Measurement of higher education in the census and current population survey

We examine measurement error in the reporting of higher education in the 1990 Decennial Census and the post-1991 Current Population Survey (CPS). We document that measurement error in the reporting of higher education is prevalent in Census data. Further, these errors violate models of classical measurement error in important ways. The level of education is consistently reported as higher than it is (errors are not mean 0), errors in the reporting of education are correlated with covariates that appear in earnings regressions, and errors in the reporting of education appear correlated with the error term in a model of earnings determination. Thus, neither well-known results on classical measurement error nor recent models of nonclassical measurement error are likely valid when using Census and CPS data. We find some evidence that the measurement error is lower in the CPS than in the Census, presumably because first interviews are generally conducted in person.