Abstract
To explore whether changes in the selection into full-time work among German men were a driver in the rise in wage inequality since the mid-1990s, we propose a modification of selection-corrected quantile regressions. Addressing Huber and Melly’s (J Appl Econom 30(7):1144–1168, 2015) concerns, this modification allows us to estimate the effects of selection with respect to both observables and unobservables. Our findings show that those employed in 1995 would have had lower wages in 2010 than those employed in 2010 and wage dispersion would have been higher, suggesting that full-time workers have become less heterogeneous over time.
Highlights
Germany experienced a considerable increase in wage inequality until 2010 (Dustmann et al 2009; Card et al 2013; Möller 2016; Biewen et al 2018)
Our second question addresses the changes over time: How would wage inequality have developed if selection into full-time employment had not changed over time?
Regarding the substantive economic research question, BFS analyze the gender wage gap accounting for selection into employment among females while this paper investigates the role of selection into employment in explaining the increase in wage inequality among males
Summary
Germany experienced a considerable increase in wage inequality until 2010 (Dustmann et al 2009; Card et al 2013; Möller 2016; Biewen et al 2018). With the control function approach augmented by a transformation of the dependent variable, we estimate quantile regressions which are corrected for selective movements between unemployment and full-time work. In a recent important paper, Arellano and Bonhomme (2017) suggest a copula based method to provide consistent estimates of quantile regressions with selection correction They estimate quantile regressions while assuming a fixed copula between the conditional rank in the wage distribution and the rank in the error term of the selection equation. Similar to the assumed copula in Arellano and Bonhomme (2017), these studies assume a specific functional form regarding the link between the selection equation and the distribution regression This link (modeling the so-called selection sorting effect) is allowed to vary as a function of a linear index of the covariates, and the coefficients in the index are estimated.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have