BACKGROUND AND EXPECTATIONS

Although partisanship has declined in some advanced democracies (Dalton and Wattenberg Reference Dalton and Wattenberg2002; Dassonneville and Hooghe Reference Dassonneville and Hooghe2018; Schmitt and Holmberg Reference Schmitt and Holmberg1995), it nevertheless remains a powerful force. This is the case even in countries where aggregate partisanship is low (Bankert, Huddy, and Rosema Reference Bankert, Huddy and Rosema2017). And, although partisanship is among the most stable of political attitudes, substantial evidence indicates that it does respond to external context. For example, elite polarization (e.g., Adams, Green, and Milazzo Reference Adams, Green and Milazzo2012; Hetherington Reference Hetherington2001; Lupu Reference Lupu2015; Thornton Reference Thornton2013), the number of parties (e.g., Bowler, Lanoue, and Savoie Reference Bowler, Lanoue and Savoie1994; Holmberg Reference Holmberg, Dalton and Klingemann2007; Huber, Kernell, and Leoni Reference Huber, Kernell and Leoni2005; Karp and Banducci Reference Karp and Banducci2008; Katz Reference Katz1980; Sartori Reference Sartori1976), and institutional features (e.g., Gerring, Thacker, and Moreno Reference Gerring, Thacker and Moreno2009; Schmitt Reference Schmitt, Bartles and Bellucci2009; Singh and Thornton Reference Singh and Thornton2013) have all been shown to influence partisanship or partisan strength.

Our focus is on the effects of elections on patterns of partisanship. From the earliest studies of voting behavior and campaigns (Berelson, Lazarsfeld, and McPhee Reference Berelson, Lazarsfeld and McPhee1954; Lazarsfeld, Berelson, and Gaudet Reference Lazarsfeld, Berelson and Gaudet1944), it has been argued that elections activate existing predispositions, and this argument is consistent with more recent evidence (e.g., Gopoian and Hadjiharalambous Reference Gopoian and Hadjiharalambous1994; Holbrook Reference Holbrook1996; Stimson Reference Stimson2004). As Campbell (Reference Campbell2000) argues, election campaigns “restore” partisanship, and evidence from the US indicates that campaigns make partisan attitudes more pronounced (e.g., Grant, Mockabee, and Monson Reference Grant, Mockabee and Monson2010). Further, US panel data indicate that ambivalence, a dimension of attitude strength (Miller and Peterson Reference Miller and Peterson2004), declines over the course of the election (Rudolph Reference Rudolph2011). Similarly, as election day nears, uncertainty toward parties declines (Peterson Reference Peterson2014), perceptions of the candidates polarize (Miller and Shanks Reference Miller and Shanks1982), and the relationship between partisanship and vote intentions strengthens (Bartels Reference Bartels, Brady and Johnston2006). Evidence also suggests that partisan intensity responds to the presence of an election (Allsop and Weisberg Reference Allsop and Weisberg1988; Brody and Rothenberg Reference Brody and Rothenberg1988; Huckfeldt, Sprague, and Levine Reference Huckfeldt, Sprague and Levine2000). As per Iyengar and Simon (Reference Iyengar and Simon2000, 159), nearly six decades of evidence shows that “the principal impact of the campaign will be to push partisans into their respective corners.”

In line with this literature, as an election—and by extension political parties—becomes less salient, so should party identification. This leads to three testable hypotheses. First, we hypothesize that as an election becomes less salient, one should be less likely to identify with a party. Similarly, as parties become less salient in the minds of the public, partisan strength should decline for many of those who continue to identify with a party. Therefore, our second hypothesis is that the strength of one’s attachment decreases as an election becomes less salient.

As a corollary, we expect that, because partisanship becomes less prominent as elections become less salient, it should have a smaller influence on subsequent political evaluations. Consequently, our third hypothesis is that the link between partisanship and political evaluations will weaken as elections become less salient.

DATA AND METHODS

Our data come from the Comparative Study of Electoral Systems. The CSES questionnaire is embedded in national post-election surveys in dozens of countries, and it asks cross-nationally comparable questions about partisan attachments and evaluations of political actors.

To examine whether elections affect partisanship, researchers benefit from a measure of the degree to which an election is salient in an individual’s mind. Many survey projects ask respondents whether or not they witnessed or participated in a variety of activities associated with an election, such as campaigning for a candidate or voting, and it is possible to create an indicator of election salience from responses to such questions. However, correlations between such self-reported indicators and partisanship could be spurious. For example, they may surface solely because individuals who engage with political campaigns and vote are more educated, and education is itself linked to partisanship. Further, such correlations do not provide evidence of the direction of a relationship; perhaps partisans are more likely to campaign for candidates and to vote.

To overcome such limitations, we leverage the fact that the dates on which CSES respondents are interviewed within each post-election survey are likely unrelated to partisanship and its correlates, and we use the number of days between the day of the election and the date on which a respondent was interviewed to capture election salience. To account for the fact that increases in this delay will matter soon after the election than several months after, we take the natural logarithm. We call the resulting variable time since election. Under the assumption that respondents in each survey are as if randomly treated (Dunning Reference Dunning2008) with an interview date, we can interpret the effects of time since election within a given survey as causal. ^{Footnote 1}

Because we pool the CSES data, it is critical that we account for the fact that survey firms operate under varying constraints and schedules across countries. For example, all interviews in one country may occur within a few weeks following the election, whereas in another country, the interviews might take several months to complete. Perhaps citizens in countries that get surveys into the field relatively quickly tend to be more partisan—a possibility because such countries may also have, on average, longer experiences with electoral democracy. Left unaddressed, this would hamper our ability to identify the effects of time since election. ^{Footnote 2} Thus, we analyze the data in a multilevel framework. In particular, in addition to estimating a random intercept for each CSES election survey in each country, we allow the effect of time since election to vary randomly across surveys. ^{Footnote 3}

We include in our sample all CSES election surveys for which the data needed to measure our key variables are available. These 86 surveys are listed in Appendix A, along with information on the minimum, average, and maximum time from the election to an interview for each survey. We show the predicted random effects for each survey in Appendix B.

PLAUSIBILITY OF THE IDENTIFYING ASSUMPTION

When leveraging naturally occurring treatments—in our case, interview timing—it is important to examine whether the treatment process worked in a manner that precludes the need to control for potential confounders (cf. Sekhon and Titiunik Reference Sekhon and Titiunik2012). In particular, we recognize the possibility that survey agencies could roll out their interviews in a way that makes our assumption of as if random interview date assignment implausible. For example, people in cities may tend to be interviewed sooner than people in rural areas. If city dwellers are also more partisan, on average, this could account for our findings. To examine the plausibility of our assumption of as if random assignment of interview dates within surveys, we check whether several potential confounders predict the time between the election and the date of interview.

As shown in Appendix D, within each survey, we find only trivial relationships between individual characteristics and the date on which one was interviewed. There, we also scrutinize the potential links between the degree of democracy and country-level wealth and interview timing across countries, finding null relationships. Thus, we are confident that the relationships shown in Figures 1–3 are truly due to the time from the election to the interview. To further demonstrate this, we show that our results are robust to models that include a slate of control variables in Appendix D.

Additionally, about 84 percent of the respondents in our sample were interviewed solely by telephone or in person. For the other 16 percent, interviews were either done entirely by mail or conducted by mail in conjunction with a telephone or face-to-face interview. Allowing respondents to mail back their questionnaires on their own time potentially harms our assumption of as if random interview timing—perhaps those with partisan ties tend to mail back questionnaires more quickly than those without. To account for this, we re-estimate our models with those who were not surveyed solely by telephone or solely in person removed from the sample. Results, which are shown in Appendix E, are very similar to those of our main models.

TIME SINCE AN ELECTION AS A MEASURE OF ELECTION SALIENCE

Although the time passed since an election is perhaps an uncontroversial gauge of election salience, it is worth probing the validity of our key causal variable. To do so, we rely on the fact that individuals are more likely to provide ideological placements of themselves and political parties when elections are salient—and therefore the information environment is rich (e.g., Aldrich et al. Reference Aldrich, Schober, Ley and Fernandez2018). If the time since election is inversely related to election salience, it should also be inversely related to the likelihood that survey respondents will provide self-placements and placements of political parties along an ideological scale. We show that this is indeed the case in Appendix F.

SUPPLEMENTARY MATERIAL

To view supplementary material for this article, please visit https://doi.org/10.1017/S0003055418000722.

Replication materials can be found on Dataverse at: https://doi.org/10.7910/DVN/03CDTK.