Reducing non-response in longitudinal surveys by improving survey practice

My thesis is about the prevention of unit non-response in longitudinal surveys. Non-response can lead to error in survey estimates, meaning they give a biased picture of the true value in the population. The main sources of non-response in surveys are non-contact and refusal. For longitudinal surveys, which follow the same people over time, non-location is an additional source of non-response. Non-response over time in a longitudinal survey is often referred to as attrition. There are two broad approaches to dealing with non-response in surveys. The first is to make statistical corrections to the survey estimates to take account of non-response error. The second is to try to prevent non-response by improving how surveys are designed and conducted. My research takes a prevention approach, and addresses all three of the major components of non-response in longitudinal surveys. I use randomised experiments to evaluate fieldwork interventions designed to reduce non-response and prior wave data to evaluate the impact of fieldwork interventions on non-response bias. I use data from large-scale, high quality datasets: the UK Millennium Cohort Study and the Innovation Panel of Understanding Society, the UK Household Longitudinal Study. My main findings are: refusal conversion can be an effective way of reducing non-response both in the immediate and longer term and can lead to some modest reductions in non-response bias; improving the design of the covering letter used on between-wave mailings can improve return rates from sample members with lower levels of education and those who speak languages other than English at home; respondent characteristics are related to the success of both office and field tracking; sample members who respond to an invitation to make their own interview appointment require less fieldwork effort overall and they are more likely to do this when a financial incentive is offered