Lang : English
  • English
  • Français
  • हिन्दी
  • Português
  • Español
close× Call Us

Brief 41: Investigating Unrevealed Biases in Ethnic Voting Behavior in Ugandan Public Opinion Surveys

In order to understand the effect of public opinion survey conditions on responses, this study randomly varies two elements of the survey protocol: (1) respondents are randomly selected to either privately or publicly state their vote choice and (2) respondents are randomly selected to be “primed” to the salience of ethnicity. Respondents are significantly more likely to state support for a coethnic candidate when ballots are private and significantly less likely to state support for a coethnic candidate once they have been primed to recognize ethnicity as a salient characteristic.

Read Full Study

Category: Elections

Tags: Uganda, survey, ethnicity, voter behavior, polling, public opinion, coethnic

Date of Publication: Friday, August 12, 2016

EGAP Researcher: Elizabeth Carlson

Partners: Photo Credit Mark Johnston via Creative Commons 2.0

Geographical Region: Africa

Research Question:

Do public opinion survey respondents systematically conceal their true preference for coethnic politicians because they recognize the undesirability of a choice based on ethnic affiliation?


Preparer: Nicholas Kuipers



Across Africa, observers have frequently noted that voters tend to support politicians of their own ethnicity. Studies confirming this phenomenon often rely on public opinion surveys and correlate stated ethnicity with stated vote choice. Research in Africa and elsewhere has shown that ethnicity is sensitive, however; voters tend to conceal their affinity for politicians of their own ethnic group because they recognize the undesirability of such preferences. To what extent are our current measures of ethnic voting—as captured by public opinion surveys—biased by this effect?


Research Design:

To test the extent to which respondents conceal their true preferences, the author designs and implements a survey experiment in Uganda with 753 respondents. The author randomly sampled census enumeration areas. Then, surveyors recruited respondents from every fifth household in four directions from the center of the enumeration areas. Respondents were then randomly assigned to one of four treatment categories: (1) unprimed to ethnicity and vote choice is private, (2) unprimed to ethnicity and vote choice is public, (3) primed to ethnicity and vote choice is private, or (4) primed to ethnicity and vote choice is public. Respondents were asked to cast a vote for one of three hypothetical presidential tickets whose characteristics vary along five dimensions—ethnicity, education, prior office, track record, and platform.  



The author finds statistically significant evidence that respondents conceal their true preferences for coethnic candidates. When vote choice is private and respondents are unprimed to ethnicity, hypothetical coethnic candidates win 60% of the time, while non-coethnic candidates win only 45% of the time. Meanwhile, the author finds evidence that, as respondents are forced to publicly reveal their vote choice or are intentionally primed to consider the salience of ethnicity, they reveal a greater preference towards candidates with characteristics other than a shared ethnic background—education, prior office, track record, and platform. 


Policy Implications:

There are two related policy implications to this research. First, the most straightforward implication of this research has to do with the way in which practitioners structure public opinion surveys so that they do not get biased estimates of support for a politician or policy. Particularly in developing countries where public opinion surveys are often conducted face-to-face, the architects of the questionnaires should ensure that respondents’ preferences are understood to be private, even to the surveyor. Second, and relatedly, better structuring public opinion surveys to arrive at unbiased estimates of preferences can increase accountability and representation Policy interventions that induce politicians to meet incorrectly measured preferences can unintentionally reduce welfare.