Tax evasion and measurement error: A comparison of survey income with tax records

The main constraint for empirical research on tax evasion is, unsurprisingly, the lack of suitable data, this being even more pronounced for developing countries. This paper provides estimates of the individual-specific determinants of tax evasion based on a unique dataset combining a household income survey and tax records for Estonia. The nature of data linkage implies high quality matching with no consent bias and, hence, allows to assume that survey income represents respondent’s true income subject to measurement error. Unlike earlier studies attributing all such income discrepancies either to tax evasion or measurement error, I model each process in a joint framework. Focusing on employment income, the identifying assumption made is that measurement error is not related to the sector where the individual works while the same does not hold for tax compliance. Specifically, it is assumed that taxes cannot be evaded in certain sectors.