Standardized versus Spontaneous Translation of Survey Questions: An Analysis of Kenyan DHS Data

Alexander Weinreb, Hebrew University of Jerusalem
Mariano Sana, Louisiana State University

Demographic surveys implemented in multilingual settings routinely rely on a template questionnaire and its standardized translation to the main languages spoken by the respondents. In some settings, however, the number of languages is too large to make this practice feasible. We analyze data from the Kenya 1998 DHS, in which about a quarter of interviews were conducted in a language other than that of the questionnaire held by the interviewer, so that the interviewers had to rely on their own spontaneous translation of the questions rather than on the project’s standardized translation. We explore the effects of spontaneous translation on the proportion of variance that is across interviewers (the intraclass correlation coefficient rho) and on systematic differences in responses (bias). Preliminary results suggest that, although the former tends to increase with spontaneous translation, there are no clear bias effects. In general, results are heterogeneous and vary on a variable-by-variable basis.