An analysis of speaking fluency of immigrants using ordered response models with classification errors

We develop parametric models that incorporate misclassification error in an ordered response model and compare them with a semiparametric model that nests the parametric models. We apply these estimators to the analysis of English-speaking fluency of immigrants in the United Kingdom, focusing on Lazear's theory that due to learning or self-selection, there is a negative relation between speaking fluency and the ethnic minority concentration in the region. Specification tests show that the model allowing for misclassification errors outperforms ordered probit. All models lead to similar qualitative conclusions, but there is substantial variation in the size of the marginal effects.