How Compelling is Compulsory Voting? A Multilevel Analysis of Turnout

Abstract

In a recent contribution to Political Behavior (30:455–467), Panagopoulos, using aggregate turnout data, shows that individuals living under compulsory voting rules are most likely to go to the polls when penalties for abstaining are both strict and routinely enforced. In this project, I expand on the work of Panagopoulos by simultaneously examining both election-level and individual-level factors. I use a broad sample of 36 countries, some with compulsory voting and some with voluntary rules, which provides a more detailed understanding of the correlates of turnout. Results indicate that the presence and severity of compulsory rules do indeed affect turnout, while personally held characteristics, including age, education, income, and political efficacy remain critical to an individual’s turnout decision calculus.

Appendix: Adjustments for Overreporting

Table 1 makes it clear that the percentage of individuals who reported voting generally exceeds officially reported turnout. Due to faulty recollection or social desirability, individuals often incorrectly report having participated in an election (Bernstein et al. 2001; Karp and Brockington 2005). If overreporting is indeed systematic, inferences drawn from individual-level turnout data will be biased.

Survey weights are generally used to correct for oversampling in a population. For example, imagine a population is 50% female and 50% male, but a sample taken by a researcher is 25% female and 75% male. To correct for this, each observation is weighted by its true population proportion divided by its observed sample proportion. In this example, each female in the sample is weighted with a value of 2 (50%/25%) and each male a value of 2/3 (50%/75%).

To correct for discrepancies between official and reported turnout, I use data on actual and reported turnout rates (see Table 1) to apply the same weighting strategy used to correct for oversampling. For example, in the 2005 Albanian election overreporting was very pronounced, with 91.5% of individuals claiming to have voted, while official turnout was only 59.6%. To correct for this disparity, each respondent who reported voting was assigned a weight of 59.6/91.5 ≈ .652, and each individual who reported abstaining was assigned a weight of 40.4/8.5 ≈ 4.75. I applied such weights to each individual in each election in the sample, effectively making reported turnout equivalent to official turnout. I re-estimated Models 3–5 of Table 3, this time accounting for inaccurate reporting with logistic regression models that weight each individual’s contribution to the results appropriately and account for the clustering of the individuals within elections.

The results of each re-estimation are given in Table 4. There is a near uniform increase in the significance levels of the macro-level variables. This is likely because the survey-weighted models do not explicitly model election-effects and thus may overstate the impact of the election-level variables. Nevertheless, it is again shown that compulsory rules, their severity, and their enforcement, boost turnout. It is also shown in Model A2 that enforcement of compulsory rules may affect turnout independent of sanctions; the fear of getting caught staying home may boost turnout propensity independent of the penalties for abstention. Finally, age, education, income, and efficacy again maintain a strong positive effect on turnout across each model.

Table 4 The correlates of individual-level turnout (with survey weights)