🇪🇺🇩🇪 Published Today in @bjpols.bsky.social 🇮🇹🇫🇷 How a voting advice application affected voting behavior in three large-scale field experiments: shorturl.at/2ekBj TLDR of our study (with @simonhix.bsky.social & @rlachat.bsky.social) below 👇 1/14

Voting advice applications (VAAs) are now common tools ahead of elections in many democracies, in particular in multi-party settings. For example, in Germany, about a third of the electorate (>20 Mio. Voters) consult the Wahl-o-Mat VAA ahead of federal elections. 2/14

How do these tools affect voting behavior? Their intended and declared purpose is to help voters make better informed vote choices, i.e., to help them vote for their ideologically most aligned party. But do they have further downstream effects (positive or negative) beyond this purpose? 3/14

Given inconsistencies in findings between experimental and observational studies on the one hand, and several shortcomings in past VAA studies (most prominently: insufficient statistical power in experiments & unadjusted confounding in observational studies) on the other hand… 5/14

… we set out to re-test these hypotheses with an improved design in Germany, France & Italy: a pre-post panel study during the 2024 European elections, where half of our 6000+ participants were randomly encouraged to fill out a VAA ahead of the election. 6/14

We worked with EuroMPmatch, a multilingual VAA which presents respondents with 20 policy proposals that the EP voted on in the past, documents their opinions on these policies, and then reports (in percentages) their agreement with each relevant party across all policy areas. 7/14

Shortly after the election, we surveyed all participants again, measuring their voting behavior and knowledge. We did not find any evidence that VAAs increased voter turnout, the amount of voteswitching, or political knowledge (neither intent-to-treat, nor complier average causal effects). 8/14

Lastly, we were able to obtain the full records from EuroMPmatch, allowing us to merge our survey data with the VAA data itself. This allowed us to compare VAA recommendations with actual voting behavior. 9/14

Using this data and a regression discontinuity design with proportional agreements between users and parties as the running variable, we estimated the local average treatment effect of becoming the top-recommended party on receiving a given users vote. 10/14

We show that the top-recommendation spot increases the probability of receiving a vote by 2-7pp depending on the specification (beyond the inherent endogenous correlation between policy agreement and voting propensity). 11/14

This suggests that VAAs, while not increasing the quantity of voteswitching, do increase the quality of voteswitching: while many non-users switch their votes ahead of elections too, VAAs make it more likely for users to switch precisely to the party they are ideologically most aligned with. 12/14

At the same time, we fortunately did not find any evidence that VAAs are ideologically manipulative. In the aggregate, they did not push voters into one ideological direction or to any particular parties (except for two leftwing French parties trading votes within one camp). 13/14

In sum, our results suggest that VAAs are effective precisely (and only) for their intended purpose: guiding voters to make a more informed vote choice for the party they are ideologically most aligned with. 14/14