Combining multiple imputation and control function methods to deal with missing data and endogeneity in discrete-choice models