The wage and employment dynamics of minimum wage workers.

Macpherson, David A.

1. Introduction

An important issue in the policy debate over the merits of increasing the minimum wage is the duration of minimum wage employment. At one extreme, if minimum wage workers were entry-level workers who quickly accumulate skills that push their wages above the minimum, a minimum wage hike would have a relatively short term effect on any given worker's income. On the other extreme, if all minimum wage workers find themselves in "dead end" jobs that provide no opportunities for wage growth, a hike in the minimum wage could have long-term effects on their incomes.

There are several studies that provide insights into the wage growth of minimum wage workers. Using 1984-4985 data from the Survey of Income and Program Participation (SIPP), Smith and Vavrichek (1992) show that over 60% of workers earning the minimum wage in 1984 were earning more than the minimum wage one year later. For those with wage gains, the typical wage increase was approximately 20%. Long (1999) finds similar results using the 1992-1993 SJPP. Using National Longitudinal Survey data from the early 1980s, Schiller (1994) finds that, after entering a minimum wage job, only 15% of continuing workers were still earning the minimum wage after three years.

While most minimum wage workers realize sufficient wage growth to rise above the minimum wage relatively quickly, a significant minority does not. The existing literature provides some insight into which workers are most likely to be "stuck" at the minimum. A fairly consistent finding is that less educated workers and part-time workers are less likely to rise above the minimum. However, the understanding of the processes at work in determining the extent of wage growth is limited.

Perhaps the most-studied determinant of wage growth for minimum wage workers is the extent of on-the-job training. Hashimoto (1982) argues that a hike in the minimum wage could reduce the amount of training workers receive and thereby reduce their subsequent wage growth. More recently, however, Acemoglu and Pischke (1999) argue that, in noncompetitive labor markets, a minimum wage hike could actually increase employer-provided training. Empirical evidence on the effects of a minimum wage hike on training is mixed. (1)

This study uses 20 years of short panel data sets on minimum wage workers to improve the understanding of the wage and employment dynamics of minimum wage workers in several respects. First, it examines the degree of labor market attachment of minimum wage workers by computing transition rates into and out of minimum wage employment. The transitions could be the result of entry or exit from the labor market or from wage increases or decreases. The evidence reveals that minimum wage employment is short lived for many workers. Second, among minimum wage workers who continue employment, we explore alternative explanations for wage growth. Our study examines the importance of job training, job switching, and changes in nonwork conditions that might lead workers to accept jobs that are not the best match for their skills. We find that switching jobs is vital to significant wage growth among minimum wage workers, particularly for young workers who find themselves in "low-training" occupations. We also demonstrate t hat finishing school improves wage growth, partly by increasing the chance that a worker makes a job change. Reenrolling in school has the opposite effect.

2. Data

The monthly Current Population Survey (CPS) Outgoing Rotation Group (ORG) files from January 1979 through December 1999 allow construction of a series of 20 separate two-year panel data sets on minimum wage workers. The CPS is structured so that a given household is sampled four consecutive months, not interviewed for eight months, and then interviewed for another four consecutive months. When the household leaves the sample at the end of the first or last four-month period of interviews, it is part of an ORG. The matched ORG files provide information on a person at the beginning and end of a one-year period. (2)

We construct two samples of minimum wage workers from the CPS-ORG files. (3) The first includes all workers who earned the minimum wage in the reference week of the first year of the two-year panels, regardless of their earnings or employment status in the second year. This sample provides information on the wage growth, job change, and employment behavior of minimum wage workers. The second data set includes all workers who earned the minimum wage in the second year of the two-year panels, regardless of their earnings or employment status in the first year. These data allow us to investigate the path into minimum wage employment.

In addition to the minimum wage samples, we create a comparison sample that includes workers earning above the minimum wage in the first year of the panels.

Table 1 provides sample means for some key demographic characteristics for workers earning the minimum wage in the first year of the panels. The first-period minimum wage sample has 33,520 workers and the comparison sample 923,752 workers earning above the minimum wage.

Compared to workers earning above the minimum wage, workers in minimum wage jobs are younger and less educated and work fewer hours. Relative to workers earning above the minimum, minimum wage workers are eight times more likely to be under age 21, nearly four times as likely to have no high school degree, and six times more likely to work less than 20 hours per week.

3. Transitions to and from Minimum Wage Employment

In this section, we examine transitions into and out of minimum wage employment. In Table 2, the frequencies of the four transition rates into and out of minimum wage employment are listed. The transitions into minimum wage employment are from nonemployment or from employment that paid a wage at, above, or below the minimum wage. The transitions out of minimum wage employment are to no employment and from minimum wage employment to employment paying at, above, or below the minimum wage.

Minimum wage jobs are often thought of as "entry-level" jobs. The evidence in our data supports this view. Among the workers earning the minimum wage in the second year of our panels, 39.4% were not employed in the prior year. Among the comparison sample of people earning above the minimum wage in the second period of the sample, only 8.5% reported no employment in the prior year. Hence, minimum wage workers are approximately five times more likely to be entrants from a spell of nonemployment than those earning above the minimum. Among the minimum wage workers who were not employed in the prior year, the most common reasons reported for nonemployment were that they were enrolled in school (44%), were unemployed (24%), or were doing housework (16%).

While a large share of minimum wage workers is beginning a new spell of employment, the majority (about 60%) is not. Approximately one-quarter (22.6%) were minimum wage employees in the prior year, and another one-quarter (27.8%) were earning above the minimum wage. Among people working in both periods, this implies that the chance of earning above the minimum wage in the prior year is nearly one-half. Hence, while several studies emphasize that the majority of minimum wage workers rises above the minimum in a short period of time, it is also true that a substantial share of minimum wage workers previously earned above the minimum wage.

Approximately 10% of the workers earning the minimum wage in the second year of the panels were earning less than minimum in the prior year. One explanation for workers earning less than the minimum wage is that minimum wage laws do not apply to all employees. Under current legislation, workers under the age of 20 can be paid a "subminimum" wage for their first 90 days of employment. There are also exemptions for disabled workers, full-time students, and students enrolled in vocational education programs. Workers who receive tip income (e.g., waiters and waitresses) can currently be paid as little as $2.13 an hour if their reported tip income is sufficient to bring their total hourly income to $5.15. Consistent with these exemptions is the fact that, among workers earning the minimum wage in the second year of the panels, the waiter and waitress occupations had the greatest fraction of workers being paid below the minimum in the prior year. (4) Also, minimum wage workers under the age of 21 are more likely to have been paid less than the minimum in the prior year.

Minimum wage employment is short lived for the vast majority of workers. Among workers earning the minimum wage in the first year of the panels, only 21.6% of minimum wage workers are still earning the minimum by the end of the year. Nearly one-half (47.2%) of the first-year minimum wage workers report rising above the minimum wage by the end of the year. A small percentage (7.3%) remains employed but earns a wage below the minimum in the second year, and approximately one-fourth (23.9%) leave employment by the second year. The fraction of workers leaving employment by the second year is only one-third as high (8.6%) in the comparison sample of workers earning above the minimum wage in the first period.

4. Determinants of Wage Growth

In this section, we examine the distribution of wage growth for people who are earning the minimum wage in the first year of the CPS-ORG panels. While the earlier analysis reveals that the majority of workers continuing employment from one year to the next will rise above the minimum wage, the extent of wage growth varies dramatically across minimum wage workers, the type of job in which they start, and their propensity to change employers, occupation, and industry. We also compare the wage growth of minimum wage workers with a comparison sample of workers earning above the minimum wage in the first year of the panels.

Table 3 provides a comparison of real wage growth of minimum wage workers with the comparison sample of people earning above the minimum wage. (5) Median real wage growth for the comparison sample of workers earning above the minimum wage in the first year of the panels is 1.3% over the 1980-1999 sample period. This is substantially below the 3.5% median wage growth for workers earning the minimum wage in the first year of the panels. (6) Among minimum wage workers who rise above the minimum by the second year of the panels, median wage growth is 14.8%. Reflecting the decline in the real value of the minimum wage over the sample period, workers who earned the minimum wage in both years of the panels experienced a median reduction in real wages of 3.6%. (7)

There are at least three distinct ways that on-the-job experience could lead to wage growth for minimum wage workers. First, workers may receive job-specific skills making them more productive in their jobs with their current employer. Second, the minimum wage job may provide general skills that allow the worker to move into jobs with greater earnings potential. However, it is conceivable that a switch of employers could be necessary to obtain a match that rewards these newly acquired skills. Third, it is possible that wage growth occurs because a worker is in a minimum wage job only because temporary circumstances preclude accepting a job with greater earnings potential. When circumstances change, the worker moves to an employer that can make better use of his skills (e.g., a person enrolled in school may be capable of acquiring a job with greater skill requirements and higher pay but chooses the minimum wage job because of greater flexibility in hours or a shorter commute. On leaving school, he switches to a job that has higher skill requirements and pays more).

Depending on the source of wage growth, changing employers will have differential effects on wage growth. If the source of wage growth is the accumulation of employer-specific skills, a switch of employers will retard wage growth. If the accumulation of general skills is the source, employer switching would have a positive effect if it leads to a better match. Alternatively, if switching employers improves the match between skills and job requirements, employer switching will improve wage growth.

In an investigation of the determinants of wage growth, Gottschalk (2001) finds that switching employers enhances wage growth, particularly for less educated workers. We extend this line of reasoning by examining the importance of switching employers to the wage growth of minimum wage workers. We also examine the importance of other personal and job attributes that might impact the extent of wage growth as well as the importance of employer switching to wage growth.

A change of employer in the CPS data can be measured two ways. Our first measure is based on the subset of panels that contain information on employee tenure. (8) If a person who reports work in both years of the panel indicates she has less than a year of tenure, we assume that she switched employers. (9) Among workers earning the minimum wage in the first year of the panels, we estimate that 30.7% switch employers by the second year. (10) The rate of job switching among workers earning above the minimum wage is less than one-third as high (9.5%).

As a second measure of job switching, we compare the three-digit occupation and industry reported in the two years of the panels. This measure of job switching has an important advantage in that all the panel data sets contain industry and occupation information, allowing for a much larger sample for data analysis. A shortcoming of the industry/occupation-based measure is that there is measurement error in the classification methods. (11)

Changes in industry and occupation provide a different kind of information than the tenure-based measure. The tenure-based measure reflects a change in employers, regardless of whether there is a significant change in job duties. Job switches based on changes in both industry and occupation reflect a change in both employer and job type. Among workers who report employment in both years of the panels, we estimate that 34.9% of those earning the minimum in the first year of the panel change both industry and occupation. In the comparison sample of workers earning above the minimum wage in the first year and continuing employment into the second year, only 18.8% report a change in both occupation and industry. Minimum wage workers are almost twice as likely as other workers to change occupation and industry.

The effect of employer switching on wage growth is illustrated for minimum wage workers and the comparison sample in Table 4. The estimates imply that changing employers or occupation and industry is associated with significantly higher wage growth for minimum wage workers but slightly lower wage growth for workers earning above the minimum wage. A switch of employers improves wage growth by 3.6 percentage points for minimum wage workers but worsens wage growth by 1.6 percentage points for workers in the comparison sample. A switch of both occupation and industry enhances wage growth by 10.8 percentage points for minimum wage workers but reduces wage growth by 0.1 percentage points for workers in the comparison sample. Employer switching, especially when combined with a change in occupation and industry, is particularly important to the wage growth of minimum wage workers. This suggests that the accumulation of general skills or improvement in match quality is more important than the accumulation of firm-spec ific skills. More evidence on this point is provided here.

Another potential source of unusual wage growth are changes in nonwork circumstances that improve the chance of accepting employment that is a better match to the worker's skills. One example would be when a worker either finishes or starts a spell of school enrollment. To examine this, we restrict the sample to observations with enrollment information--workers age 24 or less in surveys from 1984 forward. We then divide workers into four groups depending on their enrollment status at the beginning and end of the panels and compute median wage growth.

Information on the importance of school enrollment for minimum wage workers is given in Table 4. School enrollment is common for many minimum wage workers. For workers under age 24 earning the minimum wage in the first year of the panels, 43.0% report being enrolled in school in both years, 15.6% report being enrolled in the first year but not the second, and 4.7% report no enrollment in the first but enrollment in the second.

Workers who stay enrolled or start enrollment have lower wage growth than those who either finished a spell of school enrollment or were not enrolled in either period. This pattern holds for minimum wage workers and those earning above the minimum wage, and the size of the enrollment effects are similar. It thus appears that school enrollment restricts a worker's earnings potential and that completion of an enrollment spell leads to above-average wage growth.

Changes in work hours might also account for large changes in wages for workers. Unfortunately, we cannot determine whether observed changes in hours are voluntary. However, we expect that a willingness to increase from part-time to full-time work would increase access to higher-paying jobs. In the reverse direction, a desire to cut hours may reduce access to high-paying jobs. To investigate the validity of this hypothesis, we divided workers into five groups depending on the change in hours worked per week between the first and second year of the panels. Wage growth is substantially higher for those who start at the minimum wage and increase work hours. For example, median wage growth for those who increase work hours by 20 or more hours per week is 20.6%. This compares to median growth of 4.4% for those who changed hours worked by fewer than 10 hours per week. The relationship between changes in hours and wage growth is much weaker in the comparison sample. Moreover, the relationship between wage growth and either sharp increases or decreases in hours worked is the opposite of that found for minimum wage workers.

Given that many of the factors affecting wage growth will be correlated with worker characteristics, multivariate analysis is necessary to sort out the separate effects of each factor. We use median regression methods to further investigate the determinants of wage growth for minimum wage workers. (12) Particular attention is paid to the variation in wage growth in the different types of minimum wage jobs, the importance of training, and the extent to which job changes, changes in school enrollment, and changes in hours worked affect wage growth.

Table 5 presents the results of median regressions estimating the determinants of the percentage change in real wages for the minimum wage and comparison sample. The sample is restricted to workers who are employed in both years of the two-year panels. In addition to variables describing the worker and the job, year dummies are included to control for macroeconomic conditions and changes in the level of the federal minimum wage. State dummies are included to control for state-specific effects.

The regression estimates reveal that for both the minimum wage and the comparison sample, wage growth rises with education. However, the effect of education is much sharper in the minimum wage sample. Compared to workers with a high school degree, wage growth among workers with a college degree is 9.2 percentage points higher for workers in the minimum wage sample but only 1.0 percentage point higher for workers in the comparison sample. In both samples, wage growth rises and then falls with age and is greater for full-time than part-time workers.

The 16 occupation dummies included in the model were formed by selecting the 15 three-digit occupation groupings that employ the largest number of minimum wage workers. The 16th occupation grouping (listed as "other") includes all other occupations. The largest minimum wage occupation (cashiers) is the reference group in the regression. There is statistically significant variation in wage growth across the occupations. Comparing the minimum wage occupations with the highest and lowest wage growth reveals a difference of 12.4 percentage points in the minimum wage sample and a difference of 6.5 percentage points in the comparison sample. The ranking of wage growth across occupations is very similar in the minimum wage and comparison samples. (13)

To provide further insight into why wage growth differs across workers and occupations, we estimate several additional specifications of the wage growth equation. Our first objective is to determine the importance of job training and job switching in determining wage growth. The job-training measures reflect the percentage of workers in a given three-digit occupation reporting that they received training for their job through their employer ("firm training") or by someone other than the employer ("outside training"). (14) The measure of "job change" indicates whether the worker switches industry and occupation. (15)

To discern the effect of job training and job switching on wage growth, variations of the median wage growth equation presented earlier are estimated for the minimum wage and comparison samples. In the first specification, we add controls for job switching and training. A potential concern with this specification is that training or job changes could be endogenous in the wage growth equation. For example, Neumark and Wascher (2001) cite evidence that firms may select the workers with higher levels of ability for training. If ability is not adequately controlled for, this could lead to an upward bias in the estimated effect of training on wage growth. A similar problem would emerge if those with greater wage growth potential were more likely to change industry and occupation. For example, if workers who have skills that are not fully utilized in their current job are more likely to switch industry and occupation, our estimate of the job-switching effect could partly reflect this. Since the available data do no t allow us to properly address these potential endogeneity problems, the results must be interpreted with some caution.

With these caveats in mind, the estimates for specification 1 in Table 6 imply that, other things being the same, median real wage growth is 5.8% higher for workers who change industry and occupation. This is in stark contrast to workers earning above the minimum wage in the first period, where a change in industry and occupation has a slightly negative but statistically insignificant effect. One might interpret this difference as an indication that a disproportionate share of minimum wage workers are in jobs that are a poor match to their skills and can thus benefit substantially from a change in industry and occupation.

Firm-provided job training has a larger positive effect on wage growth for minimum wage workers than for workers earning above the minimum wage. Job training provided outside of the firm has a positive effect on the wage growth of minimum wage workers but a statistically insignificant negative effect on the wage growth of workers earning above the minimum wage. Given evidence that minimum wage increases reduce access to training for minimum wage workers, a minimum wage hike could have adverse effects on workers' long-term wage prospects. (16)

According to job-matching theories of wage growth, job switching early in a worker's career is particularly important to wage growth. (17) When a worker first begins employment, he continues switching employers until an acceptable match is made. With each switch, the quality of the match (and the corresponding wage rate) improves. Eventually, a good match is found, and the returns to job switching diminish.

To determine whether the effect of job matching on wage growth diminishes with age, the wage growth models are estimated with the addition of interactions between the job switch variable and dummies representing the worker's age. The estimates indicate that, for minimum wage workers, changes in occupation and industry enhance wage growth for workers in all age categories. However, the effect of job changes rises until the worker is in his mid-20s and then begins to fall. Whereas a job change enhances wage growth by 13.2% for a worker between the ages of 22 and 25, its effect is only 0.9% among workers age 65 or more. Switching occupation and industry has much smaller effects on wage growth in the comparison sample of workers earning above the minimum wage, and there is no consistent pattern across age-groups.

The effect of a job change on wage growth should vary depending on the type of training and skills that a worker acquires on the initial job. If training provides skills that are firm specific, a job change may worsen wage growth. Alternatively, if the first job is not a good match to the worker's skills, a job change could have a positive effect on wage growth. To examine whether training improves or worsens the returns to a job change, the third specification in Table 6 adds interactions between the job-training and job-change variables. The results for minimum wage workers reveal that a job change has a greater positive effect on wage growth when workers leave an occupation that had low training levels. In occupations where there is no firm or other training, median wage growth is 8.2% higher for workers who switch industry and occupation. On the other hand, at the maximum value of firm and other training observed in the minimum wage sample (.722 and .638, respectively), estimated wage growth is slightly h igher for workers who do not change industry and occupation. For the comparison sample of workers earning above the minimum wage, a similar pattern emerges. Training enhances wage growth most for workers who stay within the same industry and occupation, and a job change enhances wage growth most for workers who were in occupations with low training levels. However, the effects are generally smaller in the comparison sample. A job change from an occupation with no training enhances wage growth by 8.2% in the minimum wage sample but only 1.3% in the comparison sample.

Since there is evidence that the effects of job training and employer switching differ by sex, we also estimated wage regressions by sex. (18) For the sake of brevity, we did not include all the additional results in the tables since the pattern is easy to summarize. For both men and women, the basic patterns found in the pooled sample still hold up. Namely, both training and a switch in employer and occupation enhance wage growth. Also, the interaction effect between training and a switch of employers is negative for both sexes. The major difference is that the effects of firm-provided training, job changes, and the interaction effects are nearly twice as large for men. However, training done by someone outside the firm has nearly twice the effect for women as men.

To determine whether changes in school enrollment have an important effect on wage growth, we estimate a fourth specification with the sample restricted to workers with enrollment information and include dummy variables indicating whether the person continues, stops, or starts an enrollment spell over the two-year panel. The reference group includes people who are not enrolled in either year of the panel. The estimates reveal that minimum wage workers who complete a spell of school enrollment have 5% higher median wage growth than the reference group, and the effect is approximately the same among workers earning above the minimum wage in the first year of the panels. The estimates also imply that staying in school or returning to school reduces wage growth for both minimum wage workers and workers in the comparison sample.

For some people, increasing work hours could have a large effect on their ability to acquire a job that pays above the minimum wage. For others, the minimum wage may be the most they can earn regardless of whether they work full or part time. We expect that a person's earning opportunities would be improved if they are willing to take a full-time job and that the effect would be especially large for those with higher levels of education. The hypothesis is that part-time employment would impose a larger wage penalty on more educated workers. To investigate these hypotheses, we estimate a wage growth regression with interactions between the education dummies and the change in hours between the two periods. For minimum wage workers, the effect of a 20-hour increase in work hours on median wage growth ranges from a low of 2% among workers with eight or fewer years of education to a high of 32% for those with more than 16 years of education. For workers earning above the minimum wage, the picture is quite differen t. In five of the six education groups, increases in work hours are associated with lower wage growth, and the negative effect is greatest among the most-educated workers. The negative effect observed in the most-educated workers could reflect the endogeneity of work hours. Namely, if a worker experiences a decrease in wages, he may respond by increasing work hours.

A possible concern with our analysis is that wage growth is observed only for those workers who continue employment. Consequently, the wage regressions could contain a sample selection bias. (19) There are well-known techniques for correcting sample selection bias in a linear regression model. A shortcoming, however, is that the corrected estimates are typically sensitive to distributional and identifying assumptions, and the statistical properties of the sample correction methods in a median regression model are not yet developed. (20)

With the previously mentioned caveats in mind, we examined the importance of sample selection in the minimum wage sample using the two-step procedure popularized by Heckman (1976). (21) We failed to reject the null hypothesis of no sample selection bias in four of the six wage growth specifications estimated previously. In the two cases where there was evidence of sample selection, it pointed toward negative selection bias, suggesting that the workers with unobservables leading to below average wage growth are more likely to continue employment. However, we are not confident in the corrected estimates since the sign of the sample selection bias was not robust to alternative specifications of the wage equation and statistically insignificant at the .10 level in four of the six specifications. It is important to note, however, that correcting for sample selection bias had virtually no impact on the estimates that were the focus in this paper. For the minimum wage sample, correcting for sample selection had a sm all effect on the estimated coefficients presented in Table 6, and our empirical results are robust to corrections for sample selection. (22)

6. Summary and Conclusions

Most workers earning the minimum wage in a given year will either earn more than the minimum wage in the following year or exit employment. While the majority of minimum wage workers will not be earning the minimum wage a year later, almost 25% of workers earning the minimum wage in a given year were earning above the minimum wage in the previous year.

Improved matches through job switching are particularly important to the wage growth of some minimum wage workers. Compared to workers earning above the minimum wage, minimum wage workers reap especially large benefits from a change in industry and occupation, though the effects diminish sharply after age 45.

The importance of a change in industry and occupation to wage growth depends on the type of occupation that a worker starts in. While job training is relatively uncommon in occupations with large numbers of minimum wage worker, it has especially large effects on the wage growth of minimum wage workers. If a minimum wage worker finds himself in an occupation with a low training level, a job change is especially important to wage growth. However, if there is a very high level of training in the original occupation, a switch of occupation and industry could adversely affect wage growth. This result emphasizes the importance of understanding how minimum wage hikes affect training levels. While a hike in the minimum wage may benefit workers in the short run, it could hurt their prospects for future wage growth if firms cut back on training levels.

Data Appendix

The data for this study are drawn from the 252 monthly Outgoing Rotation Group (ORG) Current Population Survey (CPS) files from January 1979 to December 1999. In the CPS. eight panels are used to rotate the sample each month. A sample unit is interviewed for four consecutive months and then, after an eight-month rest period, for the same four months a year later. Each month, a new panel of addresses, or one-eighth of the total sample, is introduced.

The outgoing rotation groups (ORGs) include the people that are in either rotation group 4 or rotation group 8 (i.e., the subsamples that will be leaving for the eight-month rest period or permanently). Since 1979, the people in the ORGs were asked questions from an earnings supplement providing information on union status, weekly earnings, hourly earnings, and hours worked. Individuals potentially can be identified for the same month in consecutive years: that is, individuals in rotation group 4 in year I can be matched to individuals in rotation group 8 in year 2.

Matching people across years in the CPS was accomplished as follows: From the ORGs, data files were created for pairs of years (e.g., rotation 4 in January 1992 and rotation S in January 1993). Within each file, individuals were sorted on the basis of household ID, year, gender, and age. To be considered an acceptable match, a rotation 8 individual in year 2 had to be matched with a rotation 4 individual in year 1, with identical sex, household ID, survey month, and an age difference between 0 and 2. (23) If more than one person in year I can be matched to a given individual in year 2, additional variables (e.g., marital status, education) are used to find the correct match. If it is impossible to find a unique match in year 1 for an individual in year 2, the observation is deleted. Starting in 1994. the CPS included reliable individual identifiers that simplified matching individuals across time. (24)

Two minimum wage samples arc used in the analysis. The first includes wage and salary workers earning exactly the minimum wage in the first year of the two-year panels. (25) The second includes wage and salary workers earning exactly the minimum wage in the second year of the two-year panels. The minimum wage is defined as the greater of the federal or state minimum wage and was computed for each month over the sample period. The wage rate is defined as the reported hourly wage for workers paid by the hour. For workers not paid by the hour, the hourly wage rate is imputed by dividing usual weekly earnings divided by usual weekly hours. Only 0.6% of workers earning the minimum wage in the first year of the two-year panels have imputed earnings. Because of frequent job switching by minimum wage workers, however, 9.0% of continuing workers have imputed wages (i.e., are not paid by the hour) in the second year. (26) While imputations create the potential for measurement error, workers with imputed wages are inclu ded in the analysis to avoid dropping those workers that rise above the minimum wage (or fall to the minimum) by switching between jobs that pay by the hour and salaried positions. As noted in the text, these workers have higher-than-average wage changes.

Table 1

Sample Means for Workers in the Current Population Survey by Minimum
Wage Status (a)

Variable Workers above Minimum Workers at Minimum

Education
 Less than high school degree 12.8% 47.6%
 High school degree 38.7% 33.6%
 Some college 23.5% 16.3%
 College degree 17.1% 2.2%
 Graduate degree 7.8% 0.3%

Age
 16-18 2.5% 34.2%
 19-21 4.1% 15.6%
 22-25 7.6% 8.7%
 26-35 27.7% 13.1%
 36-45 27.2% 9.7%
 46-55 19.7% 7.4%
 56-64 9.4% 6.4%
 65-99 1.9% 4.9%

Weekly hours worked
 1-9 1.2% 7.6%
 10-19 3.3% 23.6%
 20-29 6.3% 26.1%
 30-34 11.1% 9.6%
 35 or more 78.1% 33.0%
Sample size 923,752 33,520

(a)The sample means are for workers in the first year of the two-year
panels of the outgoing rotation groups in the 1979-1999 Current
Population Survey. A worker's wage rate in the first year of panel data
sets is used to classify them according to whether their wage was at or
above the minimum wage.

Table 2

Transition Rates for Minimum Wage Workers

Workers earning minimum wage in
 second year of panels
 With wage rising to minimum 10.2%
 With wage remaining at minimum 22.6%
 With wage falling to minimum 27.8%
 Entering employment 39.4%
 Sample size 32,166

Workers earning minimum wage in
 first year of panels
 Whose wage rises above minimum 47.2%
 Whose wage stays at minimum 21.6%
 Whose wage falls below minimum 7.3%
 Who leave employment 23.9%
 Sample size 33,520

Table 3

Median Real Wage Growth of Minimum Wage Workers by Type of Transition
(a)

 Median Real
Group Wage Growth

All workers earning above minimum wage in first year 1.3%
Minimum wage workers in first year of panels
 All minimum wage workers in first year 3.5%
 Workers who earn above minimum in second year 14.8%
 Workers who earn minimum in second year -3.6%
 Workers who earn below minimum in second year -11.0%

(a)The data are drawn from the Current Population Survey Outgoing
Rotation Group panels for the years 1979-1999.

Table 4

Wage Growth by Minimum Wage Status (a)

 Above Minimum Wage
 Median Wage Sample
 Growth Size

All 0.9% 923,752
Subgroups of workers
 Same employer in both years 1.8% 24,286
 Change in employer between years 0.2% 2653
 t-statistic for equality -4.85
 Same industry of occupation in 0.9% 750,441
 both years
 Change industry and occupation 0.8% 173,312
 between years
 t-statistic for equality -2.6
 Start school enrollment in second 1.3% 3434
 year
 Enrolled in school both years 2.1% 13,352
 Not enrolled in school either year 3.9% 35,499
 Stop enrollment in second year 8.67 6387

Change in hours worked per week
 between years
 Decrease 20 or more 2.0% 18,277
 Decrease 10 to 19 6.5% 45,065
 Between decrease of 9 and 0.8% 806,522
 increase of 9
 Increase 10 to 19 -1.6% 39,467
 Increase 20 or more -2.0% 14,422

 At Minimum Wage
 Median Wage Sample
 Growth Size

All 3.5% 25,494
Subgroups of workers
 Same employer in both years 4.7% 474
 Change in employer between years 8.3% 210
 t-statistic for equality 2.25
 Same industry of occupation in 3.3% 15,015
 both years
 Change industry and occupation 14.1% 8042
 between years
 t-statistic for equality 72.56
 Start school enrollment in second 9.6% 289
 year
 Enrolled in school both years 7.4% 2620
 Not enrolled in school either year 13.9% 2229
 Stop enrollment in second year 15.4% 948

Change in hours worked per week
 between years
 Decrease 20 or more 5.2% 510
 Decrease 10 to 19 4.9% 1365
 Between decrease of 9 and 4.4% 16,062
 increase of 9
 Increase 10 to 19 9.4% 3002
 Increase 20 or more 20.6% 2118

(a)All data are from two-year panels created from outgoing rotation
groups in the Current Population Surveys for the years 1979-1999. The
wage rate for the first year of the panel is used to classify workers as
earning at or above the minimum wage.

Table 5

Determinants of Median Real Wage Growth (a)

 Workers Earning above
 Minimum Wage in First Year
 Coefficient t-statistic

Constant -0.038 -5.35
Education (elementary school
 reference group)
 High school dropout 0.005 1.83
 High school degree 0.010 4.63
 Some college 0.014 6.17
 College degree 0.020 8.50
 Graduate degree 0.021 8.02

Age (16-18 reference group)
 19-21 0.002 0.66
 22-25 0.003 0.82
 26-35 -0.013 -3.94
 36-45 -0.023 -6.84
 46-55 -0.030 -8.87
 56-64 -0.034 -9.78
 65-99 -0.043 -10.07
Female 0.006 6.91

Hours worked (1-9 reference group)
 10-19 0.031 7.18
 20-29 0.042 10.25
 30-34 0.038 9.45
 35+ 0.049 12.57
Public sector employee 0.003 2.50

Occupation
 Waiters and waitresses -0.058 -8.38
 Cooks -0.009 -1.90
 Sales workers, other commodities -0.001 -0.13
 Janitors and cleaners -0.016 -3.48
 Miscellaneous food preparation -0.010 -1.39
 occupations
 Stock handlers and baggers 0.001 0.15
 Textile sewing machine operators -0.028 -4.35
 Food counter, fountain, and -0.001 -0.06
 related occupations
 Nursing aides, orderlies, and -0.005 -1.03
 attendants
 Waiters'/waitresses' assistants -0.022 -2.28
 Maids and housemen -0.015 -2.37
 Sales workers, apparel 0.007 0.90
 Farmworkers -0.034 -4.88
 Secretaries -0.007 -1.88
 All other occupations -0.016 -5.01

Race and ethnicity (white, non-
 Hispanic reference group)
 Black -0.009 -5.71
 Other nonwhite -0.006 -1.85
 Hispanic -0.008 -3.79

Marital status (never-married
 reference group)
 Married, spouse present 0.001 0.80
 Ever married, no spouse present 0.001 0.46
 Sample size 184,729

 Workers Earning Minimum
 Wage in First Year
 Coefficient t-statistic

Constant -0.028 -4.08
Education (elementary school
 (reference group)
 High school dropout 0.001 2.53
 High school degree 0.028 6.78
 Some college 0.049 9.92
 College degree 0.120 15.02
 Graduate degree 0.184 10.77

Age (16-18 reference group)
 19-21 0.010 2.51
 22-25 0.026 5.32
 26-35 0.010 2.08
 36-45 0.004 0.71
 46-55 -0.004 -0.69
 56-64 -0.013 -2.2
 65-99 -0.034 -5
Female -0.028 -10.85

Hours worked (1-9 reference group)
 10-19 0.007 1.45
 20-29 0.020 4.2
 30-34 0.026 4.72
 35+ 0.055 11.43
Public sector employee -0.013 -3.56

Occupation
 Waiters and waitresses -0.046 -8.02
 Cooks -0.011 -1.89
 Sales workers, other commodities -0.002 -0.34
 Janitors and cleaners -0.003 -0.47
 Miscellaneous food preparation -0.012 -1.84
 occupations
 Stock handlers and baggers -0.004 -0.68
 Textile sewing machine operators -0.036 -5.02
 Food counter, fountain, and -0.004 -0.54
 related occupations
 Nursing aides, orderlies, and 0.007 0.89
 attendants
 Waiters'/waitresses' assistants -0.019 -2.45
 Maids and housemen -0.009 -1.13
 Sales workers, apparel 0.004 0.52
 Farmworkers -0.044 -4.56
 Secretaries 0.078 8.61
 All other occupations 0.009 2.19

Race and ethnicity (white, non-
 Hispanic reference group)
 Black -0.006 -1.6
 Other nonwhite 0.009 1.13
 Hispanic -0.018 -3.84

Marital status (never-married
 reference group)
 Married, spouse present 0.012 3.07
 Ever married, no spouse present 0.005 0.95
 Sample size 25,494

(a)The data are from the 1979-1999 Current Population Survey Outgoing
Rotation Group panels. The regressions also include year and
state-of-residence dummies.

Table 6

Expanded Models of Median Real Wage Growth (a)

 Workers Earning above
 Minimum Wage
 Coefficient t-statistic

Specification 1
 Change industry and occupation -0.001 -1.23
 % of occupation receiving firm
 provided training 0.006 1.97
 % of occupation receiving other
 training -0.006 -1.40
 Sample size 184,729

Specification 2 (job-switching
 effects by age-group)
 16-18 -0.012 -2.35
 19-21 0.000 0.08
 22-25 0.008 2.39
 26-35 0.000 0.19
 36-45 -0.002 -1.12
 46-54 0.001 0.53
 56-64 -0.009 -2.27
 >64 -0.019 -2.28
 Sample size 184,729

Specification 3
 Change industry and occupation 0.013 5.10
 % of occupation receiving firm
 provided training 0.011 3.15
 % of occupation receiving other
 training 0.001 0.14
 a*b -0.021 -2.44
 a*c -0.087 -7.75
 Sample size 184,729

Specification 4 (effect of change
 in enrollment status)
 Stop enrollment 0.046 6.09
 Stay enrolled -0.020 -2.81
 Start enrollment -0.040 -4.31
 Sample size 11,592

Specification 5 (effect of change
 in hours by education)
 Elementary school -0.0003 -1.13
 High school dropout 0.0003 1.65
 High school degree -0.0007 -7.37
 Some college -0.0008 -7.57
 College degree -0.0074 -56.63
 Graduate degree -0.0130 -74.05
 184,618

 Workers Earning
 Minimum Wage
 Coefficient t-statistic

Specification 1
 Change industry and occupation 0.058 28.15
 % of occupation receiving firm
 provided training 0.123 9.52
 % of occupation receiving other
 training 0.094 5.24
 Sample size 25,494

Specification 2 (job-switching
 effects by age-group)
 16-18 0.027 8.58
 19-21 0.097 21.16
 22-25 0.132 21.24
 26-35 0.129 24.02
 36-45 0.111 16.08
 46-54 0.037 4.69
 56-64 0.021 2.13
 >64 0.009 0.70
 Sample size 25,494

Specification 3
 Change industry and occupation 0.082 18.14
 % of occupation receiving firm
 provided training 0.164 10.48
 % of occupation receiving other
 training 0.102 4.78
 a*b -0.153 5.80
 a*c 0.001 0.04
 Sample size 25,494

Specification 4 (effect of change
 in enrollment status)
 Stop enrollment 0.057 6.70
 Stay enrolled -0.023 -3.17
 Start enrollment -0.011 -0.89
 Sample size 6,646

Specification 5 (effect of change
 in hours by education)
 Elementary school 0.001 4.12
 High school dropout 0.003 18.54
 High school degree 0.004 23.55
 Some college 0.006 27.70
 College degree 0.012 24.21
 Graduate degree 0.018 14.34
 25,479

(a)All specifications also include the controls listed in Table 5.
Specifications 2, 4, and 5 also include the job-switching and
job-training measures.

Received August 2001; accepted April 2002.

(1.) For example, Lazear and Miller (1981) and Grossberg and Sicilian (1999) find mixed effects of a minimum wage hike on training. Neumark and Wascher (2001) find that higher minimum wages reduce the level of employer-provided training.

(2.) Additional details on the data set are provided in the Data Appendix.

(3.) The minimum wage is defined for each month and state in the sample. When a state has a minimum wage that exceeds the federal minimum, the state minimum wage is used. The wage rate is defined as the hourly wage rate for those that report being paid by the hour. For those who are not paid by the hour, we impute the hourly wage by dividing weekly earnings by weekly hours. Details on the frequency of imputations are included in the Data Appendix.

(4.) The occupation category with the second-largest fraction of workers being paid below the minimum wage is the "food counter, fountain, and related occupations."

(5.) Since the CPS data are monthly, real wages are calculated using the monthly series for the consumer price index for urban consumers.

(6.) If workers with imputed wages in the second year of the panels (9.0% of the sample) are dropped from the analysis, median wage growth drops to 3.0%. Workers who switched from hourly to nonhourly jobs experience much higher wage growth (31.0%).

(7.) If workers with imputed wages in the first year of the panels are dropped from the sample, the median reduction in real wages is reduced to 4.1%.

(8.) The panels were matched CPS supplements that contained data on employee tenure. The analysis includes the following months: January 1983, May 1983, January 1987, May 1988, January 1991, April 1993, February 1996, February 1998, and February 2000.

(9.) Unfortunately, it appears that this classification method has some error. In the January and February surveys, workers with fewer than three years of tenure are asked to provide tenure in months. Apparently, workers round their answers because there is significant heaping in the data at 12, 24, and 36 months.

(10.) For this part of the analysis, workers earning below the minimum wage in either period are eliminated because the reasons for earning below the minimum vary between noncoverage, potential measurement error in the earnings variable, and tip credits that allow covered workers to be paid less than the minimum. The qualitative nature of the results is not changed by this exclusion.

(11.) See, for example, Mellow and Sider (1983) and Macpherson and Hirsch (1995).

(12.) Median regression methods were chosen to reduce the effect of extreme outliers in the data and the fact that wage growth is truncated from below given the requirement of a minimum wage.

(13.) The Spearman rank correlation coefficient of the coefficients on the occupation dummies is .71 with a p-value of 0.003.

(14.) The occupation-specific training measures were created from the January 1983 and 1991 CPS. The firm training variable represents the percentage of workers within a given three-digit occupation who report participation in either formal or informal company training with the current employer. The other training measure represents the percentage of workers receiving training from someone other than the employer white on the current job. The variables were matched to the individual data using time-consistent three-digit Census occupation and industry codes. The 1970 codes were matched to 1980 codes using the mapping included in U.S. Bureau of the Census (1989). The minor differences between the 1990 and 1980 codes were resolved on the basis of a 1992 Census Bureau memorandum.

(15.) A list of representative occupations ranked by training intensity is available on request. All but two of the 15 occupations with the largest number of minimum wage employees ranked in the bottom one-third of occupations ranked by either firm or outside training intensity.

(16.) Neumark and Wascher (2001) suggest that a 10% increase in the minimum wage would reduce the incidence of formal training by 18% (1.8 percentage points) among 20- to 24-year-olds.

(17.) See, for example, Altonji and Shakotko (1987) and Topel (1991).

(18.) See, for example, Loprest (1992) and Olsen and Sexton (1996).

(19.) If, for example, workers with greater potential for wage growth are more likely to continue employment, wage growth in the sample of continuing workers would overstate potential wage growth for all workers. The sample selection could also bias the estimated effects of explanatory variables.

(20.) Vella (1998) provides a recent review of the methodology for correcting sample selection biaaes and the shortcomings of the various approaches.

(21.) In the first step, a probit model of the decision to continue employment is estimated. The control variables include all those in the wage regression specified in Table 6 plus controls for the number of children in the household in various age categories, the type of living quarters, and whether the worker is either the household head or the spouse of the household head. The probit estimates are used to generate a correction factor (the inverse Mills ratio) that is added to the wage regression model as an additional explanatory variable to correct for sample.

(22.) Correcting for sample selection changed 24 of the 25 coefficient estimates presented in Table 6 by less than 0.01. The largest change was for the coefficient on the "stop enrollment" variable, which dropped from 0.057 to 0.043 on correcting for sample selection.

(23.) Since surveys can occur on different days of the month, age change need not equal 1.

(24.) There are several reasons why matches might not be found for a given individual. The most important reasons for a match failure include (i) a household moves: (ii) an individual moves out of the household, or (iii) the Census is unable to reinterview a household or obtain information on the individual. Perrachi and Welch (1995) analyze the attrition rates in matched March CPS files and find that the match rate is lowest among those in their early 20s. Sample sizes are reduced further because of partial panels in 1984-1985, 1985-1986. 1994-1995, and 1995-1996 due to changes in Census location identifiers during 1985 and 1995.

(25.) The data source for the federal and state minimum wages is the Monthly Labor Review.

(26.) Among workers earning the minimum wage in the second year, 0.6% had imputed wages in the second year, and 10.4% had imputed wages in the first year of the panel.

References

Acemoglu, Daron, and Jorn-Steffen Pischke. 1999. Minimum wages and on-the-job training. NBER Working Paper No. 7184.

Altonji, Joseph, and R. Shskotko. 1987. Do wages rise with job seniority? Review of Economic Studies 54:437-9.

Gottschalk, Peter. 2001. Wage mobility within and between jobs. Boston College Working Paper No. 486, April.

Grossberg, Adam J., and Paul Sicilian. 1999. Minimum wages, on-the-job training, and wage growth. Southern Economic Journal 65:539-56.

Hashimoto, Masanori. 1982. Minimum wage effects on training on the job. American Economic Review 72:1070-87.

Heckman, James J. 1976. The common structure of statistical models of truncation, sample selection and limited dependent variables and a simple estimator for such models. Annals of Economic and Social Measurement 5:475-92.

Lazear, Edward P., and Frederick H. Miller. 1981. Minimum wages vs. minimum compensation. In Report of the Minimum Wage Study Commission. Washington, DC: U.S. Government Printing Office, pp. 347-80.

Long, James B. 1999. Updated estimates of the wage mobility of minimum wage workers. Journal of Labor Research 20:493-504.

Loprest, Pamela 5. 1992. Gender differences in wage growth and job mobility. American Economic Review 82:526-32.

Macpherson, David A., and Barry Hirsch. 1995. Wages and gender composition: Why do women's jobs pay less? Journal of Labor Economics 13:426-71.

Mellow, Wesley, and Hal Sider. 1983. Accuracy of response in labor market surveys: Evidence and implications. Journal of Labor Economics 1:331-44.

Neumark, David, and William Wascher. 2001. Minimum wages and training revisited. Journal of Labor Economics 19:563-95.

Olsen, Reed Neil, and Edwin A. Sexton. 1996. Gender differences in the returns to and acquisition of on-the-job training. Industrial Relations 35:59-77.

Peracchi, Eranco, and Finis Welch. 1995. How representative are matched cross-sections? Evidence from the Current Population Survey. Journal of Econometrics 68:153-79.

Schiller, Bradley R. 1994. Moving up: The training and wage gains of minimum wage entrants. Social Science Quarterly 75:622-36.

Smith, Ralph B., and Bruce Vavrichek. 1992. The wage mobility of minimum wage workers. industrial and Labor Relations Review 46:82-8.

Topel, Robert. 1991. Specific capital, mobility, and wages: Wages rise with job seniority. Journal of Political Economy 99:145-76.

U.S. Bureau of the Census. 1989. The relationship between 1970 and 1980 industry and occupation classification systems. Washington, DC: U.S. Bureau of the Census.

Vella, Francis. 1998. Estimating models with sample selection bias. Journal of Human Resources 33:127-69.

William E. Even *

* Department of Economics, Miami University, Oxford, OH 45056, USA; E-mail evenwe@muohio.edu.

David A. Macpherson +

+ Department of Economics, Florida State University, Tallahassee, FL 32312, USA; E-mail dmacpher@)coss.fsu.edu; corresponding author.

We are grateful to the referees for several useful comments that helped improve the paper.