THE IMPACT OF MINIMUM WAGE INCREASES BETWEEN 2007 AND 2009 ON TEEN EMPLOYMENT A Thesis submitted to the Faculty of the Graduate School of Arts and Sciences of Georgetown University in partial fulfillment of the Requirements for the degree of Master of Public Policy in Public Policy By Jiashan Cui, B.A. Washington, DC April 12, 2012
Copyright 2012 by Jiashan Cui All Rights Reserved ii
THE IMPACT OF MINIMUM WAGE INCREASES BETWEEN 2007 AND 2009 ON TEEN EMPLOYMENT Jiashan Cui, B.A. Thesis Advisor: Gurkan Ay, Ph.D. Abstract This paper analyzes the employment effects of minimum wage increases at the federal level between 2007 and 2009 among U.S. teenage workers (ages 16-19). Using the pooled crosssectional data from the Current Population Survey Outgoing Rotation Groups, and following a model by Burkhauser et al. (2000a, 2000b), I find that the impact of minimum wage increases on the overall teen employment is negative and statistically significant. After specifically controlling for the recent recession, the estimated elasticity for in teen employment is slightly smaller than previous studies. I also compare the employment effects of minimum wage increases between male and female teenage workers, among white, black and Asian teenage workers. I find that employment effects do not concentrate on any particular group. There has been increasing pressure on raising the minimum wage floor as the economy starts to recover slowly. The results suggest that this policy recommendation might potentially hinder the recovery of employment. iii
ACKNOWLEDGMENTS This thesis is dedicated to my family in China for their love and support. I would like to thank my thesis adviser, Professor Gurkan Ay, for his invaluable guidance and advice. I am grateful to Professor Barbara Schone, Professor Adams Thomas and Mike Barker for their help. JIASHAN CUI iv
TABLE OF CONTENTS Abstract... iii Introduction... 1 Literature Review... 4 Empirical Model... 10 Theoretical Explanation... 10 Empirical Model... 11 Data Description... 12 Data... 12 Variables... 13 Empirical Results... 17 Baseline Analysis... 17 Robustness Checks... 23 Real minimum wage.... 23 Lagged employment effects.... 26 Alternative error corrections and random effects model.... 28 Sub-Sample Based on Demographic Characteristics... 30 Conclusions and Future Research... 33 Conclusions... 33 Future Research... 34 Appendix... 35 References... 40 v
Introduction Under the Fair Labor Standards Act (FLSA) the U.S. government first set the national minimum wage at 25 cents per hour in 1938. Since then the federal rate has kept rising, and an increasing number of states have started to set state minimum wages above the federal level. The fact that minimum wages and economic conditions at state level vary across the country provides a quasi-experiment opportunity, allowing us to examine the impact of minimum wage increases over time using state-specific data (Neumark & Wascher, 2006, p. 3). (See Figure A1. and Table A1 in Appendix for state and federal minimum wages between 1996 and 2010) Economists have long been interested in the impact of minimum wage legislation on employment. However, despite extensive empirical research, debate on the employment impact of minimum wage increases is still unsettled, and the results conflict: Most researchers have argued that minimum wage increases reduces aggregate employment. However, in contrast to those findings, a few researchers find either insignificant effects or positive impacts of minimum wage increases on employment in general and teen employment in particular. Prior to the most recent increases from 2007 to 2009, the federal minimum wage rate had not changed for almost 10 years since 1998. The U.S. government raised the federal mandate first to $5.85 in 2007, and then kept raising the rate for two consecutive years: to $6.25 in 2008 and to $7.25 in 2009 an overall increase over the entire period of 41 percent. Figure 1 shows both the real and nominal values of the federal minimum wage since 1988. 1
Nominal Dollars 8.00 7.00 6.00 5.00 4.00 3.00 2.00 1.00 0.00 4.50 4.00 3.50 3.00 2.50 2.00 1.50 1.00 0.50 0.00 Real Dollars Year Federal Minimum wages are nominal values Federal Minimum wages are real values Source: Generated by the author based on information of federal minimum wage from the Department of Labor. Note. a I use the CPI-U index to adjust for inflation and the federal wage levels are re-scaled such that CPI-U index 1988=100. Figure 1. Trends of Nominal and Real Federal Minimum Wages from 1988 to 2010 a The recent minimum wage rate increases coincided with the 2007-2009 recession, which has been defined as the longest and deepest recession of the post-world War II era (Labonte, 2010, p. 1). The unemployment rate jumped from 4.8 percent in the late 2007 to 10 percent in 2009, and remained at 9.7 percent in 2010 (Katz, 2010, p. 2). And the fiscal and monetary policies by the U.S. government in response to the situation have been characterized as the most aggressive ones in history (Blinder & Zandi, 2010, p. 1). Hence, it is arguable that estimates from previous studies might not be applied here if specifications from those studies fail to 2
control for the macroeconomic changes and policy interventions during the recession. Now while the economy and employment are still under slow recovery, legislators are facing more pressure of raising the minimum wage. In fact, several states, e.g. New York, New Jersey, and Connecticut, have started to push for higher state mandates; and advocates have been arguing for higher minimum wage at the federal level (Greenhouse, 2012). Therefore, it is worth reexamine the employment effect of minimum wage increases under the recent economic downturn, especially segregate the impact of the minimum wage policy changes on employment from influences of the other macroeconomic and policy factors, which can serve one of the references for further changes on minimum wage policy. In this thesis, I focus on the impact of minimum wage increases between 2007 and 2009 on teen employment. Teenage workers, in particular, are more likely to be adversely affected by this legislation since they have relatively less work experiences and a large proportion of them are expected to work at low-paid jobs (Card, 1992a, p. 23). I also studied the impacts within the teen workers according to gender and race. 3
Literature Review In contrast to the impact of minimum wage increases on the aggregate employment, this paper, I investigate how teenage employment is affected. The literature on the impact of minimum wage increases on the aggregate employment rate is very broad. Yet the existing literature on this topic is not conclusive regarding the employment effect of minimum wage increases on teenagers. Neumark and Wascher (2006) wrote a comprehensive review about researches on the impact of minimum wage increases on employment; therefore, instead of listing of the broad literature, this section mainly lays out how the literature evolves overtime and the main research results. Earlier empirical research confirmed the economic theory that minimum wage increases adversely affect the aggregate employment rate. Charles Brown, Charles Gilroy and Andrew Kohen (1983) found that a 10 percent increase in minimum wage reduces teenage employment by 1 percent. Their findings also showed that the negative effect on employment among young adults (ages 20-24) is more modest: they concluded that a 10 percent increase in the minimum wage correspond to a probable 0.25-percent decrease in young adult employment. Wellington (1991) pointed out that, compared to the real dollar value of minimum wage in 1981, the real minimum wage in 1986 had actually decreased by about 20 percent, which provides a natural experiment to test the hypothesis that a decrease in the minimum wage increases employment (Wellington, 1991, p. 28). He also finds an adverse impact of minimum wage increases on employment, but a smaller one than Brown et al. (1983) had previously found. He concluded that a 10 percent increase in the minimum wage reduce teen employment by approximately 0.60 percent. 4
In the Brown et al. and Wellington studies, minimum wages were generally measured by the Kaitz Index, which is the ratio of the nominal minimum wage divided by the average hourly wages that are weighted by the proportion of workers covered by the minimum wage for each industry (Brown, Gilroy, & Kohen, 1983, p. 11). Beginning in the late 1980s, an increasing number of states set their state minimum wage rates above the federal level. These regional variations among states also allowed researchers to empirically analyze the employment effects of minimum wage increases. These analyses using regression covariates rather than the Kaitz Index as a measure of the minimum wage were later characterized as the new minimum wage research (Neumark & Wascher, 2006, p. 4). Most of those studies used time-series and panel data techniques to measure the impact of the minimum wage at the national level, and they continued to confirm that minimum wage increases have a modest but negative impact on employment. Neumark and Wascher (1992), and Neumark et al. (2004) confirmed that raising minimum wage would adversely affect teen employment. They found that a 10 percent increase causes approximately up to a two-percent employment decline among teenagers. Neumark et al. (2004) also argued that low-wage workers, in particular, were adversely affected by such a policy change, in terms of their working hours, income, and aggregate employment. Deere et al. (1995) measured the impact of minimum wage increases on aggregate employment at the federal level. They used two binary variables to measure federal minimum wage increases in 1990 and 1991, which assumes non-linear effects of minimum wage increases on employment (Sabia, 2009, p. 322). The results confirmed the conventional prediction of a negative correlation between minimum wage increases and aggregate employment. Their results 5
also confirmed their assumption that the impact of minimum wage increases would be larger for groups with higher proportions of low-wage workers. Burkhauser et al. (2000a) examined the impact of minimum wage increases on youth employment at the federal level and find a significant and negative effect on the teenage employment. Their fixed effect model included state dummies to account for the variation across states, and month dummies to account for the seasonal variation as the model used by Card and Kruger (1995). They argued that the inclusion of year controls explain why Card and Kruger (1995) got insignificant results, which should be excluded from the model. They also ran regressions based on the specifications by Deere et al. (1995), the results of which confirmed their original estimates. They also controlled for lagged minimum wage effects and different business cycles. They found that inclusion of recession dummy and a 12-year lagged minimum wage variable slightly increase the magnitude of the elasticity. Sabia (2006) focused on the employment impact of minimum wage increases on retail and small business. His results showed a negative and statistically significant impact of minimum wage hikes on teen employment even after controlling for time trends. Sabia (2009) looked at the employment effect using the national data, and his results were consistent with his findings on the retail and small business that minimum wage increases have a negative and statistically significant impact on teen employment, with the presence of year controls. Sabia argued that more states are setting their state minimum farther above the federal rate, and the variation across states is large enough to help identify the impact of minimum wage increases on teen employment. Despite these extensive researches, however, the literature on the impact of minimum wage increases is inconclusive. Several studies found insignificant or even positive employment 6
effects of minimum wage increases. For example, to capture the effect of minimum wage increases, Card (1992a) grouped states according to the fraction of teenage workers that were affected by the policy change. He examined the impact of the April 1990 minimum wage increase on teenage employment, and found no significant evidence of a negative effect. However, other researchers later pointed out that his measurement of minimum wage increases is subject to inflation; as a result, the estimates cannot accurately measure the employment effects associated with minimum wage increases. (Neumark & Wascher, 2006, p. 15) Other studies focused on the employment effects of minimum wage increases within particular states or industries. Their underlying assumption is that states/industries that do not experience minimum wage increases can serve as valid control groups to compare with the employment outcomes of similar groups that are affected by the policy change (Neumark & Wascher, 2006, p. 10). Most of those studies found a negative impact of minimum wage increases on employment. Using a difference-in-differences approach, Card (1992b) studied the employment effect of a minimum wage increase in 1988 on low-skilled teenage workers in California. He used a set of states (Arizona, Florida, Georgia, New Mexico and Texas) as a comparison sample, which had similar employment trends as California in 1987, yet did not experience the policy changes. His results showed a positive correlation between the minimum wage increase and teen employment; the employment rate of teenage workers in California actually rose faster than that of the comparison groups after the minimum wage increases (Card, 1992b). However Card s study has been criticized on the ground that the control groups might be inadequate. In particular, given the geographic distance between California and the comparison group, critics argue, it is likely that they experienced different demand trends (Neumark & Wascher, 2006, p. 11). 7
Using first differencing examination, Katz and Krueger (1992) looked at the employment effect on fast food restaurants in Texas after the federal minimum wage changes in 1990 and 1991. In their study, minimum wage changes were measured by the proportional increases needed by firms to comply with the new minimum. Their results showed a positive growth in youth employment in response to the minimum wage increase. Their data came from a telephone-survey conducted by the authors of fast food restaurants in Texas. This survey method has raised questions from researchers about the validity of their data and, as a result, their research outcomes (Neumark & Wascher, 2006, p. 30). Using difference-in-differences estimation, Card and Krueger (1994) studied the employment effect of minimum wage increases on the fast food industries in New Jersey in 1992. They considered Pennsylvania as a valid control group given its geographic proximity to New Jersey and the different minimum wage policies of the two states during that period. Their results show either increased employment rate associated with minimum wage increases or no employment effect. Again, their results are estimated based on their self-collected data, the reliability of which is a big concern (Neumark & Wascher, 2000, pp. 1362-63). Card and Krueger (2000) re-conducted this comparative study looking at employment effects of the federal minimum wage increase in Pennsylvania in 1996, which was not binding for New Jersey. This gave the authors to eliminate possible systematic differences between the two states that might have influenced the employment effect in their first study. In response to the criticisms about the reliability of their survey data, they used employer-reported ES-202 files from the Bureau of Labor Statistics (BLS) instead of a telephone survey. Their results echoed their initial conclusion that modest changes in minimum wage have little systematic effect on employment (Card & Krueger, 2000, p. 1398). Later researchers pointed out, however, that the 8
employment trends in Pennsylvania and New Jersey in 1996 before the policy changes were not consistent with each other, which would make Card and Kruger s difference-in-differences identification strategy problematic (Neumark & Wascher, 2006, p. 35). Welch (1995) argued that the prediction of negative impact of minimum wage increases might not be able to be applied for an individual industry or sector, because the employment effects from those case studies might simply reflect the particular factor intensities in those individual industries (Welch, 1995, p. 847). Given the considerations over the internal and external validity of previous case studies, this study will focus on the impact of the federal minimum wage increases from 2007 to 2009 on aggregate teen employment across the county, using the latest CPS data through December 2010. 9
Empirical Model Theoretical Explanation This analysis focuses on the impact of minimum wage increases on teen employment. Two economic models explain the impact of the policy change. The neoclassic model assumes perfectly competitive labor and product markets, under which individual employers take market prices as given. So each firm hires as many workers as they need at the market wage. The minimum wage regulation sets the wage level above the market equilibrium rate, raising the cost of labor relative to the cost of capital, ceteris paribus. This increase in labor costs leads firms to substitute capital for labor in the production process. Therefore, increasing minimum wage results in a decrease on employment (Stigler, 1946). The second monopsony model argues that firms that hire workers at minimum wages still have some market power over wage rates. As a result, increases in minimum wage, if they are not above the market equilibrium, will compress the profits of employers, but they will also have a positive effect on employment: higher wages will encourage more people to enter the labor force (Neumark & Wascher, 2008). Hence the question of the net impact of the minimum wage on employment remains to be decided empirically. 10
Empirical Model To examine the employment effect of minimum wage increases, I estimate the following model used by Burkhauser et al. (2000a, 2000b): = + + + ++ + (1) where is the ratio of teenage employment to the teen population in state i in month j in year t; controls for any state-specific, time-invariant characteristics; measures state-invariant year effect at the national level; is a set of explanatory variables (the same used by Card and Krueger, 1995). Since I use monthly data, monthly effects are included into the model to control for unmeasured seasonal changes in employment that are not due to minimum wage policy changes. is the prevailing state minimum wage in state i in month j in year t. Therefore, reflects the impact of minimum wage increases on the change of aggregate teen employment ratio. 11
Data Description Data This analysis uses data from the Current Population Survey (CPS) published by the Bureau of Labor Statistics. The CPS surveys a nationally-representative sample of household survey and provides comprehensive information about the U.S. labor market that is suitable for this analysis (Labor Force Statistics from the Current Population Survey, 2012). Consistent with the research of Burkhauser et al. (2000a, 200b), I use monthly data from the CPS s Merged Outgoing Rotation Groups (MORG) data from January 1996 to December 2010 (CPS Merged Outgoing Rotation Groups, 2012). The major advantage of using this monthly data over the annual March CPS data that states might adjust state mandates on minimum wages more than once within a year, which can be only captured by monthly data; and that more information about changes in the labor market within a year can be included in the analysis. This advantage leads to a more accurate estimate of the impact of minimum wage increases (Burkhauser, Couch, & Wittenburg, 2000a, p. 659). I construct state-month observations from the individual-level CPS data. This process involves generating model variables for each year, and collapsing the database from the individual level to the state level for each year. On average, there are 612 state-month observations in each year, which constitutes a sample of 2,448 observations from January 2007 to December 2010. The sample size increases to 9,180 when I expand the observations from January 1996 to December 2010. The CPS data contain population weights assigned to each observation. These weights are taken into account when I collapse the data to state-month level to make the analysis sample nationally representative. 12
Information about historical minimum wages at state-year level is gathered from the Department of Labor, and merged into the state-month dataset above. For wage variables, I also use the Consumer Price Index - All Urban Consumers (CPI-U) from Bureau of Labor Statistics to adjust wage variables for inflation (Consumer Price Index- All Urban Consumers, 2012). The information about different business cycles is available from the website of National Bureaus of Economic Research (US Business Cycle Expansions and Contractions, 2010). Variables Table 1 below provides definitions of variables used in my model. Table A2 in the Appendix shows the annual weighted means and standard deviations of variables in Table 1 between 1996 and 2010 at the national level. The dependent variable is the aggregate employment ratio among teenagers, which is measured as the percentage of employed teenage workers among all teens in a state. Teenage workers who are self-employed and working in agriculture sector are excluded from the calculation. The national employment ratio for teenage workers is 32.5% for 2007-2010, and 39% for 1996-2010. In Table A2, the ratio of teen employed is present by years, and we can observe that the employment ratio has been decreasing over time. The key independent variable in this paper is the natural log of the prevailing minimum wage in a given state. For each state, the prevailing minimum wage is the higher of the state or federal minimum wages in that particular month. All prevailing minimum wages are measured at the nominal value. The log prevailing minimum wage has been increasing overtime, which is consistent with the trend of minimum wage policy overtime. 13
The additional control variables included in the model are the natural log of the average hourly wage of prime-age adults, share of teenagers in a state, and the state unemployment rate of prime-age male adults. The natural log of average wages of prime-age adults measures the average hourly earnings for people at age range between 25 and 54. Average hourly wages are calculated based on the earnings information in the survey. For hourly workers, I use the reported hourly earnings; and for non-hourly workers I divide the usual weekly earnings by the usual working hours to get the average hourly earnings. The natural log of average wages of prime-age adults has been increasing from 1996 to 2010; and the national average between 2007 and 2010 is slightly higher than the number using the expanded data 1996 2010. The share of teenagers is defined as the share of teenager population over the working age population (ages 16-61) in a state. The ratio at the national level remains relative constant: about 9% for the time frame of 1996 2010 and 2007 2010. The unemployment rate is measured as the ratio of unemployed prime-age male adults (ages 25-54) over the overall working age population (ages 16-61). I generate three sets of dummy variables to account for state-specific fixed effects, seasonal changes, and yearly time trends, State dummies capture the state characteristics that do not change over time. I include monthly dummy variables to account for the seasonal changes in the labor market. In addition, I generate year dummies to control for any unobserved stateinvariant changes at the national level that might bias the model estimate. Burkhauser et al. (2000a, 2000b) suggest that controls for year effects should be excluded from the specification, because including year effects might remove most of the variation in minimum wage at the federal level, and few variations in the minimum wage are left for the analysis. However, other 14
researchers indicate that exclusion of year controls might be problematic if the observed macroeconomic controls in the model are not sufficient to account for unobserved national trends that are both correlated with the changes in minimum wages and employment situation in the labor market. Hence, I estimate two models based on Equation 1: one with controls for year effects and another without these controls to test for this concern. 15
Table 1 Definitions of Variables Used in the Econometric Analysis and Weighted Means and Standard Deviation a Variable Definition 2007-2010 1996-2010 Ratio of employed Log state minimum Ratio of teenage (ages 16-19) Employment to teenage population in a given state The natural log of the greater of the state or federal minimum wage Log adult wage The natural log of the wages of prime-age adults (ages 25-54) Share of teenagers Unemployment rate The share of teenagers in the overall working-age population (ages 16-61) The prime-age male (ages 25-54) unemployment rate in the state 0.325 (0.128) 1.930 (0.109) 2.989 (0.127) 0.090 (0.017) 0.062 (0.040) 0.390 (0.137) 1.736 (0.161) 2.819 (0.189) 0.092 (0.020) 0.043 (0.031) Month effects State effects Year effects Recession A set of dummy variables equal to one for each month in the year A set of dummy variables equal to one for each state A set of dummy variables equal to one for each year Dummy variable equal to one in the month that economy was officially in a recession Number of observations Number of observations 2448 9180 Source: Computed by the author from the CPS Outgoing Rotation Groups data. Note. a Defined by author based on Burkhauser et al. (2000a, 2000b); weighted means are calculated by the author from the CPS Outgoing Rotation Group data; heteroskedasticity-robust standard errors are in parentheses. 16
Empirical Results In this section, I first analyze a series of models, by gradually adding more controls from Equation 1, to see whether the control variables still serve as valid controls in the analysis. Then I conduct various sensitivity tests to check the robustness of the estimates. Eventually I test the employment effects of minimum wage increases on sub-samples of different demographic characteristics, such as race, gender and ages, to see whether I can find similar patterns among those sub-groups, or if the policy effect is highly driven by certain groups within teenage workers. Baseline Analysis Table 2 presents the coefficients from a series of models based on Equation 1 using the pooled CPS s MORG data. Model 1, 2 and 3 use the most recent data 2007-2010 sample. Model 4 and 5 expand the observations from January 1996 to December 2010. This thesis applies time-series analysis technique using panel data, which potentially will raise the problems of autocorrelation and heteroskedasticity. One of the assumptions for simple and multiple regression analysis is independence of error terms, meaning that the error terms are uncorrelated with each other. Violation of this assumption will not affect the magnitude of the coefficients, but standard errors can be underestimated because of this problem, which will hinder the reliability of the inferences. For analysis using time series data, this problem is usually present in the form of serial correlation that the error in one period will affect the error in the subsequent periods, which will in turn influence the significance of estimates. Results from the test for the presence of first-order autocorrelation confirm this potential concern. Heteroskedasticity refer to a problem that random variability in the error term will potentially 17
bias the significance level of estimates, which can also occur in time series regression model. Therefore, in this section each model uses Feasible Generalized Squares (FGLS), correcting for common autocorrelation via the Prais-Winsten, and calculating heteroskedasticity-robust error (Wooldridge, 2008, pp. 429-31, 452, 455-56). 18
Table 2 Effects of Nominal Minimum Wage Increases on the Ratio of Teen (Ages 16 19) Employment to Teen Population a 2007-2010 1996-2010 Variable Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Log state minimum b -0.205*** c -0.241*** -0.051-0.045** -0.056*** -0.042** (0.037) d (0.029) (0.036) (0.019) (0.021) (0.019) Log adult wage b 0.037 0.012 0.056* 0.001 0.025** -0.001 (0.026) (0.032) (0.032) (0.017) (0.070) (0.018) Share of teenagers 0.039-0.125-0.144-0.043-0.029-0.046 (0.13) (0.126) (0.124) (0.067) (0.070) (0.067) Unemployment rate -0.47*** -0.413*** -0.129*** -0.304*** -0.380*** -0.300*** (0.072) (0.06) (0.072) (0.046) (0.047) (0.046) Month effects Yes Yes Yes Yes Yes Yes State effects Yes Yes Yes Yes Yes Year effects Yes Yes Yes Yes Recession dummy Yes Recession 1 dummy Yes Recession 2 dummy Yes 0.116 0.454 0.482 0.458 0.371 0.459 Number of observations 2448 2448 2448 9180 9180 9180 Elasticity -0.587-0.770-0.101-0.106-0.126-0.099 Note. a The dependent variable is the ratio of employed teenagers to the teenage population; b Minimum wage and adult average wage are nominal values; c * ** *** d Heteroskedasticity-robust standard errors are in parentheses. 19
Model 1 presents estimates controlling for the natural log of adult hourly earnings, share of teenagers, unemployment rate among prime-age males and monthly adjustments. The estimate of log of prevailing state minimum wage shows negative and statistically significant impact on the employment ratio among teenage workers. The estimated coefficient is -0.205, statistically significant at 1-percent level. The employment elasticity represents the percentage changes in the employment ratio in response to a 1-percent increase of the minimum wage variable, which is calculated as β/ where is the average teen employment rate during the sample period. (Even & Macpherson, 2010) The estimated elasticity of employment effect is -0.587, which implies that on average, a 10% increase in minimum wage might lead to a 5.87 percent decrease in teen employment ratio. In Model 2, I add state dummies to all the specifications in Model 1 to control for the variations across states that are relatively stable over time. The coefficient of the minimum wage variable drops slightly to -0.241, and the magnitude of elasticity of the minimum wage increases from -0.587 to -0.770, which imply that part of change in the employment ratio can be contributed to time-invariant state-specific characteristics. Model 3 adds controls for year effects to the Model 2 variables. After controlling for year effects in the model, the estimated coefficient of the prevailing minimum wage becomes insignificant. This finding is consistent with Burkhauser et al. (2000a, 2000b); they argued that this is because year effects explain most of the variation in the minimum wage variable. Year effects control for any state-invariant macro-economic situation and/or policies that apply for every state, which means the minimum wage increases at federal level, will at least be partly removed from the minimum wage variable. Sabia (2009) argued that federal minimum wage changes will not equally affect each state, given the fact that a number of states set their state 20
mandates at various levels. Thus, the remaining variation in the minimum wage variable is due to state minimum wages that are higher than the federal mandate. To confirm this assumption, however, cross-state differences in a longer time period need to be analyzed. Since 1997, an increasing number of states have set their state minimum wages above the federal level. I, therefore, expanded the analysis to the period between January 1996 and December 2010. This model is labeled as Model 4 in Table 2, and the estimated minimum wage coefficient is found to be negative and significant. I find that the elasticity estimate is - 0.106, which implies that a 10-percent increase in prevailing minimum wage reduces employment among teenage workers by one-percent. Burkhauser et al. (2000a, 2000b) argued for the importance of controlling for the impact of different business cycles on employment. According to the National Bureau of Economic Research Business Cycle Dating Committee, two periods officially defined as recessions fit the time frame of this paper: March 2001 November 2001 and December 2007 June 2009 (US Business Cycle Expansions and Contractions, 2010). However, when the federal government raised the minimum wage both in October 1996 and December 1997, the economic climate was more robust, which will potentially improve the precision of the estimates of employment effect by including observations under different business cycles (Burkhauser, Couch, & Wittenburg, 2000a, p. 668). It could be argued that, the specifications in Model 4 the unemployment rate of prime-age male, average hourly earnings of prime-age adults and year effects might be not sufficient to control for all unobservable macroeconomic changes under different economic situations. Therefore, Model 5 includes a dummy variable for recession, and the estimate on the minimum wage variable shows an increase in negative magnitude (from -0.045 to -0.056), which is significant at the 1-percent level. 21
Model 5 uses one single recession indicator in the specification to control for the impact of different business cycles on employment. To make the model outcomes more accurate, I include two recession dummies, because, compared to the 2001 recession, the 2007-2009 recession was more severe in both duration and impact and may have had different impacts on teen employment. To address this concern, I use two separate recession dummy variables representing each of the two recessions in Model 6. Both the estimated coefficient of the prevailing minimum wage and the elasticity measure decreased significantly in absolute value: the coefficient increases from -0.056 to -0.042, and the elasticity raise from-0.126 to -0.099. In using a single recession indicator, I assume that other macro-economic and policy controls in the specifications are sufficient to control any unobservable components in the recent recession that differ from the short recession in 2001. The changes imply that previous estimates might overstate the employment impact of minimum wage increases among teenage workers due to insufficient control in the specifications. As the estimates in Table 2 show, with sufficient variations in the minimum wage variable, minimum wage increases have a negative and statistically significant impact on aggregate employment among teenage workers, and the elasticity of employment is in a range from -0.099 to -0.126. I conclude that Model 6 might be more suitable for the analysis of the impact of current minimum wage increases between 2007 and 2009, because Model 6 is able to control for any unobservable macro-economic component or/and policy that can potentially bias the estimate, and the estimate would be more accurate. 22
Robustness Checks In this section, I report regression results of a series of sensitivity checks to test the robustness of the estimates above. I first re-estimate the models in Table 2 with wage variables measured at real values, to see whether I get similar patterns of changes in the coefficients of minimum wage variable; later lagged employment effects of minimum wage increases are also allowed based on Model 4 6 in Table 2, to see whether changes in the estimates still hold with different controls for different business cycles. For the last part, I estimate the employment effect of minimum wage increases by exploring alternative ways of correcting for standard errors in Model 6 (Table 2). Real minimum wage. I first re-estimate each model included in Table 2 using wage variables after adjusting the wage variables (prevailing minimum wage and adult wage) for inflation using the CPI-U index from the Bureau of Labor Statistics. The results of the inflation-adjusted models presented in Table 3 are similar to results of the models using the nominal values. Whether the real or nominal values are used, the main conclusions of the model remain the same: An increase in prevailing minimum wage reduces employment among teenage workers. For example, when using real values between 2007 and 2010, the coefficient of minimum wage variable is -0.243 and remains statistically significant at the 1-percent level in Model 7. In the corresponding model that uses nominal values (Model 2), this estimate was -0.241. The significance of the minimum wage estimate changed only in one occasion between the models using the nominal wages and the models using the real wages: between Model 3 and Model 8. In Model 3, the minimum wage coefficient is estimated to be -0.051. Although not 23
significant, it was very close to being significant. When inflation-adjusted minimum wage variables are used in Model 8, the coefficient estimate becomes -0.061 and significant at 10- percent level. Models 9 to 11 present estimates using observations from January 1996 to December 2010, where the point estimate is around -0.045 and the employment elasticity is about -0.1. 24
Table 3 Effects of Real Minimum Wage Increases on the Ratio of Teen (Ages 16 19) Employment to Teen Population a 2007-2010 1996-2010 Variable Model 6 Model 7 Model 8 Model 9 Model 10 Mode 11 Log state minimum b -0.206*** c (0.039) d -0.243*** (0.033) -0.061* (0.036) -0.047** (0.019) -0.046** (0.019) -0.044** (0.019) Log adult wage b (0.026) (0.032) (0.032) (0.017) (0.017) (0.017) 0.009 0.034 0.048-0.001-0.002-0.003 Share of teenagers 0.042 (0.129) -0.120 (0.126) -0.144 (0.124) -0.042 (0.067) -0.043 (0.067) -0.046 (0.067) Unemployment rate -0.442*** (0.071) -0.469*** (0.060) -0.129* (0.072) -0.304*** (0.046) -0.303*** (0.046) -0.300*** (0.046) Month effects Yes Yes Yes Yes Yes Yes State effects Yes Yes Yes Yes Yes Year effects Yes Yes Yes Yes Recession dummy Yes Recession 1 dummy Yes Recession 2 dummy Yes 0.114 0.449 0.482 0.458 0.458 0.459 Number of observations 2,448 2,448 2,448 9,180 9,180 9,180 Elasticity -0.570-0.758-0.123-0.113-0.111-0.107 Note. a The dependent variable is the ratio of employed teenagers to the teenage population; b Minimum wage and adult average wage are nominal values; c * ** *** ; d Heteroskedasticity-robust standard errors are in parentheses. 25
Lagged employment effects. Some researchers (Neumark and Wascher 1994; Baker et al. 1999; Burkhauser et al. 2000a, 2000b; Campolieti, Gunderson and Riddell 2006; and Sabia 2009) argued that the effect of minimum wage increases on employment might not appear immediately. It takes time for firms and employees to adjust to the changes. It is, therefore, important to introduce a lagged minimum wage variable into Model 4, 5 and 6, which include a 12-month lagged value of minimum wage. The estimates are reported in Model 11, 12 and 13, respectively. With the lagged employment effect present, the estimated negative impacts of the prevailing minimum wage on teen employment are still statistically significant but more concentrated around -0.140, which implies that minimum wage increases will have longer employment effect, and that the inclusion of a lagged minimum wage will improve the precision of the analysis. I also observe that the coefficient of lagged minimum wage variable itself is not statistically significant different from zero. Sabia (2009) argues that this might due to the statistical relation between year effects and the lagged minimum wage. 26
Table 4 Effects of Nominal Minimum Wage Increases on the Ratio of Teen (Ages 16 19) Employment to Teen Population with Lagged Employment Effects: January 1996- December 2010 a Variable Model 11 Model 12 Model 13 Log state minimum b -0.055** c -0.055** -0.054** (0.026) d (0.026) (0.026) Log state minimum lagged 1 year b 0.001 0.003 0.004 (0.025) (0.025) (0.025) Log adult wage c 0.005 0.004 0.003 (0.018) (0.018) (0.018) Share of teenagers -0.047-0.048-0.050 (0.069) (0.069) (0.069) Unemployment rate -0.317*** -0.317*** -0.313*** (0.047) (0.047) (0.047) Monthly effects Yes Yes Yes State effects Yes Yes Yes Year dummies Yes Yes Yes Recession dummy Yes Recession 1 dummy Yes Recession 2 dummy Yes 0.463 0.464 0.464 Number of observations 8,568 8,568 8,568 Elasticity -0.140-0.141-0.137 Note. a The dependent variable is the ratio of employed teenagers to the teenage population; b Minimum wage and adult average wage are nominal values; c * ** *** ; d Heteroskedasticity-robust standard errors are in parentheses. 27
Alternative error corrections and random effects model. Another potential criticism is that different error corrections might affect the significance of the policy impact. For example, I report all the estimates correcting for common heteroskedasticity; however, it is suggested by Bertand, Duflo, and Mullainathan (2004), and Sabia (2009) suggested that the estimates should allow for correlation of observations within each state, so the standard errors should be clustered at state level, which might subsequently alter the significance level of the estimates. To test these concerns, I estimate four regressions using all specifications of Model 5 in Table 2. Corrections in the error terms should not substantially alter either coefficients or elasticity of the minimum wage variable. Therefore, I mainly focus on changes in the significance level. In Model 14 standard errors are clustered at state level, which allows for correlation of variables within the same state. Model 15 assumes there is a cross-sectional correlation in the error term across the panel variable state, while still correcting for common heteroskedasticity. Model 16 assumes the autocorrelation problem is state-specific, hence, the estimates are regression results controlling for state-specific autocorrelation. The estimates remain statistically significant at the 10-percent level, which confirms that the negative impact of minimum wage is statistically significant. State effects control for time-invariant components that are state-specific. In Model 17 of Table 5, instead of assuming those time-invariant characteristics are state-specific, I use a random effect model, assuming those unobserved time-invariant components follow a normal distribution. Consistent with previous estimates, the coefficient of the minimum wage variable is still negative and remains statistically significant at the 10-percent level. I also find the employment elasticity is -0.099, which is the same as the estimate in Model 6 in Table 2. 28
Table 5 Effects of Nominal Minimum Wage Increases on the Ratio of Teen (Ages 16 19) Employment to Teen Population with Alternative Error Corrections: January 1996 December 2010 a Variable Model 14 Model 15 Model 16 Model 17 Log state minimum b -0.042* c -0.030** -0.034** -0.042* (0.025) d (0.014) (0.017) (0.025) Log adult wage b -0.001 0.007-0.002-0.001 (0.019) (0.013) (0.016) (0.019) Share of teenagers -0.046-0.037-0.006-0.047 (0.054) (0.048) (0.059) (0.054) Unemployment rate -0.300*** -0.243*** -0.312*** -0.310*** (0.054) (0.036) (0.043) (0.055) Month effects Yes Yes Yes Yes State effects Yes Yes Yes Yes Year effects Yes Yes Yes Yes Recession1 dummy Yes Yes Yes Yes Recession2 dummy Yes Yes Yes Yes 0.459 P-value for P-value for Chi-2=0.000 Chi-2=0.000 0.466 Number of observations 9180 9180 9180 9,180 Elasticity -0.099 - -0.097-0.099 Note. a The dependent variable is the ratio of employed teenagers to the teenage population; b Minimum wage and adult average wage are nominal values; c * ** *** ; d Heteroskedasticity-robust standard errors are in parentheses. 29
Sub-Sample Based on Demographic Characteristics The estimates above have consistently demonstrated negative and statistically significant impacts of minimum wage increases on employment among teenage workers. I also analyze the employment effect on sub-groups with specific demographic characteristics to examine whether the employment effects are similar across different demographic groups, or the negative impact is highly skewed towards a certain group among teenage workers. Tables 6 and 7 present estimates of Equation 1 for different demographic groups, based on gender, race and ages. Given that all state-month observations in my analysis are aggregated from the CPS data at the individual level, I cannot directly use binary indicators to compare the coefficient of each racial group. Instead, I run separate regressions for each sub-sample to compare the impact of prevailing minimum wage on teen employment. The estimates shown in Tables 6 and 7 do not show the same pattern as the estimates before. The coefficient of minimum wage variable for male is -0.023, and the estimate for female teenage workers is -0.047, which is not statistically significant. In Table 7, the coefficient is - 0.038, 0.001 and 0.056 among teenage workers that are White, Black and Asian, respectively. It is possible for these groups to be jointly significant while separately insignificant: By looking into the policy impact on sub-groups, the analysis is only using a smaller sample, and the estimates lose their statistical powers from smaller Number of observations. Another possible explanation for this is that negative employment effect of minimum wage increases is more due to distributional issues. According to BLS in 2007 (Characteristics of Minimum Wage Workers: 2007, 2008), among hourly paid workers the percentage of workers with wages at or below the prevailing minimum wage did not vary too much among white, black and Asian workers; while there were more female teenage workers who were making at or below the prevailing minimum 30
wage (about 3 percent) than male teenage workers (about 1 percent). The estimates here are actually quite consistent with the distribution of minimum wage workers across different demographic characteristics. Table 6 Effects of Nominal Minimum Wage Increases on the Ratio of Teen (Ages 16 19) Employment to Teen Population by Gender: January 1996 - December 2010 a Variable Male Female Log state minimum b -0.023-0.047 (0.027) d (0.038) Log adult wage b -0.012-0.010 (0.027) (0.028) -0.128 0.288* Share of relevant population (0.131) (0.163) Unemployment rate Month effects Yes Yes State effects Yes Yes Year effects Yes Yes Recession 1 dummy Yes Yes Recession 2 dummy Yes Yes 0.180 0.114 Number of observations 9180 9178 Elasticity -0.059-0.099-0.366*** c -0.271*** (0.062) (0.073) Note. a The dependent variable is the ratio of employed teenagers to the teenage population; b Minimum wage and adult average wage are nominal values; c * ** *** ; d Heteroskedasticity-robust standard errors are in parentheses. 31
Table 7 Effects of Nominal Minimum Wage Increases on the Ratio of Teen (Ages 16 19) Employment to Teen Population by Race: January 1996 - December 2010 a Variable White Black Asian Log state minimum b -0.038 0.001 0.056 (0.031) d (0.069) (0.115) Log adult wage b -0.001-0.061*** c -0.209** (0.030) (0.069) (0.101) Share of relevant population -0.105 (0.071) -0.043 (0.257) -1.188* (0.650) Unemployment rate -0.349*** -0.183-0.33 (0.055) (0.164) (0.207) Month effects Yes Yes Yes State effects Yes Yes Yes Year effects Yes Yes Yes Recession 1 dummy Yes Yes Yes Recession 2 dummy Yes Yes Yes 0.214 0.031 0.03 Number of observations 9148 6548 4015 Elasticity -0.112 0.109-0.101 Note. a The dependent variable is the ratio of employed teenagers to the teenage population; b Minimum wage and adult average wage are nominal values; c * ** *** ; d Heteroskedasticity-robust standard errors are in parentheses. 32
Conclusions and Future Research Conclusions This paper measures the impact of minimum wage increases on teenage employment. Consistent with previous analysis, the results reflect a negative and significant employment effect on teenage workers. I find that the elasticity of teenage employment with respect to minimum wages is in a range of -0.1 to - 0.14, which is slightly lower than previous researchers found using CPS s MORG data. For example, Burkhauser et al. (2000a) found the elasticity is between -0.2 and -0.6, and Sabia (2009) found a -0.2 to -0.3 range. Part of the differences is due to the different datasets each study uses: this paper uses a more updated dataset than those of previous researchers, under which an increasing number of workers are earning higher wages than the minimum wages. For example, according to the information from the BLS, workers who were earning at or below the prevailing Federal minimum wage took up 3 percent of the overall hourly-paid workers; while in the 2006 before the government starting to raise the federal minimum wage, the share dropped to 2.2 percent (Characteristics of Minimum Wage Workers: 2006, 2007). I also examined the employment effect of minimum wage increases by gender (female, male) and by race (white, black and Asian) separately, but none of the estimate is statistically significant. Only a small fraction of workers in the US labor market are earning minimum wages, and a large share of workers will not be affected by the minimum wage policy changes. For workers who are not affected by this policy change, the coefficient of the minimum wage variable should be zero, but the coefficient of the minimum wage variable measures the average impact on all teenage workers, and the observations from minimum wage earners are much smaller than the 33
number of worker who will not be affected by this policy change, which means that the estimate might be biased towards zero and the estimate here might underestimate the negative impact of minimum wage increases on teen employment rate (Neumark & Wascher, 2008, pp. 83-89). Recently, voices in government and civil society have called for further increases in minimum wage (Editor, 2012). From a policy perspective, this might be a concern. The labor market is still in recovery and it will take time for the employment rate to reach its level in December 2008. (Stiglitz, 2012) The empirical results in this paper l confirm earlier findings of a negative impact of minimum wage increases on employment, which means that right now increases in minimum wage might worsen the existing problem in the U.S. labor market. Future Research One of the reasons that researchers analyze the policy impact among teenage workers is that we expect a large share of teenage workers earning minimum wages, and as a result, they might be hurt the most by further policy changes (Card, 1992a, p. 23). However, some researchers argue that now in the U.S. labor market most teenage workers are working for wages higher than the minimum wage and should not be affected by the policy changes. Hence, an estimate of the aggregate employment ratio across states will give a biased estimate of the real policy impact. They further argue that to get an accurate estimate, we should look specifically at teenage workers who are directly affected by this policy change. This requires matching the individuals in the CPS data across years and calculating the exact elasticity of the employment effect of minimum wage increases on teenage workers. This task requires an amount of time that exceeds the time frame of this thesis. Nonetheless, it could be an interesting topic for my future research. 34