Social influence and political mobilization: Further evidence from a randomized experiment in the 2012 U.S. presidential election
From left to right estimates are based on a regression with heteroskedasticity-robust sandwich standard errors of validated vote on 1) an indicator variable for those in both the banner and feed condition, 2) an indicator variable for being in the banner and feed conditions and a control variable indicating whether the subject voted in 2010, 3) an indicator variable for being in the banner and feed conditions, stratified by voter behavior in 2010, 4 & 5) two indicator variables, one for being in the banner condition and one for being in the feed condition, stratified by voter behavior in 2010. Regressions in 1)– 3) exclude individuals who received only one of the two treatments so the comparison is both vs. neither. These results suggest the banner and feed condition combined to yield a 0.24% increase in voter turnout, and that it likely depended on both mechanisms to generate this increase. Asterisks indicate p<0.05.