practices relative to non adopters. Since differences in firm and workforce characteristics are also
substantial, the effect of the practices on wage dispersion will be more rigourously established using
OLS regressions in which proper controls for firms and workers characteristics are included. Before
presenting the regression analysis results, the next section discusses limitations of the NES data.
3.3 Data limitations
As it is a plant-level dataset, the NES does not have detailed variables on establishment’s workforce
characteristics. In particular, workers’ average experience or tenure are not available. These variables
affect individual wages and may explain a substantial part of the manager-production workers wage dif-
ferentials. Given that the paper focuses on the analysis of variations in the manager/production worker
wage ratio across establishments, it is important to first verify whether or not the lack of control for
average experience modifies the cross-establishment variations in wage differentials. Using information
on workers potential experience from the CPS in the years 1993 and 1996, the manager/production
worker wage ratio is estimated across industries and firms of different sizes. The inclusion of workers
experience does not significantly affect the estimated ratio (see Appendix A.2 for details).
A second important limitation of these data set is that there is no information on the date the
firm started to adopt the practice. It is therefore difficult to identify properly the effects that would
be specifically related to the timing of adoption (learning effects at adoption as well as differences in
experience with the given practices). The analysis in this paper is based on the assumption that all
firms that adopted the practices, have done so during the same period within the last 25 years.
Finally, although the NES 1996 sample has a longitudinal component of about 900 companies that
also participated in the survey of the NES 1993, the longitudinal link is not publicly available. On the
other hand, previous studies using the NES that have exploited the longitudinal aspect of the NES have
not found much differences in the results when using OLS or fixed-effect estimations (Black and Lynch
(1997) and (2000)). Using the longitudinal format of the NES may not constitute an improvement in
the quality of the results if the number of workplace practices switchers is very small which is likely to
be the case given the years analyzed (mid-90s) and the short period of time between the two years.
As a result, the analysis hereafter uses the cross-sectional aspect of the data to identify the effect
of the practices through inter-establishment variations. The methodology however does not take into
account effects due to firm-specific unobserved heterogeneity. On that last point, the analysis in this
paper based on wage ratios rather than wage levels has the advantage of controlling for a given type of
firm-specific unobserved heterogeneity. Indeed, if the unobserved heterogeneity term is constant over
time and affects similarly the wages of all categories of workers, it will drop out of the wage ratio
equation.
12