This paper demonstrates empirically that the limited dependent variable technique for sample selectivity bias correction provides two very different estimates of the gap between union and non-union wages. Our findings, based on the Canadian Survey of Work History, suggest that selectivity...