Age, Experience and the Contextual Determinants of Turnout: A Deeper Look at the Process of Habit Formation in Electoral Participation

Abstract

I test two potential mechanisms driving habitual voting: direct self-reinforcement of turnout and social ageing. I do so by critically examining a hypothesis proposed by Franklin who argues that the impact of electoral context on turnout declines as citizens experience more and more elections. I thus attempt to test whether the effects of context on turnout are indeed mediated by what is called “electoral experience” or simply by age. As, in most cases, those two are nearly perfectly correlated, I utilize survey data from Switzerland where women were enfranchised late and thus many female respondents did not experience any elections until reaching relatively old age. My test suggests that it is not experience that mediates the impact of context on turnout. I find weak but suggestive ageing effects. Yet, those effects are surprisingly non-monotonic. The impact of context on turnout is relatively strong for the youngest eligible citizens to decline for those who have entered their 30s. It sharply increases, however, when people are, roughly, in their mid-40s to decline again late in life.

Acknowledgements

I am grateful to Peter Selb for kindly sharing his data set on electoral competition in Switzerland, and to the Swiss Centre of Expertise in the Social Sciences (FORS) for granting their permission to use the Swiss Electoral Studies data for the purpose of this research. This research would also not have been possible without an inspiring discussion I had years ago with Cees van der Eijk. Last but not least, I thank Michael Marsh and two anonymous reviewers from this journal for their comments on an earlier draft of this article.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1. It is worthwhile emphasizing at the outset that the term “habitual voting” might be potentially misleading. In psychology, the term “habit” is reserved for behaviours that are characterized by repetition due to “automaticity” (Moors and De Houwer 2006). This sort of behaviours is most usually performed on a daily basis, if not more frequently, and thus tend to be difficult to give up. Elections, by contrast, are relatively rare events. Therefore, the term “habitual voting” should be interpreted as a metaphor aimed at depicting the developmental and inertial aspects of electoral participation rather than as a result of an attempt to classify voting as just another “habit” in the sense known from psychology. While Aldrich, Montgomery, and Wood (2011) have tried to classify voting that way, their concept is hardly representative of what is known to be the core of the theory of “habitual voting”, particularly the ideas developed by Plutzer (2002) and Franklin (2004). (For a more in-depth critique of Aldrich et al.’s proposition, see Górecki 2013b, 150).

2. These effects are roughly similar to those obtained by Franklin and Hobolt (2011) who also relied on the regression discontinuity approach. Another study of this sort (Dinas 2012) finds a slightly stronger habit effect, comparable to that found by Denny and Doyle (2009). Meredith's result seems to be most credible though. This is because the latter author, unlike the others, used data taken from actual voting records rather than survey data prone to various types of bias (resulting most notably from turnout over-reporting and recall effects).

3. For this reason, the presence or absence of ceiling (floor) effects seems to be an empirical issue rather than one that can be solved unequivocally without reference to a particular context in which habit formation takes place. I thus leave this issue aside and return to it briefly in the empirical section (see note 13).

4. SELECTS: Swiss National Election Studies, cumulated data file 1971–2011 (data set). Distributed by FORS, Lausanne, 2013, www.selects.ch.

5. Two Swiss electoral surveys, covering the 1979 and the 1983 election, were lost. While an alternative survey is available for 1979, there is no data set covering the 1983 election. In this paper, I thus analyse 10 out of 11 elections held between 1971 and 2011. For further details, see http://www2.unil.ch/selects/spip.php?rubrique13&lang=fr.

6. Grofman and Selb's (2009) data set covers all the National Council elections held between 1971 and 2007. In my study, I add the election held in 2011. Therefore, on the basis of the formula provided by Grofman and Selb, I myself calculated district-level competitiveness figures for the latter election.

7. Another potential control variable that could be included in the model is certainly partisanship. However, party attachment is potentially endogenous to the process of habit formation in electoral participation (Dinas 2014), and thus, it could create a post-treatment bias. I thank an anonymous reviewer from this journal for pointing this out to me. At the same time, I would like to note that additional analyses (not reported here), with partisanship included in the model, yielded very similar results to those presented later in this article.

8. Due to the model being extraordinarily complex, I skip the estimates with respect to regression coefficients and standard errors. They are available from the author on request.

9. The additional experience hypothesis (H2), concerning women only, is not supported either. For reason of space constraints, the respective estimates have not been included in the article but are available from the author on request.

10. While I realize that such an approach results in my conclusions being extremely tentative and vulnerable to criticism, there are at least two reasons to argue that it might be sensible. First, there is a substantial difference in women's electoral experience between the earliest cohort and those born between 1961 and 1975. Second, focusing on just those two cohorts would not at all have changed the conclusions of the above analysis of cohort effects and, as such, it would not have strengthened the evidence for the experience hypothesis.

11. The exact test of H4 would mean a simple reshuffling of the effects demonstrated in Figure 1 and ordering them according to age group rather than according to cohort. Since these would be the exact same (pervasively insignificant) effects, such ordering is not presented in this paper but is available from the author on request.

12. All variables, including index of competition and controls, are held at their medians. The effect for women aged 18–30 is estimated for the 1961–1975 cohort, while the effects for all other age groups are estimated for the pre-1931 cohort. I am on the safe side here as women from the latter cohort have lower propensities to vote than those belonging to the former cohort, other things being equal. Had I been able to extrapolate, the increase in turnout levels between the youngest and the second youngest group would thus have been even greater than demonstrated in Figure 3.

13. At the same time, there seems to be enough room for both increase and decrease in turnout, regardless of age group. Ceiling and floor effects are therefore unlikely to operate here to any significant degree.