A Mixture Model of Errors in Twin Education Reports

Abstract

Measurement error is a potential problem with estimating the wage effect associated with first-differences in twin's education levels. To account for this, Ashenfelter and Rouse (1998) provided two reports of each twin's education level. One is the own report and the second is the sibling's report. This paper uses recent results on finite mixture models (Adams (2016)), to show this may be enough information to identify the underlying true education level without requiring additive measurement errors. The estimated returns to education are shown to be either the same or below the IV estimates presented in Ashenfelter and Rouse (1998). The strong parametric restrictions of the IV or "classical measurement error" approach may overestimate returns to education by up to 21%.