Polling in New Democracies and Electoral Malpractice: The Case of Brazil

Abstract

In many developing democracies, polling often comes with elevated levels of bias and variance. We argue that electoral malpractice can be one reason why. We build a theory and test it with data from elections between 2002 and 2014 in Brazil. We find that polling errors are larger in: (a) elections with many undecided voters and large imbalances in financial resources among campaigns; (b) the poorer Northeast region of Brazil, which is more closely associated with patronage and vote buying; and (c) low-profile, low-information elections. Our analysis serves as a cautionary tale for interpreting polling in democracies like Brazil, even if/when other sources of error in the polling industry are mitigated.