Working Paper: CEPR ID: DP2012
Authors: Christian Dustmann; Arthur van Soest
Abstract: A number of studies have established a positive effect of migrants' language proficiency on their productivity. It has been argued that these estimates are upward-biased because of the presence of unobserved heterogeneity. To obtain an accurate estimate of language effects is important since it has potentially important implications for migration policies. Using panel data on immigrants in Germany, we show that in self-reported measures of language proficiency measurement error is substantial. Our results suggest that measurement error is a more serious problem than unobserved heterogeneity, and that Least Squares underestimates the effect of speaking fluency on earnings.
Keywords: international migration; language; endogeneity
JEL Codes: F22; J24
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| measurement error in language proficiency (C83) | underestimation of the effect of language proficiency on earnings (J31) |
| unobserved heterogeneity (C21) | upward bias in OLS estimates of language proficiency on earnings (J31) |
| years of residence (R23) | earnings (J31) |
| language proficiency (G53) | years of residence effect on earnings (J31) |
| language proficiency (G53) | earnings (J31) |
| language proficiency (G53) | earnings (IV estimates) (C26) |