Artificial Extremism in Partisan Polling

• In Event: Partisan Extremism in the US

Abstract

As response rates to public opinion polls have plummeted, scholars and pollsters have worried that non-response bias skews poll results away from true opinions of the American public. Using randomized survey treatments designed to identify non-ignorable response based on Bailey (2018), this paper explores non-response in partisan samples. The general finding is that the kind of people who are eager to discuss politics very different than people less interested in talking about politics as they are much more likely to vote and to have more extreme views. At the aggregate level, this causes little bias for many political questions because the subsample of extreme left respondents is off-set by the subsample of extreme right respondents. For partisan subsamples, however, the biases are striking: Democrats who want to discuss politics are vastly more liberal in their views than Democrats who do not want to discuss politics while Republicans who want to discuss politics are vastly more conservative in their views than Republicans who do not want to discuss politics. The partisan gap reflected in polls may therefore be much larger than in reality, due to this non-response bias. The implication is that public opinion polls exaggerate partisan extremism in a way that is inaccurate and particularly unhelpful in the current political environment.

Author

• Michael A. Bailey, Georgetown University