The Plummeting Labor Market Fortunes of Teens and Young Adults
Trang Hoang_Research Paper
1. Work Incentive Effect of State Expenditures on Transfer Programs
Trang Hoang
Cornell College
May 2014
Abstract
The purpose of this paper is to examine the incentive effect of per capital expenditures on social welfare programs by state, instead of the commonly studied social assistance received by individuals. State spending on social welfare is determined by fiscal capacity (including tax revenues, federal aids, and local expenditures on antipoverty programs), need or poverty rate, and political and institutional cultures. In other words, what this paper actually looks at how the likelihood of individuals to receive social benefits based on their states’ expenditures influences their labor market behavior. The main findings indicate that living in a state with higher per capita expenditures on welfare programs will likely increase participation in the labor force and that the effect is slightly greater for whites than for non-white workers.
2. I. INTRODUCTION
The effect of social welfare programs on labor market behavior has been the topic of numerous economic researches. For example, the momentum for the Welfare Reform Act of 1996 in the United States was the concern that limitless transfer payment by the government gave no incentive to individuals to go off welfare benefits and participate in the labor force. In Europe, the “eurosklerosis” problem is mainly created by, some economists believe, the overly generous social benefits, such as unemployment benefits, food stamp programs, and mortgage- modification programs, which lead to disincentives to actively search for unemployment. Lemieux and Milligan (2004) studied a unique policy period in Quebec, Canada, where people under 30 years old received benefits 60 percentage lower than recipients with age over 30, and they pointed out strong evidence that more generous social assistance reduced employment. Rodriguez-Pose and Tselios (2012) researched on welfare regimes in European countries and came to the conclusion that the socioeconomic incentives for people to get education and work are stronger in countries with the weakest welfare systems. Even though there have been improvements, the ongoing reforms have not completely resolved the problem of discouraging work caused. Thus, the incentive effect of social welfare programs lies in the heart of transfers policy discussion and continues to a vital topic of research for many economists.
In this paper, I study the relationship between state level of social assistance and the decision to participate in the labor force, using data in 2009. The subjects studied are prime-age potential workers (age between 24 and 55) in the United States. It is important to understand that the social welfare variable being used in this model is different from the amount of social assistance individuals receive. Instead, it is the per capital dollar amount of the sum of three major welfare programs by state, including Supplemental Nutrition Assistance Program (SNAP or formerly
3. known as Food Stamp), State Unemployment Insurance (SUI), Temporary Assistance for Needy Families (TANF), and other government transfer payments, such as education and healthcare benefits. Even though it might not be effective for assessing individual cases, using per capita social assistance by state gives us a different view of looking at the issue that I will clarify below.
In a 2004 report by the U.S. Department of Health and Human Services, the three most important determinants of state spending on social welfare are fiscal capacity (including tax revenues, federal aids, and local expenditures on antipoverty programs), need or poverty rate, and political and institutional cultures. If the per capital social assistance is high, it indicates that citizens in that state is more likely to receive benefits than those in different states, given they are under similar financial and family constraints. In other words, what this paper actually looks at how the likelihood of individuals to receive social benefits based on their states’ expenditures influences their labor market behavior. In addition to the main question, I also pose questions about the differences in work incentive effect of social welfare between white and non-white potential workers.
An advantage of using per capital social welfare instead of individual assistance dollar amount is that it will offset the problem with selectivity and limited data. Given the fact the dataset of this study is one point in time and not longitudinal, studying only recipients of social benefits will create selection bias because many people are unemployed at the time they are in social programs. However, as discussed above, per capital social welfare is influenced by many factors, which more or less have some direct effect on the participation rate. This can create bias in our model.
4. II. DATA DESCRIPTION AND DESCRIPTIVE STATISTICS
1. Data Description
Most of my analysis relies on data from the Current Population Survey (CPS): Annual Social and Economic (ASEC) Supplement Survey, 2010 collected by the U.S. Census Bureau. The CPS, conducted monthly, is a labor force survey providing current current estimates of the economic status and activities of the population of the States. Specifically, the CPS provides estimates of total employment (both farm and nonfarm), nonfarm self-employed persons, domestics, and unpaid helpers in nonfarm family enterprises, wage and salaried employees, and estimates of total unemployment. In addition to the basic CPS questions, respondents were asked
questions from the ASEC, which provides supplemental data on poverty, geographic mobility/migration, and work experience. Comprehensive work experience information was given on the employment status, occupation, and industry of persons aged 15 and over. The universe of the 2010 CPS version I work with is the civilian non-institutional nonmilitary population of the States living in housing units. The total number of respondents is more than 155,000, but restricted for age (between 24 and 55), my model uses around 90,000 observations.
To complement the data from the CPS 2010, I use the Statistical Abstract of the United States 2012, which is comprehensive summary of statistics on the social, political, and economic organization of the United States. The Abstract allows me to retrieve data on social welfare expenditures by state for different programs and state population. There social programs included are SNAP, TANF, SUI, and other government transfer payments, which mainly consist of medical payments, retirement and disability benefits, and educational and training assistance payments. For both the CPS and the Statistical Abstract, I use data from the year 2009.
5. 2. Descriptive Statistics
The population of this study is made up of 53% of females and 79% of white prime-age, potential workers. The participation rate is 82%, slightly lower than the rate for whites only (83%), but higher than the rate for non-white subjects (77%). Both groups have the same average level of education and experience, which stay at 13.3 years and 20.1 years, respectively. Approximately 66% of whites in the sample have married, while the statistics for non-whites is only 50%. This leads a lower average number of children under 18 for non-whites, which is 0.83 as opposed to 1.05 for whites.
Two seemingly problematic differences between whites and other races lie in non-labor income and per capital social welfare. While white workers have higher mean for non-labor income than their non-white counterparts ($42,396 and $34,612), the first group also has remarkably larger per capital social welfare ($94, 235 and $24,964). It raises a question whether people who have higher wealth/non-labor income tends to live in states with more generous welfare programs. However, the correlation between two variables is negative and negligible (-0.0027 for observations within 24 and 55 years old, both whites and non-whites), indicating that such a relationship can be considered nonexistent.
Conducting t-test for the differences in means in these variables between both groups, I find all of them statistically significant.
6. III. CONCEPTUAL FRAMEWORK
The Labor Supply Theory states that participation decision depends on non-labor income, reservation wage, and received wage. An increase in non-labor income, ceteris paribus, will increase the likelihood of an individual to drop out of the labor force. I expect to see that per capital social assistance has the same effect on labor market behavior as non-labor income. However, since our main explanatory variable does not behave exactly as individual unearned incomes, it is predictable to see results different from my expectation.
Apparently, there are many family and personal factors that affects an individual choice of participating in the labor market. In this model, I control for education, years of experience, number of children under 18, gender, race, non-labor income, and marital status. For years of experience, I use the quadratic function, which is commonly employed in most labor supply papers. Except education and experience, the effect on participation rate of other variables is pretty ambiguous.
IV. REGRESSION MODEL AND RESULTS
Consider the regression model:
(1)
where INLF is the dummy variable for labor force participation of an individual between 24 and 55 years old, SOWE is variable for per capita social assistance, and SWRC is the interaction term between SOWE and A_FEMALE. represents other explanatory variables, which are further explained in Table 2.
7. Under the first model, are both insignificate at 10% level (t-stats are 1.60 and -0.68). I test for heteroskedasticity using the special case of the White test ̂ ̂ ̂
where ̂ stands for fitted values and ̂ is the observed errors from equation (1). Since the statistic is 7864.60, and is extremely close to 0, we reject the null hypothesis and conclude that there is heteroskedasticity in this model. However, since some of the fitted value are outside the unit interval, I do not adjust the model using weighted least squares but stay with OLS with robust standard errors. Fortunately, since the sample size is very large, we can assure that the coefficients are fairly efficient.
Now the coefficient on SOWE is statistically significant under 10% level (t-statistic = 1.66), but of is pretty high (0.475) and its t-statistic is only -0.71. Note that t-statistics are small probably because the coefficients are practically infinitesimal, and not because their standard errors are large. However, when dropping SWGD, there is not much change in the results, and therefore I decide to remove this variable from the model.
I also run two regression models for whites (2) and non-whites (3) separately using similar structure and conduct a Chow test for differences in regression functions between two groups. The OLS results are reported in Table 3 and discussed with more details below.
From the table, we can see an $1000 increase in SOWE increases a person’s likelihood to participate by .0008%. The same result is seen in the second model, when all observations are whites only. However, in the last model, the effect goes up to 0.0015%, which is still very small.
8. Model (1) also tells us that an additional year of education raise participation rate by 2.37%, while it is 2.26% for whites and 2.87% for non-white workers. After the 18th in the labor force, another year of experience actually decreases a person’s likelihood to participate. For whites, the variable has the maximum effect at 18.5 years and it is 20 years for non-whites. The coefficient on A_WHITE in (1) is .056, which means being white will increase the chance of participating by 5.6%. As non-labor income increases by $1000, INLF decreases by .07% for an average person. While being married raises the likelihood of participating by 5.74 percent points, being a female decreases it by a quite significant amount at 11.58%. The effect of A_FEMALE on INLF is even stronger for white workers (12.84%) but much less for the other group (6.87%). Having another child under 18 years old reduces INLF by 1.53% on average and 1.91% for whites. The effect of A_FOWNU18 in (2) is not statistically significant (t-stat = 0.28). Notice that in every variable which has a positive coefficient, the effect on INLF in the first group is relatively smaller compared to the average, while for factors with negative effect on INLF, the magnitude this group has is at least as large as the average.
I conduct a Chow test against the null hypothesis that there is no difference between the two groups. Our -statistic at (9, 90432) degree of freedom is 27.96, which means we reject the null hypothesis at 1% level of significance. Note that even though there is heteroskedasticity in our model, for a Chow test, normality is not needed for asymptotic analysis.
In order to easily compare the effect of different variables, I standardize the model using beta coefficients. From Table 4, we can see that education has the most effect on labor force participation, followed by the gender variable. One standard deviation increase in years of education increases the standard deviation of INLF by .1813. A_FEMALE increase by one standard deviation, similarly, raises INLF by .1491 standard deviation. In all three models,
9. SOWE has the smallest beta coefficient (.0131, .0143 and .0112, respectively). In model (3), A_FOWNU18 has its beta coefficient at .0023. However, since this variable’s coefficient is statistically insignificant, we do not include it in our analysis.
The sign of SOWE is somewhat different from our expectation. This indicates that there are factors other than the non-labor income effect that influences how SOWE behaves. There are several possible explanations. First, the per capital social welfare variable consists of many social welfare programs, which may have different effects on work incentive. For example, social programs on education and training are likely to boost participation rate, while unemployment benefits might have opposite effects. However, most programs are more complicated in a sense that they provide supplemental non-cash service to welfare recipients, such as career orientation and job training as in the case of TANF. Recipients of TANF are also required to work after 24 months in the program, which might increase labor force participation rate.
Second, our cross-sectional data does not allow us to follow individuals before and after social programs, which hinders us from drawing precise conclusion about the effect of social welfare on work incentive. And third, as mentioned in the introduction of this paper, state expenditures on social programs depend on its fiscal capacity. Higher SOWE might indicate that the state has larger tax revenues, probably as a result of high participation rate, which raises the problem with simultaneity bias.
10. V. CONCLUSION
This paper studies the ceteris paribus effect of the per capita social welfare by state on prime-age potential workers’ decision to participate in the labor force. I use OLS estimation to compute the effect, controlling for several family and individual variables. The main finding is that living in a state with large expenditure on welfare programs will increase the likelihood of participating regardless of races. However, the magnitude of this effect is not particularly significant, making it difficult to draw an assertive conclusion.
Previously discussed, the use of per capita social assistance has a couple of complications, since it is determined by different socio-economic and political characteristics of each state. Based on this paper, future researches can look into a method to identify and separate these factors in order to study the work incentive effect of social program under different perspectives. One possible way to do that is instead of using individual data, we can control for state characteristics to test the relationship between per capita dollar amount spent on social welfare by state and participation rate in each state. This will considerably reduce our number of observations to only 50 states and create difficulty when correcting for heteroskedasticity and bias in hypothesis tests, but in return, we gain more consistency between the explanatory and dependent variables.
Another potential improvement for this research is to test for the effect of specific social welfare programs. Since those programs target at different groups of the population and serve different purposes, they probably have contradicting effects on participation decision. However, such a study might require panel data and the study of the before and after behavior of benefits recipients, which leads to a different approach from the one used in this paper.
11.
12. References
Black, Dan A., Terra G., McKinnish, and Seth G. Sanders (2003), “Does the availability of high- wage jobs for low-skilled men affect welfare expenditures? Evidence from shocks to the steel and coal industries,” Journal of Public Economics, Vol. 87, pp. 1921-1942.
Blank, Rebecca M. (2002), “Evaluating welfare reform in the United States,” Journal of Economic Literature, Vol. 40, No 4, pp. 1105-1166. Lemieux, T., & Milligan, K. (2004). Incentive effects of social assistance: A regression discontinuity approach. Center for Labor Economics, University of California, Berkeley Working Paper No.77, May.
Hoynes, H., & Schanzenbach, D. (2012). Work Incentives and the Food Stamp Program. Journal Of Public Economics, 96(1-2), 151-162. doi:http://dx.doi.org/10.1016/j.jpubeco.2011.08.006
Rodriguez-Pose, A., & Tselios, V. (2012). Welfare Regimes and the Incentives to Work and Get Educated. Environment And Planning A, 44(1), 125-149. Wooldridge, J. M. (2006). Introductory Econometrics: A Modern Approach (5th ed.). Mason, OH: Thomson/South-Western.
13. Table 1: Summary Statistics Variable ALL WHITES NON-WHITES Participation rate Per capita social assistance Years of education Years of experience Percentage of whites Non-labor income Percentage of marriage Percentage of females Number of children under 18
14. Table 2: Variable Descriptions INLF = 1 if participate, 0 otherwise SOWE per capita social benefits (in thousands) SWRC SOWE*A_FEMALE A_YRSED number of years of completed education A_EXP years of labor market experience A_EXP2 square of years of market experience A_WHITE = 1 if white, 0 otherwise NONLAB non-labor income (in thousands) A_MARRIED = 1 if married, 0 otherwise A_FEMALE = 1 if female, 0 otherwise A_FOWNU18 number of children under 18 years old
15. Table 3: OLS Results. Dependent Variable: INLF Independent Variables ALL (1) WHITES (2) NON-WHITES (3) SOWE A_YRSED A_EXP A_EXP2 A_WHITE NONLAB A_MARRIED A_FEMALE A_FOWNU18 ** INTERCEPT Observations R-squared Note: 1.The quantities in parentheses below the estimates are the robust standard errors. 2. * indicates statistical siginificance at 10% level of confidence only. 3. ** indicates statistical insignificance at 10% level.