[Journal:] Zeitschrift für ArbeitsmarktForschung - Journal for Labour Market Research [ISSN:] 2510-5027 [Volume:] 42 [Year:] 2009 [Issue:] 2 [Pages:] 121-139
Linking data from different sources can be used to compensate for the weaknesses characteristic for the single sources and therefore significantly increase the potential for analyses. This requires that linking data implies no consequences that cancel out the advantages. Considering this background the contribution analyses to which extent we face such consequences when linking administrative data of the 'Bundesagentur für Arbeit' (BA) to survey data on the level of individuals. Due to reasons of data protection the respondents have to be asked to allow linking their survey data to their administrative data. This necessity may result in a kind of 'linkage bias', as it may be, that their willingness to admit correlates with other characteristics relevant for the research question. This may bias results. Using data of a study, in which persons were asked, whether they allow linking the survey data to administrative data, we analyse whether there is selectivity in admission regarding a number of characteristics that are known to influence respondent behaviour. Using multilevel analysis we show that only some characteristics are correlated with allowing data linking: Women, foreign persons, persons with low income and persons interested in protecting private information as for example on income or social benefits tend to be underrepresented in the restricted sample using the linked data set. In order to investigate the consequences of this selection for specific research questions we propose a simple test based on a 'seemingly unrelated estimation' and present two examples.