How Language Shapes Public Opinion: Survey Experimental Evidence in Kenya

Abstract

How does survey language affect survey responses? In many parts of the world, people use multiple languages within the same country. Yet, existing literature suggests that linguistic minorities often systematically misrepresent not only their political opinion but also their recollection of objective facts. Borrowing from the existing literature in political science, psychology, and survey research, we theorize that language influences survey responses mainly through two channels: first, by priming political knowledge and contexts, and second by provoking social desirability bias. To test these expectations, we conduct an original survey experiment in Kenya by randomly administering the survey in either English or Swahili - the two official languages of the country. We find that first, respondents who randomly receive their surveys in English as opposed to Swahili are more likely to correctly answer a battery of political knowledge questions and report higher interest in politics more generally. Second, those who are in the English language treatment group are more likely to report experiences of participation in formal politics, such as voting and community gathering, but, at the same time, less likely to report experiences of participation in informal politics, such as protests and demonstrations. Third, we also document evidence of social desirability bias, but only in dimensions relevant to the difference between the English and Swahili languages. Additional tests provide indirect evidence that the effects of language on political knowledge, interests, and participation are driven mostly by priming effects and not by social desirability bias; that we are indeed capturing the effect of language and not some other politically salient identities – on survey responses; and that language effects are mostly direct, rather than being mediated by some heightened sense of anxiety, or other feelings