Indeed, some of the variables in the process-produced data of extraordinary reliability, but some of them contain defects like missing values and/or inconsistent information or sequences. By the example of information on educational achievement provided in the process-produced research-data by the Research Data Centre (FDZ) of the Federal Employment Agency at the Institute for Employment Research we show the dimension of the problem and we explain how we recommend for dealing with these problems.
The reliability of information on educational achievement in the process-produced IAB-data has been distrusted because they originate from administrative sources such as the notification process of the social security system and from several internal procedures of the German Federal Employment Agency (BA) since this information is unverified and misreporting has no consequences concerning obligations or claims out of the social security neither for the employer nor for the employee. Previous research use deductive imputation methods that make use of the panel structure of the data set (Fitzenberger et al. 2005; Drews 2006, Wichert/Wilke 2010). All approximation methods take more or less plausible assumptions as a starting point. However, until now there was no way to test the reliability of the corrected education variable arising from the different approximation methods.
Now, by linking the process-produced data with survey-data which conducted the individual educational history of respondent in every detail (ALWA-ADIAB) we are able to explicitly check which approximation method result in most reliable information on educational achievement.