Activating the Right Electorate: Internet Use, Nativism, and Voting for the AfD

Abstract

The internet is regularly blamed to play a major role in the surge of populist radical right parties such as the Alternative für Deutschland (AfD). While previous studies have already established a generally positive association between internet use and a higher propensity to vote for the AfD, we know little about the conditionalities of this effect. Based on theories about cognitive dissonance, I argue that the high choice environment of the internet offers populist radical right actors a tool to efficiently activate voters with a nativist political predisposition. Using data from the German Longitudinal Election Study in 2016/17, I find that using the internet for political information is associated with a substantively higher propensity to vote for the AfD only among individuals who already had a political predisposition that was consistend with the party’s political agenda. The results highlight the important role of widespread far-right worldviews in the electorate for the rise of a populist radical right party in Germany. The internet mainly worked as an amplifier as it helped the AfD to activate its potential electorate.

The author is indebted to Rüdiger Schmitt-Beck, Simon Ellerbrock, Christiane Grill, Anne Schäfer, Harald Schoen, Klara Raiber, Franziska Quoss, and the participants of the 2020 AK Wahlen conference for their helpful feedback on a draft version of this article. The editor and reviewers of this volume provided constructive criticism that greatly improved this article.

Appendix A. Model Fits

In the following, I test whether the inclusion of the assumed interaction effect fits better as a potential data generating process than a model without it. The estimates for the latter are displayed in Model 1 in Table 1. Interestingly enough, predictions from the simpler model already depict an interactive relationship between the two independent variables when one predicts probabilities. A second difference between the two poles of the outgroup index and the two internet usage scenarios predicts a difference of 31% points (CI_2,5% = 24.5%; CI_97,5% = 39.0%) in the propensity to vote for the AfD that is solely due to the change in internet usage. This makes it another case where the interactive relationship is not visible by the mere reporting of the estimates and their p-values, but the compression of the logit model already accounts for (see Berry et al. 2010). Including an interaction term to control for the possibility of a false positive as Rainey (2016) suggests, only made the effect more pronounced. A comparison of goodness-of-fit indices further establishes that the interaction model fits the data better than the baseline model. A comparison of the respective model fits shows that the second model both in terms of the log-likelihood and the AIC (Log—Likelihood₁ > Log—Likelihood₂; AIC₁ < AIC₂) outperforms the baseline. Computing the AIC-weights, it can be inferred that the interaction model is almost 52 times more likely to be the better fitting one of the two (see Wagenmakers and Farrell 2004). An ANOVA test (χ²) with the two models further supports the finding that the interaction model is better in explaining the variance in the data (p < 0.002).