Journal of Public Economics 88 (2004) 1567 1588 www.elsevier.

com/locate/econbase

The incidence of personal income taxation: evidence from the tax reform act of 1986
Jeffrey D. Kubik *
Department of Economics, 426 Eggers Hall, Syracuse University, Syracuse, NY 13244, USA Received 21 October 2000; received in revised form 13 August 2002; accepted 4 September 2002

Abstract This paper studies the short-run incidence of personal income taxation in the US by examining how the wage structure shifted after the Tax Reform Act of 1986. I calculate the marginal tax rate for the median worker of each occupation during the 1980s to determine the professions whose workers were most affected by the tax change. When controlling for trends over time in the wage structure by occupation, I find that individuals in occupations that experienced large decreases in their median marginal tax rates received lower pre-tax wages after 1986 as the number of workers and the hours worked in those professions increased. The results suggest that consideration of the wage effects of tax reforms is important when using these reforms to identify behavioral parameters and evaluating their distributional consequences. D 2002 Elsevier B.V. All rights reserved.
Keywords: Personal income taxation; Tax Reform Act

1. Introduction Most analyses of the short-run incidence of the personal income tax are very simple. The supply and demand shifts induced by changes in income tax policy are generally assumed not to affect the pre-tax wages of workers. Because workers then bear the full burden of the tax, investigations of the distributional consequences of personal income tax reforms need only be concerned with the direct effects of tax rate changes on workers. Redistribution from wage movements caused by tax policy innovations can be ignored. But basic models of the labor market suggest that this assumption is potentially strong. In a simple framework, the pre-tax wages and incomes of workers are endogenously
* Tel.: +1-315-443-9063; fax: +1-315-443-1081. E-mail address: jdkubik@maxwell.syr.edu (J.D. Kubik). 0047-2727/$ - see front matter D 2002 Elsevier B.V. All rights reserved. doi:10.1016/S0047-2727(02)00182-2

1568

J.D. Kubik / Journal of Public Economics 88 (2004) 15671588

determined by tax policy. For an economy with workers of different skills, the pre-tax wage of each skill will depend on the quantity of labor supplied at all skill levels because the supply of one skill affects the marginal productivity of each type of labor. If a tax reform alters the quantity of labor supplied of at least one of the skill levels, the wage structure will change. The magnitude and direction of this shift depends on the size of the labor supply response to the tax change and the structure of the production process. The goal of this paper is to measure whether changes in the personal income tax affect the US wage structure, using evidence from the Tax Reform Act of 1986. First, I estimate how the marginal tax rate of the median worker in each occupation changed because of the tax reform. Then, I track the wages of workers by occupation before and after 1986. Under simple assumptions about the labor supply response of workers to tax changes and the specification of the aggregate production function, the size of the wage change for workers in an occupation caused by the tax reform will depend on the magnitude of the marginal tax rate change of workers in that occupation compared to workers in other jobs. Therefore, I relate the wage changes of workers in occupations after 1986 to the tax rate change of the median worker in that occupation caused by the new income tax policy. The Tax Reform Act of 1986 is particularly well suited for studying the effect of tax changes on the US wage structure. The legislation significantly lowered marginal tax rates for high-income individuals while low-income workers were relatively unaffected by the reform. To the extent that workers in the same occupation earn similar incomes and therefore face similar tax circumstances, workers in high-income professions experienced large marginal tax rate declines compared to workers in low-income jobs, providing useful variation for identifying the effect of the tax change on wages by occupation. Using data from several years of the Current Population Survey, I find that the pre-tax wages of occupations whose workers faced sharp declines in marginal tax rates after the 1986 tax reform fell compared to the wages of workers in other occupations when controlling for trends in the wage strucure by occupation over the 1980s. The estimates imply that an occupation with a median worker who experienced a 10 percentage point marginal tax rate decline after the 1986 reform, about the largest median decline caused by the tax change, encountered between a 1.3 and 2.5% decrease in its median wage. Also, the number of hours worked by individuals, especially by women, in these occupations increased significantly compared to other occupations, suggesting that shifts in the relative supply of labor caused these relative wage changes. The results indicate that the gains to high-income workers of lower marginal tax rates after the Tax Reform Act of 1986 were partially offset by declines in the pre-tax wages of workers in high-income occupations. Therefore, this work suggests that it is important to take into account wage effects when evaluating distributional issues of income tax policy, both when reviewing the effects of a specific tax reform and in an optimal tax setting. To evaluate how personal income tax changes influence an individual, consideration must be given not only to the particular tax circumstances of the individual but also to how other people in the labor market, especially workers in the same occupation, are affected by the reform. Also, there is a large literature that attempts to use tax reforms to identify behavioral parameters such as labor supply elasticities. Much of this work assumes that the tax reform does not alter the wage distribution of workers when calculating these parameters. My results indicate that such an assumption might be problematic in many situations.

J.D. Kubik / Journal of Public Economics 88 (2004) 15671588

1569

The paper proceeds as follows. In the next section, I discuss some of the theoretical work on the incidence of the income tax and consider some issues of how the Tax Reform Act of 1986 affected the labor market. Section 3 describes the data and empirical strategy for estimating the effect of tax policy changes on the wage structure. The results of this analysis are presented in Section 4, and Section 5 concludes.

2. Preliminary considerations Much of the theoretical work on the role of the personal income tax in the labor market is examined within an optimal tax framework. A long and storied literature investigates the design of an optimal income tax system when the tax distorts an individuals choice between labor and leisure. However, most of this work assumes that the labor supply decisions of individuals influenced by tax policy do not affect the wage rates that those individuals face (see, for example, Mirrlees, 1971; Sheshinski, 1972; Atkinson, 1973). Feldstein (1973) is the first optimal tax study to consider a model in which the wages of workers are endogenously determined by their labor supply decisions. He derives the structure of the optimal linear income tax when production requires two types of labor: skilled and unskilled.1 Redistribution from wage changes occurs when tax policy affects the relative supply of skilled and unskilled labor.2 Stern (1982) and Stiglitz (1982) use a model with two types of labor similar to Feldstein (1973) to examine the optimal nonlinear income tax.3 The mechanism by which tax rate changes shift the wages of workers is the same in all of these optimal tax studies. The tax reform affects the labor supply decisions of individual workers of the two skill levels differently, altering the relative supply of skilled to unskilled labor. Given standard assumptions about the structure of the aggregate production function, the pre-tax wage increases for the type of labor that becomes relatively more scarce and falls for the labor that becomes relatively more abundant.4
1 Feldstein (1973) assumes that there is a fixed number of workers of each skill. Each worker decides how much labor to supply. Christiansen (1988) examines a somewhat similar model in which a fixed number of workers move between two occupations based upon economic incentives including tax policy. Workers make no other labor supply decisions. The wages of the two occupations depend on the relative supply of workers in those two jobs. 2 In Feldsteins linear tax framework, the redistribution from wage changes caused by tax policy reinforces the direct redistribution from the tax. Therefore, he concludes that the optimal linear income tax is very similar whether or not the wage effects are considered. However, Allen (1982) argues that the structure of the optimal linear income tax significantly changes if endogenous wages are considered when Feldsteins assumption of a Cobb-Douglas production function is relaxed. 3 Unlike the standard findings about the shape of the optimal non-linear tax schedule, they conclude that skilled (high wage) workers should face a negative marginal tax rate, and unskilled (low wage) workers should face a positive marginal tax rate. This structure is optimal because the negative marginal tax rate on skilled workers increases their labor supply, which then increases the wage of unskilled labor. 4 This is a description of how tax changes can affect the labor market in the short-run. In the long-run, a personal income tax reform might affect capital accumulation in the economy if individuals change their savings decisions. This change in capital accumulation could also affect the labor market. Therefore, the long-run incidence of the personal income tax can be much different than the short-run. This paper concentrates on the short-run. See Feldstein (1974) and Kotlikoff and Summers (1979) for discussions of tax incidence within a growth model.

1570

J.D. Kubik / Journal of Public Economics 88 (2004) 15671588

This can be illustrated with a simple one good model. Production of the good is carried out by skilled (LS ) and unskilled labor (LU ) with the production function: Q FLS ; LU The supply of skilled labor depends on the after-tax wage: LS LwS 2 1

The government decides to place a tax on skilled labor. For simplicity, assume that the supply of unskilled labor is fixed. Assuming that all of the good can be sold in the market at a price equal to one, the after-tax wage for skilled labor is: wS 1 FS LS ; LU where s is the tax rate. The wage for unskilled labor is: wU FU LS ; LU 4 3

By totally differentiating Eqs. (2) and (3) and assuming that there was no previous tax on skilled labor, the effect of the imposition of this new tax on the after-tax wage of skilled labor can be written: dwS d wS FSS LS S 1  wS 5

where gS is the skilled labor supply elasticity. By assuming constant returns to scale, this can be simplified to: dwS d w S u S 1  6

where r is the elasticity of substitution between skilled and unskilled labor, and u is the unskilled labor share u wU LU =Q. Eq. (6) demonstrates that the incidence of the tax depends importantly on the elasticity of substitution in the production process and the labor supply elasticity of skilled labor. When skilled and unskilled labor approach being perfectly substitutable (r l), then firms will not employ skilled labor if its pre-tax wage increases after the tax change, causing all of the tax to be borne by skilled labor. On the other hand, as skilled and unskilled labor become less substitutable, then skilled labor bears less of the tax, and the pre-tax wage of skilled workers will increase after the tax change. The more production can adjust to the tax, the more skilled labor bears it. If skilled labor supply is perfectly inelastic (gS 0), all of the tax will be borne by skilled labor. As skilled labor supply becomes more elastic, skilled workers cut back their labor after the tax, causing the after-tax wage of skilled workers not to fall by the full

J.D. Kubik / Journal of Public Economics 88 (2004) 15671588

1571

amount of the tax. The more that workers can adjust to the tax with their labor supply, the less of the tax they bear. Therefore, except in extreme circumstances, the model suggests that the pre-tax wages of workers in an occupation affected by a tax increase will increase as workers adjust their labor supply. The larger the tax change, the greater that pre-tax wages will change. This framework can be used to roughly predict how personal income tax changes from the Tax Reform Act of 1986 affected the wage distribution. The legislation simultaneously lowered the marginal tax rate of high-income individuals while broadening the tax base. Together, these two tax rule changes were designed to be approximately distributional and revenue neutral. The top marginal tax rate was lowered from 50 to 28%, a decline of over 40%. To balance this tax decrease for high-income individuals, many of the exclusions and deductions often used by these taxpayers, such as the 60% capital gains exclusion and some deductions for IRAs and interest expenses, were eliminated. Most lower income individuals were relatively unaffected by the reform. Marginal tax rates in lower brackets were similar before and after the new tax law, and few low-income taxpayers before 1986 claimed the exclusions and deductions that were discontinued after the tax changes.5 If there is a high correlation in the training and earnings potential of people in the same occupation, then the Tax Reform Act of 1986 affected people who work in the same job similarly. Workers in high-skilled occupations experienced sharp declines in marginal tax rates because they earned high incomes, and most workers in low-skilled positions experienced only small changes in marginal tax rates. For example, most lawyers tax rates would have fallen compared to the marginal tax rates of most short-order cooks after 1986.6 If these changes in marginal tax rates caused by the Tax Reform Act of 1986 affected workers in various occupations differently, then the tax reform might have altered the wage structure of workers across occupations. As suggested by the model, high-skilled professions that experienced increases in the quantity of labor supplied because of declining marginal tax rates would have lower pre-tax wages compared to workers in other occupations. This wage decrease would have partially counterbalanced the increase in after-tax wages that workers in those professions enjoyed because of the 1986 tax change.

5 An increase in the personal exemption removed a number of very low-income households from the tax rolls. This tax change was the most important aspect of the 1986 reform for low-income individuals. See Hausman and Poterba (1987) for a more complete discussion of how the 1986 tax law changes affected different individuals. 6 Of course, a married couples marginal tax bracket depends on the combined income of both spouses. Many people in low-skilled occupations might be married to a person in a high earning job, lessening the withinoccupation correlation of tax status. However, it is well known that the educational level of spouses is highly correlated (see, for example, Pencavel, 1998). Such assortative mating suggests that a person in a low-skilled occupation is more likely to marry another person in a low-skilled occupation. Also, the tax status of a person depends on both labor and non-labor income. Many people in low-skilled jobs might have large earnings from other sources, again decreasing the correlation of tax status of people in the same occupation. However, because income from labor and non-labor sources tends to be highly correlated for the non-elderly, this again would not be typical.

1572

J.D. Kubik / Journal of Public Economics 88 (2004) 15671588

This prediction depends critically on the assumption that changes in marginal tax rates have an effect on the labor supply decisions of individuals. This effect can come in two forms. First, individuals can alter their participation and hours of work decision. There is a large literature that measures the effect of changes in net wages on labor force participation decisions. Although there is still some controversy in this literature, most of the work examining men shows that wage changes have little impact on their work decisions (see, for example, Pencavel, 1986; Triest, 1990). Therefore, it is unlikely that labor force participation changes by men could cause large wage changes after the 1986 tax reform. However, much of the literature on the effect of net wage changes on the labor supply of women, especially married women, finds a much larger effect (see, for example, Hausman, 1985; Heckman, 1993; Eissa, 1995; Blundell et al., 1998). Changes in the hours worked or participation of women in high skilled occupations after the Tax Reform Act of 1986 are a much better candidate for altering the wage distribution. Second, a tax reform can cause a worker to switch occupations. Decreasing the rate of the highest tax bracket might encourage a worker to move to a higher paying position. For example, in a high tax regime, a lawyer might decide to teach (because there are many nonpecuniary benefits to teaching). However, when marginal tax rates fall, the opportunity cost of teaching increases and the person might decide to devote more time to her law practice. In this example, the tax reform does not necessarily change the aggregate hours worked by the individual, but the quantity of labor supplied to different occupations changes. There is not much of a literature on how tax policy affects occupation choice, so it is not possible to gauge how large of an effect net wage changes have on occupation switching. However, along with changing the labor force participation decisions of women, this is another potential means by which tax changes can affect the wage distribution. This background discussion of how to think about the incidence of personal income taxation illustrates several issues that need to be addressed in the subsequent empirical work. First, I need to show that there were sizeable differences in how the Tax Reform Act of 1986 affected the marginal tax rates of workers across occupations. If workers in all occupations were affected similarly by the reform, then a shift in relative wages would be unlikely. Second, redistribution from tax reforms caused by wage changes occurs because the relative supply of labor in different occupations changes. If I find that the wage structure changed after the Tax Reform Act of 1986, I need to relate that shift to changes in relative labor supply across occupations to confirm that there is a link between the tax reform and the wage structure. Now I turn to a discussion of the data I use to gauge how the tax changes affected wages of workers by occupation and my empirical strategy for measuring the relationship.

3. Descriptions of the data and the empirical strategy This section explains the methods and data used to measure the role of the Tax Reform Act of 1986 in shifting the US wage distribution. First, I specify the calculations for determining how the tax changes affected workers in different occupations. Then, I describe the measurement of wages by occupation over time. Finally, I present the regression framework used to relate the tax changes to the wage changes by occupation.

J.D. Kubik / Journal of Public Economics 88 (2004) 15671588

1573

Fig. 1. Marginal tax rate change distribution, 1985 to 1988.

3.1. The data on the effect of the 1986 tax reform on marginal tax rates on workers by occupation To estimate how the marginal tax rates of individuals changed because of the 1986 tax reform, I calculate how different tax rules over time affected a set of households that represent the population of working age adults. That is, I take a representative sample of individuals and compute what their tax rate would have been before the 1986 reform and what it would have been after the reform based on their income, demographic characteristics, and the tax laws. My sample of individuals is from the 1987 March Current Population Survey (CPS). Containing a large amount of demographic information about a nationally representative sample of US households, the 1987 March CPS includes detailed data on the 1986 annual income and occupation of respondents. For each adult between the ages of 18 and 65 in the 1987 March CPS, I calculate the marginal tax rate the person would have faced in 1985 and 1988 based on his or her family income and other characteristics. The tax calculator I use takes into account an individuals marital status, the number of dependent children in the family, and total family income. Given the parameters of the US tax code for 1985 and 1988, it then produces the estimated marginal tax rate of the individual for those years. It assumes that all families take the standard deduction and married couples file jointly.7 I adjust the 1986 income amounts reported in the 1987 March

7 The tax calculator used to make these marginal tax rate estimates is from Dickert-Conlin and Chandra (1999). I use this calculator to compute the marginal tax rate of each adult in a family. Therefore, both a husband and wife will be in the sample even though they have the same marginal tax rate.

1574

J.D. Kubik / Journal of Public Economics 88 (2004) 15671588

Fig. 2. Median marginal tax rate change distribution by occupation, 1985 1988.

CPS for each family by the Consumer Price Index to convert total family income into 1985 and 1988 dollars for the tax computations.8 Fig. 1 presents the distribution of the estimated changes in marginal tax rates for individuals from 1985 to 1988 because of the Tax Reform Act of 1986 using this tax calculator and the sample of adults from the 1987 March CPS. There are large differences in how the tax law changes affected individuals. The estimates suggest that almost onehalf of the 1987 March CPS respondents would have enjoyed a declining marginal tax rate because of the reform, with almost 13% of the sample experiencing a decrease of over 10 percentage points. Approximately 25% of the CPS sample would have experienced no change while the rest would have encountered some increase in marginal tax rates.9 Because there is large variation in how individuals were affected by the Tax Reform Act of 1986, I now turn to see whether workers in the same occupations tended to be influenced similarly by the tax changes. Only if there are significant differences across professions in how the tax reform altered marginal tax rates, could there be any change in the wage distribution after 1986. I calculate the median marginal tax rate change from 1985 to 1988 by occupation at the three-digit level.10 Fig. 2 displays the distribution of

Instead of using the 1987 March CPS for all of the data to calculate tax rates, I could use data from the CPS collected for the year I am computing the tax rate. For example, I could use the 1986 March CPS (which collects 1985 income information) to calculate the 1985 marginal tax rates and the 1989 March CPS (with 1988 income data) for the 1988 tax rates. However, changes in the tax rates from 1985 to 1988 would then depend not only on the tax law changes but also from changes in the composition of the samples (possibly in response to the tax law changes). To isolate the effect of the tax reform, I calculate tax rates for all years from the same sample of households. 9 This breakdown is very similar to the estimated changes in marginal tax rates reported in Hausman and Poterba (1987) using information from the 1983 IRS Public Use Tax Return data file and the TAXSIM model of the National Bureau of Economic Research. 10 The CPS uses the three-digit Occupational Classification from the 1980 Census from 1983 to 1991. I calculate the median tax rate change for 468 different occupations. Several other studies examine labor markets at this level of occupation detail using the CPS. See, for example, Gruber and Krueger (1991).

J.D. Kubik / Journal of Public Economics 88 (2004) 15671588

1575

median marginal tax rate changes by occupation using the 1987 March CPS data. There are large differences across workers by occupation in the tax rate change caused by the tax reform, but, not surprisingly, the distribution of median marginal tax rate changes across occupations is not as widely dispersed as the distribution of changes across individuals. About 65% of the occupations experienced a decrease in median marginal tax rates, and about 20% encountered higher median tax rates. Returning to my previous example about lawyers and short-order cooks, I estimate that the median decrease in marginal tax rate for lawyers after 1986 is 10 percentage points, about the largest decline for any occupation, while the median short-order cook experienced a 4 percentage point increase in marginal tax rate.11 3.2. The data on wage changes of workers by occupation With these estimates of how the Tax Reform Act of 1986 affected workers in different occupations, I now construct measures of how the wages of workers in these occupations changed over time. The Merged Outgoing Rotation Group Files of the CPS are an ideal source for this wage data. The data set contains wage information on a large representative sample of the US population.12 Also, because the occupation codes are the same in the outgoing rotation group data as the March CPS, it is simple to create a merged data set of the wage and marginal tax rate information by occupation. To obtain several years of wage information before and after the Tax Reform Act of 1986, I use this CPS data annually from 1983 to 1990 to calculate the median wage of workers by occupation.13 I could just compute the median hourly wage of all adults in the CPS data by occupation for each year during this period. However, I am worried that changes in wages of workers by occupation over time might be caused by changes in the composition of the workforce and not by the wage effects of tax policy. For example, if declining marginal tax rates cause more married women to enter the labor force in highincome occupations, then the median wage of those occupations could decline because a large number of inexperienced workers are entering that profession and not because of changes in the wages of continuing workers. To avoid this measurement issue, I only examine the wages of a group of workers that has the most stable attachment to the labor force: men between the ages of 25 and 55. Assuming that these men do not alter their labor force participation in response to tax policy changes, shifts in median wages of these workers by occupation after the tax reform should reflect the effects of the changes in the labor supply of workers to these occupations and not changes in the composition of the workers in the occupation.
I measure the effect of the tax reform using the change in the median tax rate of the workers in an occupation instead of, for example, the change in the average tax rate to avoid the possibility that my measure is substantially affected by a few outliers in an occupation. In my empirical work, I show how sensitive my results are to using averages instead of medians. 12 I use the NBER CPS Labor Extract version of this outgoing rotation group CPS data. With wage data on almost 200 000 individuals per year, it is the best data set for calculating wages of three-digit occupations over time. 13 Because the wage data in the CPS is top-coded, I avoid the issue of how to deal with the missing upper tail of the wage distribution by using medians instead of means.
11

1576

J.D. Kubik / Journal of Public Economics 88 (2004) 15671588

Using this group of men, I create a data set of 3744 observations of the median wages of workers by occupations (468 occupations8 years). I then merge the information on the marginal tax rate changes of workers by occupation to these wage data. By linking these wage and tax data, I create a panel data set to examine how wages changed because of the Tax Reform Act of 1986. Table 1 presents some descriptive statistics of this data set. The table shows the variation over time and across occupations in the log median wage of 25 55-year-old men by occupation, the median marginal tax rate of workers by occupations, and the change in marginal tax rates of workers by occupation because of the tax reform. 3.3. Regression framework My regression specification should be designed to measure whether the relative wage structure by occupation changed after 1986. The model indicates that occupations whose workers were on average directly affected by a tax decrease because of the Tax Reform Act of 1986 should have experienced a substantial decrease in pre-tax wages compared to workers in other occupations. This suggests a difference-in-difference type estimation strategy where I compare the wages of workers before and after 1986 in occupations greatly affected by the tax changes to the wage changes to the workers of other professions. The specification is, logWagei;t  b1 MTR Changei After 1986t Occupation Dummiesi Year Dummiest Occupation Dummiesi Linear Trendt ei;t log(Wagei; t) is the log wage of the median man between the age of 25 and 55 in occupation i in year t. MTR Changei is the estimated median change in the marginal tax rate of adults between 1985 and 1988 in occupation i, and After 1986t is an indicator that the observation is after 1986. Occupation Dummiesi is a set of dummy variables corresponding to the 468 different occupations, Year Dummies t is a set of year dummies, and Occupation DummiesiLinearTrendt is a set of linear trends for each occupation. ei;t is an error term. The occupation dummies control for fixed differences in wages across jobs, and the year dummies control for aggregate trends in wages. The separate linear trends by
Table 1 Descriptive statistics of the data set Mean (1) Log(median wage of occupation) Median marginal tax rate of occupation Change in the median marginal Tax rate of occupation (1985 1988) 2.47 [0.32] 20.92 [6.88] 2.12 [4.05] 10th Percentile (2) 2.04 15.00 8.00 50th Percentile (3) 2.48 18.00 1.00 90th Percentile (4) 2.85 28.00 3.00

Notes: data on 468 occupations from 1983 to 1990 (3744 observations). Standard deviations are in brackets. Observations are weighted by the number of 25 55-year-old men in the occupations in 1986.

J.D. Kubik / Journal of Public Economics 88 (2004) 15671588

1577

occupation control for any constant changes over time in the relative wage structure by occupation. For example, there is a large literature that documents significant changes in the wage structure during the 1980s, most notably the large increase in the relative wages of highly educated workers (see, for example, Bound and Johnson, 1992; Levy and Murnane, 1992; Katz and Murphy, 1992; Murphy and Welch, 1992). This literature surveys the potential causes of these wage shifts, usually concentrating on the effects of changes in the supply and demand for different types of labor over this time period.14 Because most individuals in high tax brackets tend to be highly educated and workers in lower tax brackets tend to have less education, changes in the relative wages between high-income and low-income occupations after 1986 might be caused by changes in the demand for skill skill instead of tax policy. If the Tax Reform Act of 1986 caused the wages of workers in high-income occupations to fall, then this trend in the returns to education might mask this effect. Failure to properly account for the effect of changes in the returns to education by controlling for trends in the wage structure will bias my estimates toward not finding that workers in occupations that experienced a large decline in marginal tax rates experienced lower wages after 1986. When presenting my empirical results, I will show how the inclusion of these linear trends by occupation affects my estimates, examining if they affect the results in the predicted direction. After adding these controls, the coefficient of interest is b1, which measures how the log median wage of workers in an occupation after the 1986 tax reform is affected by the marginal tax rate change the workers in that occupation experienced. Put differently, it gauges whether the wages of workers in occupations that experienced large reductions in marginal tax rates changed relative to a linear trend after 1986 differently than the wages of other occupations. Several important assumptions are needed for this empirical strategy to work. First, it is necessary to assume that the change in the marginal tax rate of the median worker in an occupation captures the effect of the tax reform on the individuals who are considering entering or changing their labor supply in that occupation. If the marginal worker in a profession is much different than the median in terms of tax status, then this empirical approach will not accurately measure how the Tax Reform Act of 1986 altered the relative quantity of labor supplied to different professions and the wages of workers in those occupations. Therefore, I need to assume that because the education and skills necessary to be in an occupation are similar for everyone, the tax status of the median worker in an occupation is close to the tax status of the average person considering changing her labor supply in a profession. Another assumption of this identification strategy is that the 1986 tax reform affected wages only because of shifts in the relative supply of workers in different occupations. However, a tax reform might also alter how an employee wants to be compensated by an employer, possibly changing her wage. Feldstein (1995) argues that the Tax Reform Act of 1986 caused high-income workers to shift their compensation away from non-taxable fringe benefits and toward taxable compensation (including wage income), increasing

14 Both Katz and Murphy (1992) and Murphy and Welch (1992) find that changes in the wage structure in the 1980s are consistent with a linear trend in the demand for skilled versus unskilled labor.

1578

J.D. Kubik / Journal of Public Economics 88 (2004) 15671588

taxable income. To the extent that this compensation shifting occurs when marginal tax rates change, there is a negative relationship between tax rates and taxable wages of workers not caused by shifts in the labor market. If this tendency for high-income workers to adjust their compensation to changes in marginal tax rates is large, then it will also bias my results against finding a positive effect of marginal tax rates on the wages of workers by occupation.

4. Results This section presents the estimates of the effect of the Tax Reform Act of 1986 on the wage structure. First, I describe how changes in marginal tax rates after 1986 affected the median wages of workers by occupation. Next, I show how the tax reform altered the quantity of labor supplied to different occupations. Then, I explore the effect of including additional controls in the regression specification, concentrating on alternative methods of controlling for changes over time in the relative demand for skilled versus unskilled workers. Finally, I examine some specification checks to test the robustness of my results. 4.1. The effect of the tax reform on wages Table 2 shows the estimates of the effect of changes in marginal tax rates on wages of workers by occupation. Column (1) displays the results of the regression specification of Eq. (7).15 The coefficient on the change in the median marginal tax rate after 1986 is positive and statistically significant from zero, indicating that workers in occupations that enjoyed a decrease in marginal tax rates because of the tax reform experienced decreasing wages compared to other occupations. The size of the coefficient indicates that the median wage of workers in an occupation that had a median marginal tax rate decline of 10 percentage points fell by about 2.5% after 1986. To determine the importance of adding linear time trends by occupation to the regression specification, I present the regression results omitting those variables in column (2) of Table 2. The coefficient on the change in the median marginal tax rate after 1986 is now negative, although its magnitude is very small. Removing the linear trends has the predicted effect on the results under the assumption that the demand for skilled workers is increasing throughout the 1980s. Increasing wages of workers in high education occupations because of these trends in the demand for workers of different skills mask the effect of the tax reform on the wages of workers in high skilled occupations. Accounting for these trends (column (1)) removes the downward bias of the coefficient of changes in marginal tax rates on wages. My estimates when linear trends are removed from the specification are very similar to Moffitt and Wilhelm (2000), who find using individual data that the wages of workers who experienced a substantial decline in their tax rates after 1986 experienced rising pre-tax wages. Their work also does not control for trends in wages.
15 The regression is weighted by the population of 25 55-year-old men in each occupation in 1986. This population is estimated using the sample weights of the 1986 CPS merged outgoing rotation group file.

J.D. Kubik / Journal of Public Economics 88 (2004) 15671588 Table 2 The effect of changes in marginal tax rates on the median wages of workers by occupation (1) Change in the median marginal tax Rate of occupationafter 1986 indicator Year effects Occupation effects Linear time trends for occupations Observations 0.2514 (0.1014) Yes Yes Yes 3744 (2)

1579

0.0656 (0.0547) Yes Yes No 3744

Notes: the dependent variable is the log median wage of occupations by year of 25 55-year-old men. Standard errors are in parentheses. The regressions are weighted by the number of men in each occupation in 1986.

The regression results of the effect of marginal tax changes from the Tax Reform of 1986 on the median wages of workers by occupation are consistent with the predictions of models that consider the wage effects of tax reforms when controls for trends in the wage structure are included. These findings suggest that it is important to consider not only how an individual is directly affected by a potential tax reform from changing marginal tax rates, but also how the reform affects the labor market the person works. For example, consider a lawyer who, due to unusual family or business circumstances, is in a low tax bracket. The Tax Reform Act of 1986 would not have greatly affected his marginal tax rate. However, because most of his co-workers experienced large increases in after-tax wages, his wage would have fallen. He enjoyed none of the direct benefits of the tax change but suffered from the wage effect. On the other hand, the short-order cook who, again for odd family or business reasons, was in a high tax bracket would enjoy a lower marginal tax rate after the 1986 tax changes and would also experience a higher wage because most of his co-workers marginal tax rates rose after the reforms. He benefits from both the direct tax effect and the wage effect. These results also have important implications for empirical work that uses tax reforms to identify behavioral parameters. Many studies use changes in the tax system to obtain plausibly exogenous variation in after-tax wages of workers. Well-known work has used this wage variation to measure labor supply elasticities (see Eissa, 1995, 1996), implicitly assuming that the incidence of the tax changes is completely borne by the individual workers. Heckman (1996) criticizes this empirical strategy for several reasons, including the fact that it ignores potentially important trends in the wages of workers affected by the tax reform when measuring these labor supply elasticities. My results suggest that these wage trends are empirically important and should be considered when calculating how workers are affected by a tax reform. 4.2. The effect of the tax reform on employment If the wage movements that I measure are the effects of the 1986 tax reform, then I should observe shifts in the relative supply of labor of workers across occupations consistent with these wage changes. Labor supply should increase in occupations that experienced a large decrease in marginal tax rates after 1986. Using the merged outgoing rotation group data from the CPS, I calculate the total quantity of labor supplied by occupation over time; my measure is the aggregate number of hours worked in an

1580

J.D. Kubik / Journal of Public Economics 88 (2004) 15671588

occupation in a week each year.16 Besides calculating the total labor supply by occupation using this measure, I can also compute it for different demographic groups by occupation. Because any significant changes in labor supply due to tax changes are expected to be from women, I calculate the labor supply measure separately for men and women. To gauge the effect of marginal tax rate changes on labor supply by occupation, I estimate a regression equation that is similar to my previous specifications, but now the dependent variable is my labor supply measure. logHoursi;t 1 MTR Changei After 1986t Occupation Dummiesi Year Dummiest Occupation Effectsi Linear Trendt ei;t 8

log(Hoursi; t) is the log of the aggregate weekly hours worked by those in occupation i in year t, and the other variables are the same as before. The coefficient of interest, b1 , measures whether the number of hours worked in occupations that were greatly affected by the tax reform increased or decreased relative to other professions.17 The estimates of the regression specification of Eq. (8) are presented in Table 3. First, I examine how marginal tax rate changes affect the hours worked of all workers in column (1).18 The coefficient on the change in the median marginal tax rate for workers in an occupation after 1986 is negative and statistically significant from zero, implying that the aggregate hours worked of those in professions that enjoyed lower marginal tax rates increased compared to other occupations. The size of the coefficient suggests that hours worked in an occupation that experienced a 10 percentage point decrease in its median marginal tax rate grew by about 3.2% compared to the workers in other occupations.19 Because most of the literature on the effect of wages on labor supply indicates that women are more sensitive to tax changes than men, I next investigate how the 1986 tax changes affected the labor supply of women and men separately. Column (2) displays the effect of changing marginal tax rates on the labor supply of women.20 The tax reform had a large effect on the number of women participating in professions that experienced a large
16 To calculate the aggregate hours worked in an occupation in a year, I use the usual number of hours that respondents work in a week and the CPS sample weights. 17 Because I am not taking into account the wage effects of the tax change when parameterizing the tax change on workers in an occupation, this coefficient should not be interpreted as some sort of labor supply elasticity. is simply a measure of whether the change in the labor supply of workers in an occupation because of the tax reform is consistent with the movements in their wages. 18 This regression is weighted by the total number of workers in each occupation in 1986 using the sample weights from the 1986 merged outgoing rotation group CPS file. 19 When the linear time trends by occupation are not included in the regression specification, the coefficient on the change in the median tax rate for workers in an occupation after 1986 is larger in absolute value than the estimate including these trends. There is still the same qualitative relationship; aggregate hours grew in occupations that experienced decreases in median marginal tax rates. 20 Because there are more male-dominated occupations than female-dominated occupations, there are some professions that have no women workers in the CPS some years. Therefore, the sample size of the regressions examining women only is smaller than other estimations. In my sample, I only include occupations that have women workers in all 8 years of the sample. Including occupations that have women in the sample fewer than 8 years does not qualitatively change the reported results. The regression is weighted by the number of women in the occupations in 1986.

J.D. Kubik / Journal of Public Economics 88 (2004) 15671588

1581

Table 3 The effect of changes in marginal tax rates on the cumulative number of hours worked per week by occupation All (1) Change in the median marginal tax Rate of occupationafter 1986 indicator Year effects Occupation effects Linear time trends for occupations Observations 0.3235 (0.1052) Yes Yes Yes 3744 Women only (2) 0.4393 (0.1503) Yes Yes Yes 3064 Men only (3) 0.1061 (0.1239) Yes Yes Yes 3744

Notes: the dependent variable is the log of the number of hours worked per week in an occupation. Standard errors are in parentheses. The regression in column (1) is weighted by the number of all workers in occupations in 1986. The regression in column (2) is weighted by the number of women in occupations in 1986, and column (3) is weighted by the number of men in occupation in 1986.

decline in marginal tax rates after 1986. In contrast, the effect of marginal tax changes on the employment of men is shown in column (3). There is little effect of the tax changes on mens participation. The coefficient on the change in the median marginal tax rate after 1986 is small and not statistically different from zero.21 The differences in the results between men and women are comforting because they are consistent with the literature on the sensitivity of labor supply to wage changes. But they are also reassuring because they demonstrate that the composition of prime-age men working in different occupations does not respond to the tax changes. When using the wages of 25 55-year-old men to measure the wage effects of the 1986 tax reform, I needed to assume that their labor force participation decisions were not influenced by the reform. The results in Table 3 support this assumption. 4.3. Additional controls The results of the empirical work are consistent with the hypothesis that the 1986 tax reform altered the relative supply of workers in different occupations and changed the wage distribution. After controlling for trends over time, I find that workers in occupations that experienced a large decrease in marginal tax rates after 1986 increased the quantity of labor they supplied and experienced a decrease in wages. Now I need to address some concerns about my empirical strategy. One potential worry is that the composition of 25 to 55 year old men in an occupation is changing over time in a way that affects its median wage and is correlated with the changes in marginal tax rates after 1986 but not caused by those tax changes. For example, the median worker in high-skilled jobs might be getting younger on average after 1986 compared to other professions, causing the median wage of workers in those occupations to fall if there is a positive association between experience and wages.22
As an alternative measure of labor supply, I have used the total number of workers in an occupation instead of the total hours worked. When this labor supply measure is used, I obtain very similar results. 22 If these changes are constant over the 8-year period of the sample, then they are controlled for by the linear trend for each occupation in the regression specification. Changes in the composition of the prime-age sample of men are most problematic if there are changes in trends that start after 1986.
21

1582

J.D. Kubik / Journal of Public Economics 88 (2004) 15671588

To account for this potential problem, I control for the demographic composition of workers in occupations over the 8-year sample period in the regression specification. My strategy is to use a two-step estimation procedure. In the first stage, I regress an individuals log wage on a number of demographic characteristics. The sample for this regression is again men between the ages of 25 and 55 from the merged outgoing rotation group sample of the CPS. I use the residuals of this regression to calculate the median log wage of workers by occupations by year. Then my second stage regression is just Eq. (7) but with this new measure of workers wages by occupation. The CPS contains a wealth of demographic information about respondents. The attributes I use in the first stage wage regression are educational attainment, age, race and marital status. The regression equation is: logWagej;t  b1 High School Graduatej;t b2 Some College Educationj;t b3 College Graduatej;t b4 Some Graduate Eduationj;t b5 Aged 30 34j;t b6 Aged 35 39j;t b7 Aged 40 44j;t b8 Aged 45 49j;t b9 Aged 50 55j;t b10 Whitej;t b11 African Americanj;t b12 Marriedj;t ej;t 9

where log(Wagej; t) is the log wage of individual j in year t. High School Graduatej; t is a dummy for completing high school, Some College Educationj; t is an indicator for having completed fewer than 4 years of college, College Graduatej; t is a dummy for completing 4 years of college, and Some Graduate Educationj; t is a dummy for having more than 4 years of college education. Table A.1 in Appendix A displays the first stage regression estimates. Column (1) of Table 4 shows the regression results using the specification of Eq. (7) but the dependent variable is calculated using the residuals of this first stage regression. The coefficient on the change in the median marginal tax rate of workers in an occupation is slightly smaller than before; however, it is still positive and statistically different from zero. The magnitude of the coefficient suggests that workers in an occupation that experienced a 10 percentage point decrease in its median marginal tax rate after 1986 had about a 2% decline in pre-tax wages. Therefore, my results do not appear to be driven by changes in the observable composition of workers in occupations. One limitation of this two-step estimation technique is that the effects of demographic characteristics on wages are constrained to be the same over time. With the large changes in the skill premium observed in the 1980s, I need to allow the effects of some characteristics, especially education, to vary over time. I repeat my two-step estimation technique, allowing the effect of the education and age variables in the first stage regression to be different each year over my 8-year sample. Column (2) of Table 4 presents the results of the second stage of the estimation. Again, the effect of changes in the median marginal tax rate of workers in an occupation on wages falls slightly, but qualitatively the results are very similar to before. In my regression specification of Eq. (7), I include linear trends for each occupation to control for changes over time in the wages of workers by occupations not related to the Tax Reform Act of 1986, including changes in the demand for skilled workers in the 1980s. By controlling for demographic characteristics in the first stage of my two-stage

J.D. Kubik / Journal of Public Economics 88 (2004) 15671588

1583

Table 4 The effect of changes in marginal tax rates on the median wages of workers by occupationcontrolling for demographic characteristics The effect of demographic controls constant over time (1) Change in the median marginal tax Rate of occupationafter 1986 indicator Year effects Occupation effects Linear time trends for occupations Observations 0.2006 (0.0887) Yes Yes Yes 3744 The effect of education and age allowed to vary over time (2) 0.1721 (0.0882) Yes Yes Yes 3744 (3) 0.1288 (0.0462) Yes Yes No 3744

Notes: the dependent variable is the median residual by occupation and year of a regression of the log wage on demographic characteristics of 25 55-year-old men. Standard errors are in parentheses. The regressions are weighted by the number of men in each occupation in 1986.

estimation procedure and allowing the effect of those characteristics to vary over time, I am also controlling for these changes in the returns to education. Therefore, I can examine what happens when I remove the linear trends from my two-step estimation technique to see if I obtain different results from when I previously omitted the linear trends. The results of this regression are in column (3) of Table 4. The coefficient on the effect of changes in the median marginal tax rate of workers in an occupation is again smaller but positive and statistically significant from zero. The magnitude of the coefficient indicates that a 10 percentage point decrease in the median marginal tax rate of workers in an occupation was followed by a 1.3% decline in wages in that profession. Therefore, controlling for changes over time in the demand for skilled workers either by including linear trends by occupation or by this two-step procedure both produce the same qualitative results; however, the effect of changing marginal tax rates on pre-tax wages is about twice as large when linear trends are included in the regression specification. 4.4. Specification checks Finally, I examine the robustness of my empirical results to alternative estimation specifications. One potential concern about my previous empirical work is how I parameterize the effect of the Tax Reform Act of 1986 on workers in occupations as the difference between the their marginal tax rate under the 1988 tax rules and their marginal tax rate under the 1985 rules. An alternative measure of the effect of the tax reform is the difference in the average of the median marginal tax rates of workers by occupation of the first 4 years after the 1986 reform and the average of the median marginal tax rates for workers in an occupation during the 4 years preceding the tax reform. That is, using my sample of adults from the 1987 March CPS, I calculate the median marginal tax rates of workers in all occupations for the years 1987 through 1990 and take the average for workers in each occupation over those years. I do the same thing calculating the marginal tax rate for the years 1983 to 1986. The difference in those two

1584

J.D. Kubik / Journal of Public Economics 88 (2004) 15671588

averages is my new measure of how the 1986 tax reform affected workers in an occupation. Column (1) of Table 5 presents the regression results of the effect of the tax changes on the wages of workers in occupations using this new measure. The estimated effect of a change in marginal tax rates on wages is very similar to my previous results, suggesting that my findings are not being driven by exactly how I parameterize the tax change. In my discussion of my regression specification in Section 3, I argued that an important assumption of my empirical work is that the measure of the change in the median marginal tax rate of workers in an occupation because of the Tax Reform Act of 1986 is capturing the effect of the tax reform on the marginal workers making their labor supply decisions. The results above and other work in the literature suggest that women are the workers most sensitive to these tax changes. Therefore, if the effect of the 1986 tax reform is systematically different for women in an occupation than the median worker, then my regression might be misspecified. To examine this potential problem, I calculate my measures of the change in the median marginal tax rate of workers by occupation using only women instead of the entire population of 18 65-year-olds. Using this new measure of the effect of the tax reform, I estimate Eq. (7) in column (2) of Table 5. The coefficient on the effect of changes in the median marginal tax rate on wages is slightly larger than my previous results, indicating that using the change in the marginal tax rate of the median worker as a gauge of the effect of the 1986 tax reform is not problematic. I am also concerned that my regression results that include linear trends by occupation might be measuring the fact that the growth in the wages of skilled workers to unskilled workers was slowing down over the 1980s instead of measuring the effect of the 1986 tax reform on wages. If the increases in the returns to education were decelerating over time for reasons other than tax policy, then my empirical strategy might attribute those changes in wage trends as a response to the tax reform. Under this scenario, I should find that the wages of high skilled workers fall compared to other workers after 1986, and that their wages also fall after other years in the late 1980s. Therefore, I test whether I obtain the same results using another sample period in the late 1980s instead of my 1983 to 1990 sample centered on 1986.
Table 5 The effect of changes in marginal tax rates on the median wages of workers by occupationspecification checks Median MTR of occupation Average MTR change (1) Change in the marginal tax rate of Occupationafter reform indicator Year effects Occupation effects Linear time trends for occupations Observations 0.3016 (0.1110) Yes Yes Yes 3744 MTR change of women (2) 0.3149 (0.1096) Yes Yes Yes 2976 1986 1991 sample (3) 0.0083 (0.1088) Yes Yes Yes 2838 Average MTR of occupation (4) 0.2153 (0.0611) Yes Yes Yes 3744

Notes: the dependent variable is the log median wage of occupations by year of 25 55-year-old men. Standard errors are in parentheses. The regressions are weighted by the number of men in each occupation in 1986.

J.D. Kubik / Journal of Public Economics 88 (2004) 15671588

1585

I use the 6-year period between 1986 and 1991, pretending that the tax reform occurred in 1988 instead of 1986.23 Therefore I have three pre-reform years (1986 1988) and three post-reform years (1989 1991). I estimate a regression using the specification of Eq. (7), but my median wage of workers by occupation data are from this later sample period.24 The results of this estimation are presented in column (3) of Table 5. The coefficient on the change in the median marginal tax rate of workers in an occupation after the tax reform (1988 in this instance) is very small and not statistically different from zero. Therefore, the change in the trends in wages appear to occur around 1986, suggesting that they are related to the timing of the tax reform. Lastly, my measure of the effect of the tax reform on workers by occupations is parameterized as the difference in the median tax rates of people in a given occupation before and after the tax reform. I examine the sensitivity of my results to using these median tax rate measures by estimating Eq. (7) using the difference in the average tax rates of workers in an occupation as the independent variable of interest. The final column of Table 5 presents the estimates of Eq. (7) using this new measure of the effect of the tax reform on workers by occupation. The estimated effect of the tax change on the pre-tax workers is very similar to my previous results, indicating that my results are not sensitive to whether I use medians or averages.

5. Conclusion This paper studies the incidence of the personal income tax by examining how wages changed for workers across occupations after the Tax Reform Act of 1986. From data on a nationally representative sample of adults and information on tax rules over time, I calculate estimates of how workers in different occupations were affected by the tax changes to estimate whether the wages of workers in occupations most affected by the tax reform shifted relative to the wages of other professions. I find that, after controlling for trends in the wage structure by occupation over the 1980s, occupations whose median worker enjoyed a decline in marginal tax rates because of the 1986 reform experienced a decline in pre-tax wages after 1986. These declines are associated with increases in the relative supply of labor in those professions. These findings suggest that the effect of the 1986 tax reform on an individual depended not only on the individuals tax circumstances but also how other workers in the persons profession were affected by the changes. When assessing the distributional consequences of tax reforms, the wage effects of the tax changes should be considered. The supply shifts in the labor market caused by 1986 tax reform did affect wages. It is also important to consider these wage effects when using the after-tax wage variation of tax reforms to identify behavioral parameters. Most of the work using tax reforms assumes that the pre-tax wages of workers are unaffected by tax changes. The
23 I am unable to use data past 1991 because the coding of three-digit occupations change between 1991 and 1992 in the CPS data. 24 For my measure of the effect of the tax reform on different occupations, I still use the difference of the occupations median marginal tax rate in 1988 and 1985.

1586

J.D. Kubik / Journal of Public Economics 88 (2004) 15671588

results of this study indicate that this assumption might lead to biased estimates of these parameters. Although this analysis shows that there were wage effects associated with the tax law changes of 1986, these results should not be interpreted as predictions of how the labor market will respond to other potential tax changes. A major drawback of the reduced form approach of this analysis is that the findings cannot easily be generalizable. Other tax reforms that affect different groups of workers might have somewhat different effects. Future work should give a more general picture of how the labor market responds to the labor supply shifts associated with tax reforms. Acknowledgements I am grateful to Stacy Dickert-Conlin, Gary Engelhardt, Doug Holtz-Eakin, Tom Kniesner, Hisahiro Naito, Raymond Robertson, John Yinger and seminar participants at the University of Illinois, Syracuse University and the National Tax Association meetings for helpful comments. I also thank Stacy Dickert-Conlin for the use of her programs for calculating marginal tax rates.

Appendix A . The effect of demographic characteristics on wages

Table A.1 High school graduate Some college education College graduate Some graduate education Between the ages of 30 and 34 Between the ages of 35 and 39 Between the ages of 40 and 44 Between the ages of 45 and 49 Between the ages of 50 and 55 White African-American Married Observations 0.2695 (0.0022) 0.3910 (0.0024) 0.5843 (0.0027) 0.6311 (0.0027) 0.1169 (0.0021) 0.1995 (0.0022) 0.2618 (0.0024) 0.2922 (0.0026) 0.2917 (0.0026) 0.0910 (0.0038) 0.0739 (0.0045) 0.1517 (0.0017) 448,558

Notes: the dependent variable is the log of the wage of men between the ages of 25 and 55. The regression specification also includes year effects.

J.D. Kubik / Journal of Public Economics 88 (2004) 15671588

1587

References
Atkinson, A.B., 1973. How progressive should the income tax be? In: M. Parkin (Ed.), Essays on modern economics. Allen, F., 1982. Optimal linear income taxation with general equilibrium effects on wages. Journal of Public Economics 17, 135 143. Blundell, R., Duncan, A., Meghir, C., 1998. Estimating labor supply responses using tax reforms. Econometrica 66, 827 861. Bound, J., Johnson, G., 1992. Changes in the structure of wages in the 1980s: an evaluation of alternative explanations. American Economic Review 82, 371 392. Christiansen, V., 1988. Choice of occupation, tax incidence and piecemeal tax revision. Scandinavian Journal of Economics 90 (2), 141 159. Dickert-Conlin, S., Chandra, A., 1999. Taxes and the timing of births. Journal of Political Economy 107, 161 177. Eissa, N., 1995. Taxation and Labor Supply of Married Women: The Tax Reform Act of 1986 as a Natural Experiment, NBER Working Paper No. 5023. Eissa, N., 1996. Labor supply and the economic recovery Tax Act of 1981. In: Feldstein, M., Poterba, J. (Eds.), Empirical Foundations of Household Taxation. MIT Press for the National Bureau of Economic Research, Cambridge, MA. Feldstein, M., 1973. On the optimal progressivity of the income tax. Journal of Public Economics 2, 357 376. Feldstein, M., 1974. Tax incidence in a growing economy with variable factor supply. Quarterly Journal of Economics 87, 551 573. Feldstein, M., 1995. The effect of marginal tax rates on taxable income: a panel study of the 1986 tax reform act. Journal of Political Economy 103, 551 572. Gruber, J., Krueger, A.B., 1991. The incidence mandated employer-provided insurance: lessons from workers compensation insurance. In: Bradford, D. (Ed.), Tax Policy and the Economy 5. MIT Press, Cambridge, MA. Hausman, J., 1985. Taxes and labor supply. In: Auerbach, A.J., Feldstein, M. (Eds.), Handbook of Public Economics, Vol. 1. North Holland, Amsterdam. Hausman, J., Poterba, J.M., 1987. Household behavior and the tax reform act of 1986. Journal of Economic Perspectives 1, 101 119. Heckman, J.J., 1996. Comment on labor supply and the Economic Recovery Act of 1981. In: Feldstein, M., Poterba, J. (Eds.), Empirical Foundations of Household Taxation. MIT Press for the National Bureau of Economic Research, Cambridge, MA. Heckman, J.J., 1993. What has been learned about labor supply in the past twenty years? American Economic Review Papers and Proceedings 83, 116 121. Katz, L., Murphy, K.M., 1992. Changes in relative wages 1963-1987: supply and demand factors. Quarterly Journal of Economics 107, 35 78. Kotlikoff, L.J., Summers, L.H., 1979. Tax incidence in a life cycle model with variable labor supply. Quarterly Journal of Economics 93, 705 718. Levy, F., Murnane, R.J., 1992. US earnings levels and earnings inequality: a review of recent trends and proposed explanations. Journal of Economic Literature 30, 1333 1381. Mirrlees, J.A., 1971. An exploration in the theory of optimal income taxation. Review of Economic Studies 38, 179 208. Moffitt, R.A., Wilhelm, M., 2000. Taxation and the labor supply decisions of the affluent. In: Slemrod, J. (Ed.), Does Atlas Shrug? The Economics Consequences of Taxing the Rich. The Russell Sage Foundation and Harvard University Press, New York and Cambridge. Murphy, K.M., Welch, F., 1992. The structure of wages. Quarterly Journal of Economics 107, 285 326. Pencavel, J., 1986. Labor supply of men: a survey. In: Ashenfelter, O., Layard, R. (Eds.), Handbook of Labor Economics. Vol. 1. North Holland, Amsterdam. Pencavel, J., 1998. Assortative mating by schooling and the work behavior of wives and husbands. American Economic Review Papers and Proceedings 88, 326 329.

1588

J.D. Kubik / Journal of Public Economics 88 (2004) 15671588

Sheshinski, E., 1972. The optimal linear income tax. Review of Economic Studies 39, 297 302. Stern, N., 1982. Optimum taxation with errors in administration. Journal of Public Economics 17, 181 211. Stiglitz, J.E., 1982. Self-selection and Pareto efficient taxation. Journal of Public Economics 17, 213 240. Triest, R.K., 1990. The effect of income taxation on labor supply in the United States. Journal of Human Resources 25, 491 516.