Assessing and improving survey data quality in sub-Saharan Africa

High-quality survey data is the foundation for informed and valid decision-making, policymaking, and research. In low- and middle-income countries (LMICs), surveys rely heavily on interviewers to administer questions to respondents who have varying degrees of literacy and who live in areas without stable telephone or internet access. Another characteristic of LMICs, especially in sub-Saharan Africa, lies in their multi-ethnic and multilingual nature. This doctoral thesis examines survey data quality using a Zambian panel survey through the lens of the interviewer, exploring gender-of-interviewer effects (chapter 2), connecting interviewer performance to translation quality through different modes of translations (chapter 3), and examining language switching as a problematic interaction in multilingual surveys (chapter 4). To estimate interviewer effects, the thesis leverages the random assignment of interviewers to respondents in each of the three panel waves conducted in 2016, 2018, and 2019. The analysis combines the survey data with two additional sources of data: (a) the self-administered interviewer surveys capturing key interviewer characteristics, such as age, gender, and language proficiency; and (b) the behavioural coding of interactions between respondents and interviewers during the survey interview. The findings of Chapter One demonstrate non-negligible interviewer effects for behavioural, knowledge, and attitudinal questions, which call for a more nuanced approach to studying social desirability bias. Chapter Two shows a considerable reduction of interviewer effects from the first wave (which had no scripted questionnaire translation) to the subsequent waves (which employed a pretested translation into the lingua franca of the three study regions). Chapter Three examines the frequency of language switching across different questions and two different regions demonstrating its co-occurrence with other problematic interactional behaviours that are known to influence survey data quality. The latter two studies highlight and recommend good practices in questionnaire translation and interviewer selection and training in multilingual settings.