The value of sibling data for identifying the causal effect of schooling on wages hinges on our ability to eliminate biases due to the mismeasurement of schooling. Analysts typically assume errors in schooling reports are "classical." In this study, we use generalized method of moments to estimate the parameters of a range of measurement error models, including forms of both classical and mean-reverting error models; we estimate the models using a sample of identical twins and a sample of non-twin siblings. The results of likelihood ratio-type tests reveal that variants of classical measurement error models fit both datasets about as well as more flexible models.
©2011 Walter de Gruyter GmbH & Co. KG, Berlin/Boston