Conferences

Field interviewers for Statistics Netherlands are expected to adhere to a fieldwork strategy, that dictates that one of the first two visits to a sample unit should be in the evening, that the first visit should be in the first half of the fieldwork period, that non-contacts should be visited six times, and that visits should be spread over time of day and days of the week. It can be shown that when this strategy is followed, response rates can be up to 25 percentage points higher than when the strategy is not followed. Likewise, the number of visits needed is substantially less when interviewers adhere to the strategy.

Although this is an interesting phenomenon in its own right, more interesting is how this behaviour influences data quality. If there is a difference between addresses where the strategy is followed, with a high response as a consequence, and addresses where the strategy is not followed, with a high chance of nonresponse, chances are that bias is introduced. By linking paradata on fieldwork behaviour to registries with a rich variety of background variables (e.g., ethnic origin, household composition, job holders, income, etc.) and area properties (e.g., number of inhabitants per postal code, number of people of foreign ethnic origin, number of people without work, etc) I aim to study the relation between fieldwork behaviour and sample unit properties, to see if and how interviewers introduce bias by their fieldwork behaviour. The second question I will try to answer, concerns the direction of the causality between following fieldwork strategy, high response rates and costs.