Language Fluency and Earnings: Estimation with Misclassified Language Indicators

We use panel data from the German Socio Economic Panel to estimate the determinants of language fluency of immigrants, and its impact on earnings. Self reported measures of language proficiency contain substantial reporting errors. We specify a panel data model which takes explicitly account of misclassification. We extend the existing literature on misclassification of categorical dependent variables by distinguishing between time persistent and time varying misclassification errors, using panel data. The repeated information on language fluency allows us also to distinguish between cohort effects and exposure effects. We then add a wage equation to the model and estimate it jointly with the speaking fluency equation. In this way, we take account of the two problems that may bias OLS estimates: misclassification errors and correlated unobserved individual heterogeneity in wages and speaking fluency. We nd that both have important consequences for the estimated effect of speaking fluency on earnings.