The impact assessment of social and labour policies is mainly based on direct inquiry on their recipients. In the case of net impact assessments, one has to observe a group of untreated individuals that is very homogeneous with respect to the treated. However, the very high cost of direct surveys limits their scope, as it prevents large sample size, hindering the reliability of estimates. The access to databases created for different purposes (fiscal, administrative, monitoring) could overcome the above limitations, with much convenience in terms of cost and data quality. This paper presents a validation exercise based on a very broad and complex employment database, the Comunicazioni Obbligaorie (COB), that collects employers’ compulsory notifications in the case of changes in a job contract. This exercise was possible thanks to an extraordinary condition, giving contemporaneous access to two different data sources, the COB database and a direct survey on recipients of vocational training policies in the Piedmont Region. The paper discusses the main differences between employment indicators evaluated on direct survey data and on COB data. Inconsistencies involve the 20% sample. In addition, such inconsistencies do distort the net impact assessment, since the key variables for placement outcomes vary significantly. Hence, whenever administrative databases are used as a source for socioeconomic analyses, extreme caution and critical thinking must be used, verifying the reliability of estimates deriving from their use and comparing results with other sources. In the lack of such preliminary validation, which should be carried out in close collaboration with the regional or national authorities managing the information systems, we run the risk not only to uncritically accept information producing systematic bias, but also to prevent the development of procedures that are essential for the system development.
Keywords: Impact Assessment, Validation, Administrative Data, Survey, Multivariate Analysis