On the robustness of social-circle surveys: Respondent selection issues, egocentrism, and homophily

Abstract

Asking people about the preferences of people in their social circles tends to yield more accurate estimates of population preference distributions than does asking each respondent about their own preference. This is likely because the former approach taps into people's knowledge about others and thereby generates an implicit super sample that includes non-sampled members of participants' social circles. The present paper makes two contributions. First, it uses a set of simulation studies to argue that the superiority of social-circle surveys can be expected to be robust in the face of respondent selection issues (e.g., non-response and coverage bias), people being highly fallible about other people's preferences (egocentric bias), and people largely surrounding themselves with those who share their preferences (homophily). Second, it reports on a survey experiment offering preliminary evidence that egocentric bias in particular can be reduced significantly through a simple survey prompt. In closing, the paper also discusses the relationship between social-circle questions and the type of closely related expectation questions (e.g., “Who do you expect will win the election?“) typically found on prediction markets — markets for placing bets on future or otherwise unknown events — which also tend to outperform traditional polls.