SAMPLE SELECTION BIAS In that case we could use least squares to fit the model as usual. Y i unobserved if for 13