Who Minimum Wage Increases Bite: An Analysis Using Monthly Data from the SIPP and the CPS.

Richard V. Burkhauser [*]

Kenneth A. Couch [+]

David C. Wittenburg [++]

We use monthly data from the Survey of Income and Program Participation and the Current Population Survey to estimate the effect of the minimum wage. Minimum wage increases significantly reduce the employment of the most vulnerable groups in the working-age population--young adults without a high school degree (aged 20-24), young black adults and teenagers (aged 16-24), and teenagers (aged 16-19). While we also find that minimum wage increases significantly reduce the overall employment of young adults and teenagers, these more vulnerable subpopulations are even more adversely affected.

1. Introduction

Neoclassical competitive models of firm behavior predict that wage increases reduce the quantity of labor demanded by firms and, holding the wage rate constant, that the least valued workers are the first fired or the last hired. [1] Minimum wage legislation exogenously increases the price of labor, and data from periods when the minimum wage changes can be used to empirically test its effect on the employment of vulnerable workers. However, a new body of research using such data suggests that minimum wage increases do not reduce employment and may even increase it (Card 1992a, b; Katz and Krueger 1992; Card, Katz, and Krueger 1994; Card and Krueger 1994, 1995). From a theoretical perspective, this "new economics of the minimum wage" has been characterized as an assault on the law of demand (Ehrenberg 1995).

If modest increases in the minimum wage have no employment effects, then the appropriateness of this method in helping the working poor is strictly a distributional issue. [2] However, if minimum wage increases reduce employment and if the jobs lost are concentrated among the vulnerable groups the policy claims to assist, then policy makers must consider this unintended consequence. Hence, estimating the elasticity of employment with respect to minimum wage increases is more than simply an empirical test of economic theory.

In this paper, we use monthly data from the Survey of Income and Program Participation (SIPP) and the Current Population Survey (CPS) to examine the effect of raising the minimum wage on employment. In contrast to much of the new minimum wage literature, we find that minimum wage increases significantly reduce employment. Moreover, we find that the elasticity of demand for labor with respect to increases in the minimum wage is greatest for the most vulnerable groups in the working-age population--young adults who have low levels of education, young black adults and teenagers, and teenagers. We also find that minimum wage increases reduce the overall employment of young adults and teenagers, though the estimated elasticities are smaller:

2. The New Economics of the Minimum Wage

The new minimum wage literature is dominated by studies that find that minimum wage increases have an insignificant or, in some cases, a positive and significant effect on the employment of young adults and teenagers (aged 16-24) or on other subgroups within that population (Card 1992a, b; Katz and Krueger 1992; Card, Katz, and Krueger 1994; Card and Krueger 1994, 1995). Most of these studies have now generated replies arguing that raising the minimum wage significantly decreases employment in these populations (Neumark and Wascher 1992, 1994, 1996, in press; Deere, Murphy, and Welch 1995; Taylor and Kim 1995; Burkhauser, Couch, and Wittenburg in press).

Because these new studies of the minimum wage vary widely in their techniques, we focus our discussion on those that use CPS data and employ an estimation strategy related to the one we use in this paper (Neumark and Wascher 1992, 1994, 1996; Card, Katz, and Krueger 1994; Card and Krueger 1995; Deere, Murphy, and Welch 1995; Burkhauser, Couch, and Wittenburg in press). [3] Specifically, we review studies that use a pooled, time-series, cross-sectional approach to estimate the employment effects of minimum wage increases on vulnerable groups, such as teenagers, with CPS data.

These studies use variation across states and over time in the minimum wage to identify employment changes. The typical reduced-form equation is

[E.sub.it] = [[alpha].sub.0] + [MW.sub.it][beta] + [X.sub.it][gamma] + [T.sub.t][tau] + [S.sub.i][delta] + [[epsilon].sub.it], (1)

where [E.sub.it] is the ratio of employed teenagers to the teenage population, [MW.sub.it] is a variable representing the minimum wage, [X.sub.it] is a set of explanatory variables, [T.sub.t] is a set of year dummy variables, and [S.sub.i] is a set of state dummy variables. The key coefficient of interest is [beta], which represents the effect of minimum wage increases on employment.

Neumark and Wascher (1992) estimate the effect of the minimum wage using the Kaitz index to measure minimum wage increases and find that this variable has a negative and significant effect on employment. [4] Based on the estimated coefficient on the Kaitz index, they conclude that the employment effect of the minimum wage on teenagers and on young adults and teenagers is negative and significant. They report elasticities ranging from -0.1 to -0.2 for teenagers and from -0.15 to -0.2 for young adults and teenagers. While the sign of the elasticities is consistent with the view that minimum wage increases cause employment decreases, they do not find consistent evidence that the effects are larger among younger, and presumably less skilled, teenage workers. They note the possible theoretical inconsistency of this result and discuss it at some length in later work (Neumark and Wascher 1996).

In an interchange with Neumark and Wascher (1994), Card, Katz, and Krueger (1994) (and later, Card and Krueger 1995) argue that the variables Neumark and Wascher (1992) use, particularly the Kaitz index, are inappropriate and drive their results. As an alternative to the Kaitz index, which Card, Katz, and Krueger (1994) argue is a flawed measure of the effect of minimum wage increases, they use the logarithm of the higher of the state or the federal minimum wage. [5] When Neumark and Wascher (1994) use this same model, they also find that minimum wage increases have no significant effect on teenage employment. Card and Krueger (1995) expand their argument regarding the misspecification of the Neumark and Wascher (1992) model and once again find no evidence that increases in the minimum wage reduce teenage employment. In some specifications, they actually find the effect on teenage employment is significantly positive.

Deere, Murphy, and Welch (1995) also use pooled repeated cross-sectional data from the CPS to estimate the impact of minimum wage increases, but their specification differs in some important ways from the studies discussed above. Deere, Murphy, and Welch (1995) estimate the effect of minimum wage increases over the period from 1985 through 1993. They construct intervals of monthly data from 1985 through 1993 that begin on the first day of April of each year and end on the last day of March during the following year. For each interval, they create annual averages of variables. These intervals correspond with the federal minimum wage increases that occurred in April 1990 and April 1991. Unlike the studies discussed above, they ignore the effects of state minimum wage increases and only focus on federal changes.

Another important difference between the Deere, Murphy, and Welch (1995) approach and others in this literature is that their specification excludes year effects. The minimum wage variables they use are year dummies that capture the effect of the 1990 and 1991 federal minimum wage increases. The first minimum wage dummy variable equals one in the period that the minimum wage was $3.80 per hour (from April 1990 to March 1991). The second minimum wage dummy variable equals one in the period that the minimum wage was $4.25 (April 1991-March 1993). Over the remaining periods in their data, the minimum wage was $3.35, and those year effects are captured in the constant. Hence, the two key dummy variables of interest measure the effect of raising the minimum wage relative to $3.35. [6]

They estimate the effect of the minimum wage on the employment of six demographic groups. They report employment declines of -4.8, -6.6, and -7.7%, respectively, for teenagers (aged 15-19) who are male, female, and black and employment declines of -1.5, -2.5, and -4.4%, respectively, for high school dropouts (aged 20-54) who are male, female, and black as a result of the increase in the minimum wage from $3.35 to $3.80 per hour in 1991. Although they do not calculate elasticities in their work, the implied elasticities given the 13.4% minimum wage increase from $3.35 to $3.80 are -0.37, -0.49, and -0.56, respectively, for teenagers who are male, female, and black and -0.12, -0.19, and -0.32, respectively, for high school dropouts who are male, female, and black.

Burkhauser, Couch, and Wittenburg (in press) examine each of the studies discussed above and conclude that the main difference between the Card and Krueger (1995) specification (and other papers in the literature based on Eqn. 1) and the Deere, Murphy, and Welch (1995) specification is in the interpretation of the year effects. Deere, Murphy, and Welch (1995) use a set of year dummies to capture the effects of federal minimum wage increases. In contrast, Card and Krueger (1995) include a minimum wage variable and individual year controls in their analysis. Conceptually, Card and Krueger (1995) attempt to distinguish between individual year effects and minimum wage increases, while Deere, Murphy, and Welch (1995) directly interpret the year variables as the federal minimum wage effect.

Burkhauser, Couch, and Wittenburg (in press) show that all of the variation associated with federal minimum wage increases is eliminated when year effects are included in Equation 1. This occurs because federal minimum wage rates are raised in discrete steps and typically do not change over a calendar year. Hence, the minimum wage effects in the Card and Krueger (1995) specification can only be identified by using the relatively small number of observations where the state minimum wage is higher than the federal minimum wage.

Because of this, studies like Card and Krueger (1995), which include year effects in Equation 1, actually estimate the marginal effect of state minimum wage increases relative to the federal level rather than the direct effect of federal minimum wage changes. While the experience of states that have a higher minimum wage rate than the federal minimum wage could be used as a proxy, Burkhauser, Couch, and Wittenburg (in press) question why one would want to do so when information on the variation in the federal minimum wage over time is available. To make this point empirically, they show that more than 90% of the variation in the key variable used by Card and Krueger (1995) to measure the minimum wage is eliminated when year effects are included and that the only remaining variation originates from differences between individual state minimum wages and the federal minimum wage level.

Burkhauser, Couch, and Wittenburg (in press) use monthly CPS data from 1979 to 1997 to re-estimate the primary Card and Krueger (1995) minimum wage specification and find that minimum wage increases significantly reduce teenage employment when year effects are excluded. They also explore the sensitivity of their results to a variety of alternative methods of controlling for the minimum wage as well as for different dimensions of time. For instance, they use alternative measures of the minimum wage, include lagged minimum wage values, and explore models similar to the one used in Deere, Murphy, and Welch (1995). They also test alternative specifications substituting recessionary dummies for year effects. Finally, they test the sensitivity of their results to alternative specifications of the error term, including corrections for state-specific autocorrelation, state-specific heteroskedasticity, and across-state correlations. Their findings do not substantively change across these alternative specifications. T heir elasticity estimates of the effect of an increase in the minimum wage on teenage employment are significant and range from -0.2 to -0.5.

This paper extends the Burkhauser, Couch, and Wittenburg (in press) analysis to other demographic groups at risk of employment loss from minimum wage increases. Specifically, we examine the effect of minimum wage increases on young adults and teenagers aged 16-24 as well as more vulnerable groups within that population of young workers. Because Card and Krueger (1995) demonstrate the sensitivity of minimum wage estimates based on Equation 1 to alternative sets of control variables, we test the sensitivity of our major results to the inclusion of different sets of control variables. All of our results come from the preferred Burkhauser, Couch, and Wittenburg (in press) specification that excludes year effects from Equation 1. [7] For key results, we provide, in the Appendixes, estimates based on a model specification similar to that of Deere, Murphy, and Welch (1995). [8]

3. Data

We use two data sources to examine the importance of minimum wage increases on employment. The first is the 1990 research panel of SIPP. The SIPP provides information on a representative sample of the U.S. population from January 1990 through May 1992. Because these data have not been used in this context before, we construct a parallel data set using monthly CPS information over the same period. We also extend our analysis using monthly CPS data from January 1979 through December 1997 in order to examine the robustness of our estimates over a longer time frame.

Our use of monthly data differs from Card and Krueger (1995), who use information from the May CPS from 1973 through 1989, and from Deere, Murphy, and Welch (1995), who use annual averages of months from the 1985 through 1993 CPS. [9] The advantage of using data for all months from the SIPP and CPS rather than one annual observation from the month of May or one annualized observation is that it increases the amount of information. The number of state-month observations potentially available from January 1990 through May 1992 from the SIPP and CPS is 1218 and 1479, respectively, versus the sample of 751 state-month observations employed by Neumark and Wascher (1992, 1994) or Card, Katz, and Krueger (1994) using May data only from 1973 through 1989. [10] In our longer (1979-1997) CPS data, the sample size increases to 11,628 state-month observations.

The 1990 SIPP

For each of four months beginning in February 1990, the Census interviewed a new rotation group that in itself was a random sample of the U.S. population for the 1990 SIPP panel. These four rotation groups were interviewed eight times at four-month intervals. Each interview contains monthly information for the preceding four months. We chronologically ordered the rotation groups to create a 29-month interval during which all four rotation groups had interview responses. The time period, from January 1990 through May 1992, includes both the April 1990 increase in the federal minimum wage from $3.35 to $3.80 per hour and the April 1991 increase from $3.80 to $4.25 per hour.

Weighting allows the SIPP data to be used in a number of different configurations while continuing to be a nationally representative sample. We chose to use only information on those individuals who responded to each of the eight surveys (over the 29-month period) and to weight accordingly.

Monthly Data from the CPS

To establish the robustness of our results, we construct a similar dataset from the monthly outgoing rotation groups of the January 1990--May 1992 CPS data. We again use weights to make the sample nationally representative. Our final set of estimates also uses monthly CPS data but from January 1979--December 1997. We report these estimates to show that our results are not unique to the 1990-1992 period.

Table 1 provides the definition of all variables used in our analysis. The weighted mean and standard deviation for each variable are provided in Appendixes 1-4. We use a set of control variables that includes the log of the higher of the federal or state minimum wage, the log of the average adult wage in the state, the proportion of each examined group's population in the state population, and the prime-age male unemployment rate in each state. [11] These control variables are defined to exactly match those in Card, Katz, and Krueger (1994), Card and Krueger (1995, Table 7.1), and Burkhauser, Couch, and Wittenburg (in press). As mentioned above, unlike Card, Katz, and Krueger (1994) and Card and Krueger (1995), we do not control for year effects in our specifications. [12]

All of the equations we estimate include controls for state effects and seasonal adjustment of the monthly data. The estimates are also weighted by the underlying state populations in each survey. We tried alternative models that used the number of observations available for constructing the state-month observations as a weight as well as unweighted models. These differences in weighting had no substantive effect on our results. [13]

Although not shown in the paper, we also explored the sensitivity of our estimates to a wide variety of alternative error corrections. We investigated the effect of corrections for common heteroskedasticity, groupwise heteroskedasticity, groupwise autocorrelation, cross-group correlations, and random effects. The resulting parameter estimates from these alternative corrections do not alter the conclusions drawn in the text. [14]

4. SIPP Estimates (January 1990-May 1992)

In each column of Table 2, we present estimates of Equation 1 for specific demographic groups. The estimates are based on SIPP data from January 1990 through May 1992. During this period, the minimum wage increased from $3.35 to $3.80 in April 1990 and to $4.25 in April 1991. At the bottom of each column, we report the elasticity of the particular group's employment to an increase in the minimum wage.

In columns (1) and (2) of Table 2, we provide estimates for the two age groups whose employment is considered to be most seriously affected by minimum wage increases-teenagers (aged 16-19) and young adults and teenagers (aged 16-24). For both groups, the minimum wage parameters are negative and significantly different from zero. One would expect the elasticity of employment with respect to the minimum wage to be larger among teenagers than among the combined group of young adults and teenagers because, on average, teenagers have less job experience. We find the expected ordering, with an estimated elasticity for teenagers of -0.566 and for the combined group of young adults and teenagers of -0.254. Furthermore, the underlying minimum wage parameters on which these elasticitises are based are significantly different from one another. [15] These elasticity estimates indicate that one would expect a 10% increase in the minimum wage to be associated with a 2.54% decline in the employment of young adults and teen agers and a much sharper 5.66% decline in teenage employment.

Within the group of young adults and teenagers, one would expect the employment elasticity from raising the minimum wage to be greater for the less skilled or for those likely to face discrimination in a market in which there is an excess supply of workers. To test these hypotheses, we divide the data on young adults and teenagers in column (2) by race in columns (3) and (4) and then segment the young adults (aged 20-24) by those with and without a high school diploma in columns (5) and (6). [16]

Our findings are consistent with expectations. The minimum wage parameters for both black and nonblack young adults and teenagers are negative, and they are both significantly different from zero and from one another. The estimated elasticity for black young adults and teenagers (-0.846) is more than four times the elasticity for nonblack young adults and teenagers (-0.181). Similarly, the minimum wage parameters for those with and without a high school diploma are negative, and they are both significantly different from zero and from one another. The estimated elasticity among high school graduates is -0.164. The comparable elasticity for those who have not received a high school degree is -0.852.

Appendix 5 contains estimates similar to those in Table 2 except that a set of dummy variables is used to represent each minimum wage value rather than a single variable constructed as the log of the actual level. All values of the minimum wage were included in the estimation, but for ease of exposition, we only report federal minimum wage values during the time frame of the table. [17] We find a significant negative effect on employment of raising the minimum wage for each group. Furthermore, the ordering of the minimum wage parameters is the same as we observe in Table 2. However, unlike Table 2, the minimum wage parameters of the different groups are not always significantly different from one another. [18]

The new economics of the minimum wage literature demonstrates the sensitivity of estimates of the employment effect of the minimum wage to specification changes. To help establish the robustness of our estimates, we conducted a range of sensitivity tests for each of the equations in Table 2. We found that, for all of our estimates, the results were insensitive to the types of specification changes other authors reported to be problematic. In Table 3, we provide an example of these robustness checks using our SIPP sample of teenagers. [19] Columns (1)-(4) of Table 3 contain estimates of Equation 1 where, across the columns, we drop, individually and in combination, the explanatory variables for the proportion of teenagers in the state population and the prime-age male unemployment rate. Across all specifications in Table 3, the minimum wage parameter is negative and significantly different from zero. We find that the estimated elasticities in each of the alternative specifications are larger than those in our base case in column (1). The estimated elasticities range from -0.566 to -0.623.

5. CPS Estimates (January 1990-May 1992)

The minimum wage parameters we obtain from the monthly STPP data across our various groups of young adults and teenagers are consistently negative and statistically different from zero. The elasticities for the groups one would expect the policy to most affect--teenagers, young blacks, and young adults without a high school education--show the greatest sensitivity to minimum wage increases. However, our estimated elasticities are above those reported in the past minimum wage literature and contrast sharply with the results of others in the new minimum wage literature.

Because the SIPP data have not been used before in studying the effects of the minimum wage, it is possible that our estimated elasticities are generated solely by some peculiarity in that sample. We address this issue by comparing our results with those from CPS monthly data for the same period, that is, January 1990-May 1992. Because of sampling variability, we did not use the CPS data to look at all the subgroups we considered in the SIPP. [20] We only construct samples for teenagers (aged 16-19) and for young adults and teenagers (aged 16-24).

Table 4 contains re-estimates of Equation 1 for these two groups using CPS data. The estimated minimum wage parameters are, negative and significantly different from zero for both teenagers (column 1) and young adults and teenagers (column 2), although the elasticity estimates are somewhat smaller than those obtained from the SIPP estimations in Table 2. However, we continue to observe the theoretically implied ordering of these elasticities, -0.456 for teenagers and -0.196 for young adults and teenagers, and the underlying minimum wage parameters on which these elasticities are based are significantly different from one another.

Some of the difference between Table 2 and Table 4 results may be related to basic differences in the sampling schemes of the two data sets. Two differences are worth highlighting. First, the CPS is a cross-sectional database that interviews different people in each month, while the SIPP is a panel database that interviews the same people every month. Because the SIPP reinterviews people, it naturally has much less sampling variability from one month to the next than the CPS. This is particularly true in our application because we only included observations of individuals present in every month of the data. [21] Second, the CPS asks employment questions about the preceding week while the SIPP asks such questions about the preceding month. Because the SIPP calendar period is longer, we believe the response is less subject to variability due to short-term phenomena such as not being employed for an entire week.

While we cannot directly measure the difference in our SLPP and CPS estimates due to design differences, there is one dimension in which we can offer some evidence. In the CPS, all states are individually identified, while the SIPP groups a number of smaller states. Our SIPP estimates exclude states that are not individually identified. [22] Our estimates may be biased if those states are systematically different from the others.

To investigate this potential bias, in columns (3) and (4) of Table 4, we re-estimate the CPS-based equations reported in the first two columns of Table 4 excluding the set of states omitted from our SIPP estimates. The parameter estimates are slightly larger than those in the first two columns, but the elasticities in columns (3) and (4) are not meaningfully different from comparable elasticities in columns (1) and (2). Although we are not able to fully explain the differences in our results using the CPS and SIPP data, the practical implication of Table 4 is that our basic conclusions using the SIPP do not substantially differ from those using similarly constructed CPS data. [23]

In Table 5, we perform the same set of robustness checks on our equation for teenagers using the CPS data that we conducted for our SIPP teenagers in Table 3. [24] Columns (l)-(4) contain estimates with various parameter exclusions. In each specification, the minimum wage parameter is negative and significantly different from zero. As in the SIPP estimates, little variation is observed in the estimated elasticities based on the alternative specifications contained in Table 5. The estimated elasticities range from -0.456 to -0.524.

6. CPS Estimates (January 1990-December 1997)

Our estimated elasticities for teenagers using either the SIPP or CPS monthly data are above the consensus range of estimates reported by Brown, Gilroy, and Kohen (1982) and Brown (1988). To determine if these results are unique to a period dominated by an economic downturn, we expand the time frame under consideration to include all years for which monthly CPS data are available. This expands our sample to cover all months from January 1979 through December 1997.

Table 6 contains estimates of Equation 1 for teenagers and for young adults and teenagers using monthly CPS data from January 1979 through December 1997. For teenagers and the combined group of young adults and teenagers, the estimated minimum wage parameter is negative and significantly different from zero. However, our coefficient estimates are somewhat smaller than in the shorter SIPP and CPS estimates. This finding is consistent with the view that the effect of minimum wage increases on employment during a recessionary period, such as the one covered in our analysis in the previous section, are consistently larger than those during a period of economic expansion.

We again observe an ordering of employment effects that is consistent with theory, but the two minimum wage parameters are not significantly different from one another. The estimated elasticities for teenagers (-0.302) and for young adults and teenagers (-0.197) are in the range of -0.1 to -0.3 reported by Brown, Gilory, and Kohen (1982). [25]

In Appendix 9, we re-estimate our Table 6 models but use inflation-adjusted values for the wage variables (log state minimum and log of average adult wage) using the Bureau of Labor Statistics' CPI--All Consumers Index. Controlling for price changes in this manner does not substantially alter our results. The negative minimum wage effect is still statistically significant, but the estimated elasticities fall to -0.248 for teenagers and -0.166 for young adults and teenagers.

In Table 7, we perform the same set of robustness checks on our equation for teenagers as we conducted in Table 5. [26] In each specification, the minimum wage parameter is negative and significantly different from zero and falls within the consensus range of estimates. Little variation is observed in the estimated elasticities when various combinations of explanatory variables are excluded from the calculations.

A final issue we explore is whether our results are dependent on using all months of the data and could be obtained using only information from the month of May, as has been done in most other studies. Table 8 presents the estimates of minimum wage increases on teenagers using annual CPS data (1979-1997) for the months of January-December. The minimum wage parameter is negative and statistically significant in the annual data for every individual month. Not surprisingly, we find that the employment effects of minimum wage increases tend to be largest during the summer months (June, July, and August) when teenage employment is at its peak. The estimated elasticities range from -0.148 (February) to -0.592 (June). We find a similar pattern of results when we compare the estimate for the month of May (the month used in many of the studies in the new literature on the minimum wage) in Table 8 to the estimate using all months in Table 6.

7. Conclusions

This paper measures the consequences of minimum wage increases for the employment of vulnerable groups in the labor market. We consistently find that minimum wage increases have a significant negative employment effect on each vulnerable group we investigate. Furthermore, we find the ordering of the magnitude of the negative employment effects is consistent with our expectations that employment of the most vulnerable groups is most adversely affected. The largest effects are observed for young adults without a high school degree, young black adults and teenagers, and teenagers.

The employment elasticities we estimate for teenagers using data from 1990 through 1992 are generally above the consensus range of estimates reported by Brown, Gilroy, and Kohen (1982). Estimates for teenagers using CPS data from 1979 through 1997 are uniformly within that range. While the 1990-1992 data cover the minimum wage increases in 1990 and 1991, they also span a major recessionary period. We believe that the estimates using those data capture the employment effect of the relevant minimum wage increases but may overstate the responsiveness of employment in other periods. Our finding of a smaller response of teenage employment to increases in the minimum wage in our 1979-1997 CPS sample is consistent with this view. The results from this longer period also demonstrate that, even in samples that include robust periods of economic growth, the effect of increasing the minimum wage on employment is negative and statistically significant for vulnerable groups.

The evidence presented in the paper is consistent with neoclassical predictions that minimum wage increases significantly reduce employment, with their largest effects felt by the most vulnerable groups of workers. This is important in the broader context of minimum wage policy. Burkhauser, Couch, and Wittenburg (1996), using simulations that assume raising the minimum wage has no effect on employment, demonstrate that minimum wage increases are not a target-effective way of helping the working poor. They find that less than 20 cents of every dollar of the increased wage bill associated with raising the minimum wage from $3.35 to $4.25 actually flowed to poor families. Here we show that ignoring behavioral effects is likely to lead to an overstatement of the efficacy of minimum wage increases in helping the working poor because the employment of more vulnerable and presumably lower income groups is more likely to be adversely affected by them.

(*.) Department of Policy Analysis and Management, Cornell University, Ithaca, NY 14853, USA; E-mail [e-mail]; corresponding author.

(+.) Department of Economics, University of Connecticut, 341 Mansfield Road, Storrs, CT 06269, USA; E-mail [e-mail].

(++.) The Lewin Group, 3130 Fairview Park Drive, Suite 800, Falls Church, VA 22042, USA; E-mail [e-mail].

This paper has been through many drafts, each of which has been improved, thanks to careful readings and comments by Charles Brown, Donald Deere, Doug Holtz-Eakin, Marvin Kosters, Alan Krueger, David Neumark, William Rodgers, Steve Ross, Scott Scrivner, David Stapleton, and William Wascher as well as seminar participants at Cornell University, Georgia Tech University, Syracuse University, Queen's College, the Bureau of Labor Statistics, and the American Economic Association Meeting (January 1998). Martha Harrison provided valuable research assistance.

(1.) The neoclassical arguments against minimum wage increases on efficiency and income distributional grounds were first fully laid out by Stigler (1946).

(2.) An existing literature examines the distributional effects of minimum wage increases. See Burkhauser and Finegan (1989), Horrigan and Mincy (1992), Burkhauser, Couch, and Glenn (1996), and Burkhauser, Couch, and Wittenburg (1996).

(3.) A more detailed discussion of the recent literature on the minimum wage, as well as the work that preceded it, can be found in Brown, Gilroy, and Kohen (1982), Brown (1988), Card and Krueger (1995), Neumark and Wascher (1996), and Burkhauser, Couch, and Glenn (1996).

(4.) The Kaitz index is the ratio of the minimum wage to the average adult wage multiplied by the percent of persons covered by the minimum wage.

(5.) Among other criticisms, Card, Katz, and Krueger (1994) show that the Neumark and Wascher estimation results are sensitive to the use of two variables. First, the construction of the variable Neumark and Wascher use to control for schooling in the CPS includes only teenagers in school and not working. This control variable for schooling not only underestimates the number of teenagers in school by excluding the intersection of teenagers in school and working but its construction is systematically related to the dependent variable measuring teenage employment. When the schooling variable is dropped from the estimates, the parameter used to measure the impact of the minimum wage, the Kaitz index, is no longer statistically significant.

A second important criticism made by Card, Katz, and Krueger (1994) concerns the use of the Kaitz index as a measure of the minimum wage. They argue that the inclusion of adult wages as a function of the Kaitz index is problematic because adult wages, which are included in the denominator of the Kaitz index, are correlated with economic activity and hence should be correlated with the teenage wage rate. Factors that lead to an increase in adult wages, however, will lead to a decline in the Kaitz index. Therefore, the Kaitz index may be negatively correlated with teenage wages and a poor measure of how the minimum wage increase affects teenagers. They demonstrate this point by regressing teenage wages on the Kaitz index. They find a negative relationship between the Kaitz index and teenage wages. However, when teenage wages are regressed on a more direct measure of the minimum wage, that is, the larger of the natural log of the state or the federal minimum wage, the expected positive and significant relations hip is found. Using this direct measure of the minimum wage and dropping the problematic schooling variable, their estimates indicate that the effect of raising the minimum wage on employment is positive and statistically significant at the 0.05 level.

(6.) Deere, Murphy, and Welch (1995) used an F-test to show that additional individual year dummies should not be included. They ignore the effects of state minimum wage rates in their analysis.

(7.) In Burkhauser, Couch, and Wittenburg (in press), we note that, in specifications that do not include individual year controls, the minimum wage variable may capture unobserved macroeconomic factors. Hence, our estimates using the 1990-1992 CPS and SIPP samples could be viewed as upper-bound estimates of the effect of minimum wage increases. As will be seen, this may explain why our minimum wage estimates for this period of economic recession are higher than those based on the 1979-1997 CPS data.

(8.) In these estimates, we replace the constructed minimum wage variable with a set of categorical variables representing each individual level of the minimum wage with $3.35 as the omitted category. Between January 1979 and December 1997, we observe 41 different state or federal minimum wage values for our data. While our actual estimates include a variable for each individual level of the minimum wage, in all the Appendixes except one, we only report the parameter estimate associated with each federal minimum wage level.

(9.) While wage information is available for the month of May back to 1973 in the CPS, the same information that was collected in May only became available on a monthly basis for each outgoing rotation group in 1979. This explains the use of the May observations from 1973 through 1989 in the work of Neumark and Wascher (1992) and others and our use of the data beginning in 1979.

(10.) Our SIPP data have fewer state-month observations than the CPS data because, unlike the CPS, eight smaller states are grouped in the SIPP. We exclude the grouped states in our analysis.

(11.) Where a wage rate is directly reported, we use this value; otherwise, we divide reported salary by the relevant hours of work.

(12.) As discussed above, Burkhauser, Couch, and Wittenburg (in press) demonstrate theoretically and empirically why year controls are inappropriate in this context.

(13.) The unweighted estimates of the minimum wage parameter, which are available from the authors upon request, are also significantly different from zero at conventional levels and tend to show larger negative employment effects than the weighted estimates we report in the paper.

(14.) Representative examples for some of these corrections are reported in Burkhauser, Couch, and Wittenburg (in press).

(15.) Assuming independence, a t-statistic for the zero null for the difference between the minimum wage parameters for teenagers (-0.246) and young adults and teenagers (-0.143) is calculated as (-0.246 - (-0,l43))/[[[(0.033).sup.2] + [(0.022).sup.2]].sup.0.5] = 2.64. Thus, one would reject the null hypotheses at the .05 level. We refer to similar tests of differences in means throughout the paper but do not report the calculation results. They are available upon request.

(16.) We exclude teenagers when comparing high school graduates and nongraduates because we do not want to include persons in the nongraduate category who fall into this classification solely on the basis of their age. Presumably, if we include all persons aged 16-24 in our analysis of these two groups, the majority of individuals aged 16-19 would not have completed their high school degree because they are still in school.

(17.) In Appendix 8, which covers the longest period of our analysis (January 1979-December 1997), we report the full set of estimated coefficients for all 41 minimum wage values. The coefficient for the shorter SIPP and CPS panel are available on request.

(18.) The parameter for the dummy variable representing a $3.80 minimum wage is always significantly different from $3.35 and its ordering is consistent with expectations, but the difference between pairs of parameters is significantly different only for high school graduates verses non-high school graduates. The parameter for the dummy variable representing a $4.25 minimum wage is always significantly different from $3.35, the ordering is as expected, and there is a significant difference between all paired parameters of interest.

(19.) Estimates similar to those in Table 3 are available upon request from the authors for each subgroup of the SIPP examined in Table 2.

(20.) In our estimations using both the CPS and SIPP data, we only use groups that we can identify across all periods in our estimations. Due to the panel nature of the SIPP, less sample variability exists in identifying some of these groups across smaller states than exists in repeated cross-sections like the CPS. As a result of sampling variability in the CPS, we are unable to generate a balanced panel of state observations without dropping a large number of states in the black and high school dropout equations.

(21.) The SIPP weights allow one to make this choice and still obtain nationally representative estimates.

(22.) These states are Vermont, Maine, Iowa, North Dakota, South Dakota, Alaska, Idaho, Montana, and Wyoming. It should be noted that, in the work of Neumark and Wascher (1992) and Card, Katz, and Krueger (1994), only 22 states are used prior to 1976, so a similar issue arises in their work.

(23.) Appendix 6 contains a parallel set of estimates to those found in Table 4 except that we substitute a set of dummy variables for the single variable that represents the logged values of the minimum wage. All values of the minimum wage were included in the estimation, but we only report federal minimum wage values during the period in the table. The parameter for the dummy variable representing a $3.80 minimum wage is always significantly different from $3.35, and its ordering is consistent with expectations, but the difference between pairs is not significant. With the exception of parameters for teenagers and young adults and teenagers in the full CPS sample, which are significantly different from one another, the same pattern holds for the dummy variable representing $4.25. As was the case in Table 4, the estimated elasticities of employment with respect to the minimum wage are again smaller in the CPS sample with all states relative to the sample that omits the states missing from our SIPP sample.

(24.) A full set of these robustness checks is available from the authors upon request.

(25.) Appendix 7 contains a parallel set of estimates to those found in Table 6 except that we substitute a set of dummy variables for the single variable that represented the logged values of minimum wage. All 41 values of the minimum wage were included in the estimation, but we only report the 7 federal minimum wage values during the period in the table. The parameter for the dummy representing each of the federal minimum wage values is significantly different from $3.35. In all cases, the parameter value for teenagers is larger than the parameter value for young adults and teenagers. But this difference is not significant for the $3.80 parameter value. In Appendix 8, we report the full set of coefficients for all the minimum wage values. In general, the signs are as expected, positive for values below $3.25 and negative for values above $3.25. However, they are not always significant.

(26.) Similar sets of robustness checks for each of the equations are available from the authors upon request.

References

Brown, Charles. 1988. Minimum wage laws: Are they overrated? Journal of Economic Perspectives 2:133-47.

Brown, Charles, Curtis Gilroy, and Andrew Kohen. 1982. The effect of the minimum wage on employment and unemployment. Journal of Economic Literature 20:487-528.

Burkhauser, Richard V., Kenneth A. Couch, and Andrew J. Glenn. 1996. Public policies for the working poor: The earned income tax credit versus minimum wage legislation. In Research in labor economics, volume 15, edited by Sol W. Polachek. Greenwich, CT: JAI Press, pp. 65-110.

Burkhauser, Richard V., Kenneth A. Couch, and David C. Wittenburg. 1996. Who gets what from minimum wage hikes: A replication and re-estimation of Card and Krueger. Industrial and Labor Review 49:547-52.

Burkhauser, Richard V., Kenneth A. Couch, and David C. Wittenburg. In press. A reassessment of the new economics of the minimum wage literature using monthly data from the CPS. Journal of Labor Economics.

Burkhauser, Richard V., and Aldrich T. Finegan. 1989. The minimum wage and the poor: The end of a relationship. Journal of Policy Analysis and Management 8:53-71.

Card, David. 1992a. Using regional variation in wages to measure the effects of the federal minimum wage. Industrial and Labor Relations Review 46:22-37.

Card, David. 1992b. Do minimum wages reduce employment? A case study of California, 1987-1989. Industrial and Labor Relations Review 46:38-54.

Card, David, Lawrence F. Katz, and Alan B. Krueger. 1994. Comment on David Neumark and William Wascher, 'Employment effects of minimum and subminimum wages: Panel data on state minimum wage laws.' Industrial and Labor Relations Review 48:487-96.

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 Review 84:772-93.

Card, David, and Alan B. Krueger. 1995. Myth and measurement: The new economics of the minimum wage. Princeton, NJ: Princeton University Press.

Deere, Donald, Kevin M. Murphy, and Finis Welch. 1995. Reexamining methods of estimating minimum wage effects: Employment and the 1990-1991 minimum wage hike. American Economic Association Papers and Proceedings 85 :232-7.

Ehrenberg, Ronald G. 1995. Review symposium: Myth and measurement: The new economics of the minimum wage, by David Card and Alan B. Krueger. Industrial and Labor Relations Review 48:827-8.

Horrigan, Michael W., and Ronald B. Mincy. 1992. Public policy changes and the distribution of income. In Uneven tides: Rising inequality in America, edited by Sheldon Danziger and Peter Gottschalk. New York: Russell Sage Foundation, pp. 251-76.

Katz, Lawrence, and Alan B. Krueger. 1992. The effect of the minimum wage on the fast food industry. Industrial and Labor Relations Review 46:6-21.

Neumark, David, and William Wascher. 1992. Employment effects of minimum wages and subminimum wages: Panel data on state minimum wage laws. Industrial and Labor Relations Review 46:55-81.

Neumark, David, and William Wascher. 1994. Employment effects of minimum and subminimum wages: Reply to Card, Katz, and Krueger. Industrial and Labor Relations Review 47:497-512.

Neumark, David, and William Wascher. 1996. Reconciling the evidence on employment effects of minimum wages: A review of our research findings. In The effects of the minimum wage on employment, edited by Marvin H. Kosters. Washington, DC: AEI Press, pp. 55-86.

Neumark, David, and William Wascher. In press. The effect of New Jersey's minimum wage increase on fast-food employment: A re-evaluation using payroll records. American Economic Review.

Stigler, George J. 1946. The economics of minimum wage legislation. American Economic Review 36(June):358-65.

Taylor, Lowell J., and Traeil Kim. 1995. The employment effect in retail trade of California's minimum wage increase. Journal of Business and Economic Statistics 13(2)(April):175-82.