BACK to Research Page | Home | Curriculum Vitae

"Measuring the Cyclicality of Real Wages: How Important is the Composition Bias?"
with Gary Solon and Robert Barsky
Quarterly Journal of Economics, Vol 109 No 1, (February 1994), 1-25.

Abstract

In the period since the 1960s, as in other periods, aggregate time series on real wages have displayed only modest cyclicality (blue starred line above). Macroeconomists therefore have described weak cyclicality of real wages as a salient feature of the business cycle. Contrary to this conventional wisdom, our analysis of longitudinal microdata indicates that real wages have been substantially procyclical since the 1960s (red dash line above). We show that the true procyclicality of real wages is obscured in aggregate time series because of a composition bias: the aggregate statistics are constructed in a way that gives more weight to low-skill workers during expansions than during recessions.

The article on JSTOR, or a slightly older NBER WP version.

The PSID data for all men and women in the balanced sample.

The OSIRIS programs for data construction and regressions.

Average real wage growth for the balanced panel of men is reported in NBER Working Paper 4202 (and can be calculated from the balanced panel available above). The time dummies used to construct the cyclicality of the real wage for the unbalanced panel (see p.13 and Table II) are:

```                        Avg PSID Real Wage Changes
(1)     (2)     (3)     (4)     (5)
Unemployment    Men     Men     Women   Women   Every1  Ch Log
Year    Rate    Change  Unwghtd Wghtd   Unwgtd  Wghtd   Wghtd   BEA Wage
1968    0.036   -0.002  0.10317 0.08416 0.09705 0.07486  0.04698  0.01480
1969    0.035   -0.001  0.08546 0.07610 0.09111 0.08776  0.05281  0.00556
1970    0.049    0.014  0.07924 0.06142 0.04743 0.04317  0.04501  0.00604
1971    0.059    0.010  0.06820 0.05035 0.06136 0.06061  0.03019  0.00863
1972    0.056   -0.003  0.08741 0.08004 0.06191 0.05594  0.06469  0.01315
1973    0.049   -0.007  0.07308 0.06733 0.06478 0.07892  0.03482  0.00771
1974    0.056    0.007  0.07035 0.05688 0.06722 0.05296  0.02019 -0.00461
1975    0.085    0.029  0.01755 0.01112 0.03955 0.05480 -0.00687  0.00410
1976    0.077   -0.008  0.06820 0.06728 0.05524 0.05601  0.04071  0.01157
1977    0.071   -0.006  0.07290 0.06343 0.04437 0.05064  0.03954  0.00640
1978    0.061   -0.010  0.07335 0.06556 0.06284 0.06267  0.00480  0.00824
1979    0.058   -0.003  0.06738 0.05335 0.07603 0.07203  0.03933  0.00048
1980    0.071    0.013  0.04567 0.05153 0.03535 0.03896  0.02264 -0.00144
1981    0.076    0.005  0.02847 0.02628 0.02723 0.02945  0.02263  0.00144
1982    0.097    0.021  0.02531 0.01882 0.05331 0.05432  0.01431  0.00571
1983    0.096   -0.001  0.05539 0.06441 0.06645 0.07673  0.05374  0.00047
1984    0.075   -0.021  0.07858 0.07981 0.03092 0.03628  0.05667  0.00094
1985    0.072   -0.003  0.05323 0.05316 0.09000 0.08601  0.03694  0.00516
1986    0.070   -0.002  0.06091 0.06990 0.06311 0.06574  0.01448  0.01421
1987    0.062   -0.008  0.05155 0.05023 0.08302 0.09988  0.03882  0.00135

Regress chlnW(n) = b(0) + b(1) trend + b(2) ChUnemp + e

b(2)-hat recovers resutlts for Tab II col 4 using chlnW(1) and
Table II col 6 using chlnW(3).  These series (de-meaned) are graphed
below.  The graph at the top of the page displays the series chlnW(5).

```