NBER Publications by Kenneth Train

We describe two methods for correcting an omitted variables problem in discrete choice models: a fixed effects approach and a control function approach. The control function approach is easier to implement and applicable in situations for which the fixed effects approach is not. We apply both methods to a cross-section of disaggregate data on customer's choice among television options including cable, satellite, and antenna. As theory predicts, the estimated price response rises substantially when either correction is applied. All of the estimated parameters and the implied price elasticities are very similar for both methods.