I conduct two different estimations suggested by BHKN. First is the linear model using ordinary least squares in the first stage and logit in the second stage. Second is the semiparametric method that maintains the linearity assumption, but the first stage is made more flexible by orthogonal polynomial transformations.