How should we translate survey questionnaires? An analysis of Kenyan DHS data

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

Virtually all survey data used to describe people's behavior and attitudes in Africa and other developing countries derive from questionnaires formulated in a "global" language (typically English or French) and then translated into a local one. Using Kenyan DHS data, this article explores the impact of two different types of translation procedures on the research enterprise. It first assesses the extent to which each differentially affects one of four indicators of measurement (i.e. non-sampling) error. It then assesses how the accumulated effects of different translation procedures on measurement error across multiple variables of interest substantively affect estimated relationships among those variables. Overall, results suggest that different translation procedures affect univariate statistics modestly, but multivariate relations more substantially. This is a new result on an underresearched topic. It also has considerable implications for the way that basic survey and census data are collected in sub-Saharan Africa.

  See paper

Presented in Session 45: Methodological issues in measuring poverty and human welfare