Identification of the Union Wage Impact using Panel Data: The British Case During 1995-2009

Georgios Marios Chrysanthou, University of Alicante

Georgios Marios Chrysanthou

This study estimates the union wage impact employing BHPS data spanning the period of 1995-2009 using balanced and unbalanced panels of male and female employees. We employ a two stage control function variant approach to estimate a dynamic model of unionism and wage determination. This estimator is flexible in its treatment of unobserved heterogeneity and reveals information regarding the economic sorting mechanism of individuals within unionised and non unionised establishments. However, it limits our ability to include lags of the dependent variable and union membership in the structural wage equation. In response to this, we use an alternative difference generalised method of moments (GMM) estimator. Further, in order to account for the endogeneity of union membership without having to rely upon the potentially restrictive assumptions of the control function variant approach, that entails estimation of the reduced form under random effects assumptions, we additionally employ the fixed effects (within transformation) estimator. We conclude that fixed effects and difference GMM estimation are inappropriate in that the restriction that unobserved individual heterogeneity is time-invariant and equally rewarded in the two sectors is rejected in all of our estimates. We find that the unobserved factors that influence union membership also affect wages and conclude that UK trade unions still play a non-negligible role in wage formation. Upon accounting for the endogeneity of union status, measurement error in union membership status and the impact of job changers we conclude that there is a positive union membership differential (defined as the differential between covered members and covered non members) and a small union coverage differential gain for covered non members suggesting the existence of free riding. Upon exclusion of the potentially endogenous occupational controls, coverage differentials become statistically insignificant while union membership differentials are estimated to be approximately 8.9 and 13.4 per cent for male and female members.