Journal of Health Economics 39 (2015) 123–134

Contents lists available at ScienceDirect

Journal of Health Economics journal homepage: www.elsevier.com/locate/econbase

The effect of state dependent mandate laws on the labor supply decisions of young adults夽 Briggs Depew ∗ Louisiana State University, United States

a r t i c l e

i n f o

Article history: Received 11 January 2014 Received in revised form 19 November 2014 Accepted 25 November 2014 Available online 4 December 2014 JEL classification: I13 I18 J22 Keywords: Health insurance Labor supply Affordable Care Act Young adults State Insurance Mandates

a b s t r a c t Prior to the Affordable Care Act, the majority of states in the U.S. had already implemented state laws that extended the age that young adults could enroll as dependents on their parent’s employer-based health insurance plans. Because of the fundamental link between health insurance and employment in the U.S., such policies may effect the labor supply decisions of young adults. Although the interaction between labor supply and health insurance has been extensively studied for other subpopulations, little is known about the role of health insurance in the labor supply decisions of young adults. I use the variation from the implementation and changes in state policies that expanded dependent health insurance coverage to examine how young adults adjusted their labor supply when they were able to be covered as a dependent on their parent’s plan. I find that these state mandates led to a decrease in labor supply on the intensive margin. © 2014 Elsevier B.V. All rights reserved.

1. Introduction The Affordable Care Act (ACA) drastically reshaped the health insurance landscape for young adults by allowing individuals under the age of 26 to remain on their parent’s employer-based health insurance plans. While many provisions of the ACA have been debated, the law to expand dependent coverage has received favorable reviews from both sides of the political aisle (H.R. 3970, 2009; H.R. 4038, 2009) as well as the general public. Early estimates of the efficacy of the law show large increases in the health insurance coverage rate of affected young adults (Antwi et al., 2013; Cantor et al., 2012b; Sommers and Kronick, 2012; Sommers et al., 2012). However, policies to expand health insurance coverage to dependent children through their parent’s employer based health

夽 I would like to thank Ozkan Eren, Price Fishback,Theresa Gutberlet, Gautam Gowrisankaran, Kei Hirano, Taylor Jaworski, Carl Kitchens, Ashley Langer, Mindy Marks, Naci Mocan, Bob Newman, Ron Oaxaca, Seunghwa Rho, Jessamyn Schaller, Kosali Simon, ToddSorensen, Dek Terrell, Tiemen Woutersen, and two anonymous referees for helpful feedback. ∗ Tel.: +12255783795. E-mail address: [email protected] http://dx.doi.org/10.1016/j.jhealeco.2014.11.008 0167-6296/© 2014 Elsevier B.V. All rights reserved.

insurance plans may have unintended consequences in the labor supply decisions of young adults. Although there has been considerable work emphasizing the role of health insurance in the labor supply decisions of various subpopulations, such as spouses and retirees, the interaction between health insurance and the labor supply decisions of young adults has been mostly overlooked, until recently. In addition to studying the health insurance effects of the ACA, Antwi et al. (2013) find that the ACA also led to a decrease in the likelihood of full-time employment and hours worked. These estimates are the first set of empirical evidence that suggest that young adults may decrease their labor supply when they have access to an outside source of health insurance. In addition to being the first set of evidence, the results of Antwi et al. (2013) are the only set of evidence on the topic. To establish this important finding on the implications of policies that expand dependent health insurance coverage, it is exigent that other studies using alternative identification strategies find similar effects. However, recent work of Slusky (2013) finds that the health insurance and labor market findings of Antwi et al. (2013) are not robust to falsification tests. Slusky (2013) argues that the identification strategy of Antwi et al. (2013) does not satisfy the parallel trends of assumption of the

124

B. Depew / Journal of Health Economics 39 (2015) 123–134

difference-in-difference framework (Abadie, 2005; Bertrand et al., 2004). In this paper I use a much richer source of variation to estimate the effect of dependent coverage mandates on the labor supply of young adults. Specifically, I exploit changes in state-level dependent coverage laws that were implemented prior to the ACA’s dependent mandate. I am able to invoke much weaker identification assumptions than Antwi et al. (2013) that enable me to avoid the identification critique of Slusky (2013). Similar to the general findings of Antwi et al. (2013), I find evidence that these state policies decreased the labor supply decisions of young adults on the intensive margin through decreases in hours worked and full-time employment. A recent article (Dillender, 2014) also uses variation in these state dependent mandate laws to analyze the implications of extending dependent heath insurance coverage to young adults on the labor market and higher education. Dillender (2014) finds that these laws are associated with higher levels of education and wages for individuals who were 18 or younger when a dependent mandate law was implemented. A key difference between this paper and the paper of Dillender (2014) is the identification strategy and thus the underlying mechanism. In Dillender (2014), individuals are only considered to be treated if they were 18 or younger at the time of the law change. Because Dillender (2014) studies the wage and education levels of individuals 26 and older, he is only able to use identifying variation from a few of the early adopting states (see Table 1 for a list of law changes). Therefore, Dillender (2014) findings provide insight on some of the long-run benefits of these policies: individuals may invest more in human capital at earlier ages and sort into better paying jobs later in their 20s. However, the identification strategy and estimation in this paper answers a more direct question of how young adults immediately adjust their labor supply when they exogenously have an alternative source of insurance coverage. The young adult population is important to policy makers because not only is it the largest uninsured group in the U.S., their decisions in regards to human capital accumulation through on-the-job employment plays an important role in the continuing development of the American economy. A young adult who is able to acquire health insurance coverage through a parent’s plan may have no need of his or her own employer-sponsored insurance. Depending on the value of health insurance to a young adult, expanding dependent coverage may cause an affected individual to change employment, reduce hours worked, or entirely exit the labor force. In the mid 1990s, states began implementing dependent mandates that extended the age threshold that was used to determine if an adult child was eligible to be on a parent’s health insurance plan. By 2010, over half the states had enacted expanded dependent coverage mandates. Insurance companies in states with no expanded coverage law generally allowed children under the age of 19 to qualify as dependents on a parent’s policy. However, if a child was a full-time student and not married then they could typically stay on their parent’s policy until the age of 23. To estimate the effect of state laws that expanded dependent coverage, I exploit the variation of policy implementation across time, across state, and across age groups within a state. These three margins of variation allow me to use a difference-in-difference-indifference (DDD) framework. Specifically, I compare the outcomes of individuals who were eligible in states who have implemented a policy to expand coverage to the outcomes of individuals within the same state who were not eligible for the policy. In addition, I control for differences between young adults in non-policy states who would either be classified as eligible or ineligible. I find evidence that young adults adjust their labor supply on the intensive margin in response to having access to a parent’s health insurance

policy. In the main set of analyses, I only use age to assign eligibility. However, states often set other criteria such as marital status or student status to determine eligibility status. Not using the full set of eligibility criteria allows me to avoid the potential endogeneity problem of individuals selecting into treatment. Unfortunately, this empirical strategy leads to attenuation bias in the point estimates of interest. By including additional eligibility criteria (marital status, student status, and own children) in the assignment of eligibility I find much larger point estimates. However, these results are likely being driven by selection into treatment in addition to the effect from access to dependent coverage. The estimated results are robust along many important dimensions. First, the main set of analyses find similar but attenuated results as the set of analyses that uses the full set of eligibility criteria. Differences between the two would raise suspicion of spurious effects because the mechanism is working in the same direction across the two sets of analyses. Second, the mechanism is consistent across gender for both the effect on health insurance and the effect on labor supply decisions. Specifically, the results suggest that state policies that expand dependent health insurance coverage caused females to have a higher take-up rate of health insurance than males. This finding is consistent with females having a higher demand for health insurance. Differences in demand can likely be attributed to greater health care costs of females than males aged 19–29 and to greater risk aversion of females than males. Consistent with the differential in health insurance take-up across gender, the analyses of the labor supply response find that females were more likely to adjust their labor supply than were males. Third, the results are robust when the Great Recession years are omitted from the analysis, suggesting that contemporaneous shocks from the Great Recession are not driving the results. Lastly, through a falsification test that randomly assigns the years that states implement an expanded dependent coverage law, I find no effect on the labor supply outcomes of young adults. This paper contributes to the literature in health economics and labor economics in a number of ways. It shows that access to health insurance plays an important role in the labor supply decisions of young adults. It does this using much weaker identification assumptions than the previous paper by Antwi et al. (2013). In addition, the results in this paper are informative in understanding potential effects of the ACA and speaks directly to the differences found in previous work (Antwi et al., 2013; Slusky, 2013). In terms of dependent mandate laws, the results in this paper add to the work of Monheit et al. (2011) and Levine et al. (2011) who studied the effects of state-level dependent mandates on health insurance coverage by both analyzing a longer time horizon of state policies than the two previous papers and by studying how the state mandates had differential effects across gender. Finally, in contrast to previous work that has shown the potential benefits from dependent mandate laws (Levine et al., 2011; Antwi et al., 2013; Sommers et al., 2012; Dillender, 2014), this paper highlights some of the negative consequences that also accompany such laws.

2. Dependent health insurance and state laws In 2008, young adults between the ages of 19 and 29 made up 17 percent of the population but accounted for 30 percent of the 46 million uninsured individuals in the U.S. (Nicholson et al., 2009). Although the health insurance rate is high for children, it drops significantly after age 18 because individuals who receive health insurance through public programs such as SCHIP are excluded from the programs once they turn 19 years of age. The sharp decline in health insurance coverage at age 19 has been well documented

B. Depew / Journal of Health Economics 39 (2015) 123–134

125

Table 1 State dependent coverage laws. State

Full year

Eligibility criteria a

Colorado Connecticut Delaware Floridab Idahob Illinois Indiana Iowab Kentucky Louisianab Maine Maryland Massachusettsb Minnesota Missouri Montana New Hampshire New Jerseyb New Mexico New York North Dakotab Pennsylvania Rhode Island South Dakotab Texasb Utah Virginia Washington West Virginia Wisconsin

Implemented

Maximum age

2006 2009 2008 2008 2008 2010 2008 2009 2008 2009 2007 2008 2007 2008 2008 2008 2007 2006 2003 2010 1995 2010 2007 2005 2005 1995 2007 2009 2007 2007

24 25 23 24 24 25 23 No limit 25 23 24 24 25 24 24 24 25 29 24 29 25 29 24 23 No limit 25 24 24 24 26

Student

Yes Yes

Yes Yes

Not married Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

No children

Yes

Yes

Yes

Yes Yes Yes Yes

Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Yes

Yes Yes Yes

Sources: Cantor et al. (2012a), National Conference of State Legislatures (2010), and Various State legislature laws and Insurance memos. a Full year implemented is the first full calendar year the policy was implemented. b Florida expanded the maximum age to 29 in 2008. Idaho law allows for non-students through the age of 20. Iowa law allows for non-students through the age of 24. Louisiana law allows for non-students through the age of 20. Massachusetts’ law allows for non-students through the age of 20. New Jersey increased the maximum age to 30 in 2009. North Dakota law allows for non-students through the age of 21. South Dakota increased the age limit for students to 30 in 2007. Texas law allows non-students until the age of 25.

in the health economics literature (Levine et al., 2011; Anderson et al., 2012; Cardella and Depew, 2014).1 Prior to the ACA’s dependent mandate, unless a state policy had been implemented to expand dependent coverage, most employer plans did not cover dependents after the age of 18 unless they were enrolled in a college or university as a full-time student. If a dependent was a full-time student, than he or she was typically covered through the age of 22.2 Data from the 2008 National Health Interview Survey shows that ineligibility because of age or student status was the second most common reason behind the high cost of insurance for why a young adult aged 19–29 did not have health insurance. Beginning in 1995 and continuing throughout the past decade, state legislatures implemented dependent mandates that required health insurance plans to more generously cover child dependents. This was executed by extending the age limit in which young adults could be covered on their parent’s plans.3 The objective

1 Levine et al. (2011) study the impact of individuals who age out of SCHIP on health insurance outcomes. Anderson et al. (2012) and Cardella and Depew (2014) use this age cut-off in a regression discontinuity framework to study the effects of health insurance coverage on the use of medical services and self-reported health, respectively. 2 Employers had strong financial incentives from the tax code to follow these age limits (Levine et al., 2011). 3 State dependent mandates generally apply to all regulated health insurance markets and the state health benefit plans for public employees. Idaho’s law applies only to individual and small-group markets. Minnesota’s law does not include state health benefit plans (Cantor et al., 2012a).

of these polices were to decrease the uninsured rate of young adults. By January of 2010, 30 states had implemented state laws to expand dependent coverage. In addition to expanding the age limits, states also set additional eligibility criteria that often included other factors such as student status, marital status, and whether an individual has his or her own children. Therefore, these state policies are often more limited in scope than the federal law under the ACA that does not incorporate student status, marital status, and dependent’s children in the eligibility criteria. Table 1 presents the implementation year and the more common eligibility criteria of the state dependent coverage laws. When states enacted laws to expand dependent coverage, firms were typically required to abide by the new law upon renewal of the policy. Using tax records (Form 5500) from firm welfare plans in 2009, I found that the majority of health benefit plans renew in January. For this reason, in Table 1 the indicated implementation year is for the following January if the policy was implemented in any month other than January. Table 1 also indicates the state’s eligibility criteria that dependents must satisfy to be eligible as a dependent for health insurance. The maximum age in Table 1 represents the oldest age before becoming ineligible. As shown in Table 1, a few states did not set an upper age limit for individuals enrolled in school.4

4 Table 1 provides a simplified view of the roll-out of the state dependent mandates. As footnoted in the bottom of the table, some states expanded their previous law after their first law was implemented. See Cantor et al. (2012a) for more details on these state laws.

126

B. Depew / Journal of Health Economics 39 (2015) 123–134

The efficacy of state mandates to expand dependent coverage is substantially mitigated by the fact that these state health insurance regulations do not apply to all insurance policies. First, these state laws are not applicable to public insurance. Therefore, parents on Medicare or Medicaid are not able to provide dependent coverage for their children. Second, self-insured firms (firms that assume the financial risk by paying claims directly instead of contracting with an insurance carrier) do not have to comply with state health insurance regulation. Under the federal Employee Retirement Income Security Act of 1974 (ERISA), self-insured firms are exempt from most state-level health insurance regulations, including state benefit mandates. Two previous papers have studied the impact of expanding dependent coverage laws on the insurance rates of young adults. Monheit et al. (2011) uses a difference-in-difference (DD) estimation strategy by comparing differences in the health insurance rates of young adults before and after a state implemented an expanded dependent coverage law. Monheit et al. (2011) find no statistically significant evidence to support that young adults are more likely to be insured as a result of the state policies, rather reallocation in the type of coverage. Specifically, they find that treated young adults were now more likely to be covered as a dependent and less likely to be covered by their own employer-sponsored policy. The other study, by Levine et al. (2011), more fully exploits the potential impact of the policy by considering whether individuals were actually eligible within a state.5 Using a similar DD estimation strategy as Monheit et al. (2011), Levine et al. (2011) similarly find that individuals 19–24 years of age were not more likely to have health insurance coverage as a result of the expanded coverage laws. However, when they interact their DD estimator with an indicator for eligibility they find a significant increase in health insurance coverage for those who were eligible. 3. Health insurance and labor supply Much of the prior literature on health insurance and the labor market has primarily focused on job mobility, single mothers on Medicaid, spousal insurance coverage, and retiree health insurance. Cooper and Monheit (1993) and Madrian (1994) find strong evidence of “job-lock” due to employer-sponsored health insurance. However, research has shown that the expansion of public health insurance has had little or no affect on labor supply. Strumpf (2011) uses state variation in the introduction of Medicaid in the 1960s and 1970s and finds no effect on labor force participation. Yelowitz (1995) found that Medicaid expansion in the 1980s and 1990s that reduced work disincentives caused an increase in labor force participation of single mothers. However, more recent research by Ham and Shore-Sheppard (2005) finds no relationship between labor force participation and this expansion in Medicaid. Meyer and Rosenbaum (2001) finds that the Earned Income Tax Credit plays a large role in the labor supply decisions of single mothers, while Medicaid had little or no effect on employment decisions. In reference to the recent health care reform, Pohl (2014) uses a structural model of labor supply to analyze how single mothers will respond to mandated employer-sponsored health insurance and expansions to Medicaid from the ACA. He finds increases on both the extensive and intensive margin. A long string of literature has pointed to the fact that married women are likely to decrease their hours worked and labor force participation when they are able to receive health insurance through their husband’s plan (Buchmueller and Valletta, 1999;

5 As discussed later in this paper, the additional eligibility criteria used by Levine et al. (2011) is endogenous because young adults could select into treatment.

Olson, 2002; Royalty and Abraham, 2006). In addition, there has been a number of studies that show that health insurance plays an important role in the employment decisions of individuals near retirement (Blau, 1994; Blau and Gilleskie, 2001).6 However, prior to Antwi et al. (2013) no paper had attempted to study the interaction of labor supply and access to health insurance for the young adult population. Laws that allow young adults to access to their parent’s health insurance as dependents can alter the labor supply decisions of affected young adults on a number of dimensions. Without an outside source of health insurance through a parent, a young adult may sort into jobs that provide health insurance. These types of jobs nearly always require full-time employment. Therefore, a dependent mandate may cause one to decrease hours worked, switch to a different job, or exit the labor force. These outcomes are important because accumulated on-the-job experience is an important factor for future wages. However, breaking the link of employerbased health insurance may also allow individuals to sort into jobs where they are more productive. Therefore, a dependent mandate law may have immediate unintended welfare benefits in the labor market through better employee–employer matching. In addition to the labor market, a dependent mandate may also effect the human capital decisions of young adults. An individual may enroll or continue in school if they can access a parent’s health insurance plan instead of relying on an employer for affordable heath insurance. The potential welfare implications of dependent mandates are complex because they involve significant changes on a number of margins. Changes in health insurance take-up does have a significant role in welfare analysis, however, there is an associated cost with increasing the number of individuals who are covered on health insurance plans. Depew and Bailey (2014) find that the ACA’s dependent mandate resulted in higher premiums for plans that covered children. Specifically, they find that because of the dependent mandate, plans that covered children increased 2.5 percent. This estimate is slightly larger, although not statistically different, than the 2.2 percent increase in the number of individuals in plans that cover children (Depew and Bailey, 2014). In addition to premiums, a significant portion of the long-run general equilibrium welfare effects of laws that expanded coverage to young adults will depend on whether it leads to more skilled or less skilled workers in the labor market. 4. Data and empirical strategy The primary data sources for this paper are the 2001–2010 American Community Surveys (ACS) (Ruggles et al., 2001-2010). The ACS is a nationally representative sample that provides information on respondents labor market outcomes and other demographic variables. Unfortunately, prior to 2008 the ACS did not ask respondents about health insurance outcomes. Prior papers that have studied the effect of expanded dependent coverage laws on health insurance take-up have used the Current Population Survey (CPS) (Monheit et al., 2011; Levine et al., 2011) and the Survey of Income and Program Participation (SIPP) (Antwi et al., 2013). The key advantage of the ACS over the CPS or the SIPP is the number of respondents surveyed. The next section, which explains the identification strategy, will more extensively discuss the importance of using a large sample in the data analysis. To be short, the identification strategy will rely on a coarse measure of eligibility to avoid selection into treatment. Therefore, a large sample size is needed

6 See Blau and Gilleskie (2008) and Strumpf (2010) for additional evidence of the role of health insurance on the employment decisions of older individuals.

B. Depew / Journal of Health Economics 39 (2015) 123–134

.3

.5

.4

Full-time Employment .6 .7 .8

Full-time Employment .5 .6 .7

.9

Males

.8

Females

127

-5

0 Year

5

-5

0 Year

5

Control Ages - Treated State Control Ages - Control State Treated Ages - Control State Treated Ages - Treated State Fig. 1. Full-time employment by treatment and control groups.

because the estimated effects of the state laws will be attenuated estimates of the true effects.

4.1. Identification strategy To identify the causal effect from expanded dependent coverage mandates one must control for systematic shocks to the labor market that are correlated with the policy changes of states. To do this, I exploit the variation of policy implementation and age eligibility requirements, as listed in Table 1,7 by comparing the change in outcomes of individuals who meet the age criteria in states that have implemented a policy (treated group) to individuals within the same state who did not meet the age criteria (control group). In addition, I flexibly control for differences in young adults in non-treated states (those who did not adopt a policy to expand dependent coverage) who would have been either eligible or ineligible if their state had adopted a policy. This estimation strategy is the difference-in-difference-in-difference (DDD) framework that is similar to other labor- and health-related studies, such as Gruber (1994). The DDD strategy has weaker assumptions than the commonly used difference-in-difference (DD) estimation strategy. While the DD estimator requires strict assumptions on the pre-treatment trends of the control and treated groups at the state-year level, the identification assumption of the DDD estimator is that there exists no contemporaneous shock that effects the relative outcomes of the eligible group in the same state-years as the law (Gruber, 1994). In the context of this paper, contemporaneous shocks at the state-year level would violate the DD framework, however, in the DDD framework, only contemporaneous shocks at the age-state-year level are problematic. Thus, the assumption

7 In the estimation, I use the policy changes described in Table 1. I do not use variation from the ACA because the federal policy does not vary at the state-age or state-year levels.

of similar pre-treatment trends in the DDD model is much easier satisfied because of the additional within state-year control group. Fig. 1 displays the trends in full-time employment, conditional on being employed, both for females and males. Each state is centered on the year the policy was implemented (year 0). Since the timing of policy implementation and eligibility criteria vary by state, graphical analysis of the pretreatment trends in control states is not straightforward. The control state trends were constructed by duplicating the number of observations in the control states by the number of policy adopting states (thirty). Each of the duplicated control state observations was then assigned to one of the thirty treated states. Eligibility criteria were than used from the assigned treated state to place an individual from a control state into one of two groups: control ages or treated ages. As displayed in Fig. 1, female full-time employment has a different trend for treated ages in treated states relative to control ages in treated states. However, this difference in trends will be differenced out in the analysis because there is a similar difference in trends across treated ages and control ages in control states. For males, it appears that all of the trends are similar. To statistically evaluate whether the trends are different, the expanded sample was limited to preimplementation observations and regressions were ran separately by gender for each outcome on an indicator for treated age, an indicator for treated state, a linear trend, a two-way interaction for each paired regressor (three in total), and a three-way interaction of the three regressors. The coefficient on the three-way interaction is the parameter of interest. It measures the difference in the linear trend for treated individuals in treated states, after properly conditioning out differences in trends in the control groups. To account for the duplicate observations in the control states, two-way clustered standard errors (Cameron et al., 2011) were calculated on the individual and state. For each outcome, the results show that there is no statistically significant difference in trend for treated individuals, prior to treatment. It is also worth noting that Fig. 1 shows no significant change in the trend of full-time employment after policy implementation for treated individuals in treated states. This

128

B. Depew / Journal of Health Economics 39 (2015) 123–134

is consistent with the results reported in the paper, which found a small effect of the law on full-time employment. As suggested in Section 2, the eligibility criterion of states was set on a number of different margins. One strategy to estimate the effect of the state dependent coverage laws is to use all of the eligibility criteria outlined in Table 1. However, this strategy is problematic because individuals can select into treatment by enrolling in school or by delaying marriage. Therefore, if one observes reductions in labor supply under this framework, the result may be caused by contemporaneous effects from selection into treatment and not from access to parental health insurance coverage. To circumvent this selection problem, it is important that I only use exogenous eligibility criteria. The changes in state policies to expand dependent coverage from the years 2001 to 2010 provide a very suitable setting to identify the causal effect for two reasons: (1) unlike the new federal law under the ACA, states implemented the policies to expand coverage at different points in time. Therefore, it is of less concern that systematic time shocks are driving changes in labor market decisions of young adults. (2) The age cutoff for a dependent to be treated by the policy varied across states. This variation in the age eligibility requirement allows me to condition out contemporaneous effects that may be specific to certain age groups of the young adult population. In all, this across- and within- time, state, and age variation provide a robust estimation setting to reassure that the results from the state policy changes are being driven from expanded coverage laws and not contemporaneous effects. 4.2. Empirical framework The general DDD regression equation for individual i with eligibility e in state s at time t is yiest = ˛te + se + ıts + (eligibleiest × lawst ) + Xiest ˇ + εiest .

(1)

The DDD parameter coefficient of interest is . eligible is an indicator that takes the value of one for an individual who satisfies the eligibility criteria set by the state. lawst is an indicator that takes the value of one for a state that has an expanded coverage law in place at time t. Xiest is vector of observable characteristics specific to the individual. Included in X are controls for exogenous characteristics of the individual: race (white, black, or other) and gender. εiest is an unobserved term specific to the individual that effects the outcome. ˛te is an eligibility-year fixed effect,  es is an eligibility-state fixed effect and ıts is a year-state fixed effect. In the empirical analysis, standard errors are clustered on the state to allow for common unobserved shocks to occur to state residents over time.8 To avoid the identification problem caused by selection, I assign eligibility based on the age criteria set by the state. Therefore, eligibleiest , in Eq. (1), takes the value of one if individual i is below the age cutoff set by state s at time t. Similarly, ˛te is simply defined as a age-by-year fixed effect and  se is defined as a state-by-year fixed effect. Only using age to assign eligibility causes significant measurement error because the assignment of eligibility based solely on age causes a very coarse measure of actual eligibility. Therefore, this coarse measure assigns many individuals as being eligible for treatment that are not actually treated by the policy. This source of measurement error will cause significant attenuation bias in the point estimate of interest. Because of the significant attenuation bias that will exist, it is important to employ a data set with a large number of observations in order to have enough power to potentially find a statistically significant effect.

8

See Wooldridge and Imbens (2007) for a more complete discussion of DDD models with multiple groups and time periods.

Prior to testing the hypothesis that individuals alter the labor supply because of access to dependent coverage, it is important to consider how effective these state laws were in expanding dependent health insurance coverage. The studies of Levine et al. (2011) and Monheit et al. (2011) suggest that individuals treated by the policy are more likely to be insured as a dependent. However, neither Levine et al. (2011) nor Monheit et al. (2011) used the full horizon of state laws prior to the ACA. Because I use state laws until the implementation of the ACA, this study incorporates almost twice as many states that have implemented laws than the studies of Levine et al. (2011) and Monheit et al. (2011). Furthermore, since I employ a different empirical strategy than the empirical strategies applied in the two previous studies, it is important to establish that these state laws increased the likelihood of an individual below the age cutoff to have coverage as a child dependent.9 To test how the state laws to expand dependent health insurance coverage affected health insurance coverage as a child dependent, I apply the regression framework described Eq. (1) for the binary outcome of whether an individual aged 19–29 has health insurance coverage as a dependent on a parent’s plan. There are three estimation details that are worth noting. First, I use linear probability models to estimate the parameters of Eq. (1). This strategy is consistent with the other related work on this topic (Antwi et al., 2013; Monheit et al., 2011; Levine et al., 2011; Slusky, 2013). Second, the regression analysis includes state-byyear fixed effects, state-by-age fixed effects, and year-by-age fixed effects, there are over 1100 fixed effects. Therefore, to conserve space I report only the DDD estimate of Eq. (1).10 Third, for all reported regression estimates, standard errors are clustered at the state level to adjust for within-state correlation over time. 5. Results 5.1. Coverage as child dependent The mechanism of interest that causes individuals to adjust their labor supply is the expansion of dependent coverage laws that allow young adults to access their parent’s health insurance policy. However, affected young adults may not actually be insured at a higher rate, rather, they are just more likely to be insured as a child dependent. Unfortunately, the ACS did not begin asking respondents about health insurance status until 2008. Therefore, to estimate the effects of the expanded dependent coverage laws on child dependent health insurance coverage, I estimate Eq. (1) using young adults ages 19–29 years old from 2001 to 2010 March supplement of the Current Population Survey (CPS) (King et al., 2010) because the ACS did not begin collecting information on health insurance until 2007. I define an individual to have coverage as a child dependent if they indicate that they receive employer-based health insurance as a dependent from a parent or guardian. Furthermore, I assign child dependent coverage to individuals who are not married, not living with a parent, and receive coverage from someone outside of the home. The CPS only records a dependent’s health insurance policyholder if the policyholder lives in the same home as the dependent. Columns 1 and 2 of Table 2 report the point estimates and standard errors of the DDD estimator of interest for insurance coverage as a child dependent and any source of health insurance, respectively. When females and males are pooled together (the top panel of Table 2), the effect of a state expanded dependent

9 Neither of the two previous studies Levine et al. (2011) and Monheit et al. (2011) took advantage of the within state differences across ages. 10 The author is happy to provide the estimated Stata.ster from the analysis.

B. Depew / Journal of Health Economics 39 (2015) 123–134 Table 2 Effect of state dependent mandate laws on health insurance coverage. Age criteria only Dependent coverage

Any HI coverage

Dependent coverage

Any HI coverage

(1)

(2)

(3)

(4)

0.0108 (0.0098) [.7061] 258,612

0.0633*** (0.0132)

0.0247** (0.0099)

258,612

258,612

0.0259** (0.0112) [.1805] 133,834

0.0093 (0.0083) [.7407] 133,834

0.0699*** (0.0153)

0.0227** (0.0092)

133,834

133,834

0.0199** (0.0095) [.1893] 124,778

0.0119 (0.0113) [.6690] 124,778

0.0523*** (0.0121)

0.0252* (0.0129)

124,778

124,778

Females and males: DDD 0.0232** (0.0091) [.1848] 258,612 N Females: DDD

N Males: DDD

N

All eligibility criteria

a

The data consists of individuals ages 19–29 and is from 2001 to 2010 Current Population Survey. The dependent variable in columns 1 and 3 is an indicator for health insurance coverage as a dependent on a parent’s plan. The dependent variable in columns 2 and 4 is an indicator for any source of health insurance. In columns 1–2, the included regressors are an indicator for race, sex, and the sets of interactions: year-by-age, year-by-state, and age-by-state. Columns 3–4 additionally controls for student-by-state, year-by-student, year-by-married, year-by-children, state-bystudent, state-by-married, and state-by-children. b Standard errors clustered on the state are presented in parentheses. The means of the dependent variables are presented in square brackets. c * 0.10, ** 0.05 and ***0.01 denote significance levels.

coverage law causes a 2.3 percentage point increase on the likelihood of health insurance coverage as a child dependent. From 2001 through 2010, 18.5 percent of young adult aged 19–29 had coverage as a child dependent. Therefore, a 2.3 percentage point increase corresponds to a 12.4 percent increase in dependent coverage. A 2.3 percentage point increase is also consistent with the findings in the previous two studies that analyzed the impact of the state dependent mandate laws on health insurance take-up. Monheit et al. (2011) found a 1.52 percentage point increase in employer-based dependent coverage by comparing individuals across states. They do not use additional eligibility criteria in the analysis. By using the same DD strategy, Levine et al. (2011) report a 2.2 percentage point increase for the outcome of private health insurance coverage.11 Males and females have different health care costs, aversion to risk, and types of employment. Therefore, their decisions that involve health insurance may be significantly different from each other. Neither Monheit et al. (2011) nor Levine et al. (2011) studied the effect of state dependent mandate laws by gender. However, Antwi et al. (2013) show that females were more affected in regards to dependent coverage by the federal dependent mandate under the ACA. Column 1 of Table 2 reports the estimates for females and males in the middle panel and bottom panel, respectively. The point estimates suggest that females were more affected by the state laws, 2.6 percentage points relative to 2.0 percentage points. However, these point estimates are not statistically different from each other at standard testing levels. The elasticity of the point

11 Levine et al. (2011) report larger estimates for a difference-in-difference-indifference model, however, it is not clear if they control for two-way interactions. Instead, it appears that they just interact the difference-in-difference estimator with an indicator of eligibility based on age, student status, marital status and child status of the state’s criteria.

129

estimates suggests that the effects across gender are even more disparate because only 18.0 percent of females aged 19–29 are covered as child dependent relative to 18.9 percent of males. Therefore, the point estimates suggest that child dependent coverage increased by 14.4 percent for females and only 10.6 percent for males. Although the underlying mechanism of interest is dependent coverage take-up, it worth noting couple interesting points on the findings of any health insurance coverage. Sommers et al. (2012) find that under the ACA, health insurance increased more for males than females. Although not statistically different from each other, the effect from the state laws on any health insurance is also larger for males than females, as presented in columns 2 and 4 of Table 2. By comparing the point estimates for dependent coverage to the point estimates for any insurance coverage, it appears that there is a significant amount of substitution between sources of insurance. This is consistent with Antwi et al. (2013), who also found that the child dependent coverage increased significantly more than any source of insurance coverage. Furthermore, Table 2 suggests that this substitution pattern is more salient for females than males (also found in Antwi et al. (2013)). It is plausible that females adjusted their labor supply more than males by substituting out of their own employer-based plan through changes on the extensive margin or intensive margin. The finding of relatively small point estimates on dependent health insurance coverage, as displayed in the first column of Table 2, suggest that the analysis of labor supply outcomes using the same empirical framework will lead to relatively small potential effects. This highlights the importance of having a large sample size. It is not only the measurement error caused by the coarse assignment of eligibility based on age that is driving all of the attenuation bias. Recall, self-insured firms, which cover approximately half of the workers with employer-sponsored health insurance, are exempt from many state regulations, such as expanded dependent coverage mandates. In sections below I further explore the potential extent of the measurement error cause by a coarse definition of eligibility and discuss the proper interpretations of the estimates under the counterfactual of having self-insured firms also being affected by a dependent mandate, as under the ACA’s dependent mandate. 5.2. Labor supply outcomes The results found in Table 2 show that individuals are more likely to have health insurance coverage as a child dependent after a law that expanded dependent coverage was in place. However, was the changes in health insurance enough to induce changes in labor supply on either the extensive or intensive margin? To estimate the effect of the expanding dependent coverage on the labor supply of young adults, I follow the reduced-form empirical strategy presented in Section 4.2 for the labor supply outcomes: labor force participation (LFP), employment, full-time employment conditional on being employed,12 and the log of hours worked conditional on being employed. The top panel of Table 3 report the DDD point estimates and standard errors for the labor supply outcomes of all individuals (females and males pooled together) aged 19–29 from 2001 to 2010 ACS. The second and third panels report the results for females and males, respectively. Columns 1 and 2 of Table 3 report the effect of the state laws on labor force participation and employment, respectively and columns 3 and 4 report the effect of the state laws on full-time employment and the log of hours

12 Full-time employment is measured by an indicator variable that takes the value of one if the individual reports that they usually work 35 or more hours per week.

130

B. Depew / Journal of Health Economics 39 (2015) 123–134

Table 3 Effects of state dependent mandate on labor market outcomes: eligibitiy age only. All LFP

Employed

Full-time

ln(Hours)

(1)

(2)

(3)

(4)

−0.0010 (0.0022) [.6899] 2,826,314

−0.0081** (0.0033) [.7183] 1,949,993

−0.0071* (0.0040) [3.5434] 1,949,993

−0.0035 (0.0024) [.7412] 1,421,138

−0.0009 (0.0031) [.6652] 1,421,138

−0.0085** (0.0038) [.6526] 945,347

−0.0093* (0.0049) [3.4725] 945,347

0.0012 (0.0033) [.8078] 1,405,176

−0.0016 (0.0038) [.7150] 1,405,176

−0.0078* (0.0042) [.7801] 1,004,646

−0.0052 (0.0049) [3.6102] 1,004,646

Females and males: −0.0009 DDD (0.0018) [.7743] 2,826,314 N Females: DDD

N Males: DDD

N

Job = 1

a The data consists of individuals ages 19–29 and is from 2001 to 2010 American Community Survey. Columns 1 and 2 include all individuals. Columns 3 and 4 include individuals who are employed. Eligibility status is determined only by age. Included regressors are an indicator for race, sex, and the sets of interactions: year-by-age, year-by-state, and age-by-state. b Standard errors clustered on the state are presented in parentheses. The means of the dependent variables are presented in square brackets. c * 0.10, ** 0.05 and ***0.01 denote significance levels.

worked, respectively. In general, Table 3 reports very small point estimates. This is partially due to the attenuation bias and the relatively small effects reported in Table 2. In general, the point estimates suggest a decrease in labor supply because all but one of the point estimates in the table is negative. The point estimate on the pooled regression for labor force participation is very small and close to zero. However, the point estimate for females is negative but the point estimate for males is positive. For females, the point estimate suggests that the expanded dependent coverage laws decreased labor force participation by 0.35 percentage points. This point estimate is not statistically different from zero at standard testing levels, but is statistically different from zero at the 85 percent confidence level. Regardless, the point estimate is not very economically significant because approximately 74 percent of females participate in the labor force. The results for the effect on employment are similar to those of labor force participation. All three sets of results report negative point estimate but the standard errors are relatively large in magnitude. These results suggest that one cannot reject the null hypotheses that state laws that expanded dependent coverage did not cause individuals to drop out of the labor force and exit employment. Although the empirical strategy does not provide evidence that individuals altered their labor supply on the extensive margin, it may be the case that they altered their labor supply on the intensive margin by shifting from full-time employment to part-time employment or by decreasing their hours worked. Columns 3 and 4 report the point estimates and standard errors for the outcomes of full-time employment and the log of hours worked, respectively. Similar to the point estimates in the first two columns, the point estimates reported in columns 3 and 4 are negative. However, unlike the results on the extensive margin, the results on full-time employment and the log of hours worked are mostly statistically different from zero. For the pooled regression, Table 3 reports that the state laws decreased full-time employment for individuals who were employed by 0.81 percentage points. This result is statistically significant at the 5 percent level. The results that are conditional

on gender show a similar finding. The effect on females and males is 0.85 percentage points and 0.78 percentage points, respectively. These results are statistically significant at the 5 percent level for females and the 10 percent level for males. From 2001 to 2010 approximately 65.3 percent of females and 78.0 percent of males had full-time employment. Therefore, the point estimates suggests that full-time employment decreased by 1.3 percent and 1.0 percent for females and males respectively. The effects of the state laws are similar for the outcome of hours worked. From the pooled regression, I find that individuals decreased their hours worked by 0.71 percent. This point estimate is statistically different from zero at the 10 percent level. When analyzed separately by gender, the results suggest that females decreased their hours by 0.92 percent and males decreased their hours by 0.52 percent. However, only the effect on females is statistically different than zero. 5.2.1. Using variation in the fully-insured workforce The effectiveness of the state mandates are mitigated by the fact that the mandates do not apply to self-insured firms. Through ERISA, self-insured firms are not regulated by state level policies such as dependent mandates. Alternatively, fully-insured firms are affected by the state dependent mandate laws. Using data from the 2001-2010 Insurance Component of the Medical Expenditure Panel Survey (MEPS) (Agency for Healthcare Research and Quality, 2009), I find that of private sector workers that were enrolled in health insurance, 47 percent were at a firm that was fully-insured. This 47 percent represent 29 percent of the workforce. Furthermore, I find a considerable amount of variation across states and time in this value. For a state-year from 2001-2010, the minimum percent of fully-insured is 26 percent, the maximum is 80 percent, and the standard deviation is 8.4 percent. In terms of share of the work force, the previous three figures correspond to 16 percent, 56 percent, and 5.7 percent, respectively. In this subsection I apply the variation in the share of the workforce at fully-insured firms into the regression analysis.13 Specifically, I reestimate Eq. (1) with the addition of interacting the DDD estimator with the share of the workforce that is at fullyinsured firms at the state-year level. Therefore, I am able to exploit the intensity of the treatment and further test the robustness of the results in Table 2. The results from the interacted DDD are found in Table 4 and show a similar pattern to that found in Table 2. Each point estimates sign in the same direction and there is an overall similar level of statistical inference. The results in Table 4 suggest that if the entire work force was employed at fully-insured firms, than the effect of a state law on an individual’s full-time employment is negative 2.9 percentage points and the effect on hours worked is 2.4 percent. These results are statistically different from zero at 10 percent level and 5 percent level, respectively. 5.2.2. Using additional eligibility criteria Alone, the results above seem to suggest that the state laws had limited economic significance in the labor market. However, this conclusion is at least partially caused by the attenuation bias in the assignment of eligibility. Incorporating additional eligibility criteria into the analysis can alleviate part of this measurement error. However, the additional eligibility criteria, as displayed in Table 1, are likely endogenous because individuals can select into treatment by enrolling in school, delaying marriage, or not having children.

13 I appreciate the feedback from an anonymous referee who suggested to use variation in the fully-insured workforce.

B. Depew / Journal of Health Economics 39 (2015) 123–134 Table 4 Fully-insured effect of state dependent mandate on labor market outcomes: eligibitiy age only. All Employed

Full-time

ln(Hours)

(1)

(2)

(3)

(4)

−0.0098 (0.0079) 2,826,314

−0.0290** (0.0120) 1,949,993

−0.0244* (0.0140) 1,949,993

Females and males: −0.0063 DDD × Share (0.0072) 2,826,314 N Females: DDD × Share N Males: DDD × Share N

−0.0155 (0.0103) 1,421,138 0.0010 (0.0120) 1,405,176

Table 5 Effect of state laws on student status, marital status, and children.

Job = 1

LFP

−0.0077 (0.0122) 1,421,138 −0.0143 (0.0141) 1,405,176

−0.0291** (0.0140) 945,347 −0.0297** (0.0148) 1,004,646

Student

Married

Have children

(1)

(2)

(3)

−0.0105 (0.0121) [.2883] 2,826,314

−0.0029 (0.0055) [.2296] 2,826,314

0.0020 (0.0038) [.3445] 1,421,138

−0.0127 (0.0129) [.3344] 1,421,138

−0.0032 (0.0054) [.3072] 1,421,138

0.0042 (0.0032) [.2927] 1,405,176

−0.0082 (0.0114) [.2418] 1,405,176

−0.0024 (0.0061) [.1512] 1,405,176

Females and males: DDD 0.0032 (0.0027) [.3187] N 2,826,314 Females: DDD

−0.0327* (0.0178) 945,347

N Males: DDD

−0.0169 (0.0174) 1,004,646

a

131

N

The data consists of individuals ages 19–29 and is from 2001 to 2010 American Community Survey. The table reports the estimates from the interaction of the DDD estimator and the share of fully-insured workers in the state. Columns 1 and 2 include all individuals. Columns 3 and 4 include individuals who are employed. Eligibility status is determined only by age. Included regressors are indicators for race, sex, and the sets of interactions: year-by-age, year-by-state, and age-by-state. b Standard errors clustered on the state are presented in parentheses. c * 0.10, ** 0.05 and *** 0.01 denote significance levels.

The data consists of individuals ages 19–29 and is from 2001 to 2010 American Community Survey. Each set of estimates is from a separate regression. Eligibility status is determined only by age, state and year. Included regressors are indicators for race, sex, and the sets of interactions: year-by-age, year-by-state, and age-bystate. b Standard errors clustered on the state are presented in parentheses. The means of the dependent variables are presented in square brackets. c * 0.10, ** 0.05 and *** 0.01 denote significance levels.

I directly test for selection into treatment by implementing the same empirical strategy that is outlined in Section 4.2 for the outcomes of student status, marital status, and an indicator for own children. The results for these outcomes are reported in Table 5. Columns 1, 2 and 3 report the effect on being a student, married, and having children, respectively. The top panel reports the point estimates for a pooled sample of females and males, the middle panel reports the point estimates for females, and the bottom panel reports the point estimates for males. Although none of the point estimates are statistically different from zero, the results suggest that selection into treatment is likely occurring. Specifically, all of the point estimates sign in the direction that is consistent with individuals selecting into treatment. The point estimates for the three specifications for being a student are positive which suggest that these laws may have caused individuals to enroll in school or remain in school for a longer duration.14 This result is consistent with Dillender (2014) who finds that individuals 18 or younger at the time of reform were more likely to have higher years of education. However, these results potentially shed further insight by suggesting that the effect is occurring across all affected individuals and not just those who were 18 or younger when a law was implemented. To further analyze the effect on student status, I allowed the DDD estimator to vary by age group. Although not statistically different from zero, I found that the potential increase in student status is caused by individuals ages 23–25.15 The point estimates for the three specifications for being married are negative and the point estimates for the three specifications for having a child are also negative. To add to the dialog that young adults were likely selecting into treatment, the point estimates reported in Table 5 are not economically insignificant. In all,

these results suggest that using the additional eligibility criteria will lead to endogenous estimates of the causal effect of dependent health insurance coverage on the labor supply. Therefore, estimates derived using the additional eligibility criteria can be interpreted as the state policy effects that additionally capture other aspects of the laws, such as incentives to delay marriage or enroll in school, in addition to the isolated effect of dependent coverage. To obtain the point estimates using the additional eligibility criteria I must flexibly control for the characteristics that determine eligibility status in a state. This is done by interacting the three types of variation: year, age, and eligibility. Therefore, with the proper included interactions, the regression equation of interest is

14 Given the relatively small change in labor supply on the intensive margin, it worth questioning whether there is enough power in the data to detect if a specific proportion of the individuals who lower their labor supply go to school. By analyzing a series of power tests and using a type II error of.10, the data has enough power if 35% or more of individuals who reduced their labor supply enrolled in school. 15 It is worth noting that one should not expect to observe an increase in student status for individuals 19–22 because prior to any law change, young adults of ages 19–22 had access to parental health insurance if they were students.

a

yiamkfst = ˛ta + ˛tm + ˛tk + ˛tf + sa + sm + sk + sf + ıts + (eligibleiamkfst × lawst ) + Xiamkfst ˇ + εiamkfst .

(2)

˛ta , ˛tm , ˛tk and ˛tf are the time by eligibility fixed effects where a represents age, m represents marriage, k represents children and f represents full-time student. Similarly,  sa ,  sm ,  sk and  sf are the state by eligibility fixed effects. Included in the vector of observed characteristics, X, are indicators for race and gender. Columns 3 and 4 of Table 2 report the results from the estimation of Eq. (2) for the outcome of dependent health insurance coverage and any health insurance coverage, respectively. The results displayed in columns 3 and 4 are approximately three to four times larger than the results displayed in columns 1 and 2. Qualitatively, the results are similar as they show an increase in the likelihood of coverage as a child dependent. The pooled results suggest that young adults affected by the law were 6.3 percentage points more likely to have coverage as a dependent. This estimate is significantly larger than the estimate of 2.3 percentage points when only age is used to determine eligibility. Furthermore, by incorporating the additional eligibility criteria in the regression framework, the statistical significance of the results significantly increased. Each of the point estimates from the three specifications (pooled, female and male) is statistically different from zero at the 1 percent level. Table 6 reports the results for the labor market outcomes when the additional eligibility criteria are used in the analysis. In general,

132

B. Depew / Journal of Health Economics 39 (2015) 123–134

Table 6 Effects of state dependent mandate on labor market outcomes: all eligibility criteria. All LFP

Employed

Full-time

ln(Hours)

(1)

(2)

(3)

(4)

−0.0020 (0.0072) 2826314

−0.0265*** (0.0074) 1949993

−0.0251*** (0.0073) 1949993

−0.0150* (0.0085) 1421138

−0.0115 (0.0070) 1421138

−0.0370*** (0.0075) 945347

−0.0365*** (0.0084) 945347

0.0007 (0.0055) 1405176

−0.0013 (0.0042) 1405176

−0.0224*** (0.0067) 1004646

−0.0194*** (0.0059) 1004646

Females and males: −0.0032 DDD (0.0087) 2826314 N Females: DDD N Males: DDD N

Job = 1

a

The data consists of individuals ages 19–29 and is from 2001 to 2010 American Community Survey. Columns 1 and 2 include all individuals. Columns 3 and 4 include individuals who are employed. Eligibility status is determined only by age. Included regressors are indicators for race, sex, and the sets of interactions: year-by-age, yearby-state, age-by-state, student-by-state, year-by-student, year-by-married, yearby-children, state-by-student, state-by-married, and state-by-children. b Standard errors clustered on the state are presented in parentheses. c * 0.10, ** 0.05 and *** 0.01 denote significance levels.

the results are qualitatively consistent with the results presented in Table 3. For the most part, the effect on the extensive margin is negative, small and not statistically significant. The one exception is the effect on the labor force participation of females which is statistically significant at the 10 percent level and suggest that females were 1.5 percentage points less likely to participate in the labor force. The estimated effects on the intensive margin are negative and statistically significant at the 1 percent level for all of the specifications. The pooled specification suggests that employed individuals are 2.65 percentage points less likely to have full-time employment and 2.51 percent decline in hours worked. Similar to the results found on health insurance, the results suggest that females are more affected by the state laws than males. Specifically, the point estimates reflect a 3.7 percentage point decrease full-time employment for females and a 2.2 percentage point decrease for males. Similarly, hours worked decreased by 3.7 percent for females but only 1.9 percent points for males. When one considers that 65.3 percent of employed females have full-time employment relative to 78.0 percent of employed males, the results suggest that a 3.7 percentage point decrease for females corresponds to a 5.67 percent decrease in full-time employment. Similarly, a 1.9 percentage point decrease for males corresponds to a 2.44 percent decrease in full-time employment. 5.3. Discussion on robustness Prior to performing additional sensitivity tests it is worth noting the general robustness of the results. The effect of the state laws on health insurance coverage as a dependent, displayed in Table 2, suggest that females were more affected than males. The results in both Tables 3 and 6 are consistent with this dialog of differential effects by gender. As suggested, if the results were being driven by other contemporaneous factors then it is not obvious why this difference across gender would remain consistent through the analysis. Furthermore, there is economic reasoning to why females are more likely to be affected than males. For the age group of 19–29-year-olds, females have higher health care costs because of reproductive biology (Mustard et al., 1998) and are typically more risk averse (Croson and Gneezy, 2009). Therefore, females are likely

to have a higher demand for health insurance than males. Prior to a state’s policy reform, it is probable that females, because of higher demand for insurance coverage, were more likely than males to condition their labor supply decisions on access to health insurance coverage by seeking employment opportunities that offered health insurance. Through the expanded dependent coverage laws, females now have an alternative source of health insurance and therefore are more able to adjust their labor supply decisions, relative to males, as a consequence to the policy. In addition, I analyzed heterogeneous effects by motherhood to test whether women with children are more tied to the labor market. I found that the decrease in labor supply on the intensive margin is largely operating through women without children. However, it should be noted that young adult women with children are three times more likely to receive health insurance through Medicaid and thus the dependent mandate laws would have a more limited effect for this group. Using the federal dependent mandate of the ACA, Antwi et al. (2013) found similar results to those presented in this paper. Specifically, both papers found that young adults adjust their labor supply on the intensive margin but not on the extensive margin. Antwi et al. (2013) results suggest a 2.2 percentage point decrease in fulltime employment and a 4.75 percent decrease in hours worked. Although these estimates are significantly larger than the 0.8 percentage point decrease on full-time employment and a 0.7 percent decrease in hours worked found in Table 3 of this paper, the differences are not as significant when you scale up the point estimates to reflect that the state laws only affected fully-insured firms. By exploiting variation in the share of the fully-insured workforce, one can directly compare the effect of the state laws to those of the ACA. In 2010, just prior to the implementation of the ACA’s dependent mandate the MEPS suggests that 59.8 percent of the private sector workforce is at either an fully-insured or self-insured firm. Therefore, by multiplying the point estimates in Table 4 with 59.8 percent, one can calculate the labor supply effects from the state laws under the counterfactual that self-insured firms were also affected, similarly to the ACA. These calculations suggest that the effects on labor supply would be approximately 1.7 percentage points decrease in full-time employment and approximately 1.5 percent decrease in hours worked. Although these estimates are still slightly smaller than the findings in Antwi et al. (2013), there is additional measurement error caused by the coarse assignment of eligibility. Therefore, depending on the size of the remaining measurement error, the point estimates would be further scaled up. One way of scaling up the DDD point estimates on labor supply is to divide them by the DDD point estimates on dependent health insurance. Because the analysis of labor supply and dependent health insurance use the same set of regressors, the quotient from dividing the DDD estimate from a labor supply outcome by the DDD estimate from dependent health insurance coverage is equivalent to regressing a labor supply outcome on dependent health insurance coverage and instrumenting with state law changes. Table 3 reports a DDD estimate for full-time employment of −.0081 and Table 2 reports a DDD estimate for dependent health insurance coverage of.0232. Therefore, the implied split sample instrumental variable (IV) estimate is −.349. Using the same calculation from the point estimates in Antwi et al. (2013), their paper suggest an implied IV estimate on full-time employment of −.315. However, the result for log hours is not as similar in the two papers. In this paper, the implied split sample IV estimate for log hours is −.306 and for Antwi et al. (2013) the implied IV estimate is −.676. Although the empirical strategy controls for age-year fixed effects and state-year fixed effects, it is still a concern that the results presented in the paper are a derivative of the Great

B. Depew / Journal of Health Economics 39 (2015) 123–134

133

Table 7 Robustness checks. Excluding great recession Eligibility: age only

Eligibility: all criteria

% Full ins. interaction

Full-time

ln(Hours)

Full-time

ln(Hours)

Full-time

ln(Hours)

Full-time

ln(Hours)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

−0.0068 (0.0055) 1,424,322

−0.0221*** (0.0065) 1,424,322

−0.0207*** (0.0068) 1,424,322

−0.0383** (0.0158) 1,424,322

−0.0231 (0.0206) 1,424,322

−0.0004 (0.0063) 1,949,993

0.0034 (0.0067) 1,949,993

−0.0123** (0.0051) 688,748

−0.0112 (0.0073) 688,748

−0.0332*** (0.0072) 688,748

−0.0313*** (0.0086) 688,748

−0.0479** (0.0202) 688,748

−0.0423 (0.0280) 688,748

−0.0034 (0.0062) 945,347

0.0018 (0.0079) 945,347

−0.0078 (0.0055) 735,574

−0.0028 (0.0058) 735,574

−0.0180*** (0.0056) 735,574

−0.0163*** (0.0049) 735,574

−0.0303 (0.0205) 735,574

−0.0060 (0.0222) 735,574

0.0026 (0.0073) 1,004,646

0.0049 (0.0063) 1,004,646

Females and males: −0.0099** DDD (0.0042) 1,424,322 N Females: DDD N Males: DDD N

Placebo analysis Placebo analysis

a

The data consists of individuals ages 19–29 and is from 2001 to 2010 American Community Survey. Estimates presented in columns 1–6 exclude data from 2008 to 2009. Columns 1 and 2 and 7 and 8 report the DDD estimates with eligibility status being determined only by age. Columns 3 and 4 report the DDD estimates with eligibility status being determined by all eligibility criteria. Columns 5 and 6 report the estimates from the interaction of the DDD estimator (eligibility determined by age) and the share of fully-insured workers in the state. Columns 7 and 8 report the DDD placebo estimates from randomly assigning the implementation year to the states that introduced a dependent mandate law. b Standard errors clustered on the state are presented in parentheses. c * 0.10, ** 0.05 and *** 0.01 denote significance levels.

Recession and not the state mandate laws. Specifically, the DDD estimation strategy is compromised if there exists contemporaneous shocks at the age-year-state level. This is an important concern because recent research by Hoynes et al. (2012) suggest that young adults are more negatively affected by recessions. However, because the age thresholds of the dependent mandate laws vary across states and because states passed laws at different times, it is likely that any bias is eliminated through the empirical strategy. To further test the robustness of the results, I re-estimate Eq. (1) and drop the years of the Great Recession.16 The National Bureau of Economic Research states that the Great Recession lasted from December of 2007 through June of 2009. Columns 1–6 of Table 7 report the DDD results for labor supply on the intensive margin with the Great Recession years excluded from the analysis. In general, the results are similar to the main results presented in Tables 3, 4 and 6. Although some of the point estimates that were statistically significant when the Great Recession years were not omitted become statistically insignificant in the new table, the same general pattern emerges: individuals decreased their labor supply on the intensive margin and the effect was larger and more robust for females than males. To further ensure that contemporaneous effects are not driving the results, I present a simple falsification test. I do this by randomly assigning the year that states implemented a dependent coverage mandate. Therefore, I replicate the empirical analysis which originally derived the results displayed in Table 3 with placebo implementation dates across states. Similarly, I use the same empirical specification that is described by Eq. (1). The results from the placebo regressions for full-time employment and log hours worked are reported in columns 7 and 8 of Table 7. There are three main points to highlight from the placebo results. First, none of the point estimates are statistically different from zero at standard testing levels. Second, the sign of the point estimates is not consistently negative across the different specifications. In fact, the three reported point estimates for the effect on hours worked are positive. Third, the point estimates on the

16

This is the same strategy that was applied in Dillender (2014).

intensive margin are relatively small when compared to the previous results reported in Table 3.

5.4. Further discussion of the results One concern is that these laws may cause other general equilibrium effects in the labor market. In particular, these laws may induce changes in the labor demand for young adults. On the margin, these laws may cause the employer’s cost of labor to decrease for young adults because an individual below the age threshold may not enroll for their own employer-based coverage if they have coverage through a parent. Therefore, all else equal, these conjectures suggest it is advantageous for firms that offer employer-based health insurance to employ young adults who are more likely to have an alternative source of health insurance coverage. Although I am unable to disentangle labor demand effects from labor supply effects, the two effects are working in opposite directions. Therefore, the suggested effects on labor demand would make it even more difficult to find an effect on labor supply. Another concern is that these laws may have induced changes in the labor force participation of parents. Given the small changes in dependent health insurance coverage, it is unlikely that changes in parental labor supply are playing a significant role. Using data from the Survey of Income and Program Participation (US Census Bureau Survey of income and program participation, 2012), I empirically investigated the potential labor supply effects of parents by comparing the labor supply outcomes of adults with children who would be affected by the law to both the labor supply outcomes of adults without children or adults with young children. I found no evidence that suggested that parents with children who were affected by the state dependent coverage laws adjusted their labor supply in response to the polices.17

17 This is consistent with Antwi et al. (2013) who similarly found no evidence that parents adjusted their labor supply in response to the federal dependent mandate.

134

B. Depew / Journal of Health Economics 39 (2015) 123–134

6. Conclusion Little is known about the role of health insurance on the labor supply decisions of young adults, despite the large amount of literature on other subpopulations in the U.S. This paper shows that health insurance plays a role in the labor supply decisions of young adults. The results suggest that expanding dependent health insurance coverage from state policies decreased the labor supply of eligible young adults on the intensive margin. The finding that young adults respond on the intensive margin and not on the extensive margin is consistent with the results of Antwi et al. (2013). This paper contributes to the literature on a number of dimensions. First, it uses much weaker identification assumptions than the previous paper on the topic (Antwi et al., 2013) that shows that health insurance effects the labor supply decisions of young adults. Understanding the interaction of labor supply and health insurance for the young adult population is particularly important because young adults have the highest rate of uninsurance in the U.S. and their decisions about labor supply have lasting effects both on their own wage trajectory and on the continuing development of the U.S. economy. Second, this paper builds on the previous work of Monheit et al. (2011), Levine et al. (2011), and Dillender (2014) by using a more complete time horizon of states that implemented policies to expand dependent coverage. Finally, the results in this paper are of particular importance because they highlight some of the potential outcomes that may occur from the recent expansion in dependent health insurance coverage under the ACA. References Abadie, A., 2005. Semiparametic difference-in-differences estimators. Rev. Econ. Stud. 72, 1–19. Agency for Healthcare Research and Quality, 2009. Medical Expenditure Panel Survey-Insurance Component Tables i.a.2.a and i.b.2.b. http://meps.ahrq.gov/ mepsweb/data stats/quick tables search.jsp?component=2&subcomponent=1 Anderson, M., Dobkin, C., Gross, T., 2012. The effect of health insurance coverage on the use of medical services. Am. Econ. J. Policy 4 (1), 1–27. Antwi, Y.A., Moriya, A.S., Simon, K., 2013. Effects of federal policy to insure young adults: evidence from the 2010 affordable care act dependent coverage mandate. Am. Econ. J. Econ. Policy 4 (June (4)), 1–28. Bertrand, M., Duflo, E., Mullainathan, S., 2004. How much should we trust differencein-difference estimates? Q. J. Econ. 119 (1), 249–275. Blau, D.M., 1994. Labor force dynamics of older men. Econometrica 62 (1), 117–156. Blau, D.M., Gilleskie, D.B., 2001. Retiree health insurance and the labor force behavior of older men in the 1990. Rev. Econ. Stat. 83 (1), 64–80. Blau, D.M., Gilleskie, D.B., 2008. The role of retiree health insurance in the employment behavior of older men. Int. Econ. Rev. 49 (2), 475–514. Buchmueller, T.C., Valletta, R.G., 1999. The effect of health insurance on married female labor supply. J. Hum. Resour. 34 (1), 42–70. Cameron, C.A., Gelbach, J.B., Miller, D.L., 2011. Robust inference with multiway clustering. J. Bus. Econ. Stat. 29 (2), 238–249. Cantor, J.C., Belloff, D., Monheit, A.C., DeLia, D., Koller, M., 2012a. Expanding dependent coverage for young adults: Lessons from state initiatives. J. Health Polit. Policy Law 37 (1), 99–128. Cantor, J.C., Monheit, A.C., Derek, D., Kristen, L., 2012b. Early impact of the affordable care act on health insurance coverage of young adults. Health Serv. Res. 47 (5), 1773–1790. Cardella, E., Depew, B., 2014. The effect of health insurance coverage on the reported health of young adults. Econ. Lett. 124, 406–410.

Cooper, P.F., Monheit, A.C., 1993. Does employment-related health insurance inhibit job mobility? Inquiry 30 (4), 400–416. Croson, R., Gneezy, U., 2009. Gender differences in preferences. J. Econ. Lit. 47 (2), 448–474. Briggs, D., James, B., 2014, December. Did the Affordable Care Act’s Dependent Coverage Mandate Increase Premiums?, Technical report, Working Paper. Dillender, M., 2014, July. Do more health insurance options lead to higher wages? Evidence from states extending dependent coverage. J. Health Econ. 36, 84–97. Gruber, Jonathan, 1994. The incidence of mandated maternity benefits. Am. Econ. Rev. 84 (3), 622–641. Ham, John C., Shore-Sheppard, Lara D., 2005. Did expanding medicaid affect welfare participation? Ind. Labor Relat. Rev. 58 (3), 452–470. Hoynes, H., Miller, D.L., Schaller, J., 2012. Who suffers during recessions? J. Econ. Perspect. 26 (Summer (3)), 27–48. H.R. 3970, 2009, October. Medical rights and reform act of 2009. In: 111th Congress, 1st Session. H.R. 4038, 2009, November. Common sense health care reform and affordability act. In: 111th Congress, 1st Session. Miriam, K., Steven, R., Trent Alexander, J., Sarah, F., Genadek, K., Schroeder, M.B., Trampe, B., Vick, R., 2001–2010. Integrated Public Use Microdata Series, Current Population Survey: Version 3.0. Levine, P.B., McKnight, R., Heep, S., 2011. How effective are public policies to increase health insurance coverage of young adults? Am. Econ. J. Econ. Policy 3 (1), 129–156. Madrian, Brigitte C., 1994. Employment-based health insurance and job mobility: is there evidence of job-lock? Q. J. Econ. 109 (1), 27–54. Meyer, B.D., Rosenbaum, D.T., 2001. Welfare, the earned income tax credit, and the labor supply of single mothers. Q. J. Econ. 116 (3), 1063–1114. Monheit, A.C., Cantor, J.C., DeLia, D., Belloff, D., 2011. How have state policies to expand dependent coverage affected the health insurance status of young adults. Health Serv. Res. 46 (1), 251–267. Mustard, C.A., Patricia, K., Kozyrskyj, Anita, Mayer, T., 1998. Sex differences in the use of health care services. N. Engl. J. Med. 338 (23), 1678–1683. National Conference of State Legislatures, 2010,September. Covering Young Adults Through Their Parents’ or Guardians’ Health Policy. http://www.ncsl.org/issuesresearch/health/dependent-health-coverage-state-implementation.aspx Nicholson, J.L., Collins, S.R., Mahato, B., Gould, E., Schoen, C., Rustgi, S.D., 2009, August. Rite of Passage? Why Young Adults Become Uninsured and How New Policies Can Help, 2009 Update. Technical Report 1310, The Commonwealth Fund. Olson, Craig A., 2002. Do workers accept lower wages in exchange for health benefits? Do workers accept lower wages in exchange for health benefits? J. Labor Econ. 20 (2), 91–114. 2014. Vincent Pohl Medicaid and the Labor Supply of Single Mothers: Implications for Health Care Reform. Working Paper. Royalty, A.B., Abraham, J.M., 2006. Health insurance and labor market outcomes: joint decision-making within households. J. Public Econ. 90 (8–9), 1561–1577. Steven, R., Trent Alexander, J., Genadek, K., Goeken, R., Schroeder, M.B., Sobek, M., 2001–2010. Integrated Public Use Microdata Series: Version 5.0 [MachineReadable Database]. Slusky, D., 2013. Revisiting the Effects of the Affordable Care Act. Technical report. Princeton University http://www.princeton.edu/slusky/SLUSKY Sommers, B.D., Kronick, R., 2012. The affordable care act and insurance coverage for young adults. J. Am. Med. Assoc. 307 (9), 913–914. Sommers, B.D., Buchmueller, T.C., Decker, S.L., Carey, C., Kronick, R., 2012. The affordable care act has led to significant gains in health insurance and acces to care for young adults. Health Aff. 32 (1), 165–174. Strumpf, E., 2010. Employer-sponsored health insurance for early retirees: impacts on retirement, health, and health care. J. Health Econ. 10 (2), 105–147. Strumpf, E., 2011. Medicaid’s effect on single women’s labor supply: evidence from the introduction of medicaid. J. Health Econ. 30 (3), 531–548. 2012. US Census Bureau Survey of Income and Program Participation. http://www.census.gov/sipp/ Wooldridge, J., Imbens, G., 2007. What’s New in Econometrics? Difference-inDifferences Estimation. http://www.nber.org/WNE/lect 10 diffindiffs.pdf Yelowitz, A.S., 1995. The medicaid notch, labor supply, and welfare participation: evidence from eligibility expansions. Q. J. Econ. 110 (4), 909–939.