The Fallacy of Composition Bias in the RealWage Cyclicality Puzzle

Composition bias in aggregate wages is often a scapegoat for the apparent unresponsiveness of wages over the cycle. Since Bils (1985) and in particular Solon et al. (1994), who find that that real wages are highly pro-cyclical a general consensus has emerged that the observed ‘mild’ cyclicality in real wages is due to composition effects which cause counter-cyclical biases because low wage jobs are the first to be destroyed during recessions (Pissarides, 2009). In this paper, it is argued that the results of Solon et al. (1994) and other papers using similar techniques cannot possibly disentangle the true effect of composition bias. This is because the assignment of fixed weights used to keep the composition of the work force constant is arbitrary and imposes a particular direction to the bias. Thus, rather than determining the bias it only serves to show the possible magnitude once having assumed the way the bias works. As in Blundell et al. (2003) we can unravel the bias into three interpretable parts. That is biases due to individual movement in and out of work, changes in the variation of hours worked and changes in the variance of wages over the cycle. The findings show that aggregate real wages become cyclically less responsive over the cycle and no evidence of ‘counter-cyclical’ composition bias.