Lower Houses, 1890-1964. Canberra: Australian National University Press.117Iyengar, Shanto and Donald R. Kinder. 1987. News That Matter: Television and AmericanOpinion. Chicago, IL: University of Chicago Press.Jacobs, Alan M. and Scott J. Matthews. 2017. “Policy Attitudes in Institutional Context: Rules,Uncertainty, and the Mass Politics of Public Investment.” American Journal of Political Sci-ence 61(1):194–207.Jaensch, Dean. 1994. Power Politics: Australia’s Party System. 3 ed. St Leonard: Allen andUnwin.Johnston, Richard, Andre Blais, Henry E. Brady and Jean Crete. 1992. Letting the PeopleDecide: Dynamics of a Canadian Election. Stanford, CA: Stanford University Press.Kalla, Joshua L. and David E. Broockman. 2018. “The Minimal Persuasive Effects of Cam-paign Contact in General Elections: Evidence from 49 Field Experiments.” American Polit-ical Science Review 112(1):148–166.Kasara, Kimuli and Pavithra Suryanarayan. 2015. “When Do the Rich Vote Less Than the Poorand Why? Explaining Turnout Inequality across the World.” American Journal of PoliticalScience 59(3):613–627.Katz, Richard S. and Peter Mair. 1995. “Changing Models of Party Organization and PartyDemocracy: The Emergence of the Cartel Party.” Party Politics 1(1):5–35.Kedar, Orit. 2005. “When Moderate Voters Prefer Extreme Parties: Policy Balancing in Par-liamentary Elections.” American Political Science Review 99(2):185–199.Kim, Young Mie, Jordan Hsu, David Neiman, Colin Kou, Levi Bankston, Soo Yun Kim,Richard Heinrich, Robyn Baragwanath and Garvesh Raskutti. 2018. “The Stealth Media?Groups and Targets behind Divisive Issue Campaigns on Facebook.” Political Communica-tion 35(4):515–541.Krause, Werner, Denis Cohen and Tarik Abou-Chadi. 2019. “DoesAccommodation Work? Mainstream Party Strategies and the Suc-118cess of Radical Right Parties.” Working Paper, available at:https://www.dropbox.com/s/1g5wicx6peae2gs/KrauseCohenAbouchadi2019.pdf?dl =0(lastaccessedApril22, 2019).Kriesi, Hanspeter. 1999. Movements of the Left, Movements of the Right: Putting the Mobi-lization of Two New Types of Social Movements into Political Context. Cambridge Studiesin Comparative Politics New York: Cambridge University Press pp. 398–424.Krosnick, Jon A. and Donald R. Kinder. 1990. “Altering the Foundations of Support for thePresident Through Priming.” American Political Science Review 84(2):497–512.Kuziemko, Ilyana, Michael I .Norton, Emmanuel Saez and Stefanie Stantcheva. 2015. “HowElastic Are Preferences for Redistribution? Evidence from Randomized Survey Experi-ments.” American Economic Review 105(4):1478–1508.Leeper, Thomas J. and Rune Slothuus. 2018. “Can Citizens Be Framed? How Persuasive Infor-mation More than Emphasis Framing Changes Political Opinions.” Working Paper, availableat: https://s3.us-east-2.amazonaws.com/tjl-sharing/assets/CanCitizensBeFramed.pdf (lastaccessed on April 21, 2019) .Lenz, Gabriel. 2012. Follow the Leader? How Voters Respond to Politicians’ Performance andPolicies. Chicago, IL: University of Chicago Press.Lijphart, Arend. 1997. “Unequal Participation: Democracy’s Unresolved Dilemma.” TheAmerican Political Science Review 91(1):1–14.Lin, Winston, Donald P. Green and Alexander Coppock. 2016. “Standard Operating Proceduresfor Don Green’s Lab at Columbia.” available at: http://alexandercoppock.com/Green-Lab-SOP/GreenLabSOP.pdf(lastaccessedApril23, 2019)(V ersion1.05).Luskin, Robert C. 1990. “Explaining Political Sophistication.” Political Behavior 12(4):331–361.119Mackerras, Malcolm and Ian McAllister. 1999. “Compulsory Voting, Party Stability and Elec-toral Advantage in Australia.” Electoral Studies 18(2):217 – 233.McKinlay, Brian. 1981. The ALP: A Short History of the Australian Labor Party. Richmond:Heinemann Publishers.Meguid, Bonnie M. 2005. “Competition between Unequals: The Role of Mainstream PartyStrategy in Niche Party Success.” American Political Science Review 99(3):347–359.Meguid, Bonnie M. 2008. Party Competition between Unequals: Strategies and ElectoralFortunes in Western Europe. New York: Cambrige University Press.Meltzer, Allan H. and Scott F. Richard. 1981. “A Rational Theory of the Size of Government.”Journal of Political Economy 89(5):914–927.Mendelberg, Tali. 1997. “Executing Hortons: Racial Crime in the 1988 Presidential Cam-paign.” The Public Opinion Quarterly 61(1):134–157.Meredith, Marc. 2009. “Persistence in Political Participation.” Quarterly Journal of PoliticalScience 4(3):187–209.Meredith, Marc and Michael Morse. 2014. “Do Voting Rights Notification Laws Increase Ex-Felon Turnout?” The ANNALS of the American Academy of Political and Social Science651(1):220–249.Meredith, Marc and Michael Morse. 2015. “The Politics of the Restoration of Ex-Felon VotingRights: The Case of Iowa.” Quarterly Journal of Political Science 10:41–100.Miller, Peter and Ruth Dassonneville. 2016. “High Turnout in the Low Countries: Partisan Ef-fects of the Abolition of Compulsory Voting in the Netherlands.” Electoral Studies 44:132–143.Moon, Jeremy and Campbell Sharman, eds. 2003. Australian Politics and Government: TheCommonwealth, the States and the Territories. Cambridge: Cambridge University Press.120Morgan-Collins, Mona and Dawn Langan Teele. 2016. “Revisiting the Gender Voting Gapin the Era of Women’s Suffrage.” Paper presented at the 2016 American Political ScienceAssociation (APSA) Annual Meeting .Mudde, Cas. 2007. Populist Radical Right Parties in Europe. New York: Cambridge UniversityPress.Mudde, Cas and Cristobal Rovira Kaltwasser. 2017. Populism: A Very Short Introduction.Oxford/New York: Oxford University Press.Mutz, Diana. 2018. “Status Threat, Not Economic Hardship, Explains the 2016 PresidentialVote.” Proceedings of the National Academy of Sciences 115(19):E4330–E4339.Nickerson, David W. 2015. “Do Voter Registration Drives Increase Participation? For Whomand When?” The Journal of Politics 77(1):88–101.Norris, Pippa and Ronald Inglehart. 2019. Cultural Backlash: Trump, Brexit, and the Rise ofAuthoritarian Populism. New York: Cambridge University Press.Nyhan, Brendan, Christopher Skovron and Rocio Titiunik. 2017. “Differential RegistrationBias in Voter File Data: A Sensitivity Analysis Approach.” American Journal of PoliticalScience 61(3):744–760.Pons, Vincent and Clemence Tricaud. 2018. “Expressive Voting and Its Cost: Evidence fromRunoffs with Two or Three Candidates.” Econometrica 86(5):1621–1649.Powell, G. Bingham and Guy D. Whitten. 1993. “A Cross-National Analysis of EconomicVoting: Taking Account of the Political Context.” American Journal of Political Science37(2):391–414.Rubenson, Daniel, Andre Blais, Patrick Fournier, Elisabeth Gidengil and Neil Nevitte. 2007.“Does Low Turnout Matter: Evidence from the 2000 Canadian Election.” Electoral Studies26(3):589–597.121Selb, Peter and Romain Lachat. 2009. “The More, the Better? Counterfactual Evidence onthe Effect of Compulsory Voting on the Consistency of Party Choice.” European Journal ofPolitical Research 48:573–597.Sharman, Campbell. 2003. “Uncontested Seats and the Evolution of Party Competition.” PartyPolitics 9(6):679–702.Sharman, Campbell and Anthony Sayers. 1998. “Swings and Roundabouts? Patterns of Votingfor the Australian Labor Party at State and Commonwealth Lower House Elections, 1901-96.” Australian Journal of Political Science 33(3):329–344.Singh, Shane P. 2019a. “Compulsory Voting and Parties’ Vote-Seeking Strategies.” AmericanJournal of Political Science p. Forthcoming.Singh, Shane P. 2019b. “Politically Unengaged, Distrusting, and Disaffected Individuals Drivethe Link Between Compulsory Voting and Invalid Balloting.” Political Science Research andMethods 7(1):107–123.Smith, Rodney. 2003. Australian Politics and Government: The Commonwealth, the Statesand the Territories. Cambridge: Cambridge University Press chapter New South Wales,pp. 41–73.Stringer, Jordan. 2008. “Criminalizing Voter Suppression: The Necessity of Restoring Legiti-macy in Federal Elections and Reversing Disillusionment in Minority Communities.” EmoryLaw Journal 57(4):1011–1048.Umbers, Lachlan M. 2018. “Compulsory Voting: A Defence.” British Journal of PoliticalScience p. Forthcoming.van der Eijk, Cees and Marcel van Egmond. 2007. “Political Effects of Low Turnout in Na-tional and European Elections.” Electoral Studies 26(3):561 – 573.Vavreck, Lynn. 2009. The Message Matters: The Economy and Presidential Campaigns.Princeton, NJ: Princeton University Press.122Weller, Patrick and Jenny Fleming. 2003. Australian Politics and Government: The Common-wealth, the States and the Territories. Cambridge: Cambridge University Press chapter TheCommonwealth, pp. 12–40.Wolfinger, Raymond E. and Steven J. Rosenstone. 1980. Who Votes? New Haven, CT: YaleUniversity Press.Zaller, John R. 1992. The Nature and Origins of Mass Opinion. Cambridge: CambridgeUniversity Press.123Appendix ASupporting Materials to Chapter 2A.1 Smoothness of Density around Registration Deadline Cut-off124Figure A.1: Distribution of Births in Florida between May 5, 1990 and May 4, 1991300400500600700Number of births5/4/90 8/4/90 11/4/90 2/4/91 5/4/91Date of BirthBirths in Florida, 1990-1991The vertical line to the left indicates the day and month of the registration deadline for the 2008 presidentialelection, the vertical line to the right the day and month of the Election Day eligibility cutoff for the 2008 election.A.2 Registration Deadline Discontinuities in TexasDiscontinuities at the registration deadline are widespread across U.S. states. Figure A.2 plotsturnout for the 2008 and 2012 U.S. presidential elections for young voters in Texas. As we cansee in Figure A.2(a) for the 2008 presidential election, there clearly is a discontinuity aroundthe registration deadline in Texas, with voter turnout being significantly lower in that electionfor those who turned 18 after the registration deadline. The analysis in Texas is somewhatcomplicated, however, by a third discontinuity around the school entry cutoff date of September1. Young people to the left of this cutoff have started school a year earlier and thus, will usuallyjust have finished high school shortly before the 2008 election. I expect that due to the changesin these people’s lives (starting college or a job, leaving home) they are somewhat less likely to125vote than their peers who are slightly younger and still in high school at the time of the election.Figure A.2(b) shows that while there is a discontinuity in turnout for young Texans around theelection day cutoff in the 2012 presidential election, we also have a small discontinuity aroundthe registration deadline.Figure A.2: Turnout in Presidential Elections in TexasEligible in 2008 Non-Eligible in 2008Eligibility ThresholdRegistration DeadlineSchool Entry Cutoff0%10%20%30%40%50%60%Turnout5/4/90 8/4/90 11/4/90 2/4/91 5/4/91Date of Birth(a) Turnout in 2008 Presidential ElectionEligible in 2008 Non-Eligible in 2008Eligibility ThresholdRegistration DeadlineSchool Entry Cutoff10%15%20%25%30%35%40%Turnout5/4/90 8/4/90 11/4/90 2/4/91 5/4/91Date of Birth(b) Turnout in 2012 Presidential Election126A.3 Separate Treatment Around School Entry CutoffIn some states we can observe a slight drop in turnout to the left of the school entry cutoffdates of September 1 (Florida, Texas) or September 30 (Ohio). Given that those whose 18thbirthday is shortly before this deadline will already have finished high school by the time ofthe 2008 election, they are more likely to have left their established social networks in order tostart college or work. Having to vote during this short, but unstable time where young peopletransition from high school to the next stage in their lives seems to make them less likely tovote in their first few elections (Franklin 2004). As a result, we see a slight drop in turnout forthese voters in both the 2008 and 2012 elections. Figure A.3 illustrates this for Ohio, FigureA.2 above does so for Texas. Pennsylvania did not specify a school entry cutoff date in itsstatutes in the early 1990s and New York’s was after election day on December 1. Thereforeyouth turnout around the election day cutoff is unaffected by the school entry cutoff date inthese two states.Figure A.3: Turnout in 2012 Presidential Election in OhioEligible in 2008 Non-Eligible in 2008EligibilitySchool Cutoff25%30%35%40%45%50%Turnout5/4/90 8/4/90 11/4/90 2/4/91 5/4/91Date of Birth127A.4 Discontinuity in Daily Vote Totals for FloridaAnother way to test whether there is a discontinuity around the registration deadline cutoff isto plot the total number of votes by birthdate cohorts. This measure has the big advantagethat it is not affected by migration in and out of a state. Thus, if I still find support for myclaims about a discontinuity at the registration deadline, this will greatly increase the confi-dence in my findings. While this measure will be more non-linear than my turnout measuredue to seasonal variation in the number of births and, as a consequence, in eligible voters, wewould not expect to see any abrupt discontinuities in the total number of votes unless there isa treatment effect. Given that total births are much lower at weekends than for weekdays −something that makes the graphical inspection of discontinuities extremely difficult −, FigureA.4 plots the votes by birthdate cohorts for the 2008 and 2012 U.S. presidential elections inFlorida only for weekdays. Dropping all Saturdays and Sundays from these graphs also solvesthe problem that people directly to the left of both the registration deadline and election daycutoffs in 2008 were born on a weekend. Across all graphs, we can see the same pattern: thereis a clear discontinuity in votes at the registration deadline cutoff, but there is no discontinuityat the election day cutoff.Figure A.5 is similar to Figure A.4, but includes weekend days. In order to adjust for themuch lower number of births and, as a consequence, of votes for these days, I multiply thenumber of votes for Saturdays and Sundays by a factor of 1.23 and 1.31 respectively, thusincreasing the number of votes for these days to the average number of votes for weekdays.The overall pattern (i.e. a discontinuity around the registration deadline and no discontinuityat the election day cutoff) is unaffected by the inclusion of weekend days.128Figure A.4: Votes in 2008 and 2012 Presidential Elections in Florida - Excluding Weekends200250300350Votes9/4/90 10/4/90 11/4/90 12/4/90 1/4/91Date of Birth(a) 2012 Election, Narrow Window200250300350Votes5/4/90 8/4/90 11/4/90 2/4/91 5/4/91Date of Birth(b) 2012 Election, Wide Window0100200300400Votes9/4/90 10/4/90 11/4/90 12/4/90 1/4/91Date of Birth(c) 2008 Election, Narrow Window0100200300400Votes5/4/90 8/4/90 11/4/90 2/4/91 5/4/91Date of Birth(d) 2008 Election, Wide WindowA.5 Robustness of RD Estimates to Alternative BandwidthsTable A.1 replicates the results from Table 2.1 by using a 10-day window on either side of theregistration deadline cutoff instead of a 28-day window. Reducing the bandwidth to roughly athird of its original size produces results that are largely similar to those presented in Table 2.1above, further bolstering our confidence in the robustness of these results.129Figure A.5: Votes in 2008 and 2012 Presidential Elections in Florida - Including Weekends150200250300350Votes9/4/90 10/4/90 11/4/90 12/4/90 1/4/91Date of Birth(a) 2012 Election, Narrow Window150200250300350Votes5/4/90 8/4/90 11/4/90 2/4/91 5/4/91Date of Birth(b) 2012 Election, Wide Window0100200300400Votes9/4/90 10/4/90 11/4/90 12/4/90 1/4/91Date of Birth(c) 2008 Election, Narrow Window0100200300400Votes5/4/90 8/4/90 11/4/90 2/4/91 5/4/91Date of Birth(d) 2008 Election, Wide WindowA.6 Robustness of RD Estimates to Alternative Model Spec-ificationsTo make sure that the main results from Table 2.1 above are not a product of the estimationstrategy that is used in the main part of this article, I re-estimated the difference in turnoutbetween preregistration eligibles and postregistration eligibles by using the following standardOLS model:Yj = β0 + β1Tj + β2Dj + β3T ∗Dj + j (A.1)Yj is turnout measured as the number of people who voted over the number of people bornon a given day, Tj is an indicator for whether people turn 18 after the registration deadline or130Table A.1: Turnout in U.S. Presidential Elections for 1990 Birth Cohorts(FL) (TX) (NY) (OH) (PA)Registration Deadline Discontinuity2008Eligibles post-deadline 45.67 25.95 30.25 47.64 40.78Eligibles pre-deadline 51.77 33.22 30.60 48.72 44.45Difference -6.10 -7.28 -0.34 -1.08 -3.67[-7.94, -4.27] [-8.61, -5.96] [-1.76, 1.07] [-3.14, 0.99] [-5.66, -1.68]N 11,364 18,136 16,347 8,990 9,4432012Eligibles post-deadline 45.86 24.96 30.55 41.47 36.83Eligibles pre-deadline 49.00 26.79 28.85 38.81 37.21Difference -3.14 -1.82 2.66 1.50 -0.38[-4.97, -1.30] [-3.10, -0.55] [0.30, 3.10] [0.63, 4.69] [-2.33, 1.57]N 11,364 18,136 16,347 8,990 9,4432016Eligibles post-deadline 44.68Eligibles pre-deadline 47.53Difference -2.85[-4.68, -1.02]N 11,364Note: Table presents difference-in-proportions with 95% confidence intervals in brackets for young Americans born in 1990. Eachestimation uses a 10-day-window on either side of the registration deadline in 2008.not, Dj is the running variable indicating the distance (in days) from the registration deadline,and j indexes each birthday cohort. The interaction term T ∗ Dj allows for different slopeson either side of the registration deadline. Like the analysis in the main part of the text, I usea 28-day window on either side of the cutoff. Note that a major advantage of using an OLSmodel is that, assuming the model is correctly specified, it gives us an estimate of the treatmenteffect right at the cutoff point. This means that if the results from the OLS model are similarto the findings from the difference-in-proportions, this bolsters our confidence in our estimatesand in the choice of a 28-day window on either side of the registration deadline cutoff.131Table A.2: Turnout in U.S. Presidential Elections(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)FL TX NY OH PA FL TX NY OH PA FL2008 2008 2008 2008 2008 2012 2012 2012 2012 2012 2016Registration Beliefs -0.044∗ -0.066∗ 0.0065 -0.011 -0.042∗ -0.027 -0.020∗ 0.022∗ 0.016 -0.0018 -0.029∗(0.016) (0.011) (0.011) (0.015) (0.016) (0.014) (0.008) (0.009) (0.016) (0.015) (0.014)Days -0.043 -0.16 -0.20∗ 0.21 0.15 0.13 0.034 -0.15∗ 0.053 0.00055 0.19(0.130) (0.085) (0.071) (0.124) (0.125) (0.114) (0.068) (0.060) (0.130) (0.121) (0.109)Beliefs*Days -0.22 0.020 0.012 -0.30 -0.20 -0.27 -0.025 -0.026 -0.14 -0.082 -0.35∗(0.185) (0.120) (0.121) (0.176) (0.177) (0.161) (0.096) (0.102) (0.183) (0.171) (0.155)Constant 0.52∗ 0.33∗ 0.30∗ 0.49∗ 0.44∗ 0.49∗ 0.27∗ 0.29∗ 0.39∗ 0.37∗ 0.48∗(0.012) (0.008) (0.006) (0.011) (0.011) (0.010) (0.006) (0.005) (0.012) (0.011) (0.010)Observations 56 56 56 56 56 56 56 56 56 56 56R2 0.58 0.84 0.38 0.06 0.29 0.24 0.23 0.19 0.06 0.03 0.27Table displays OLS estimates with a 28-day window (25 for NY) on either side of the cutoff. Standard errors in parentheses. ∗ p < 0.05132Across the different models, the pattern is largely similar to the results from difference-in-proportions. For the 2008 elections, effect sizes are still large and statistically significantfor Florida, Texas and Pennsylvania. However, we do no longer find an effect of registrationbeliefs for New York. Based on the OLS model, the effect size for 2008 is now somewhatsmaller for both Florida and Texas. Registration beliefs decrease turnout for those who turn18 right after the registration deadline in 2008 by 4.4 percentage points for Florida and by 6.6percentage points for Texas. Note that the OLS estimates are smaller than the results from boththe difference-in-proportions with the wide 28-day and the narrow 10-day window on eitherside of the registration deadline, potentially pointing to difficulties in appropriately estimatingvoter turnout with OLS if the running variable (i.e. birthdates) is discrete and the estimationwindow is narrow (Nyhan, Skovron and Titiunik 2017, 751). The effect of registration beliefsfor the same group of voters in both Florida and Texas for the 2012 election are very similar insize to the estimates in the main text from difference-in-proportions.133A.7 Robustness of Persistence of Effect to Varying Band-widthsIn this section I show that the discontinuity in turnout that we still find four years later betweenthose who turn 18 before and those who turn 18 after the registration deadline is robust tovarying bandwidth specifications around the RD cutoff for Florida. Figure A.6 plots the size ofthe discontinuity at the registration deadline for 22 year-olds in the 2012 presidential electionfor different bandwidths. We can see that the size of the difference in turnout between thesetwo groups is largely unaffected by the choice of the bandwidth.Figure A.6: Varying Bandwidths Around Registration Deadline Cutoff-. effect21522 44 66 88 110 126 149 171 193Bandwidth (Days)CIEstNote: Effect of being born before the registration deadline of October 7, 2008 on 2012 voter turnout in Florida.134A.8 Information on Registration Requirements Across U.S.StatesWhile every U.S. state with voter registration allows people under 18 to register if they willbe 18 by Election Day, the precise age that is required to register varies across states. In somestates, young people can register as soon as they turn 16 or 17, in others, they have to wait untilthey are 17 years and 10 months old, and for many states, people are eligible to register if theywill be 18 by the next general election. However, the precise age requirement in a state seemsless important than the clarity by which these rules are conveyed. While only experimental ev-idence can test the causal effect of the clarity of information on a person’s decision to registerand vote in an election, the clarity with which this age requirement is conveyed in the electioncode or on registration forms may serve as a proxy for the degree to which voters may be uncer-tain about the exact age requirement to register in a given state. Among states analyzed in moredetail in this article, some election codes and/or their registration forms (Florida, Oklahoma,Texas) are somewhat less clear about the exact age requirement to register in that state, sug-gesting that in these states there might potentially exist more uncertainty among voters aboutwhether they have to be 18 by the registration deadline or by Election Day. The election codesin these states read as follows (emphasis added):Florida: 97.041 Qualifications to register or vote. “(1)(a) A person may become a regis-tered voter only if that person: 1. Is at least 18 years of age; 2. Is a citizen of the UnitedStates; 3. Is a legal resident of the State of Florida; 4. Is a legal resident of the county in whichthat person seeks to be registered; and 5. Registers pursuant to the Florida Election Code. (b)A person who is otherwise qualified may preregister on or after that person’s 16th birthday andmay vote in any election occurring on or after that person’s 18th birthday.”Oklahoma: 26.A1.VI.Section 4-101 - Persons Entitled to Become Registered Voters “Ev-ery person who is a qualified elector as defined by Section 1 of Article III of the OklahomaConstitution shall be entitled to become a registered voter in the precinct of his residence”135Oklahoma Constitution, Article 3, paragraph 1 “Subject to such exceptions as the Legisla-ture may prescribe, all citizens of the United States, over the age of eighteen (18) years, whoare bona fide residents of this state, are qualified electors of this state.”Texas: Sec. 13.001. ELIGIBILITY FOR REGISTRATION. “(a) To be eligible for regis-tration as a voter in this state, a person must: (1) be 18 years of age or older; (2) be a UnitedStates citizen; (3) not have been determined by a final judgment of a court exercising probatejurisdiction to be: (A) totally mentally incapacitated; or (B) partially mentally incapacitatedwithout the right to vote; (4) not have been finally convicted of a felony or, if so convicted,must have: (A) fully discharged the person’s sentence, including any term of incarceration,parole, or supervision, or completed a period of probation ordered by any court; or (B) beenpardoned or otherwise released from the resulting disability to vote; and (5) be a resident of thecounty in which application for registration is made. (b) To be eligible to apply for registration,a person must, on the date the registration application is submitted to the registrar, be at least17 years and 10 months of age and satisfy the requirements of Subsection (a) except for age.”136Table A.3: Voter Registration Information Across Seven U.S. StatesPotentially Ambiguous Registration Information:Florida: “To register in Florida, you must be a U.S. citizen, a Florida resident, at least 18 years old(you may pre-register at 16 or 17, but you cannot vote until you are 18).”Texas: “To be eligible for registration as a voter in this state, a person must be 18 years of age* orolder; be a United States citizen; not have been determined by a final judgment of a court exercisingprobate jurisdiction to be [...]” (Harris County)Oklahoma: “Who can register – You can register to vote if you are both a citizen of the UnitedStates and a resident of the State of Oklahoma, and 18 years old or older.”Idaho: “To register to vote in Idaho you must:be a U.S. Citizen,have resided in Idaho and in the county for thirty (30) days prior to the day of election, andbe at least 18 years old.Clear Registration Information:New York: “To register you must [...] be 18 years old by the end of this year; [...]”Ohio: “You are qualified to register to vote in Ohio if [...] you will be at least 18 years old onor before the day of the general election [...]”Pennsylvania: “I declare that [...] I will be at least 18 years old on the day of the next election. [...]”Sources: Voter registration forms, websites of Departments of State or of County Commissioners for Election.Emphasis added.* In a footnote on the Harris County commissioner of elections’ website and in apparent contradiction to theabove statement, it is stated that people who are 17 years and 10 months old are eligible to register in this county.A.9 Voter Registration Form from FloridaThe election code and voter registration forms in Florida state that citizens have to be 18 yearsold in order to register in this state (see Figure A.7). While the election code and registrationforms mention the possibility to preregister, they fail to explain how preregistration works orwhat precise steps citizens have to take to preregister. In personal communication with theSecretary of State’s office and several election commissioner of election offices in Florida onDecember 12, 2018, they confirmed to me that to preregister in Florida, a person simply needsto submit their voter registration form before turning 18. No other steps are required and nospecific forms for voter preregistration exist.137Figure A.7: Voter Registration Form from Florida Florida Voter Registration Application   Part 1 – Instructions (DS-DE 39, R1S-2.040, F.A.C.)(eff. 10/2013)      Información en español: Sirvase llamar a la oficina del supervisor de elecciones de su condado si le interesa obtener este formulario en español.To Register in Florida, you must be:   • a U.S. citizen,  • a Florida resident, • at least 18 years old (you may pre-register at 16 or 17, but cannot vote until you are 18).  If you have been convicted of a felony, or if a court has found you to be mentally incapacitated as to your right to vote, you cannot register until your right to vote is restored.   If you do not meet any ONE of these requirements, you are not eligible to register.  Questions? Contact the Supervisor of Elections in your county:dos.myflorida.com/elections/contacts/supervisor-of-elections Visit the Florida Division of Elections’ website at: dos.myflorida.com/elections  Where to Register:  You can register to vote by completing this application and delivering it in person or by mail to any supervisor of elections’ office, office that issues driver’s licenses, or voter registration agency (public assistance office, center for independent living, office serving persons with disabilities, public library, or armed forces recruitment office) or the Division of Elections. Mailing addresses are on page 2 of this form.  Deadline to Register: The deadline to register to vote is 29 days before any election. You can update your registration record at any time, but for a Primary Election, party changes must be completed 29 days before that election. You will be contacted if your new application is incomplete, denied or a duplicate of an existing registration. Your Voter Information Card will be mailed to you once you are registered. Identification (ID) Requirements: New applicants must provide a current and valid Florida driver’s license number (FL DL#) or Florida identification card number (FL ID#). If you do not have a FL DL# or FL ID#, then you must provide the last four digits of your Social Security number (SSN). If you do not have any of these numbers, check “None.” If you leave the field and box blank, your new registration may be denied. See section 97.053(6), Fla.Stat. Special ID requirements: If you are registering by mail, have never voted in Florida, and have never been issued one of the ID numbers above, include one of the following with your application, or at a later time before you vote:  1)  A copy of an ID that shows your name and photo (acceptable IDs--U.S. Passport,  debit or credit card, military ID, student ID, retirement center ID, neighborhood association ID, or public assistance ID); or 2) A copy of an ID that shows your name and current residence address (acceptable documents--utility bill, bank statement, government check, paycheck, or other government document). The special ID is not required if you are 65 or older, have a temporary or permanent physical disability, are a member of the active uniformed services or merchant marine who is absent from the county for active duty, or a spouse or dependent thereof, or are currently living outside the U.S. but otherwise eligible to vote in Florida.  Political Party Affiliation:  Florida is a closed primary election state. In primary elections, registered voters can only vote for their registered party’s candidates in a partisan race on the ballot.  In a primary election, all registered voters, regardless of party affiliation, can vote on any issue, nonpartisan race, and race where a candidate faces no opposition in the General Election. If you do not indicate your party affiliation, you will be registered with no party affiliation. For a list of political parties, visit the Division of Elections’ website at: dos.myflorida.com/electionsRace/Ethnicity: It is optional to list your race or ethnicity.  Boxes: Please check boxes () where applicable.  CRIMINAL OFFENSE: It is a 3rd degree felony to submit false information. Maximum penalties are $5,000 and/or 5years in prison.    PUBLIC RECORD: Once filed, all information including your phone number and email address as provided become public record except for the following which can only be used for voter registration purposes: your FL DL#, FL ID#, SSN, where you registered to vote, and whether you declined to register or to update your voter registration record at a voter registration agency. Your signature can be viewed but not copied. (Section 97.0585, Fla. Stat.)  Numbered rows 1 through 7 and 12 must be completed for a new registration.                                     Florida Voter Registration Application Part 2 – Form  (DS-DE #39, R1S-2.040, F.A.C.)(eff. 10/2013)The downloadable/printable online form is available at: dos.myflorida.com/elections/for-voters/voter-registration This is:  New Registration     Record Update/Change (e.g., Address, Party Affiliation, Name, Signature)    Request to Replace Voter Information Card  1  Are you a citizen of the United States of America?             YES              NO OFFICIAL USE ONLY             FVRS No:  2  I affirm that I am not a convicted felon, or if I am, my right to vote has been restored. 3  I affirm that I have not been adjudicated mentally incapacitated with respect to voting or, if I have, my right to vote has been restored. 4 Date of Birth            (MM-DD-YYYY)   -   -     5 Florida Driver License (FL DL) or Florida identification (FL ID) Card Number If no FL DL or FLID, then provide             Last 4 digits of Social Security Number  I have NONE of these numbers.        -    -   -    -        6 Last Name   First Name Middle Name Name Suffix   (Jr., Sr., I, II, etc.): 7 Address Where You Live (legal residence-no P.O. Box) Apt/Lot/Unit City County Zip Code 8 Mailing Address (if different from above address) Apt/Lot/Unit City State or Country Zip Code 9 Address Where You Were Last Registered to Vote Apt/Lot/Unit City State Zip Code 10 Former Name (if name is changed) Gender     M    F State or Country of Birth Telephone No. (optional) (           ) 11   Email me SAMPLE BALLOTS if option is available in my county.  (See Public Record Notice above)   My email address is:  Party Affiliation  (Check only one. If left blank, you will be registered without party affiliation)   Florida Democratic Party      Republican Party of Florida          No party affiliation  Minor party (print party name):  ______________________  12Race/Ethnicity (Check only one)  American Indian/Alaskan Native       Asian/Pacific Islander  Black, not of Hispanic Origin  Hispanic                          White, not of Hispanic Origin   Multi-racial   Other:________________   (Check only one if applicable) I am an active duty Uniformed Services or Merchant Marine member   I am a spouse or a dependent of an active duty uniformed services or merchant marine member   I am a U.S. citizen residing outside the U.S.   I will need assistance with voting.                          I am interested in becoming a poll worker. Oath: I do solemnly swear (or affirm) that I will protect and defend the Constitution of the United States and the Constitution of the State of Florida, that I am qualified to register as an elector under theConstitution and laws of the State of Florida, and that all information provided in this application is true.   SIGN/ MARK HERE Date  138Appendix BSupporting Materials to Chapter 3B.1 Within-State Analysis of Treatment Effects139Table B.1: Effects of Compulsory Voting Using Within-State Variation in Treatment Intensity- Non-Labor Candidates∆ Vote Share of Non-Labor Coalition PartiesNSW QLD SA TAS VIC WA All States1927- 1912- 1941- 1928- 1924- 1936-1930 1915 1944 1931 1927 1939(1) (2) (3) (4) (5) (6) (7)∆ Turnout -.405 .460 .120 -.149 -.509 .626 .195(.357) (.365) (.251) (.372) (.616) (.484) (.136)N 77 61 24 5 25 29 221R2 .020 .058 .007 .004 .026 .060 .017Note: Table presents difference-in-differences estimates of treatment intensity (change in turnout as a result ofcompulsory voting) on change in ALP vote share. Robust standard errors in parentheses. Commonw.:Commonwealth; Elect.: Elections.Table B.2: Effects of Compulsory Voting Using Within-State Variation in Treatment Intensity- Independent Candidates∆ Vote Share of Independent CandidatesNSW QLD SA TAS VIC WA All States1927- 1912- 1941- 1928- 1924- 1936-1930 1915 1944 1931 1927 1939(1) (2) (3) (4) (5) (6) (7)∆ Turnout -.144 -.298 .070 2.637 .219 -.113 -.098(.470) (.534) (.525) (.471) (.535) (.511) (.133)N 77 61 24 5 25 29 221R2 .002 .028 .001 .750 .007 .002 .003Note: Table presents difference-in-differences estimates of treatment intensity (change in turnout as a result ofcompulsory voting) on change in ALP vote share. Robust standard errors in parentheses. Commonw.:Commonwealth; Elect.: Elections.140B.2 Within-State Analysis of Changes in Turnout and LaborSeat ShareFigure B.1: Change in Turnout and ALP Vote Share in Queensland, 1912-1915Paddington2BowenIthacaBurnadaKennedyBundabergEnoggeraMerthyrWindsorQueentonFortitude ValleyTownsvilleToombulBarcooMundingburraOxleyCharters TowersBul mbaKurilpaCookMareeCarnarvonNundahDalbyBrisbaneSouth BrisbaneChillagoeEachamRosewoodToowongWide BayBurkeMackayBurnettFitzroyMurillaDraytonMusgraveLoganKeppelPort CurtisBremerEast ToowoombaCoorooraGympieMaranoaMaryborough2HerbertMt MorganTo woombaIpswichPittsworthAubignyNormanbyNanangoBurrumRockhamptonMurrumbaMiraniWarwickCairns-20-1001020% Change in ALP Vote Share0 5 10 15 20 25% Change in Turnout141Figure B.2: Change in Turnout and ALP Vote Share in South Australia, 1941-1944Stanley2LightGawlerMount GambierGougerMurray2OnkaparingaAngasAlbert2RidleyRocky RiverBurraVictoriaStirlingFlinders2Torrens MitchamUnleyProspectEyreChaffeyBurnsideGoodwoodNorwoodAdelaide-200204060% Change in ALP Vote Share20 25 30 35 40 45% Change in Turnout142Figure B.3: Change in Turnout and ALP Vote Share in Victoria, 1924-1927UPPER GOULBURNSTAWELL AND ARARATGEELONGDUNDASMELBOURNEsLOWANGIPPSLAND NORTHGUNBOWERWARRNAMBOOLPRAHRANGIPPSLAND WESTALBERT PARKBARWONESSENDONMORNINGTONHAWTHORNRODNEYHAMPDENWALHALLABENALLASWAN HILLDANDENONGEVELYNTOORAKBOROONDARA-40-2002040% Change in ALP Vote Share10 20 30 40 50% Change in Turnout143Figure B.4: Change in Turnout and ALP Vote Share in Western Australia, 1936-1939PilbaraNorthamSussexGeraldtonAlbanyNelsonSubiacoYorkPerth NorthPerthMaylandsKatanningFremantle North-EastPerth WestMt HawthornVictoria ParkCanningPerth EastBunburyNedlandsIrwin-MooreAv nMiddle SwanLee rvilleMt MarshallYilgarn-CoolgradieFremantleKalgoorlieFremantle South-40-2002040% Change in ALP Vote Share0 10 20 30 40% Change in Turnout144Appendix CSupporting Materials to Chapter 4C.1 Question Wording of Treatments (English Translation)A. [Control:] Which party would you vote for if there was a federal election this Sunday?CDU/CSU, SPD, Gruene, FDP, Die Linke, AfD, other party, don’t know/no answer (DK/NA)B. [Treatment 1 (economic voting):] The German economy is doing very well, with unem-ployment at a historic low. The upcoming election will shape Germany’s economic prospectsfor the next years. Which party would you vote for if there was a federal election this Sunday?CDU/CSU, SPD, Gruene, FDP, Die Linke, AfD, other party, DK/NAC. [Treatment 2 (strategic voting):] The CDU/CSU is very likely to win the federal electionand form the new government. Whoever wants political change has only one option: to makethe FDP or the Greens as strong as possible in order to avoid another four years of a GrandCoalition. Which party would you vote for if there was a federal election this Sunday?CDU/CSU, SPD, Gruene, FDP, Die Linke, AfD, other party, DK/NAD. [Treatment 3 (uncertainty/leadership characteristics):] These are uncertain times and the145world order is unstable. In times like these many think it is important for Germany to have anexperienced leader. Which party would you vote for if there was a federal election this Sunday?CDU/CSU, SPD, Gruene, FDP, Die Linke, AfD, other party, DK/NAE. [Treatment 4 (social inequality):] In the upcoming election the SPD wants to make Ger-many fairer, provide more support to people who are struggling, and have the rich pay more intaxes. The CDU/CSU and the FDP oppose those changes. Thus, the outcome of the election islikely to have a direct impact on the extent of social inequality in Germany. Which party wouldyou vote for if there was a federal election this Sunday?CDU/CSU, SPD, Gruene, FDP, Die Linke, AfD, other party, DK/NAC.2 Question Wording of Treatments in Survey (German Orig-inal)A. [Control:] Welche Partei wuerden Sie waehlen, wenn am kommenden Sonntag Bundestagswahlwaere?CDU/CSU, SPD, Gruene, FDP, Die Linke, AfD, Sonstige, weiss nicht/keine AngabeB. [Treatment 1 (economic voting):] Der deutschen Wirtschaft geht es momentan sehr gut unddie Arbeitslosigkeit ist auf einem historischen Tiefstand. Der Ausgang der Bundestagswahlwird Deutschlands wirtschaftliche Zukunft fuer die naechsten Jahre entscheidend mitpraegen.Welche Partei wuerden Sie waehlen, wenn am kommenden Sonntag Bundestagswahl waere?CDU/CSU, SPD, Gruene, FDP, Die Linke, AfD, Sonstige, weiss nicht/keine AngabeC. [Treatment 2 (strategic voting):] Die CDU/CSU wird sehr wahrscheinlich als staerkstePartei aus der Bundestagswahl hervorgehen und die neue Regierung bilden. Wer einen poli-tischen Wechsel moechte, hat im Prinzip nur eine Option: die FDP bzw. die Gruenen so stark146wie moeglich zu machen, um weitere vier Jahre einer Groen Koalition zu verhindern. WelchePartei wuerden Sie waehlen, wenn am kommenden Sonntag Bundestagswahl waere?CDU/CSU, SPD, Gruene, FDP, Die Linke, AfD, Sonstige, weiss nicht/keine AngabeD. [Treatment 3 (uncertainty/leadership characteristics):] Wir leben in unsicheren Zeiten unddie Weltordnung ist instabil. In Zeiten wie diesen ist vielen wichtig, dass Deutschland voneiner Person mit weltpolitischer Erfahrung regiert wird. Welche Partei wuerden Sie waehlen,wenn am kommenden Sonntag Bundestagswahl waere?CDU/CSU, SPD, Gruene, FDP, Die Linke, AfD, Sonstige, weiss nicht/keine AngabeE. [Treatment 4 (social inequality):] Die SPD verspricht, Deutschland gerechter zu machen,Beduerftige staerker zu unterstuetzen und die Reichen staerker zur Kasse zu bitten. Die CDU/CSUund die FDP lehnen diese Aenderungen ab. Der Ausgang der Bundestagswahl wird also di-rekten Einfluss auf das Ausmass sozialer Ungleichheit in Deutschland haben. Welche Parteiwuerden Sie waehlen, wenn am kommenden Sonntag Bundestagswahl waere?CDU/CSU, SPD, Gruene, FDP, Die Linke, AfD, Sonstige, weiss nicht/keine Angabe147C.3 Summary StatisticsObs Mean St. Dev. Min Maxfemale 1786 0.52 0.50 0 1East 1786 0.24 0.42 0 1educRS 1786 0.46 0.50 0 1educGY 1786 0.33 0.47 0 1income1500 2499 1786 0.25 0.44 0 1income2500 3999 1786 0.23 0.42 0 1income4000pl 1786 0.10 0.30 0 1incomemiss 1786 0.16 0.37 0 1unionmbr 1786 0.12 0.32 0 1age18 29 1786 0.09 0.29 0 1age30 44 1786 0.22 0.41 0 1age45 59 1786 0.39 0.49 0 1age60pl 1786 0.30 0.46 0 1mit imm 1786 0.57 0.49 0 1mit terr 1786 0.13 0.33 0 1mit crime 1786 0.05 0.22 0 1pid afd 1786 0.21 0.41 0 1pid cdu 1786 0.18 0.38 0 1pid spd 1786 0.14 0.35 0 1pid fdpgr 1786 0.07 0.25 0 1pid no 1786 0.30 0.46 0 1148C.4 Results with Alternative Dependent Variable: Party Fa-vorabilityThe following table reports results with the alternative dependent variable that records respon-dents’ favorability ratings towards each of the different parties. Favorability is measured on an11 point scale, with 0 meaning that a respondent does not like a party at all and 10 that shelikes it a lot. In contrast to the vote intention question, the party favorability was not asked atthe same time (i.e. on the same screen) as the treatment message that respondents read, buttwo questions later in the online questionnaire. It was also not directly about the upcomingelection, but asked respondents to rate their party favorability in more general terms. Based onthe results reported in Table C.1, none of the treatment coefficients is statistically significant atconventional levels, possibly suggesting that the treatments were not strong enough to producesignificant shifts in respondents’ general party favorability.C.5 Results with Continuous Political Knowledge Variable149Table C.1: Party Favorability Rating in 2017 German Federal Election(1) (2) (3) (4)AfD CDU/CSU SPD FDP/GreensEconomic Voting -0.014 -0.098 -0.339 -0.230(0.252) (0.247) (0.231) (0.187)Strategic Voting 0.173 -0.038 -0.241 -0.004(0.258) (0.244) (0.232) (0.188)Uncertainty -0.194 0.002 -0.161 -0.109(0.254) (0.242) (0.232) (0.187)Social Inequality -0.153 -0.265 -0.123 -0.311(0.250) (0.241) (0.229) (0.187)Constant 4.567 4.012 5.678 5.365(0.438) (0.421) (0.413) (0.324)Controls Yes Yes Yes YesObservations 1673 1687 1679 1669R2 0.42 0.33 0.34 0.18Note: Dependent variable is party favorability in 2017 German federal election for the party indicated at the topof each column. OLS regression with control variables: indicators for region (East vs. West), age group, gender,income group, education level, party identification, and most important topic (immigration, crime, terrorism).Models control for 122 respondents who are less anti-immigrant (indicator and interaction with treatments).Coefficients of primary interest for hypothesis tests are in bold.Figure C.1: Effect of Economic Voting Treatment on AfD Support. of AfD Vote0 1 2 3 4 5Political KnowledgeEconTr=0EconTr=1Predictive Margins with 95% CIs150Table C.2: Models for 2017 Vote Intention with Continuous Political Knowledge Variable(5) (6) (7) (8)AfD CDU SPD FDP/GreensEconomic Voting -0.085 0.0025 0.046 -0.044(0.052) (0.039) (0.044) (0.028)Strategic Voting -0.033 0.084 -0.042 -0.019(0.050) (0.040) (0.044) (0.037)Uncertainty -0.040 0.067 0.017 -0.047(0.054) (0.041) (0.044) (0.029)Social Inequality -0.10 -0.0044 0.061 -0.0037(0.047) (0.037) (0.043) (0.036)KnowledgeContinuous 0.018 -0.0021 0.011 0.0046(0.011) (0.008) (0.008) (0.007)EconTr*KnowContinuous 0.027 0.020 -0.021 -0.00053(0.016) (0.012) (0.012) (0.009)StratTr*KnowContinuous 0.013 -0.015 0.00030 0.0043(0.016) (0.012) (0.013) (0.012)UncertTr*KnowContinuous 0.014 -0.0095 -0.011 0.0034(0.015) (0.011) (0.012) (0.009)SocInTr*KnowContinuous 0.035 0.0033 -0.019 -0.0022(0.015) (0.011) (0.012) (0.011)Constant 0.16 0.020 0.037 0.081(0.051) (0.041) (0.047) (0.038)Controls Yes Yes Yes YesObservations 1786 1786 1786 1786R2 0.47 0.41 0.42 0.27Note: Dependent variable is vote intention in 2017 German federal election for the party indicated at the top ofeach column. OLS regression with control variables: indicators for region (East vs. West), age group, gender,income group, education level, party identification, and most important topic (immigration, crime, terrorism).Models control for 122 of the 1,786 respondents who are less anti-immigrant (indicator and interaction withtreatments).151Figure C.2: Effect of Social Inequality Treatment on AfD Support. of AfD Vote0 1 2 3 4 5Political KnowledgeInequalTr=0InequalTr=1Predictive Margins with 95% CIs152


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



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"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items