Measuring, explaining and adjusting for cross-country differences in unit nonresponse : what can process data contribute?

The analysis of cross-national survey data can be hindered by unit non-response. It is not uncommon for the countries in a cross-national surv ey to have very different response rates. This raises awareness amongst analysts of the potential for differential non-response errors, which might bias estimates of differences between countries. Research on cross-national differences in nonresponse and nonresponse bias, however, is still scarce, partly due to a scarcity of auxiliary data permitting such analyses. Process data are auxiliary data about the data collection process and can be suitable for nonresponse analyses in cross-national surveys if they are available across countries for both respondents and nonrespondents. The process data discussed in this thesis are contact (or call-record) data and interviewer observations of the neighbourhood. This thesis investigates the role that process data play in the measurement, analysis and adjustment of unit nonresponse in cross-national surveys. I first provide an overview of existing studies of nonresponse in the cross-national setting and the role that contact data have played therein. Quality concerns raised in these studies led to the development of equivalence criteria for cross-national contact data. The second chapter investigates the comparative collection and measurement of response outcomes. I develop a conceptual framework of influences on the response outcomes available to the survey researcher, design a codeframe of response outcomes for cross-national implementation and compare the effect of two coding strategies on deriving final case outcomes. In the third chapter I use decomposition methods to explain whether cross- country differences in contact rates are due to dif ferential sample unit characteristics, differential fieldwork characteristics or a differential effect of these characteristics on contact propensity. Finally, chapter four assesses the effect of weights based on process data on reducing relative nonresponse bias. All analyses are based on data from the European Social Survey