The paper examines strategies for data collection in telephone surveys of individuals, with emphasis on the Swedish Living Conditions Survey (LCS), which currently has a nonresponse rate of more than 30%.
Surveys are often driven by an objective to reach a predefined rate of response, say 70%. This can be costly and inefficient, and in the end the goal may still not be realized. Such a fixed objective may for example motivate interviewers to focus during follow-up on groups from which response is deemed easy to obtain, rather than on, say, underrepresented groups.
The nonresponse rate is not in itself a suitable tool for guiding the data collection, including the nonresponse follow-up. Preferably, the data collection effort should be driven by a wish to realize a better balanced or more representative ultimate set of respondents, so that estimates with reduced bias can be obtained at the estimation stage.
Several indicators for balance and/or representativity of a set of respondents have been developed in the recent past. With the aid of these indicators it is possible to monitor the degree of balance in the set of respondents, during the ordinary data collection period and during the follow-up. We study some of these indicators theoretically and empirically. We make recommendations for the field work in future editions of the LCS.
Keywords: Responsive design; Balanced response set; Auxiliary information; Data collection strategy
Biography: PhD in Statistics, Stockholm University 2006.
Senoir Statistician, Process Department, Statistics Sweden, Stockholm.
Area of interest: Survey sampling methodology, especially in conection with nonresponse in data collection and in estimation.