Keywords: behavioral coding, multilingual survey methods, low- and middle-income countries

Most low- and middle-income countries (LMICs) are multiethnic and multilingual leading to the situation that both interviewer and respondent may be conversing with each other in multiple local languages. Researchers and practitioners usually assume that an interview is administered continuously in one language. However, in multilingual settings found in multi-tribal, multiethnic countries, this assumption does not always hold as both the interviewer, and the respondent have several languages at their disposal to communicate with each other in case they encounter conversational difficulties in the designated survey language. As multilingual respondents, especially those with lower education, differ in their proficiency in the survey language, some will exhibit more cognitive processing problems evidenced by audible manifestations of problematic interactional behaviours during the interview (i.e., through seeking for clarification or repetition of the question, expressions of uncertainty through pauses and fillers, audible as such as frequent and prolonged “uhm”).

The ability to choose among several languages leads to the phenomenon of language switching, wherein speakers start in one language (usually the designated survey language) but then switch to a different language - usually one for which there is no standard translation of the questionnaire. Respondents might switch languages to better communicate their survey answer or out of habit or convenience as they are more fluent in a non-survey language. In contrast, interviewers are usually trained to refrain from deviating from the survey language against their own habits and convenience. However, when the interviewer is faced with a respondent who is not adequately proficient in the designated survey language and shows signs of comprehension problems, she might decide to deviate from standardized interviewing techniques and translate survey questions without a script, which then might lead to interviewer error.

Using the interactional analysis of the interview recordings in two local languages (Bemba and Chewa), we analyze the relationship between several indicators of problematic interactional behaviours and language switching. Local coders were trained to code those behaviours for eight selected questions in a survey on financial behaviour and attitudes in Zambia on a sample of more than 800 interviews. In a second analysis step, we link language switching to interviewer effects, which is the proportion of response variance attributed to the interviewers.
This study’s objective is to document the process of switching language and its implications to survey data quality. The Zambian survey we rely on estimates that about 2 to 7 percent of interviews exhibit some form of language switching. Albeit the frequencies of language switching differ among study regions, as well as over different questions, the results show that language switching does not occur as a single phenomenon but always in co-occurrence with other problematic interactional behaviours. Thus, we can categorise language switching as another problematic behaviour indicating the breakdown of the cognitive answer process. In this sense, language switching can also be considered as a disruption to the ideal question-answer sequence.