Comparing Comparisons for Cross-Validation (CC-CV): A Proposal for Applied Survey Research Using the Example of Attitudes towards Economic Liberalism

Abstract

AbstractComparative research that relies on sampling procedures has to deal with the problem of validity and reliability of estimates derived from a chosen sample. Social scientists are not only interested in level estimates such as mean values, but also in (causal) effects estimates as in regression type models. Thus, estimates based on these theoretical explanatory models also underlie the problems of validity and reliability. Usually researchers deal with one sample for each study of interest and implicitly rely on the adequacy of the reported parameters. In cases, where the researcher have the chance to control the whole sampling and data cleaning procedures, i.e. the field of primary data generating research, this paper will not add much at first hand. In other cases, where the researcher relies on secondary data sources that are available publicly or by other contracts, the researcher cannot control the sampling and data cleaning process ex posteriori. This problem becomes specifically important in cross-national research but may also be applied to other fields. This contribution is targeted at applied researchers to realize an easily accessible method for cross-validation based on actual empirical data, i.e. it does not make use of artificially created data and does not involve complex algorithms (cf. application in Beckers 2008). In cross-national survey research, which is the research field of application discussed here, the aim is to compare countries in order to gain knowledge about both similarities and differences and the generalities and specificities of social mechanisms (Meulemann 2002; 2008; 2009).