Beschreibung

"Linking data from different sources can enrich the research opportunities in the Social Sciences. However, datasets can typically only be linked if the respondents consent to the linkage. Strategies from the secure multi-party computation literature, which do not require linkage on the record level, might be a viable alternative in this context to avoid biases due to selective non-consent. In this paper, we evaluate whether such a strategy could actually be successfully applied in practice by replicating a study based on linked data available at the Institute for Employment Research. We find that almost identical results could be obtained without the requirement to link the data. However, we also identify several problems suggesting that the proposed strategy might not be feasible in many practical contexts without further modification." (Author's abstract, IAB-Doku) ((en))