In multipurpose surveys, there is often a need for increasing the number of questions in order to assure the coverage of all issues of interest. However, increasing the length of a questionnaire can increase respondent burden, see Sharp and Frankel (1983). This then has possibly a negative influence on the response rate as shown by Marcus, Bosnjak, Lindner, Pilischenko, and Schütz (2007), as well as the quality of the responses, see Herzog and Bachman (1981). One possibility to address this problem is to decrease the length of the questionnaire using multiple matrix sampling design or split questionnaire survey design. The latter, originally developed by Raghunathan and Grizzle (1995) can be considered as an extension of the multiple matrix sampling design described by Shoemaker (1973) and Munger and Loyd (1988). In matrix sampling, every surveyed individual answers a random set of items whereas in the split questionnaire survey design the questionnaire is divided into several components with roughly equal numbers of items, i.e., a core component containing items that are considered to be vitally important (e.g., sociodemographic items) and some split components. Apart from items in the core component, which are inquired of all individuals surveyed, only a fraction of the split components are administered to random subsamples of individuals. Assigning items to split components is based on the correlation coefficients between the items. Items with higher correlations are assigned to different components. Also, the branches of an item and items containing skip patterns are assigned to the same component, because then it can be avoided that items that explain

each other very well are jointly missing for any observation. Following the split questionnaire design each sample individual receives a fraction of the long questionnaire. Data gathered from the whole sample are combined and missing data in the sample caused by the items that are not asked from sample individuals due to split design are completed using multiple imputation as discussed by Rubin (1987), see also Gelman, King, and Liu (1998), Rässler and Koller (2002) and Adiguezel and Wedel (2008).