Does receiving information on polls change voters’ beliefs, turnout or vote choices in close elections? Theory and lab experiments predict polls increase turnout, but differ on whether likely winners or likely losers vote more. There is limited well-identified evidence in real elections. I conduct a field experiment with 2,023 low-income registered voters in Johannesburg in South Africa's 2016 municipal elections. Pre-election, few believed the ANC, the dominant party, would lose, or the election would be close, although the ANC went on to suffer a surprise upset. I give two treatment groups information from a (poorly publicised) IPSOS-Mori poll. One group learn the polls are close and the challenger party, the DA, is just ahead; another group learn the ANC is just ahead. In both treatment groups, likely winners (who hear the party they supported at baseline is just ahead) turn out more than a placebo group, on both self-reported and verified measures of turnout. Likely losers are no more likely to turn out than the control. This is counter to standard pivotal voting models, which predict likely losers vote more, and provides support for models which emphasise the role of conformity or allow utility from voting for the winner. Furthermore, vote preferences across voters and non-voters shift toward winners. Both treatments increase engagement with and knowledge of the local election, suggesting polls may improve the quality of citizen engagement in new democracies.
The Political Science Seminar Series is convened by Geoff Evans, Elias Dinas and Sergi Pardos Prado. For more information on this or any of the seminars in the series, please contact firstname.lastname@example.org.