The effect of European Parliament elections on political socialisation

ABSTRACT

Do European Parliament (EP) elections have adverse effects on the political socialisation of individuals? Instilling a lasting political disaffection in impressionable voters, the supranational contest may have negative consequences for long-term political socialisation. Relying on a large cross-national dataset from 2004, I identify the causal effect of first-time eligibility and voting in the EP elections by exploiting the exogenous variation in adolescents' birth months. The results of a discontinuity design show that the elections do not politically disengage young voters or strengthen their party bonds to radical or Eurosceptic parties. Instead, the EP elections arouse their political interest in general and their European interest in particular; a long-lasting effect that persists for more than five years. Placebo tests and various robustness tests confirm the results. Considering that individuals are most receptive towards political socialisation stimuli during early adulthood, this study sheds light on the integrative potential of the EP elections.

KEYWORDS:

• European integration
• European Parliament elections
• political interest
• political socialisation
• quasi-experiment
• voting behaviour

Acknowledgements

For comments on previous versions of the paper, I would like to thank Elias Dinas, Hanspeter Kriesi and participants of the Political Behaviour Colloquium at the European University Institute. I would further like to thank the editors of the special issue and three anonymous referees for their constructive feedback. All remaining errors are my own.

Disclosure statement

No potential conflict of interest was reported by the author.

Note on contributor

Julia Schulte-Cloos is a PhD candidate in the Department of Political and Social Sciences at the European University Institute.

ORCID

Julia Schulte-Cloos https://orcid.org/0000-0001-7223-3602

Notes

1 Other studies concerned with attitudinal change regarding partisan preferences due to the act of voting have somewhat circumvented this problem by comparing previously ineligible and eligible individuals ahead of the next election (Elinder 2012). This design substantively uncouples measurement from treatment as a full electoral cycle lies in between both respective points in time.

2 I rely on the potential outcomes framework notation (Rubin 1974) to describe the causal parameters of interest, invoking the stable unit treatment value (SUTVA) assumption that an individuals' treatment condition is not affected by the treatment condition of others units (no interference). This effect is given by the intent-to-treat (ITT) estimator: ${\tau }_{ITT}=E\left[Y_i\right|Z_i=1]-E[Y_i|Z_i=0]$, with $Y_i$ the potential outcome we would observe for unit i and $Z_i$ measuring the treatment of first-eligibility in the EP elections.

3 Non-compliance with treatment assignment is one-sided as ineligible individuals in the control group are prohibited from receiving the treatment (participation in the EP election) through the legal eligibility at the age of 18. Thus, always-takers and defiers do not exist, or formally speaking $D_i\left(0\right)=0$ for all i. I then estimate the complier average treatment effect (CACE) using instrumental variable regression: $\widehat{CACE}=\frac{\hat{E}(Y_i\mid Z_i=1)-\hat{E}(Y_i\mid Z_i=0)}{\hat{E}(D_i\mid Z_i=1)}$.

4 Formally speaking, only the assignment to treatment is random and determined by a binary indicator $Z_i$, measuring whether the respondent had reached full age at the day of the EP election while $D_i\left(Z\right)$ is a binary indicator for the compliance with the treatment under assignment, or actual voting in the European contest. We can estimate the effect of voting by instrumenting the endogenous electoral participation of first-time voters. The ITT estimates (see Table 1 column 1 and 2) have a slightly larger number of observations due to missing values on the endogenous variable of voting in the EP elections. The ITT estimates correspond to the first stage of the IV regression and do not change substantively when considering the slightly smaller sample of the CACE estimates. The instrument is not a weak-instrument (p<0.01), i.e., the instrument has a sufficient correlation with the endogenous explanatory variable.

5 The EUYOUPART dataset also includes Slovakia and Austria, which are, however, excluded from the analysis in this article as both countries held another national election, namely a presidential election, during the months leading up to the EP elections, leaving only 10 (Slovakia) and 6 (Austria) treated first-time voters in the sample.

6 Formally, their eligibility is expected to be exogenous to the relevant outcome variables, $\left(Y_i\right(1);Y_i(0\left)\right)\perp D_i$, leaving us confident that the variation in treatment and control around the cut-off quasi-randomly splits the sample (Lee and Lemieux 2010: 283).

7 Table B8 in the supplemental material shows that the results are robust to genetic optimal matching, maximising the balance between treatment and control group.

8 All standard errors and confidence intervals for the different point estimates of the treatment are constructed using the Bell-McCaffrey (BM) bias adjusted robust variance estimator as recommended by Imbens and Kolesar (2016) taking the small numbers of country-clusters into account. The adjustment combines a bias-reduced form of the cluster robust standard errors with a Satterthwaite approximation for the degrees of freedom of the t-distribution, see Imbens and Kolesar (2016) for a detailed discussion.

9 The small number of individuals in treatment and control group within each country does not allow for exploring country-specific differences of the effect size.

10 At the same time, the smaller sample size makes it more likely that imbalance between treatment and control increases by chance. Therefore, the main analysis relies on a larger bandwidth.