Does measurement error bias fixed-effects estimates of the union wage effect?

This paper compares estimates of the union wage effect using cross-section and panel estimators for male manual full-time and female employees using data from the British Household Panel Survey, 1991-1997. A comparison of cross-section and panel estimates suggests that unobserved heterogeneity biases cross-section estimates upwards. However, it is also found that the divergence between estimates is overstated because measurement error biases the fixed-effects estimates downward. Reducing measurement error in the union variable by taking averages and restricting changes in union status to occur only when a change in employer and/or job takes place increases fixed-effects estimates of the union wage effect.