the Wage Distribution
David Neumark
Mark Schweitzer
William Wascher
a b s t r a c t
This paper provides evidence on a wide set of margins along which labor markets can adjust in response to increases in the minimum wage, includ-ing wages, hours, employment, and ultimately labor income. Not surpris-ingly, the evidence indicates that low-wage workers are most strongly af-fected, while higher-wage workers are little affected. Workers who initially earn near the minimum wage experience wage gains. Neverthe-less, their hours and employment decline, and the combined effect of these changes on earned income suggests adverse consequences, on net, for low-wage workers.
I. Introduction
Labor markets can adjust along a variety of margins in response to increases in the minimum wage. For example, employers may alter the number of workers employed at an establishment, or they may adjust the average number of hours worked by each employee. In addition, rms may alter the mix of workers
David Neumark is a senior fellow at the Public Policy Institute of California, a professor of economics at Michigan State University, and a research associate of the NBER. Mark Schweitzer is an economist at the Federal Reserve Bank of Cleveland. William Wascher is assistant director in the Division of Re-search and Statistics, Board of Governors of the Federal Reserve System. The authors are grateful to Scott Adams for outstanding research assistance, and to seminar participants at Michigan State, the University of Kentucky, UC-Berkeley, the Federal Reserve Bank of Cleveland, and Florida State Uni-versity for helpful comments. The views expressed do not necessarily reect the views of the Public Pol-icy Institute of California, the Federal Reserve Board, the Federal Reserve Bank of Cleveland, or their staffs. The data used in this article can be obtained beginning October 2004 through September 2007 from Mark Schweitzer, Federal Reserve Bank of Cleveland, P.O. Box 6387, Cleveland, OH 44101. [Submitted October 2001; accepted December 2002]
ISSN 022-166X; E-ISSN 1548-8004ã2004 by the Board of Regents of the University of Wisconsin System
employed following an increase in the minimum wage, essentially attempting to realign the marginal product of their workers with the wages they are paid. As a result of these adjustments, the effects of minimum wages may extend beyond work-ers whose wages are directly impacted by the higher oor. Our evidence indicates that minimum wage increases adversely affect workers initially earning near the minimum wage, but have little impact on higher-wage workers. In particular, al-though wages of low-wage workers rise, their hours and employment fall. The com-bined effect of these changes is a decline in earned income.
Past minimum wage research focuses mainly on employment effects, and fails to distinguish minimum wage effects at different parts of the wage distribution. Conse-quently, this past research provides insufcient information with which to evaluate the policy implications of raising the minimum wage, in particular whether such increases help low-wage workers. In contrast, this paper generates a richer descrip-tion of the effects of the minimum wage on labor markets, providing evidence on a wide set of the margins along which labor market adjustments to minimum wages may occur, and how the adjustments vary at different points of the wage distribution; we provide a particularly sharp focus on minimum wage effects at the lower end of the wage distribution.
II. Existing Research
Our efforts to distinguish minimum wage effects in different parts of the wage distribution differentiates our approach from most of the existing work on minimum wages, which— in order to focus on a set of relatively low-skilled workers—typically studies employment effects for teenagers or a closely related group. However, the focus on employment effects for teenagers is arguably far re-moved from the most pertinent policy questions, for at least three reasons.
First, policymakers typically are most concerned with adults working near the minimum wage, because young workers are on the early part of their experience prole and hence are likely to grow out of minimum wage jobs, while adults working at minimum wage jobs are more likely to be permanent low-wage workers. In addi-tion, teenagers are more likely to be secondary earners. Second, because many teen-agers and young adults earn wages well above the minimum, estimates of disemploy-ment effects for young workers as a whole may mask larger disemploydisemploy-ment effects for the lowest-wage workers, and thus overstate the resulting income gains experi-enced by low-wage workers. Third, the emphasis on employment effects provides too narrow a picture of the effects of minimum wages on the economic well-being of low-wage workers. On the negative side, hours could fall in response to minimum wage increases, while on the positive side minimum wages may generate wage in-creases above the minimum. We examine the consequences of minimum wages for employment, wages, and hours, as well as the overall impact on labor income.1In all cases, we isolate the effects of minimum wages in different parts of the distribution of wages.
Two recent papers move beyond the estimation of employment effects for teenag-ers or young adults to focus more sharply on workteenag-ers who are most likely to be affected by the minimum wage. Abowd et al. (1999) examine individual-level panel data for France, where the real minimum wage rose throughout their sample period (1981– 89), and for the United States, where it fell (1981– 87). They study minimum wage effects in two opposite but closely related ways: In France, they condition on initial employment and test for disemployment effects among workers who are “caught” by minimum wage increases, while in the United States they look at indi-viduals who are “released” by the falling real minimum wage. For both countries, Abowd et al. report considerably larger disemployment effects of minimum wages for workers constrained by the minimum than for workers with marginally higher wages. Currie and Fallick (1996) carry out a similar analysis using NLSY data for the United States. They estimate the employment effects of the 1980 and 1981 federal minimum wage increases, dening as the treatment group workers whose wage prior to the increase was between the old and the new minimum wage, and as the control group workers earning near but above the minimum wage. Currie and Fallick nd that workers bound by the minimum were about 3 percent less likely than the control group to be employed after the minimum wage increase; the estimated employment elasticity for workers bound by the minimum is about20.4. The elasticities esti-mated by Abowd et al. are at least as large.
Some researchers have examined other margins of adjustment to minimum wage increases. The most extensive body of research exploring other margins of adjust-ment studies the extent to which minimum wage increases lead to positive “ripple” effects on the wages of workers already earning more than the new minimum. Gram-lich (1976) originally broached this question, suggesting that standard substitution effects or union-related relative wage considerations might lead to increases in the wages of higher-skilled workers following an increase in the legislated minimum wage. Another possibility is that the labor supply of higher-skilled workers might increase as lower-skilled workers (in the same family) become disemployed or face lower hours as a result of minimum wage increases, leading to a decline in wages for higher-skilled workers. Gramlich presents evidence suggesting that minimum wage increases raise average wage rates by about twice what would be predicted from the direct impact of minimum wage increases on workers for whom the mini-mum is binding (ignoring possible employment effects). But because Gramlich relies on aggregate data, he cannot examine where in the wage distribution the wage spill-overs occur. Grossman (1983) also presents evidence consistent with ripple effects from minimum wages.
More recent analyses use empirical methods that more directly reveal the impacts of minimum wages on the wage distribution. For example, DiNardo et al. (1996) present a semi-parametric analysis of how changes in national minimum wages have affected wage inequality, while Lee (1999) examines the impact of minimum wages on the wage distribution in more detail using state-level variation. Both papers nd evidence suggestive of positive spillovers from minimum wages to other wages, as does more limited evidence in Spriggs (1993) and Card and Krueger (1995, Ch. 9), and evidence for Canada reported by Green and Paarsch (1998).
of wages changes as a result of minimum wage increases, we try to estimate the actual impact of minimum wages on workers at different points of the initial distribu-tion of wages relative to the minimum. Second, we study addidistribu-tional outcomes (hours, employment, and labor income) to provide a more comprehensive analysis of the consequences of minimum wage increases. Finally, DiNardo et al. (1996) and Lee (1999) focus more on how minimum wages sweep up workers in the bottom tail of the wage distribution, as opposed to an analysis of the effects on the wage distribution above but near the new minimum.
Much less research looks at hours effects, and what there is focuses on the probabilities of part-time and full-time employment. Gramlich (1976) nds that minimum wages reduce full-time employment and increase part-time employment of teenagers and adult males; although an overall disemployment effect is apparent for teenagers only, the switch from full-time to part-time is consistent with hours reductions for both groups. Hungerford (2000) reports that minimum wages appear to increase the proportion of involuntary part-time workers among less-educated teenagers, and among blacks across age and education categories. In contrast, Cunningham (1981) reports evidence from an earlier period suggesting that minimum wages discourage part-time employment and boost full-time employment, as do Katz and Krueger (1992) using data from fast-food restaurants in Texas.2Finally, Zavodny (2000) nds that teenagers who remain employed following a minimum wage increase tend to experience an increase in hours worked, roughly offsetting the job losses incurred by other teens.
One study that comes closer to our more comprehensive approach of estimating the effects of minimum wages on wages, employment, hours, and income is Linne-man (1982). He uses PSID data from 1973 to predict wages in 1974 and 1975, and on the basis of the predicted wages identies workers who would be bound by the increases in the minimum wage in 1974 and 1975. His ndings indicate hours (and to a lesser extent employment) reductions among constrained workers. However, he also estimates hours and employment effects for workers in various wage intervals above the minimum, nding that individuals just above the minimum (relative to workers further above the minimum) experience reduced employment rates but in-creased hours.
We have two main criticisms of Linneman’s analysis. First, his estimates of the effects of the minimum wage on income are not based on actual income in 1974 and 1975, but are imputed from the estimated hours and employment effects. As a result, the estimates take no account of the effects of the minimum wage on the wages of workers earnings more than the minimum, and the imputation method takes no account of the distribution of wage and hours effects across individuals.3Second, his approach does not provide a credible counterfactual for the experiences of the group affected by the minimum wage increase (Card and Krueger 1995, p. 224).
We address the rst set of shortcomings by looking at wage and income effects independently. We provide a credible counterfactual by using exible estimates of 2. Neumark and Wascher (2000) and Michl (2000) touch on the issue of effects of minimum wages on employment and hours in the context of the Card and Krueger (1994) New Jersey-Pennsylvania minimum wage study.
underlying wage, hours, employment, and income changes of the nonaffected popu-lation, and using state variation in minimum wages to obtain treatment and control groups. More generally, we provide a fuller characterization of minimum wage ef-fects throughout the wage distribution, and update and strengthen the analysis by taking advantage of the state-level variation in minimum wages that has been fruit-fully exploited in the new minimum wage research.
III. Data
Our basic approach is to estimate models for changes in wages, hours, employment, and income, using data on individuals in matched monthly CPS outgoing rotation group les for the period 1979– 97. While household identiers are available for doing the matching, individual identiers are not. This raises two issues. First, to ensure that we match individuals correctly, we lter the data through a procedure that uses sex and age as primary characteristics for establishing a match. A match occurs when a household identier-primary characteristics cell includes at least one pair of rst-year and second-year records. For instance, a 31-year-old fe-male in the rst year and a 32-year-old fefe-male in the same household in the second year will be placed in the same cell. In the event of multiple matches we use a set of tiebreakers, including factors such as education. This tie-breaking phase checks for the correct matches in the cell at the most detailed partition of the specied variables rst. If no match is found, then variables are systematically dropped to arrive at a match. Additional steps are conducted to match others that may have been missed in the rst step. For example, since the CPS is not necessarily conducted on the same calendar day in subsequent years, a 31-year-old female and a 33-year-old female may be the same person, but they would have been excluded in our rst step. There are two sets of months that cannot be matched to observations 12 months ahead because of changes in the sample in response to decennial Censuses of Popula-tion: July 1984– September 1985 and June 1994– August 1995.
Second, about 20 percent of the individuals in the outgoing rotation groups who can potentially be matched across years are not successfully matched, most likely because of a change in residence. When we weight the observations, we adjust the sampling weight to account for the possibility that certain individuals have a lower probability of being in the survey in consecutive years and thus are less likely to be included in our matched sample. This adjustment is based on logit model estimates of the probability of a match as a function of demographic characteristics, with the adjusted weight an estimate of the inverse of the probability of being in our matched sample of families. Of course this procedure does not correct for nonrandom match-ing that, conditional on these observables, is correlated with changes in outcomes and therefore possibly also with minimum wage changes. We conjecture that, if anything, families most adversely affected by minimum wage increases tend to move away from areas where minimum wages have increased and toward areas where they have not. If so, the bias from nonrandom matching will tend to understate any adverse consequences of minimum wage increases.
by using a reported hourly wage—as opposed to usual weekly earnings/usual weekly hours— whenever the former is reported. The weekly measure explicitly includes tips, commissions, and overtime, while the hourly wage measure is less likely to include these. Beginning with the redesigned CPS in 1994, the hourly wage measure was changed so as to explicitly exclude tips, commissions, and overtime, but we assume that the year effects will pick up the inuence of this change on average wage changes. For hours, we use usual weekly hours (which is coded as missing beginning in 1994 if the respondent indicates variable hours), and for labor income we use the product of the wage and hours. The employment variable is an indicator equal to one for workers employed during the survey week.
IV. Empirical Framework
We illustrate our strategy by focusing on the estimation of the effects of minimum wages on the wage distribution, although the discussion generalizes to the other dependent variables we consider. To motivate the research design, we rst discuss the relatively simpler issue of estimating contemporaneous effects, which illustrates our strategy of allowing the effects of minimum wages to differ across the wage distribution while controlling in a exible fashion for other sources of changes in wages. In particular, we estimate the contemporaneous effects from the specication:
(1) w 2
isym2w1isym w1
isym
5a1
^
j bj
MW2
sym2MW1sym MW1
sym
×R(w1isym,MW1sym)j
1
^
j
gjR(w1isym,MW1sym)j
1
^
j
fjR(w1isym,MW1sym)j× w1
isym MW1
sym
1X1
isymd1Miml1Sis×Yiyp1eisym.
a more restrictive fashion by including the state-level unemployment rate. Because the data set covers all months of the basic CPS, we also include calendar month dummy variables (M) to control for seasonality—summer or holiday employment, for example—that might be spuriously correlated with minimum wage changes. Finally, eis a random error term assumed to have zero expectation conditional on the regressors, and to be independent across state, year, and month cells conditional on the regressors; given that the regressors on which we condition include month dummy variables and state-year interactions, we have already controlled for many sources of nonindependence of observations across space and time. In the estimation of the regression equations, we compute standard errors that are robust to hetero-skedasticity.
Rjdenotes a set of dummy variables that describe the level of the Year 1 wage relative to the Year 1 minimum wage; these are spelled out fully in Table 1. The Rjs control for differences in wage changes at different points of the wage distribution for reasons unrelated to changes in minimum wages, including measurement error in wages that can lead to a negative correlation between measured Year 1 and Year 2 wages.4In addition, we include interactions of theRjs with the ratio of the individu-al’s wage to the minimum wage; with the interactions, we have a spline specication without restricting the lines to join at the knot points (a dummy/spline specication). These additional interactive terms allow wage changes to differ within the cells de-ned by the Rjs, hence allowing a more exible specication of underlying wage changes.
for low-wage workers essentially mirror what we obtained when outcomes were dened relative to this cell.5
Our specication of minimum wage effects is highly exible in that it uses a set of dummy variables that divide up the initial wage distribution into fairly narrow regions —especially near the minimum wage. But this exible specication imposes strong demands on the data. An alternative, therefore, is to impose some smoothness on the estimates—in particular, on how the minimum wage effect varies across the wage distribution— by using a high-order polynomial in the wage relative to the minimum wage. We therefore consider estimates using an alternative specication:
(1¢) w
We experimented with polynomials of different orders and settled on seven as the order required to capture the variation in estimated minimum wage effects. The qualitative results were not sensitive to increasing or decreasing this order somewhat. Although the remaining discussion in this section is couched in terms of Specication 1, it carries over to Specication 1¢as well.
Previous research indicates that a signicant portion of the total minimum wage effect on employment occurs with a one-year lag (Neumark and Wascher 1992; Baker et al. 1999). Some have argued that high turnover for low-wage workers im-plies that adjustments to minimum wages will occur quickly (for example, Brown et al. 1982). But because changes in technology and management needed to replace low-skilled workers may take time, the existence of lagged adjustment effects is an empirical question. In addition, our earlier research on minimum wage effects on family incomes (Neumark et al. 1998) indicated that minimum wage increases had benecial effects on low-income families contemporaneously, but adverse effects after one year. This pattern is consistent with upward wage adjustments occurring quickly, and employment and hours adjustments occurring with a lag. As this paper also looks at incomes, a similar specication seems appropriate.
A complication arises, however, in estimating lagged effects with our data: We cannot dene a comparable set of Rjs in the year prior to Year 1 (call it Year 0) because the matched CPSs only include wage data for Year 1 and Year 2.6To get around the absence of the earlier data, we instead dene theRjs that we use to identify the lagged effects based on the Year 1 wage (relative to the Year 1 minimum), and estimate the equation
5. These results are available upon request.
(2) w
In this specication, the lagged effects associated with a minimum wage increase from Year 0 to Year 1 are dened conditional on where a worker’s Year 1 wage was in the wage distribution relative to the minimum in Year 1. (The superscript 0 represents the year prior to Year 1.) This specication of the lagged effect has the same interpretation as the usual lagged effect if the individual’s wage history does not matter. That is, it reects the usual lagged effect if, conditional onw1relative toMW1, the Year 1 to Year 2 effect of the minimum wage does not depend on the path of wage rates up tow1(for example, whether an individual’s wage was atw1 all along or instead was swept up tow1by the initial minimum wage increase). To see why this assumption is implicit in the specication, consider the most general form of the regression we would use instead of Equation 2 if we actually had three years of data:
There are two differences relative to Equation 2. First, theR function is more general in that it allows the minimum wage and baseline effects to depend on wages and minimum wages in both Year 0 and Year 1. For example,Rcould be a set of dummy variables dened over a grid of values of the wage relative to the minimum wage in each of the two years. Second, and related, in the term involving the coef-cientsfj, the baseline effects depend on pairs of values of the wage relative to the minimum wage in each of the two years, rather than on just the Year 1 relative values. The problem, of course, is that we do not have data for three years.
This is potentially a strong assumption, as it implies, for example, that anticipated wage growth should be unrelated to past wage growth for two workers with the same current wage facing the same minimum wage. In fact, though, our specication allows this restriction to be relaxed in a number of ways, by allowing wage growth to vary with a large set of control variables. Furthermore, because it includes lagged effects, we view this specication as likely to better capture the effects of minimum wages than a specication with exclusively contemporaneous effects. Nonetheless, as the results described below show, the substantive conclusions depend importantly on the inclusion of the lagged effects, and thus in future research it will be worthwhile to consider alternative methods or data sources that probe the sensitivity of the results to the treatment of lagged effects.8
To this point, we have described how we estimate contemporaneous and lagged effects of minimum wages conditional on a worker’s wage relative to the minimum wage. However, combining the estimated contemporaneous and lagged effects to obtain estimates of the total effects of minimum wages on wages (or hours, employ-ment, or income) is more complicated than in the usual case. In particular, in comput-ing the sum of the contemporaneous and lagged effects we have to keep track of the contemporaneous effect, and measure the lagged effect from the point in the wage distribution that prevails after one year (either because of minimum wage changes or baseline wage changes). That is, because workers experience wage growth whether or not minimum wages increase, and many low-wage workers are on steep regions of experience or tenure proles, a worker whose wage is near the minimum wage in Year 0 may have a wage signicantly above the minimum in Year 1. As a result, the typical minimum-wage worker may move several steps up through the wage categories dened by ourRjvariables, and thus measurement of the effects of mini-mum wages on low-wage workers needs to be conditioned on expected changes in wages for other reasons. In addition, rst-year minimum wage effects may move a worker to a different part of the wage distribution, in which case the estimated lagged effect needs to be applied to the region of the wage distribution in which the worker is likely to be found one year after the initial increase.9
To estimate the total effects, we consider a set of hypothetical workers based on average characteristics and responses to minimum wage increases in each cell de-ned by theRjs. The case we consider is a one-time,cpercent increase in the mini-mum wage. We rst use our estimates of Equation 2 to predict the contemporaneous wage change for the representative worker in each cell dened by theRjs, using (4) E
1
3
w8. An alternative is to try to predict the value of the Year 0 wage (or equivalently to predict the indicators R). But such predictions would be extremely noisy, given that wage regressions do not have great predictive power, and our ability to predict whose wages are in the extremes of the distribution might be particularly weak.
where the “hats” on the Greek letters indicate estimates, and the means ofX,S×Y, M, andw1are dened for individuals in cellj. Based on these predictions, and the average value ofw1in each cell, we can obtain a predicted value of w2 (denoted w2p) for the representative worker in each cell, which will have been affected both by the minimum wage increase and the control variables.
The next step is to predict the lagged effects that occur one year later. To make this prediction we shift the superscript on the predicted wage fromw2ptow1p, use these predicted values to place workers in new predicted cells dened by theRjs (based on MW1 and w1p), and predict the lagged effect at that point (updating potential experience by one year as well). The equation used for this prediction is
(5) E
1
3
w 22wlpwlp
4
j|
MW22MW1
MW1 50,
MW12MW0
MW0 5c,Xj,S×Yj,Mj,wj lp
2
5aˆ 1bˆLj ×c1gˆj1fˆj w1lp MW1
1Xjdˆ1Mjlˆ 1 S×Yjpˆ j51,...,J. The sum of the expressions in Equations 4 and 5 is the implied two-year effect of minimum wage increases on wages for workers in states with minimum wage increases. Note that the lagged effects are based on the average response within a wage category of a set of workers whose wages responded to a minimum wage increase one year prior. Similar expressions withcset to 0 provide estimates of the counterfactual, that is, changes in the wage distribution that would have occurred without minimum wage increases. Changes in the wage distribution for these workers are predicted based on the other control variables. Note that we cannot simply sub-tract the terms involving these control variables off of Equations 4 and 5 and report the estimatedbs multiplied by the minimum wage increase, because these control variables inuencew1pin Equation 5.
Although the procedures in this section have been described with the change in the wage as the dependent variable, identical procedures can be used to estimate the effects of minimum wage increases on hours and income. And, with respect to the analysis of employment, the only difference is that we condition on Year 1 employ-ment and look at whether the individual becomes nonemployed in Year 2, so that the dependent variable is an indicator of a change in employment status. We restrict attention to those initially employed because we do not have an initial wage for those initially nonemployed. As a result, care should be taken in interpreting the results as measuring overall employment effects of minimum wages, as labor market entry could either rise or fall in response to minimum wage increases, although we would expect the change to be in the same direction as for those initially employed. In these cases, we rst estimate the contemporaneous effect using an equation corresponding to Equation 4 for the relevant dependent variable. We then predict the Year 1 wage as described above, and use an equation corresponding to Equation 5, again for the relevant dependent variable, to estimate the correct lagged effect. Combining these predicted changes for wages, hours, employment, and income yields a detailed characterization of the effects of minimum wages at different points of the wage distribution.
corre-lations of errors, together with the rather complicated way we combine predictions from different equations, makes conventional standard errors difcult to calculate. We therefore instead use bootstrap-based standard errors. Hypothesis testing is done using the realized empirical distributions of the bootstrap estimates, in each case based on 500 repetitions.10One could argue that our large sample calls for signi-cance levels for hypothesis testing considerably smaller than the conventional 5 or 10 percent levels. However, our effective sample size is overstated by the number of observations on individuals, because we identify minimum wage effects from state-year variation. In addition, we are estimating effects for narrow subgroups, introducing many parameters to allow for a exible specication and numerous di-mensions of heterogeneity. Given that we exploit the large sample size to put strenu-ous demands on the data, it is not accurate to think of the sample size getting very large ceteris paribus, relative to other research using conventional signicance levels.
V. Results
Table 1 reports descriptive statistics for the full sample, and the sub-samples dened by theRjs that break up the distribution of initial wages into cells that—especially near the minimum—are quite disaggregated. To account for rounding and slight reporting errors in wages, note that workers with wages between ten cents less and ten cents more than the minimum are dened as minimum wage workers. The gures in this table are largely as expected. With the exception of workers initially paid below the minimum and workers in the highest wage category, average hours worked per week increase monotonically with the initial wage, from the high 20s— suggestive of a fairly high proportion of part-time workers—to more than 40. Combining hours and wages, weekly labor income (dened in nominal terms) rises monotonically. Teenagers are heavily overrepresented in the lowest wage categories; although they make up only 6 percent of the sample, they comprise 36 percent of minimum wage workers and 29 percent of workers earning above the minimum but below 1.1 times the minimum. Similarly, women, blacks, and Hispanics are overrepresented among low-wage workers.
We next discuss, in turn, minimum wage effects on wages, hours, employment, and labor income. Although the regression estimates do not provide a description of the total (contemporaneous and lagged) effects of minimum wages, we begin with these estimates before moving on to more readily interpretable graphs that report these total effects for representative workers in each region of the wage distribution. We focus rst on the dummy/spline specication with dummy variables for each cell of the distribution of wages relative to minimum wages, and the associated interactions. Following a detailed discussion of these estimates, we present results from the more restrictive polynomial specication.
Ne
um
ar
k,
Sc
h
w
eitz
er
,
and
Wa
sc
her
437
Means, Overall and by Position in the Wage Distribution
Weekly Hours, Year 1
Employed Labor Nonblack/
Proportion Year 2 Income Age 16–19 Age 201Women Men Black Hispanic non-Hispanic Year 1 variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Full sample 1.00 38.8 377.3 0.06 0.94 0.47 0.53 0.11 0.06 0.83 By Year 1 wage
w,MW2$.10 0.026 32.0 86.4 0.24 0.76 0.64 0.36 0.14 0.07 0.79
MW2$.10#w#MW1$.10 0.046 28.0 95.8 0.36 0.64 0.62 0.38 0.17 0.10 0.74
MW1$.10,w#1.1×MW 0.031 30.9 114.3 0.29 0.71 0.63 0.37 0.14 0.09 0.77 1.1,w/MW#1.2 0.052 33.0 136.9 0.19 0.81 0.61 0.39 0.15 0.10 0.76 1.2,w/MW#1.3 0.032 35.8 159.7 0.13 0.87 0.63 0.37 0.14 0.09 0.78 1.3,w/MW#1.5 0.084 36.6 184.6 0.09 0.91 0.59 0.41 0.14 0.09 0.78 1.5,w/MW#2 0.168 38.9 249.4 0.03 0.97 0.56 0.44 0.13 0.08 0.80 2,w/MW#3 0.265 40.5 365.5 0.01 0.99 0.45 0.55 0.10 0.06 0.84
3,w/MW#4 0.148 41.4 531.6 0.002 0.998 0.34 0.66 0.08 0.05 0.87 4,w/MW#5 0.078 41.9 695.6 0.001 0.999 0.28 0.72 0.07 0.04 0.89
5,w/MW#6 0.043 42.1 862.8 0.001 0.999 0.24 0.76 0.06 0.03 0.91 6,w/MW#8 0.029 41.8 1,082.6 0.001 0.999 0.20 0.80 0.05 0.03 0.93
A. Wages
Columns 1 and 1¢of Table 2 report the estimates ofbandbL, respectively, in Equation 2, using the percent change in wages as the dependent variable. The contemporaneous effects (thebs) are straightforward to interpret, as they measure the percentage change in the wage resulting from a 1 percent increase in the minimum wage. The estimates reveal pronounced, statistically signicant positive effects near the minimum. In partic-ular, for workers at or just above the minimum wage, the elasticity of wages with respect to the minimum is about 0.8. The elasticity falls to about 0.25 to 0.4 for workers between 1.1 and 1.5 times the minimum. For workers below the minimum, the esti-mated elasticity actually exceeds one.11Higher into the initial wage distribution, the estimated elasticities become quite small, although some are signicant.
The estimates of the lagged effects (thebLs) also reveal some interesting patterns. In particular, especially near the minimum, the estimated coefcients are strongly negative. This negative lagged effect implies that some of the wage gains associated with minimum wage increases are “given back” in the following year. These give-backs have not been noted in previous work on the effects of minimum wages on the wage distribution. However, it is perhaps not surprising that employers take advantage of ination in subsequent years to realign wages, partly undoing the effects of legislated nominal wage increases for low-wage workers.
In the upper left-hand panel of Figure 1, we display graphically the estimated effects on the wage distribution based on the calculation described in the previous section.12In this gure (and subsequent ones) we report the effect of a one-time 10 percent increase in the minimum wage. The gure displays the differential between the percentage change in the wage experienced by workers in states with the 10 percent minimum wage in-crease, and workers in states without an increase. The light bars simply replicate the estimated contemporaneous effects that were reported in Column 1 of Table 2. Of more interest are the estimated total effects, the gray bars, which incorporate lagged effects of minimum wages. As suggested by the negative estimates of thebLs in Column 1¢of Table 1, the effects of minimum wages on the wage distribution are tempered consider-ably when lagged effects are incorporated. Near the minimum wage, the elasticity of the wage with respect to the minimum falls to about 0.4. The estimated elasticities then decline for the cells slightly higher in the wage distribution, and become negative, albeit small, for cells more than twice the minimum. The graph also displays information on the p-values associated with each estimated effect (for the null hypothesis of no effect, versus the two-sided alternative). As indicated by the presence of three or four asterisks, most of the elasticities greater than 0.1 or so (in absolute value) are statistically signicant at the 5 percent level or better.
11. We suspect that estimates for the part of the wage distribution below the minimum are less reliable for a couple of reasons, including regression to the mean in wage data erroneously reported as below the minimum, and transitions between uncovered or tipped jobs and covered jobs. The latter scenario is likely to have a positive inuence on the estimate ofbfor this cell, because the jump in the wage upon moving to a covered job will be higher the more the minimum has increased. Finally, we conjecture that minimum wage increases are followed by upward (perhaps temporary) ratcheting of minimum wage compliance, as employers and workers become better informed about prevailing minimum wages.
Ne
um
ar
k,
Sc
h
w
eitz
er
,
and
Wa
sc
her
439
Effects of Minimum Wages on Wages, Hours, Employment, and Income
Hours, Conditional
on Year 2 Weekly
Wages Employment Employment Labor Income
Current Lagged Current Lagged Current Lagged Current Lagged
Percent change in minimum wage
3dummy variables for (1) (1¢) (2) (2¢) (3) (3¢) (4) (4¢) w,MW2$.10 1.39 20.76 20.30 0.23 0.034 20.014 1.00 20.75
(0.25) (0.23) (0.13) (0.13) (0.069) (0.066) (0.37) (0.33) MW2$.10#w #MW1$.10 0.79 20.60 20.09 20.52 20.115 20.065 0.19 21.47
(0.10) (0.09) (0.11) (0.10) (0.062) (0.060) (0.20) (0.19) MW1$.10,w #1.1×MW 0.78 20.29 20.05 20.23 20.145 0.100 0.38 20.49
(0.12) (0.11) (0.10) (0.09) (0.069) (0.063) (0.21) (0.18) 1.1,w/MW#1.2 0.41 20.42 0.06 20.20 20.074 20.003 0.37 20.58
(0.08) (0.08) (0.07) (0.07) (0.048) (0.049) (0.16) (0.14) 1.2,w/MW#1.3 0.36 20.27 0.16 20.11 20.169 0.067 0.26 20.29
(0.10) (0.10) (0.07) (0.07) (0.059) (0.052) (0.16) (0.16) 1.3,w/MW#1.5 0.26 20.27 0.11 20.17 20.030 0.014 0.29 20.45
(0.06) (0.06) (0.05) (0.04) (0.036) (0.038) (0.10) (0.10) 1.5,w/MW#2 0.16 20.12 0.04 20.01 20.002 20.004 0.15 20.16
(0.04) (0.05) (0.03) (0.03) (0.025) (0.024) (0.06) (0.06) 2,w/MW#3 0.06 20.14 20.00 0.005 0.038 0.012 0.03 20.07
(0.03) (0.03) (0.02) (0.02) (0.020) (0.021) (0.05) (0.04) 3,w/MW#4 0.00 20.18 20.04 0.09 0.000 20.040 20.06 20.12
The
Journa
l
of
Hum
an
R
es
ource
s
Table 2(continued)
Hours, Conditional
on Year 2 Weekly
Wages Employment Employment Labor Income
Current Lagged Current Lagged Current Lagged Current Lagged
4,w/MW#5 0.03 20.12 20.01 0.11 0.027 0.047 0.01 20.04 (0.04) (0.04) (0.02) (0.03) (0.025) (0.027) (0.05) (0.05) 5,w/MW#6 0.08 20.23 20.05 0.11 0.018 20.010 0.10 20.14
(0.04) (0.04) (0.03) (0.03) (0.032) (0.036) (0.05) (0.06) 6,w/MW#8 0.09 20.21 20.09 0.09 0.011 0.045 0.02 20.13
(0.04) (0.05) (0.04) (0.04) (0.032) (0.037) (0.06) (0.06)
AdjustedR2 0.16 0.04 0.31 0.07
N 749,510 749,510 847,175 847,175
um
ar
k,
Sc
h
w
eitz
er
,
and
Wa
sc
her
441
Figure 1
Effects of 10 Percent Minimum Wage Increase, Full Sample, Dummy/Spline Speci-cation
Taken as a whole, these results indicate that minimum wage increases raise the wages of the lowest-paid workers. However, the evidence of wage declines for work-ers initially earning higher wages suggests either that outward labor supply shifts for higher-wage workers outweigh increases in labor demand from substitution ef-fects, or that the scale effects resulting from higher overall labor costs outweigh the substitution effects. In addition, the results in the gure indicate that wage losses at the upper end of the wage distribution are smaller in percentage terms than gains at the lower end, but larger in absolute terms. With more workers represented in some of the higher-wage cells, the ndings may suggest that minimum wages are a rather inefcient tax and transfer scheme.
B. Hours
We next turn to effects on hours (conditional on remaining employed) for workers in different regions of the wage distribution. The regression estimates are reported in Columns 2 and 2¢of Table 2. The estimates reveal little evidence of contempora-neous hours reductions for workers paid at or slightly above the minimum wage, although there is a signicant estimated decline for workers initially earning below the minimum. For workers earning between 1.2 and 1.5 times the minimum, there is evidence of moderate but statistically signicant increases in hours. The lagged effects are more striking, with signicant hours reductions for individuals at or above the minimum wage, up to about 1.5 times the minimum.
The full set of contemporaneous and total effects is displayed in the upper right-hand panel of Figure 1. For individuals below the minimum, the estimated total effect on hours is more negative than the contemporaneous effect alone. The more negative total effect occurs because the wage gains experienced by these workers (see the upper left-hand panel of the gure) put them into higher cells in the wage distribution, where, as reported in Table 2, there are lagged hours reductions. More importantly, the gure reveals hours reductions for workers initially paid at or just above the minimum wage, with elasticities near20.3; the estimates for both cells are strongly signicant. In contrast, there are no signicant total effects for higher-wage workers, although the point estimates are nearly all positive. Coupled with reductions in wages for higher-wage workers, such hours increases suggest that a higher minimum wage leads them to increase labor supply— perhaps in response to reductions in hours for low-wage family members.
C. Employment
However, as suggested by the lagged estimates in Column 3¢, and as displayed in the lower left-hand panel of Figure 1, the disemployment effects are partially offset in the second year, with the total effect becoming smaller and statistically weaker except for workers with initial wages between 1.2 and 1.3 times the mini-mum. The pattern of stronger employment effects initially, but stronger hours effects later, is consistent with employers rst laying off part-time workers, reducing xed costs of labor, and then later adjusting hours downward.
D. Income
Finally, we turn to earned income. Based on the results reported above, a higher minimum wage could have either a positive or negative effect on the earned income of low-wage workers. Low-wage workers experience wage gains as a result of mini-mum wage increases, but also experience hours and employment declines. And, as noted previously, we cannot simply use the elasticities for hours, employment, and wages to predict income effects, since we do not know the joint distributions of changes in these variables. Columns 4 and 4¢of Table 2 report the regression esti-mates. The contemporaneous effects are positive (and signicant for most cells) for workers initially earning up to twice the minimum wage. In contrast, the lagged effects are uniformly negative and quite strong, especially up to about twice the minimum.
The lower right-hand panel of Figure 1 reports the one-year and total effects. The contemporaneous effects might be interpreted as suggesting that minimum wages increase the earnings of low-wage workers; the elasticities are in the 0.2 to 0.4 range for workers initially earning up to twice the minimum and are statistically signicant. However, adding in the lagged effects reverses this conclusion. As shown by the dark bars, the total effects indicate that workers initially below the minimum, at the minimum, and up to 1.1 times the minimum experience income declines. The esti-mated effect for minimum wage workers is on the order of a 6 percent decline and is statistically signicant at the 5 percent level. The source of the reversal is clear from the other panels of the gure. Although disemployment effects are tempered, hours reductions after one year are much sharper, and the wage gains considerably weaker.
Overall, the analysis indicates that very low-wage workers are not helped and are more likely hurt by minimum wage increases. Although minimum wages bump up wages of these workers, hours reductions, in particular, interact with changes in wages in such a way that earned income declines.
In Figure 2, we show estimates based on the polynomial specication of variation in minimum wage effects throughout the wage distribution; this specication is more restrictive but imposes some smoothness on how these effects change across the distribution. For comparison purposes, we report estimated effects of minimum wages for the same cells of the wage distribution (based on the mean for each cell) as were used in the dummy variable specication.
The
Journa
l
of
Hum
an
R
es
ource
s
Figure 2
Effects of 10 Percent Minimum Wage Increase, Full Sample, Polynomial Speci-cation
increase for the lowest-wage workers, although they are somewhat smaller than in Figure 1. Also, while these positive impacts extend out a bit further into the wage distribution, the cross-over point from positive to negative effects is still at about 2–3 times the minimum. Correspondingly, the negative hours effects also extend a bit higher up into the wage distribution, with negative effects one year out evident up to 1.5 times the minimum. Similarly, the negative estimated earnings effects are the same rough magnitude as in Figure 1 for the lower-wage workers, but the nega-tive effects extend a bit higher into the wage distribution.
The only substantive difference is that the polynomial specication provides no evidence of disemployment effects one year after a minimum wage increase. Inspec-tion of Figure 1 suggests that this may owe to the more irregular pattern of employ-ment effects in the unrestricted model. But regardless, the evidence on employemploy-ment effects is rather weak using either specication.
E. Adults Only
One of the motivations we cited for looking at minimum wage effects conditional on workers’ positions in the initial wage distribution was to see whether there is evidence of labor demand reductions for skilled workers. Our focus on low-wage workers differs from past work studying specic age groups (for example, teens or young adults) that include many—but not exclusively—low-skilled workers. To assess the extent to which our results identify effects for low-skilled adults, Figure 3 reports estimates using a sample restricted to individuals aged 20 and older.13 Over-all, these estimates are quite similar to those in Figure 2, with evidence of signicant positive wage effects in the lower range of the wage distribution, but signicant negative hours and income effects for the lowest-wage workers. Thus, the negative consequences of minimum wages are not restricted to teenagers, but appear more generally for low-wage workers. We also looked at separate results for men and women, see Tables 4a and 4b. The ndings were similar for the two groups, with a slight hint that the consequences of minimum wages are worse for women.
VI. Conclusions
In this paper, we present evidence on wage, hours, employment, and labor income adjustments that occur in response to minimum wage increases. Our main contribution is to estimate these adjustments in a consistent framework that provides a relatively complete description of the effects of minimum wages at many different points of the wage distribution.
The evidence indicates that low-wage workers are most strongly affected, while higher-wage workers are little affected. Workers who initially earn near the minimum wage experience wage gains. But their hours and employment decline, and the
The
Journa
l
of
Hum
an
R
es
ource
s
Figure 3
Effects of 10 Percent Minimum Wage Increase, Adults Only, Polynomial Speci-cation
um
ar
k,
Sc
h
w
eitz
er
,
and
Wa
sc
her
447
Figure 4A
Effects of 10 Percent Minimum Wage Increase, Women, Polynomial Specication
The
Journa
l
of
Hum
an
R
es
ource
s
Figure 4B
Effects of 10 Percent Minimum Wage Increase, Men, Polynomial Specication
bined effect of these changes on earned income suggests net adverse consequences for low-wage workers. The inclusion of lagged minimum wage effects is critical in arriving at the conclusion that low-wage workers are adversely affected, as we nd that contemporaneous effects overstate the wage gains and understate the hours and income losses experienced by low-wage workers when minimum wages rise.
The results in this paper are complementary to previous work on the impact of the minimum wage on family incomes (Neumark et al. 1998; Neumark and Wascher, 2002). In particular, our nding that the earned incomes of low-wage workers decline in response to a minimum wage hike is consistent with reduced-form evidence indi-cating that minimum wages may increase the proportion of families that are poor or near-poor. In this paper we provide some insight into the transmission mechanism of minimum wage effects. A legislated increase in the minimum wage does lead to an immediate boost in the pay of low-wage workers. However, the ultimate sizes of the wage increases induced by the higher minimum typically are considerably smaller than the minimum wage hike itself, and there are hours reductions among employed workers that, coupled with small disemployment effects, generate net losses in earned income. Thus, the ndings in this paper indicate that the full range of labor market effects associated with raising the minimum wage most likely reduce the well-being of low-wage workers.
References
Abowd, John M., Francis Kramarz, Thomas Lemieux, and David N. Margolis. 1999. “Mini-mum Wages and Youth Employment in France and the United States.” InYouth Unem-ployment and EmUnem-ployment in Advanced Countries,ed. David Blanchower and Richard Freeman, 427– 72. Chicago: University of Chicago Press.
Baker, Michael, Dwayne Benjamin, and Shuchita Stanger. 1999. “The Highs and Lows of the Minimum Wage Effect: A Time-Series Cross-Section Study of the Canadian Law.”
Journal of Labor Economics17(2):318– 50.
Brown, Charles, Curtis Gilroy, and Andrew Kohen. 1982. “The Effect of the Minimum Wage on Employment and Unemployment.”Journal of Economic Literature 20(2):487– 528.
Card, David, and Alan B. Krueger. 1994. “Minimum Wages and Employment: A Case Study of the Fast-Food Industry in New Jersey and Pennsylvania.” American Economic
Review84(4):772– 93.
———. 1995.Myth and Measurement. Princeton: Princeton University Press. Cunningham, James. 1981. “The Impact of Minimum Wages on Youth Employment,
Hours of Work, and School Attendance: Cross-sectional Evidence from the 1960 and 1970 Censuses.” InThe Economics of Legal Minimum Wages, ed. Simon Rottenberg, 88–123. Washington, D.C.: American Enterprise Institute.
Currie, Janet, and Bruce C. Fallick. 1996. “The Minimum Wage and the Employment of Youth: Evidence from the NLSY.”Journal of Human Resources 31(2):404– 28.
DiNardo, John, Nicole Fortin, and Thomas Lemieux. 1996. “Labor Market Institutions and the Distribution of Wages, 1973– 1992: A Semi-Parametric Approach.” Econometrica 64(5):1001– 44.
Efron, Bradley, and Robert J. Tibshirani. 1993.An Introduction to the Bootstrap. New York: Chapman & Hall.
about Parameters, Values, and Policies: Survey Results in Labor and Public Economics.” Journal of Economic Literature 36(3):1387– 425.
Gramlich, Edward M. 1976. “Impact of Minimum Wages on Other Wages, Employment, and Family Incomes.”Brookings Papers on Economic Activity2:409– 51.
Green, David A., and Harry J. Paarsch. 1998. “The Effect of the Minimum Wage on the Distribution of Teenage Wages.” University of British Columbia. Unpublished. Grossman, Jean Baldwin. 1983. “The Impact of the Minimum Wage on Other Wages.”
Journal of Human Resources18(3):359– 78.
Hungerford, Thomas L. 2000. “Does Increasing the Minimum Wage Increase the Incidence of Involuntary Part-Time Work?” Social Security Administration. Unpublished.
Katz, Lawrence F., and Alan B. Krueger. 1992. “The Effect of the Minimum Wage on the Fast-Food Industry.”Industrial and Labor Relations Review46(1):6– 21.
Lee, David S. 1999. “Wage Inequality in the United States During the 1980s: Rising Dis-persion or Falling Minimum Wage?”Quarterly Journal of Economics114(3):977– 1023.
Linneman, Peter. 1982. “The Economic Impacts of Minimum Wage Laws: A New Look at an Old Question.” Journal of Political Economy90(3):443– 69.
Michl, Thomas R. 2000. “Can Rescheduling Explain the New Jersey Minimum Wage Stud-ies?”Eastern Economic Journal26(3):265– 76.
Neumark, David, Mark Schweitzer, and William Wascher. 1998. “The Effects of Minimum Wages on the Distribution of Family Incomes: A Non-Parametric Analysis.” NBER Working Paper No. 6536.
Neumark, David, and William Wascher. 2002. “Do Minimum Wages Fight Poverty?”
Eco-nomic Inquiry40(3):315– 33.
———. 2000. “Minimum Wages and Employment: A Case Study of the Fast-Food Indus-try in New Jersey and Pennsylvania: Comment.”American Economic Review90(5): 1362– 96.
———. 1992. “Employment Effects of Minimum and Subminimum Wages: Panel Data on State Minimum Wage Laws.”Industrial and Labor Relations Review46(1):55– 81.
Spriggs, William E. 1993. “Changes in the Federal Minimum Wage: A Test of Wage Norms.”Journal of Post Keynesian Economics16(2):221– 39.
