Administrative data has become more important for both official statistics and social science research. The 2011 Census round is one of the driving forces for this development. One of the problems that might occur is that the measurement in the administrative data is biased. We than speak of a low validity. Although this problem is often mentioned in a qualitative way, the validity is seldom quantitatively measured. In this paper, a method is presented to estimate the size of the validity of administrative variables.
Starting from the classical test theory it is supposed that the measurement of reliability can be distinguished from that of validity by repeated measurement. The validity can than be determined by using linked survey and register data which should measure the same concepts and repeat the measurement. Because repeated measurement is expensive and is not always possible, the survey and register measurement can also be conceived as two items of the same construct.
This idea is elaborated with an empirical example in which the construct validity of age, gender, educational attainment and wages is simultaneously determined. Survey data from the 11th wave of the OSA labour market supply survey are linked to administrative data from the Social Statistical Database. A linear structural equations model with a measurement component is used to compute the construct validity. The analyses reveal that the validity of the administrative measurement of age and gender is very high. Educational attainment and wages show some bias, but this bias is not higher than the bias found in the linked survey.
Keywords: Validity; Register data quality; Classical test theory; Linear Structural Equations
Biography: Bart Bakker holds two positions. The first is that of professor at the VU University Amsterdam in “The methodology of register data”, the second is at Statistics Netherlands as head of the section Socio Economic State. His research interests are in social stratification and in register methodology.