Standard Error of an Estimated Difference between Countries When Countries Have Different Sample Designs: Issues and Solutions
Peter J. Lynn, Olena Kaminska
Institute for Social and Economic Research, University of Essex, Wivenhoe Park, Colchester, Essex, United Kingdom

Standard estimation procedures for complex sample designs assume the data relates to one population and arises from one sample design. However, in cross-country studies it is often more efficient and more practical to select respondents using different sampling strategies in different countries. As a result, there is a need for estimation procedures which correctly reflect this situation. Using data from an important cross-national study, the 2007 European Union Statistics of Income and Living Conditions (EU SILC), we identify different situations and suggest estimation procedures for each. Our main focus is the estimation of differences in means and proportions when only one of two countries has a clustered design. We also consider variation in the number of selected household members (all or one). Furthermore, we compare our estimation procedures with convenient alternative suboptimal approaches an analyst may take: either not taking any clustering into account or taking only households, but not higher level clustering, into account. Our results show that in a few situations the conclusion may be sensitive to the estimation procedure, mainly when the difference between countries is small but marginally significant.

Keywords: Cross-national; Clustered sample design; Standard error; Between-country difference

Biography: Olena is a graduate of the University of Nebraska-Lincoln, USA, and is currently employed as a Survey Statistician in the University of Essex in the UK.