Specification Checks Implementation of the Estimators

elasticity of about 0.25 in Column 2: Most models of labor supply suggest that the income effect of increases in wife’s wage do not have a positive effect on husband’s hours. As for wives, greater nonlabor income is correlated with lower hours.

C. Specification Checks

1. Biases from Endogenous Marriage The analysis so far allows for assortative mating as a fixed effect. However, problems would arise if this parameterization is incorrect and there have been changes in match- ing patterns between 1980 and 1990. Changes in matching patterns can be considered as consisting of a changes in the composition of individuals who choose to get mar- ried and b changes in the spouses chosen by people who are married. It is well known that there has been a decline in the proportion of people who are married during this time period. However, one can study the observed characteristics of married and single individuals in 1980 and 1990. Proportion Married Sex Education 1980 1990 Male High school dropout 0.72 0.62 Male High school diploma 0.71 0.66 Male College degree 0.73 0.72 Female High school dropout 0.68 0.62 Female High school diploma 0.73 0.69 Female College degree 0.68 0.68 Devereux 709 Table 3 Estimates from Differenced Labor Supply Functions Men WLS G2SLS UEVE WLS G2SLS UEVE 1 2 3 4 5 6 Dependent Variable: LogAnnual Hours Logwife’s wage 0.180 0.253 0.232 0.005 0.074 0.001 0.027 0.041 0.050 0.026 0.061 0.129 Loghusband’s wage 0.046 0.054 0.074 −0.070 −0.058 −0.001 0.027 0.038 0.047 0.024 0.041 0.064 Nonlabor income −0.168 −0.269 −0.282 − 0.111 −0.314 −0.250 10,000 0.043 0.060 0.071 0.046 0.096 0.236 Husband and wife No No No Yes Yes Yes indicators Standard errors are calculated using Huber-White covariance matrix. All specifications estimated in differences. Region indicators are included in all regressions. The table shows that the proportion married has declined for both sexes with the biggest decline for individuals with less education. Thus, it is possible that the unob- served characteristics of married people have changed over time. Because the table suggests that these changes may not be similar across groups, estimates from specifi- cations without husband and wife type controls may be biased. However, even if changes in unobservables differ across groups, the estimates from specifications with husband and wife type controls will still be consistent if unobserved changes are sim- ilar within groups across regions. The second possibility is that even among married people, the choices of spouse have changed over time. It should be noted that certain types of changes are allowed for in the empirical work. The specification with husband and wife controls allows the composition within husband-wife groups to change while still identifying the labor supply elasticities using regional variation. There are differences in the proportion of matches of different types between 1980 and 1990. However, these are largely explained by the fact that there are fewer matches involving high school dropouts due to the increase in educational attainment of the population. I have also done a further specification check. Rather than use the actual wage change of husbands, I calculate husband wage changes for each woman by weighting the wage changes of different types of men where the weights are deter- mined by the distribution of husband types for that type of woman in 1980. I then compare this wage change to the average wage change of husbands for that woman type which is affected by changes in matching between 1980 and 1990. I find that the averages are almost identical. Thus, it appears that there have not been large changes in marital matching between 1980 and 1990. 2. Taxes In the analysis so far, the wages used have been gross pre-tax wages. However, the- ory suggests that labor supply decisions should depend on marginal net after-tax wages. Modeling taxes in this context is extremely difficult to do in a rigorous fash- ion because the hours decisions of individuals should depend on the marginal tax rate at each possible level of hours that they could choose to work. The marginal tax rate at the level of hours actually worked is endogenous to the labor supply choices made See Blundell and MaCurdy 1999; Heim and Meyer 2001 for detailed discussions about the difficulties of modeling labor supply with taxes. For these reasons, it is impractical to deal with taxes in a thorough fashion in this paper. Instead, in this sec- tion, I check whether my results are robust to using post-tax rather than pre-tax wage rates. Individuals pay federal, state, and payroll taxes on income. I compute the marginal federal tax rate for each couple in 1979 and 1989 using the Federal Income Tax Tables for “Married Filing Jointly” with the standard deduction. Similarly, the marginal state tax rates are calculated using the state income tax rate schedules. 19 I add the federal and state marginal tax rates to the marginal rate from the payroll taxes to get the mar- ginal tax rate from all sources. Given that income may be misreported to the Census 19. The state income tax rate schedules come from the World Almanac and Book of Facts 1979, 1980, and 1990. The Journal of Human Resources 710 or the IRS or to both and that complexities like itemized deductions are ignored in the calculations, the marginal tax rate facing any particular individual is an estimate of the true marginal tax rate faced. Hopefully, this source of error is largely averaged out by the grouping estimator. The marginal tax rate facing individuals averages 42 percent in 1979 and 34 percent in 1989. The decline is chiefly the result of the reduced federal tax rates implemented in the Tax Reform Act of 1986. The marginal after-tax wage rate is calculated as the wage multiplied by 1- τ where τ is the marginal tax rate. The results using after tax wages for husbands and wives are very similar to the results when pre-tax wages are used. The G2SLS own-wage elasticity is 0.85 0.06 for women 0.06 0.11 when husband and wife type indicators are included; the G2SLS cross-wage elasticity for women is −0.30 0.06 −0.41 0.07 when husband and wife indicators are included. The own and cross-wage elasticities for husbands are small and very similar in magnitude to the estimates using pre-tax wages. Thus, it appears that using after-tax wages instead of pre-tax wages does not change the con- clusions of the analysis.

VII. Analysis of Wages and Participation of Women