Exposing voters to non-state-provisioned information is presumed to counter incumbents’ efforts to keep voters uninformed in order to remain in power. In this study, I estimate the effect of randomized door-to-door information campaigns on voter behavior and ideology in Turkey. My design allows me to estimate heterogeneous effects of information campaigns. I find that voter response to the same campaigns increased political polarization and the effect persisted at least two years. I conclude that reducing censorship can be polarizing and, because average measures mask both positive and negative treatment effects, the impact of information campaigns on civil society is underestimated.
External Link(s)
Citation
Baysan, Ceren. 2020. "Persistent Polarizing Effects of Persuasive Communication: Experimental Evidence from Turkey." AEA RCT Registry. November 30. https://doi.org/10.1257/rct.6827-1.0.
In this study, I analyze the impact of two randomized door-to-door information campaigns that took place before a constitutional referendum in Izmir, Turkey. The information campaigns were organized and created by members of the largest party opposing the referendum and carried out by party volunteers. These information campaigns were going to be carried out regardless of any evaluation, but I designed their implementation as a randomized control trial.
Intervention Start Date
2017-04-06
Intervention End Date
2017-04-16
Primary Outcomes (end points)
Publcily available administrative data on voter behavior: neighborhood-level vote share and voter turnout from the April 2017 referendum, 2018 general election, 2018 presidential election, and 2019 local election.
Primary Outcomes (explanation)
Secondary Outcomes (end points)
Secondary Outcomes (explanation)
Experimental Design
Stratified randomization design. Randomization was stratified by quartiles of past average vote share for the main opposition party (this is equivalent to the vote share differential between the incumbent and the main opposition party). This average vote share was calculated using two general elections that were held in 2015. I submitted a pre-analysis plan to OSF before the intervention started. In this PAP, I specified estimating heterogeneous treatment effects by quartile and the covariates that would be used in all regressions.
Experimental Design Details
Randomization Method
Randomization done in office by a computer
Randomization Unit
Neighborhood
Was the treatment clustered?
No
Sample size: planned number of clusters
550
Sample size: planned number of observations
550
Sample size (or number of clusters) by treatment arms
48 to one treatment group, 52 to a second treatment group, and 450 to the control group.
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)