Accuracy and errors in self-assigned NAICS codes in tax return data

Abstract

Conventional wisdom holds that North American Industry Classification System (NAICS) codes chosen by people not experienced with the system are often mis-specified, but there has been little formal research into the scope of the problem. In this paper we explore prevalence of and patterns in misspecification in NAICS codes self-reported on two kinds of business tax forms. Errors are identified by comparing as-filed codes with codes validated by Statistics of Income. We find that over a third of codes are wrong, but that the errors are not random and often (though not always) seem to have logical reasons behind them.