Interviewer Effects In Quantitative Surveys Using A Door-To-Door Approach

Abstract

Interviewers are a principal source of error in quantitative surveys. While surveys are

often self-administered (e.g. in online surveys), it is often required to administer these face-toface.

This is the case, for example, in census surveys in low-income areas where there is little

internet penetration, like that of the quality-of-life surveys presently being conducted in

multiple countries through a residential door-to-door approach (Carr et al., 2018). In such

situations, the social interaction between the interviewer conducting the survey and the

interviewee is likely to introduce bias into the survey data collected. Interviewer effects (IE)

can influence both item non-response and answer quality, i.e., participants not providing the

true answer (Harling, et al., 2019). In an attempt to gain more representative data, this study

conducted an exploratory analysis on the possible antecedents and consequences of interviewer

effects using the Living Wage survey presently being conducted in South Africa, as the study

context. To this end, I examine the systematic biasing effects associated with deploying the

same group interviewers (n = 10), of the same ethnicity, age, and of equal gender distribution

across five sampling areas in Cape Town in a quasi-experimental design (n = 282). This study

highlighted that each interviewer is associated with a unique set of systematic bias that varies

dependent on the survey item type. Sensitive items requiring respondents to disclose personal

information were the most prone to bias, followed by interviewer-referencing and attitudinal

items sequentially. Furthermore, this study found that gender differences in the interview had

a marginal influence on the attitudes respondents are willing to share. I hope to contribute to

an understanding and critical consideration of the antecedents and consequences of deploying

human interviewers for collecting quantitative surveys, especially in a context where ethnic,

gender and political differences are loaded in social interactions and are likely to contribute to

respondents obscuring their responses.