Degree Name

Doctor of Philosophy (PhD)


School of Mathematics and Applied Statistics - Faculty of Informatics


Although interviewers play an important role in the collection of high quality data, the presence of the interviewer may result in an unintended correlation between responses in surveys. The interviewer effect can greatly increase the variance of results derived from surveys and is not considered in standard variance estimates. The impact of interviewers on total survey error should be estimated so that survey data can be interpreted appropriately. Previous studies examining the interviewer effect have generally relied on fully interpenetrated designs, in which a minimum of two interviewers are randomly allocated to each spatial area. Interpenetration provides repeated measurement of spatial areas which allows spatial and interviewer effects to be disentangled. However, conducting an interpenetrated survey is an expensive process that is not often applied in practice. This thesis presents a review of techniques for estimating the interviewer component of total survey error and establishes a general framework for examining the interviewer effect in household surveys. The framework is applied to explicitly consider the relationship between the survey design and estimation of the interviewer effect for the first time. We demonstrate how the interviewer effect can be estimated in a variety of new scenarios, including designs in which the interviewer and spatial effects were previously considered confounded. We then consider the potential gain from incorporating the longitudinal and spatial information available to the survey designer. The concept of partial interpenetration is introduced and clearly defined. Methods for estimating the interviewer effect in partially interpenetrated surveys are then considered. Partially interpenetrated survey designs will now allow us to estimate the interviewer effect and its contribution to total survey error more efficiently and cost-effectively as a regular part of the survey process. Procedures for preparing cost-optimal partially interpenetrated survey designs for the estimation of the interviewer effect are introduced and the gains from estimating the interviewer effect in partially interpenetrated surveys quantified. Regular estimation of the interviewer effect will result in higher quality surveys and will have positive implications for ongoing monitoring leading to more appropriate interviewer training, questionnaire design and cost-effective surveys.

02Whole.pdf (1533 kB)