The Value of Health Insurance and Labor Supply: Evidence from the Affordable Care Act Dependent Coverage Mandate and Early Medicaid Expansion

Transcription

1 The Value of Health Insurance and Labor Supply: Evidence from the Affordable Care Act Dependent Coverage Mandate and Early Medicaid Expansion Daeho Kim The Ohio State University September 2015 Abstract This paper examines how the differential value of health insurance affects labor supply. I exploit two sources of quasi-experimental change in health insurance provision: i) the Affordable Care Act (ACA) dependent coverage mandate; and ii) the ACA early Medicaid expansion in Connecticut, both of which were implemented in 2010 and yet induced differential values from the beneficiary s perspective. Using difference-in-differences and regression discontinuity designs, I find no evidence of the labor market impact of the ACA dependent coverage mandate despite its substantial impact on insurance coverage for young adults (i.e., employer-sponsored insurance (ESI) as a dependent child). In contrast, I find remarkable labor supply impacts of the ACA early Medicaid expansion in Connecticut. I show evidence that Connecticut s Medicaid expansion increased Medicaid coverage for low income childless adults by 6 percentage points, and as a result reduced employment rate by 4 to 6 percentage points among those low income childless adults. These differential labor supply responses are likely the result of differential valuations of the two health insurance provisions among beneficiaries. The aging out at 26 condition in eligibility and relatively high income level of eligible young adults under the ACA dependent coverage mandate led to low valuation of ESI as a dependent child and in turn no change in the labor supply of young adults, while the very low income level of the eligible population under the Medicaid expansion in Connecticut induced high value of Medicaid resulting in the reduced labor supply of low income childless adults. (JEL Codes: I13, I18, J22) I would like to thank Anna Aizer, Donald Andrews, David Blau, Audra Bowlus, Kenneth Chay, Christian Dustmann, Audrey Light, Byoung Hoon Seok, and Jesse Shapiro for valuable comments. Feng-An Yang provided excellent research assistance. I also thank Patricia Barnes at the Research Data Center (RDC), National Center for Health Statistics for help with the restricted National Health Interview Survey (NHIS) Data. The findings and conclusions in this paper are those of the author and do not necessarily represent the views of the RDC, the National Center for Health Statistics, or the Centers for Disease Control and Prevention. kim.4654@osu.edu.

2 1 Introduction Most non-elderly Americans obtain health insurance provided by their employers i.e., employer sponsored health insurance (ESI). 1 For example, in 2007, about 92% of the non-elderly under age 65 with private insurance coverage obtained their insurance through ESI (Cohen, Makuc, and Bilheimer, 2009). This implies that health insurance coverage is tightly linked to employment status. As a result, some workers may remain in their current jobs with health insurance mainly due to the fear of loss of insurance i.e., i) they would have moved to jobs that match better with their skills or job preferences even without health insurance provided at those jobs; or ii) they would have not worked at all, had there been alternative sources of health insurance untied to their employment status. 2 Therefore, the provision of health insurance untied to employment status is likely to affect labor supply, especially for those who maintain their employment in order to secure health insurance coverage, depending on their valuation of health insurance. This paper examines how the differential value of health insurance affects labor supply by exploiting two sources of quasi-experimental change in health insurance provision: i) the Affordable Care Act (ACA) dependent coverage mandate; and ii) the ACA early Medicaid expansion in Connecticut, both of which were implemented in 2010 and yet induced differential values from the beneficiary s perspective. First, the ACA dependent coverage mandate allows young adults to stay on their parents ESI as a dependent child until they turn age 26. The ACA dependent coverage mandate became effective on September 23, Prior to the ACA dependent coverage mandate, young adults lose their ESI as a dependent child typically when they turn age 19, unless they have their own-name ESI or other types of insurance (e.g., public insurance). Therefore, the ACA dependent coverage mandate is likely to affect the labor supply of young adults aged 19 to 25 as it provides them with health insurance untied to their employment status i.e., young adults can obtain health insurance through their parents ESI, instead of their own-name ESI. Although few, most of the empirical studies on the labor market impact of the ACA dependent coverage mandate have used the difference-in-differences (DD) approach, utilizing the fact that young adults aged 19 to 25 are eligible for the ACA dependent coverage mandate, while those younger than 19 or older than 25 are not (Antwi, Moriya, and Simon, 2013; Heim, Lurie, and Simon, 2014; Slusky, 2014). 3 The validity of the DD approach relies crucially on its key identification 1 ESI coverage includes ESI as an employee (i.e., own name ESI), ESI as a dependent child, and ESI as a spouse. 2 These phenomena have been considered as i) job-lock focusing on job mobility (Gruber and Madrian (1993); Madrian (1994)), and ii) employment-lock focusing on the decision to work at all (Garthwaite, Gross, and Notowidigdo (2014) among others) 3 Heim, Lurie, and Simon (2014) also use the difference-in-differences-in-differences (DDD) approach based on the fact that the ACA dependent coverage mandate affected only those young adults whose parents have private insurance. 1

3 assumption (i.e., the common trends assumption): the outcomes (e.g., employment rate) of the treatment group (e.g., aged 19 to 25) and the control group (e.g., aged 26 to 32) must be parallel over time in the absence of the treatment (i.e., prior to the ACA dependent coverage mandate). I provide evidence that this common trends assumption does not hold for labor market outcomes mainly because young adults aged 19 to 25 had been influenced disproportionately by the Great Recession prior to the ACA 4 relative to their older counterparts aged 26 to Once accounting for the pre-existing differential time trends between the treatment and control groups, I find no evidence of the labor supply impact of the ACA dependent coverage mandate, despite a substantial increase in insurance coverage (i.e., ESI as a dependent child) of 11 percentage points (or 45%). In a similar spirit, Heim, Lurie, and Simon (2014) and Slusky (2014) show no labor market effect of the ACA dependent coverage mandate once accounting for the differential time trends. 6 In contrast, Antwi, Moriya, and Simon (2013) find decreases in labor supply (full-time employment and hours of work) without accounting for differential time trends. 7 In order to further address differential pre-existing time trends, I apply a regression discontinuity (RD) design by exploiting the age discontinuities in the eligibility for the ACA dependent coverage mandate at the age of 25-and-3 rd -quarter (relative to ages of 26-and-0-quarter or older). Using data measured in age-in-quarters from both before and after the ACA dependent coverage mandate, I find no evidence of labor market effects i.e., no discontinuous changes in labor market outcomes (i.e., percent employed, percent working full time, or hours of work) at the age of 25- and-3 rd -quarter after the ACA dependent coverage mandate relative to the prior-aca mandate, despite a remarkable increase in ESI coverage as a dependent child of 7 percentage points at the age of 25-and-3 rd -quarter. Second, the ACA early Medicaid expansion in Connecticut (HUSKY D program) allows low income childless adults 8 aged 19 to 64 who were previously ineligible for Medicaid to be newly covered by Medicaid if their annual family income is at or below 56% of the Federal Poverty Level (FPL). The ACA early Medicaid expansion in Connecticut became effective on April 1, Prior to the Medicaid expansion in Connecticut, low income childless adults were not covered by Medicaid unless they were disabled or pregnant and thus most of them had to work to obtain 4 The start and end of the recession are December 2007 and June 2009, respectively (Business Cycle Dating Committee, NBER: 5 Hoynes, Miller, and Schaller (2012) show that the labor market outcomes of young adults were worse than their older counterparts during the Great Recession. I also find that during the Great Recession, the employment rate of young adults aged 19 to 25 decreased more than that of those aged 26 to 32 by 1.9 percentage points (Column (1) in Panel B of Table 3). 6 Heim, Lurie, and Simon (2014) use data from 2008 to 2010 during which the time trends of labor market outcomes (especially, on the extensive margin e.g., employment status) between the treatment and control groups are relatively parallel; and Slusky (2014) shows significant placebo effects i.e., differential time trends in labor market outcomes between the treatment and control groups even long before the ACA implementation. 7 They account for differential time trends for insurance coverage outcomes, but not for labor market outcomes. 8 Childless adults are those who do not have a child under age 19. 2

4 their own-name ESI as a source of health insurance if they wanted to have any health insurance at all. Therefore, Connecticut s Medicaid expansion is likely to affect the labor supply of low income childless adults as it provides them with Medicaid untied to their employment status. Using the DD approach i.e., comparing low income childless adults in Connecticut to those in other states in the Census Northeast region 9 after Connecticut s Medicaid expansion ( ) relative to before ( ), I find remarkable labor supply impacts of the Medicaid expansion among low income childless adults in Connecticut. I show that the common trends assumption in the context of Connecticut s Medicaid expansion is strongly supported. I then provide evidence that the Medicaid expansion in Connecticut increased Medicaid coverage for low income childless adults by 5.9 percentage points (or 30%), and as a result reduced the labor supply i.e., reductions in: i) percent employed by 3.8 to 5.9 percentage points (or 12 to 18%); 10 and ii) hours of work (per week) by 1.2 hours (or 16%). These findings are similar to those of recent studies on the labor supply impacts of Medicaid in Tennessee (Garthwaite, Gross, and Notowidigdo, 2014) and in Wisconsin (Dague, DeLeire, and Leininger, 2014). 11 However, in the study of the Oregon Health Insurance Experiment, Baicker et al. (2014) find no statistically significant impact of the Medicaid expansion on labor market outcomes, mainly due to the lack of crowd-out (i.e., no effect of Medicaid expansion on private insurance) in the Oregon experiment setting (Finkelstein et al., 2012). 12 The aforementioned differential labor supply responses i.e., no labor supply response to the ACA dependent coverage mandate among young adults aged 19 to 25, while a remarkable labor supply response to the Medicaid expansion in Connecticut among low-income childless adults are likely the result of differential valuations of the two health insurance provisions among beneficiaries. One noticeable condition in the eligibility of the ACA dependent coverage mandate is that young adults lose their eligibility (and in turn the value of ESI as a dependent child) when they turn age 26. This aging out at 26 condition in eligibility, along with the relatively high income level of eligible young adults 13, led to low valuation of insurance untied to the employment status (i.e., ESI as a dependent child) resulting in no change in labor supply of young adults. In contrast, the very 9 Other Northeastern states are Maine (ME), Massachusetts(MA), New Hampshire (NH), Jew Jersey (NJ), New York (NY), Pennsylvania (PA), Rhode Island (RI), and Vermont (VT). 10 The overall reduction of 5.9 percentage points consists of a 3.8 percentage point decrease in percent employed with ESI and a 2.1 percentage point decrease in percent employed without ESI. Also, most of the decrease in percent employed is driven by part-time workers (i.e, working less than 35 hours (per week)) by 4.8 percentage points. 11 Garthwaite, Gross, and Notowidigdo (2014) find the contraction of Medicaid in Tennessee increased the labor supply of childless adults (implied labor supply elasticity of 0.63), and Dague, DeLeire, and Leininger (2014) find the expansion of Medicaid in Wisconsin reduced the labor supply of childless adults by 2 to 24%. 12 One of the eligibility criteria requires the lottery selected individuals to have been uninsured for six months, suggesting that workers with ESI who would have most likely responded to the Medicaid expansion in their labor supply were not eligible for the Oregon Medicaid expansion program. 13 During , about 76% of eligible young adults have incomes higher than 200% of the FPL in contrast with only 36% of ineligible young adults (calculated from the NHIS). 3

5 low income level of eligible population (income at or below 56% of the FPL) under the Medicaid expansion in Connecticut induced high value of Medicaid and as a result reduced the labor supply of low income childless adults. The paper is organized as follows. Section 2 provides a brief background on the ACA dependent coverage mandate and the ACA early Medicaid expansion in Connecticut. Section 3 describes the data. Section 4 outlines an empirical strategy. Section 5 presents the empirical results, and Section 6 concludes. 2 Background on the ACA Dependent Coverage Mandate and the ACA Early Medicaid Expansion in Connecticut 2.1 The Affordable Care Act Dependent Coverage Mandate The Patient Protection and Affordable Care Act (the Affordable Care Act) of 2010 (P.L , P.L ), enacted on March 23, 2010, allows young adults to stay on their parents employersponsored insurance (ESI) as a dependent child until their 26 th birthday 14 from the plan year beginning January 1, Accordingly, I define young adults in this paper as dependent children who have not attained age 26 e.g., those aged 25 or younger (or those aged 25-and-3 rd quarters or younger when using data measured in age in quarters). Exceptions to the ACA dependent coverage mandate are: the spouse of an eligible child; a child of a child receiving dependent coverage; and a grandfathered plan. 15 The ACA dependent coverage mandate became effective on September 23, While many insurance companies agreed to start providing the dependent coverage earlier than September 23, 2010, an actual increase in ESI as a dependent child appeared to occur starting from 2011 as most new plan years of health insurance start on January 1st. Indeed, Panel A of Appendix Figure A1 shows that the ESI coverage as a dependent child increased from the first half of Prior to the ACA dependent coverage mandate, young adults lose their ESI as a dependent child typically when they turn age 19 (or age 23 if they are enrolled in college), unless they have their own-name ESI or other types of health insurance (e.g., Medicaid). Hence, the insurance coverage of young adults aged 19 to 25 is expected to increase after the implementation of the ACA dependent coverage mandate (e.g., ) relative to before (e.g., ) more than that 14 For example, if a child turns 26 on July 17, 2011, then the last day of coverage is July 16, 2011 (Federal Register, 2010). 15 A grandfathered plan is a plan or health insurance coverage existing as of March 23, 2010, which is not required to offer dependent coverage to a child under 26 who is otherwise eligible for ESI other than a group health plan of a parent for plan years beginning before January 1st, (Federal Register, 2010). 16 In the main analysis, I consider years up to 2010 as the pre ACA period, although the results do not change even if the period up to 2009 is considered the pre ACA period. 4

6 of their older counterparts (e.g., those aged 26 to 32) resulting from the increase in ESI coverage as a dependent child. 17 Therefore, the ACA dependent coverage mandate is likely to affect the labor supply of young adults aged 19 to 25 as it provides them with health insurance untied to their employment status i.e., young adults can obtain health insurance through their parents ESI instead of their own-name ESI. 18 It is noteworthy that prior to the ACA dependent coverage mandate, some states implemented their own dependent coverage mandates (Levine, McKnight, and Heep, 2011; Monheit et al., 2011; Cantor et al., 2012; Dillender, 2014; Depew, 2015). Before 2010 (prior to the ACA dependent coverage mandate), 29 states implemented their own dependent coverage mandates. 19 Although eligibility (e.g., age limit, full-time student status, etc.) varies across states, most states have the aging out condition at the age of 26. Given the possibility of heterogeneous effects, if at all, of the ACA dependent coverage mandate on labor supply, I further examine the labor market impact of the ACA dependent coverage mandate separately for states with- and without their own mandates prior to the ACA The ACA Early Medicaid Expansion in Connecticut Under provisions of the ACA, each state has the option to expand Medicaid eligibility to those adults with incomes up to 138% of the Federal Poverty Level (FPL). There is no deadline for each state to implement the Medicaid expansion. As of April 1, 2014, 27 states including the District of Columbia decided to implement the ACA Medicaid expansion starting from January 1, 2014 (Kaiser Family Foundation, 2014). It is worth noting that some states had expanded Medicaid before For example, six states have implemented the Medicaid expansion since the enactment of the ACA i.e., Connecticut and the District of Columbia in 2010; California, Minnesota, New Jersey, and Washington in 2011 (Sommers et al., 2013; Sommers, Kenney, and Epstein, 2014). 21 Connecticut is the first state in the nation to expand Medicaid coverage to low income childless adults under the ACA of The Medicaid expansion in Connecticut under the ACA was approved on June 21, 2010 and effective retroactively to April 1, 2010 (State of Connecticut De- 17 This comparison is a basis for the difference-in-differences approach widely used in most empirical studies to examine the effects of the ACA dependent coverage mandate. 18 Indeed, I exploit the fact that within the eligible age group (young adults aged 19 to 25), those young adults whose parents have employer benefits (i.e., own-name ESI) are most likely affected by the ACA dependent coverage mandate when applying the difference-in-differences-in-differences (DDD) approach in Section Those states are CO, CT, DE, GA, ID, IL, IN, IA, KY, ME, MD, MA, MN, MO, MT, NH, NJ, NM, NY, ND, OR, PA, RI, SD, TX, UT, VA, WA, and WV. 20 Recent studies find the labor market effects of the states own dependent coverage mandates (prior to the ACA): i) a decrease in labor supply of young adults on the intensive margin (Depew, 2015); and ii) an increase in wages of young adults (Dillender, 2014). 21 Note that Wisconsin also expanded Medicaid in 2009 (prior to the enactment of the ACA) to childless adults with incomes up to 200% of the FPL (Dague, DeLeire, and Leininger, 2014). 5

7 partment of Social Services, 2012). 22 In order to qualify for the Medicaid expansion in Connecticut (HUSKY D program) 23, an individual aged 19 through 64 must i) be a resident of Connecticut; ii) be a U.S. citizen or qualified immigrant 24 ; iii) not receive federal Supplemental Security Income (SSI) or Medicare; iv) not have a child under age 19; and v) meet the income limit i.e., annual income at or below 56% of the FPL. 25 For example, 56% of the FPL for a single person in 2010 was $6,064. Medicaid enrollees under this expansion receive the standard Medicaid benefit package for adults. 26 Prior to the ACA Medicaid expansion in Connecticut, low income childless adults were not covered by Medicaid unless they were disabled or pregnant and thus most of them had to work to obtain their own-name ESI as a source of health insurance if they wanted to have any insurance at all. Therefore, Connecticut s Medicaid expansion would likely affect the labor supply of low income childless adults, as it provides them with Medicaid coverage untied to their employment status. 3 Data To examine the labor supply impact of the ACA dependent coverage mandate, I use the National Health Interview Survey (NHIS) and the March Supplement to the Current Population Survey (March CPS). 27 To study the effect of the Medicaid expansion on the labor supply of low income childless adults in Connecticut, I use the American Community Survey (ACS). 3.1 The National Health Interview Survey (NHIS) I use the National Health Interview Survey (NHIS) from 2004 to 2013 to estimate the impact of the ACA dependent coverage mandate on the labor supply of young adults aged 19 to 25. The NHIS provides detailed information on health insurance coverage by payers (Private, Medicaid, Medicare, and other public), by sources of private insurance coverage (e.g., through employer or directly purchased), and by policyholders (e.g., survey respondents themselves, someone else in the 22 To be conservative, I consider years up to 2010 as the pre expansion period. 23 See State of Connecticut Department of Social Services (2013) and Connecticut Voices for Children (2014) for detailed information. 24 Ineligible immigrant groups include, but are not limited to: certain legal immigrant adults in the US for fewer than five years, undocumented immigrants, and immigrants with temporary status, such as students, temporary workers, and tourists. 25 An additional 12% income disregard is allowed for shelter costs in Region A (Fairfield County). In this paper, I focus on the income limit of 56% of the FPL (i.e., those not in Region A). Indeed, the American Community Survey (ACS) the main data set used in this paper to examine the impact of Connecticut s Medicaid expansion does not include Region A of Connecticut. 26 This includes inpatient and outpatient hospital services; physician services; laboratory services; prescription drugs; mental health services; immunizations; and emergency services. 27 Additionally, I use the Medical Expenditure Panel Survey (MEPS) to estimate the effect of the ACA dependent coverage mandate on job mobility as it provides longitudinal information on job mobility for the same individual. 6

8 family, or a person not in household). It also contains information on family interrelationships within a household. Using this information, I construct a key insurance coverage variable i.e., ESI coverage as a dependent child. For example, if a survey respondent has private insurance, the insurance is obtained through an employer but the policy is under someone else s name, and a spouse does not have own-name ESI, then I consider the respondent as being covered by ESI as a dependent child. In addition, the NHIS provides information on labor market activities for respondents 18 and older. For example, employment status last week (working for pay at a job or business, with a job or business but not at work, or working, but not for pay, at a family-owned job or business) and hours of work last week. Furthermore, to examine the potential heterogeneous effects, if any, of the ACA dependent coverage mandate on labor supply across states depending on their own mandate status prior to the ACA, I estimate the labor supply effects separately for states with- and without their own mandates prior to the ACA dependent coverage mandate. To identify states with- and without their own pre ACA mandates, the restricted variables such as state identifiers are required. 28 These confidential data were accessed through the Research Data Center at the National Center for Health Statistics. 3.2 The Current Population Survey (CPS) In order to ensure that the findings on the labor market impact of the ACA dependent coverage mandate are not driven by specific data (NHIS), I redo the main analyses using the Annual Social and Economic Supplement of the CPS (March CPS). Also, the March CPS is used to estimate the effect of the ACA dependent coverage mandate on wages 29 as it provides detailed information on earnings and a poverty measure, while the NHIS contains only broad categorical values of annual earnings. 3.3 The American Community Survey (ACS) To estimate the labor market impact of the Medicaid expansion in Connecticut among low income childless adults, I use the American Community Survey (ACS), which provides enough data for studying nine states i.e., Connecticut (treatment group) and another eight states in the Census Northeast region (control group). 30 Specifically, I use an augmented version of the ACA, IPUMS (Integrated Public Use Microdata Series), prepared by the Minnesota Population Center. (Ruggles et al., 2015) 28 State identifiers are also used to adjust the standard errors in empirical analyses (difference-in-differences and regression discontinuity) i.e., clustering standard errors at the state level over time. 29 I calculate hourly wages based on annual earnings divided by annual hours of work (hours of work (per week) annual weeks worked). 30 Other states in the Census Northeast region are Maine (ME), Massachusetts (MA), New Hampshire (NH), Jew Jersey (NJ), New York (NY), Pennsylvania (PA), Rhode Island (RI), and Vermont (VT). 7

9 The ACS-IPUMS provides, among others, information on family interrelationships within a household and an individual s annual income as well. Based on this information, I calculate family income as % of the Federal Poverty Level (FPL), a key variable for identifying the eligible population for the Medicaid expansion in Connecticut i.e., childless adults with incomes up to 56% of the FPL. Furthermore, in order to define the family unit in determining the income eligibility for the Medicaid expansion, I use a general definition of the health insurance unit (HIU) proposed by the State Health Access Data Assistance Center (SAHDAC). 31 To identify low income childless adults eligible for the Medicaid expansion in Connecticut, I restrict the sample to those aged who: i) do not have a child under age 19; ii) do not receive federal Supplemental Security Income (SSI) or Medicare; iii) are U.S. citizens or non-citizens who have been living in U.S. for 5 or more years 32 ; and iv) have annual family incomes at or below 56% of the Federal Poverty Level (FPL). 4 Empirical Strategy 4.1 The Effect of the ACA Dependent Coverage Mandate Difference-in-Differences In order to identify the effects of the ACA dependent coverage mandate, I first use a difference-indifferences (DD) approach by estimating the following regression: Y iast = α + γ a + δ t + θ Y oung a P ost t + X iastβ + λ s + U iast (1) where Y iast is the outcome variable for individual i of age a in state s in year t; γ a are age fixed effects; δ t are year fixed effects; Y oung a is an indicator variable for those young adults aged 19 to 25; P ost t is an indicator for the time period after the ACA dependent coverage mandate (e.g., ); X iast is a vector of observed characteristics including gender, race, ethnicity, marital status, health status, activity-limitation indicator, and family income; λ s are state fixed effects; and U iast is an unobserved term. The parameter of interest is θ, which measures the difference in the outcome variable between those aged 19 to 25 (treatment group) and those aged 26 to 32 (control group) after the implementation of the ACA dependent coverage mandate relative to before. as follows: In order to indirectly test the common trends assumption, I first estimate a placebo effect Y iast = α + γ a + δ t + φ Y oung a P re t + X iastβ + λ s + U iast (2) where P re t is an indicator for the time period before the ACA dependent coverage mandate (e.g., 31 See State Health Access Data Assistance Center (2012) for details. 32 I include those non-citizens as a proxy measure for the qualified immigrants under the Medicaid expansion in Connecticut. Indeed, the empirical results do not change even without using this measure. 8

10 ). If, for example, φ 0 when using data from 2004 to 2009, it implies that the treatment and control groups have differential time trends of the outcome variable even before the ACA dependent coverage mandate was implemented. Second, I test the common trends of outcome variables between the treatment and control groups by estimating the following regression: Y iast = α + γ a + δ t + ψ Y oung a T rend t + X iastβ + λ s + U iast (3) where T rend t is a linear or quadratic (or higher order) time trend. Again, for instance, if ψ 0 when using data up to 2009, it indicates differential outcome trends between the treatment and control groups prior to the ACA dependent coverage mandate. Finally, I include a time trend (T rend t ) and its interaction with a young adult indicator (Y oung a T rend t ) in equation (1) to account for differential time trends of the outcome variables. I also supplement the DD approach using the difference-in-differences-in-differences (DDD) method exploiting the fact that within the eligible age group (i.e., young adults aged 19 to 25), those young adults whose parents have employer benefits (i.e., own-name ESI) are mostly affected by the ACA dependent coverage mandate. I estimate the following DDD regression: Y iagst = α + γ a + η g + δ t + µ ag + ν at + ξ gt (4) + θ Y oung a P arentesi g P ost t + X iagstβ + λ s + U iagst where Y iagst is the outcome variable for individual i of age a in group g (i.e., those whose parents have own-name ESI) in state s in year t; γ a are age fixed effects; η g are parental ESI fixed effects; δ t are year fixed effects; µ ag are age parental ESI fixed effects; ν at are age-year fixed effects; ξ gt are parental ESI time fixed effects; P arentesi g is an indicator variable for those whose parents have own-name ESI; and λ s are state fixed effects Regression Discontinuity To further address differential pre-existing time trends, 34 I use a regression discontinuity (RD) design exploiting the age discontinuities in the eligibility for the ACA dependent coverage mandate i.e., aging-in at age 19 and aging-out at age 26. Given that the NHIS provides information on labor market outcomes only for those aged 18 and older and most 18-year-olds are still in school, I examine the discontinuity at age 25 i.e., those aged 25 are treated, while those aged 26 and older are untreated. 35 Specifically, using variables measured in age-in-quarters, I estimate the following 33 I also include state by parental ESI dummies to account for state-specific economic conditions affecting parents employment status and in turn parental ESI status; and state by year dummies for state-specific year effects. 34 I present empirical evidence of a violation of the common trend assumption of the DD approach in Section This is for ease of interpretation. For example, an increase in the outcome variable (e.g., ESI coverage as a dependent child) for the treatment group (e.g., 25-year-olds) yields a positive RD estimate when using age 25 as a 9

11 RD regression: Y it = α 0 + f (Age it c) + α 1 1 {Age it c} + f (Age it c) 1 {Age it c} (5) + δ P ost t + f (Age it c) P ost t + α 2 1 {Age it c} P ost t + θ f (Age it c) 1 {Age it c} P ost t + U it where c 25-and-3 rd -quarters, f( ) is an underlying function between the outcome and age-inquarters (e.g., second-order polynomials) which is continuous at the age of 25-and-3 rd -quarter; 1{ } is an indicator function; P ost t is an indicator for the time period after the ACA dependent coverage mandate (e.g., ). In addition, I estimate a nonparametric (local linear) regression weighted by a triangular kernel which has optimality properties at boundary points (Cheng, Fan, and Marron, 1997). 4.2 The effect of the ACA Medicaid expansion in Connecticut Difference-in-Differences To identify the effect of the ACA early Medicaid expansion in Connecticut, I estimate the following DD regression similar to (1): Y ist = α + λ s + δ t + θ CT s P ost t + X istβ + U ist (6) where Y ist is the outcome variable for individual i in state s in year t; λ s are state fixed effects; δ t are year fixed effects; CT s is an indicator variable for Connecticut; and P ost t is an indicator for the time period after the Medicaid expansion in Connecticut (e.g., ). The parameter of interest is θ which captures the difference in outcome variables (e.g., employment status) between low income childless adults in Connecticut (treatment group) and those in other Northeastern states (control group) after Connecticut s Medicaid expansion (i.e., ) relative to before the expansion (i.e., ). In order to indirectly test the common trends in outcome variables between the treatment and control groups, I include a time trend term interacted with an indicator variable for Connecticut (CT s T rend t ). If the two estimates of θ in (6) with- and without this interaction term are similar to each other, it suggests that the common trends assumption likely holds. Given that the Medicaid expansion (i.e., treatment) varies at the state level, I use state year cell level data. 36 In order to adjust for covariates that might be different across states over time, I regress the outcome variables on a vector of observed characteristics (X ist ) yielding residuals Ỹist; cutoff, whereas a negative RD estimate when using age 26 as a cutoff. 36 I present the DD estimates using both state year cell level data and individual level data in Section 5. 10

12 and then use state year cell means of these residuals Y st = 1 N st i {s,t} Ỹist as an outcome variable in the following DD regression: Y st = α + λ s + δ t + θ CT s P ost t + U st (7) where all other variables (other than Y st ) are the same as in (6). For statistical inference of θ, I use cluster-robust standard errors i.e., clustering standard errors at the state level over time to allow for unrestricted serial correlation within a state over time. Notwithstanding, another concern may arise due to the small number of clusters (i.e., nine states in this paper) with which asymptotic properties of cluster robust standard errors may not work well. Therefore, I use a wild cluster bootstrap t procedure which may lead to improved inference when using a small number of clusters (Cameron, Gelbach, and Miller, 2008). 5 Empirical Results 5.1 Summary Statistics Panel A of Table 1 shows summary statistics for young adults aged 19 to 25 (treatment group) and their older counterparts aged 26 to 32 (control group) before and after the ACA dependent coverage mandate. The treatment and control groups are quite different on observed characteristics, especially age 37 and educational attainment, which are major determinants of labor market outcomes. This implies that they might have differential trends in their labor market outcomes even before the ACA dependent mandate, which calls the common trends assumption into question. Prior to the ACA, young adults aged 19 to 25 had higher ESI coverage as a dependent child than those aged 26 to 32. This gap further increased resulting from the increase in ESI coverage as a dependent child among the treatment group whereas little change among the control group after the ACA dependent coverage mandate. However, employment rates virtually did not change and thus the pre ACA gap in employment rates remained unchanged even after the ACA dependent coverage mandate. Panel B of Table 1 presents descriptive statistics for low income childless adults in Connecticut (treatment group) and other Northeastern states (control group). The treatment and control groups are similar based on baseline characteristics other than race. Prior to the Medicaid expansion in Connecticut, the treatment group was covered by Medicaid less than the control group, while they worked more than the control group. After the expansion, however, the Medicaid coverage of the treatment group increased by 7.7 percentage points and their employment rate decreased by 6.2 percentage points. As a result, the gaps in Medicaid coverage and employment rate (hours 37 By virtue of the age eligibility of the ACA dependent coverage mandate. 11

13 of work as well) disappeared and in turn the Medicaid coverage and employment rates between Connecticut and other Northeastern states became almost the same after the Medicaid expansion. 5.2 Effects of the ACA Dependent Coverage Mandate The Effect on Health Insurance Coverage Panel A of Figure 1 plots ESI coverage as a dependent child for young adults aged 19 to 25 (solid line) and those aged 26 to 32 (dotted line) over time. It is evident that the two lines are very much parallel up to 2010, suggesting that the common trends assumption for the ESI coverage as a dependent child appears to hold strongly. Table 2 confirms this as will be explained below. Noticeably, Panel A of Figure 1 shows a dramatic increase in ESI coverage as a dependent child only for the treatment group by about 8 to 10 percentage points after the ACA dependent coverage mandate. Table 2 shows the DD estimates of the effect of the ACA dependent coverage mandate on insurance coverage. Panel A shows the estimates without accounting for differential time trends between the treatment and control groups. In Panel B, in order to test the common trends assumption, I estimate a placebo effect (φ in equation (2)) i.e., the difference in ESI coverage as a dependent child between the treatment and control groups during the pre-aca period (i.e., ) relative to an earlier pre-aca period (i.e., ). The result shows no placebo effect prior to the ACA dependent coverage mandate. Note that there were little decreases, if any, in ESI coverage as a dependent child during , consistent with Panel A of Figure 1. This is because young adults parents were likely to lose their jobs during the Great Recession, resulting in the decrease in young adults ESI coverage as a dependent child. In Panel C, I then test the common trends assumption by estimating the differential time trends (ψ in equation (3)). Again, the result suggests evidence of the common trends between the treatment and control groups. Finally, Panel D shows the DD estimates accounting for differential time trends by including a time trend (T rend t ) and its interaction with a young adult indicator (Y oung a T rend t ) in equation (1). The estimates with differential time trends are very similar to those without differential time trends (in Panel A), which strongly suggest that the common trends assumption likely holds. Overall, Table 2 shows that the ACA dependent coverage mandate increased i) ESI coverage as a dependent child by 10.8 percentage points (45% of the pre ACA rate of 23.9%) 38 ; ii) the percent with private insurance by 7.1 percentage points; and iii) the percent with any insurance by 7.0 percentage points. 39 Despite this substantial increase in ESI coverage as a dependent child, 38 This is equivalent to 3.1 million young adults newly covered by ESI as a dependent child over the three year period ( ). 39 Note that there is no crowd-out effect i.e., the increase in private insurance (ESI as a dependent child) does not affect public insurance coverage, leading to the increase in any insurance coverage by the same in magnitude (7 12

14 however, there is no discernible decrease in own-name ESI as shown in Column (2) of Panel D. This already implies little impact on labor supply, which will be explained in detail below The Effect on Labor Supply Panels B through D of Figure 1 show labor market outcomes (i.e., percent employed, percent working full time, and hours of work (per week), respectively) for the treatment and control groups over time. 40 The figures provide graphical evidence of a likely violation of the common trends assumption for labor market outcomes i.e., although the trends of the outcome variables seem to be parallel up to 2007, the outcomes show differential trends from 2008 on mainly because the treatment and control groups have been differentially affected by the Great Recession. 41 Indeed, Hoynes, Miller, and Schaller (2012) show that the labor market outcomes of young adults were worse than their older counterparts during the Great Recession. Table 3 provides evidence of a violation of the common trends assumption. Panel A shows the estimates without accounting for differential time trends. At first glance, there seems to be an effect of the ACA dependent coverage mandate on labor supply (e.g., employment rate, percent working full-time, and hours of work). 42 However, Panel B shows that even the placebo effects are as big as the estimates without accounting for differential time trends. For example, as shown in Column (1) of Panel B, the employment rate of young adults aged 19 to 25 decreased by 1.9 percentage points even before the ACA dependent coverage mandate (i.e., ) 43, which is larger in magnitude than the estimate in Column (1) of Panel A. Also, Panel C shows that the treatment and control groups have significantly differential time trends in their labor market outcomes prior to the ACA dependent mandate. Once accounting for the differential time trends, as shown in Panel D of Table 3, I find no evidence of the labor supply impact of the ACA dependent coverage mandate. In a similar spirit, Heim, Lurie, and Simon (2014) and Slusky (2014) show no effect on labor supply, once accounting percentage points). The increase in private insurance is less than that of ESI coverage as a dependent child because of a decrease in non-group private insurance. 40 Additionally, Appendix Figures A2 and A3 show the job mobility and hourly wage, respectively. 41 Panel B of Appendix Figure A1 shows that the employment rate of young adults aged 19 to 25 started decreasing more than that of those aged 26 to 32 from the first half of 2008, right after the Great Recession i.e., the start and end of the Great Recession are December 2007 and June 2009, respectively (Business Cycle Dating Committee, NBER: 42 I also examine job mobility and wages: i) job mobility because young adults might move (even while employed) to new jobs that do not offer ESI but do match better with their skills or job-preferences as they can obtain their health insurance through their parents ESI (instead of their own-name ESI); and ii) wages because young adults wages would increase as long as they do not need to keep their own-name ESI, resulting from the compensating wage differential. Appendix Figures A2 and A3, and Columns (4) and (5) of Table 3 show no impacts on job mobility nor wages. 43 This is because young adults aged 19 to 25 had been influenced disproportionately by the Great Recession prior to the ACA (i.e., ) relative to their older counterparts aged 26 to

15 for the differential time trends. 44 In contrast, Antwi, Moriya, and Simon (2013) find decreases in labor supply without accounting for differential time trends. 45 In order to assure that the finding of no labor supply effect of the ACA dependent coverage mandate is not driven by a specific data set (NHIS), I redo the analysis using the March CPS. Appendix Table A1 shows the DD estimates (from exactly the same specification as used for the estimates in Table 2) based on the data from the March CPS. 46 The results reaffirm the finding of no effect of the ACA dependent coverage mandate on the labor supply of young adults. I further examine the labor market effect of the ACA dependent coverage mandate separately for states with- and without their own dependent coverage mandates prior to the ACA 47, given the possibility that different (or even opposite) labor market responses between states with- and without their own dependent coverage mandates could result in no effect overall. Panel A of Figure 2 plots ESI coverage as a dependent child in states with their own pre-aca mandates (lines with triangles) and without their own pre-aca mandates (lines with circles). It shows that the changes in ESI coverage as a dependent child after the ACA ( ) relative to before ( ) are almost the same in magnitude between states with- and without their own pre-aca mandates. 48 The DD estimates (accounting for differential time trends) shown in Column (1) of Table 4 confirm that the increases in ESI coverage as a dependent child are the same in states with- and without their own mandates (10.6 and 10.9 percentage points, respectively). Despite these substantial increases in ESI coverage as a dependent child, however, Panel B of Figure 2 and Columns (2) through (4) of Table 4 show no labor market effect in both states with- and without their own pre-aca mandates. The same analysis using the March CPS, as shown in Appendix Figure A4 and Appendix Table A2, confirms the results, namely that there is no labor market impact of the ACA dependent coverage mandate regardless of a state s own pre-aca mandate status. To address the aforementioned differential time trends between the (age based) treatment and control groups, I exploit another eligibility criterion i.e., within the age qualified treatment group (i.e., aged 19 to 25), only those young adults whose parents have private insurance are eligible for the ACA dependent coverage mandate. Hence, young adults whose parents have own-name ESI 44 Heim, Lurie, and Simon (2014) use data from 2008 to 2010 during which the time trends of labor market outcomes between the treatment and control groups are relatively parallel; and Slusky (2014) shows significant placebo effects, suggesting the differential time trends in labor market outcomes between the treatment and control groups before the ACA implementation. 45 They do account for differential time trends for insurance coverage outcomes, but not for labor market outcomes. 46 For the placebo effects and common trend tests, I consider the time period up to 2010 (instead of up to 2009 as in the analysis using the NHIS) as the pre-aca period because the 2010 March CPS were surveyed before (or at) the enactment of the ACA dependent coverage mandate. 47 Prior to the ACA, 29 states implemented their own dependent coverage mandates (Levine, McKnight, and Heep, 2011; Monheit et al., 2011; Cantor et al., 2012; Dillender, 2014; Depew, 2015). 48 Note that in terms of level, states with their own mandates had higher ESI coverage than states without their own mandates by about 4 percentage points before the ACA. Indeed, this is consistent with the finding of Levine, McKnight, and Heep (2011). 14

16 are mostly affected by the ACA dependent coverage mandate. 49 Panel A of Figure 3 shows the ESI coverage as a dependent child for young adults whose parents have own name ESI (treatment group: solid line) and those whose parents do not have own name ESI (control group: dotted line) over time. 50 The two lines are parallel up to 2010, and then only for the treatment group after 2010 ESI coverage as a dependent child increased by 17 to 20 percentage points. Column (1) in Panel A of Table 5 presents the corresponding DD estimate of an increase in ESI coverage as a dependent child by 17 percentage points. 51 Panels B through D of Figure 3 show the labor market outcomes of the two groups, providing graphical evidence of a violation of the common trends assumption. For instance, Panel B shows that the employment rates of the two groups are extremely parallel up to 2007, but diverge from 2008 on, likely due to the disproportionate influence of the Great Recession. 52 Once these differential time trends are accounted for, the DD estimates in Columns (2) through (4) in Panel A of Table 5 show no evidence of labor market impact of the ACA dependent coverage mandate. In addition, Panel B of Table 5 presents the DDD estimates confirming the same results i.e., no effect on labor supply. It is noteworthy that eligible young adults have relatively high family incomes even prior to the ACA dependent coverage mandate. Using the NHIS, I find 76% of eligible young adults whose parents have own-name ESI have incomes higher than 200% of the FPL in contrast to 36% of those whose parents do not have own-name ESI. This relatively high income level (along with high education level) of eligible young adults likely led to the low valuation of ESI as a dependent child (i.e., insurance untied to the employment status) resulting in no change in their labor supply. In order to further address the differential time trends between the treatment and control groups (i.e., essentially between age groups: aged 19 to 25 versus aged 26 to 32), I apply a regression discontinuity (RD) design exploiting the age discontinuities in eligibility for the ACA dependent coverage mandate i.e., discontinuities at age 19-and-0 quarters and at age 25-and-3 rd quarters using outcome variables measured in age in quarters from the NHIS. I focus on the discontinuity at age 25-and-3 rd quarters (instead of age 19-and-0 quarters) because i) the NHIS provides information on labor market activities only for those aged 18 and older; and ii) there are few individuals aged 18-and-3 rd quarters (control group) and aged 19-and-0 quarters (treatment group) in the labor 49 Heim, Lurie, and Simon (2014) also consider young adults whose parents do not have employer benefits as an additional control group in their DDD approach using data from U.S. tax records. 50 Those young adults whose parents are not in household are excluded. 51 The DD estimate is adjusted for differential time trends between young adults whose parents have own-name ESI (treatment group) and those whose parents do not have own-name ESI (control group). 52 The employment rate of young adults whose parents do not have own-name ESI (the control group) decreased more than that of young adults whose parents have own-name ESI (the treatment group) partly because the control group was less educated than the treatment group i.e., 22% of the control group had less than a high school education in contrast to only 7% of the treatment group (calculated from the NHIS). Indeed, Hoynes, Miller, and Schaller (2012) show that those with lower education levels experienced decreases in employment more than those with higher education levels during the Great Recession. 15

17 force. Panel A of Figure 4 plots the changes in ESI coverage as a dependent child by age in quarters after the ACA dependent coverage mandate (i.e., ) relative to before (i.e., ). It shows the discontinuous increase in percent with ESI as a dependent child at the age of 25-and- 3 rd -quarters (relative to ages 26-and-0-quarters and older) by approximately 8 percentage points. Column (1) of Table 6 presents the RD estimates. Both parametric and nonparametric (local linear regression) 53 estimates show a significantly large increase in ESI coverage as a dependent child by 7 to 11 percentage points. If this increase in ESI coverage affects the labor supply of young adults, one may also see discontinuous changes in labor market outcomes at the age of 25-and-3 rd -quarters. However, Panels B through D of Figure 4 show virtually no discontinuous changes in any of the labor market outcomes (i.e., percent employed, percent working full time, or hours of work per week) at age 25-and-3 rd -quarters. Columns (2) to (4) of Table 6 show the corresponding RD estimates. Consistent with graphical evidence, the RD estimates (both parametric and nonparametric) show no effect of the ACA dependent coverage mandate on labor supply. Note that the nonparametric estimates are robust to bandwidth choices as Figure 5 shows the estimates are zero across different bandwidths. In sum, the empirical results (from DD and RD analyses) show no evidence of the labor supply impact of the ACA dependent coverage mandate. This is mainly because of young adults low valuation of ESI as a dependent child (insurance untied to their employment status) resulting in no change in their labor supply. First, the aging out at 26 condition in eligibility for the ACA dependent coverage mandate (i.e., young adults lose their eligibility and in turn the value of ESI as a dependent child when they turn age 26) led to the low valuation of ESI as a dependent child. Second, the relatively high income level (as well as high education level) of eligible young adults also likely induced the low valuation of ESI as a dependent child. Now, I turn to the effects of the ACA early Medicaid expansion in Connecticut explained below. 5.3 Effects of the ACA Early Medicaid Expansion in Connecticut Effects on Medicaid Coverage Panel A of Figure 6 shows Medicaid coverage of low income childless adults with family incomes at or below 56% of the Federal Poverty Level (FPL) in Connecticut (solid line) and another eight states in the Census Northeast region 54 (dotted line) from 2008 to The trends in Medicaid coverage between the treatment and control groups are very much parallel up to 2010 i.e., prior to the ACA Medicaid expansion in Connecticut. In terms of level, Medicaid coverage in Connecticut 53 I use the bandwidths suggested by Imbens and Kalyanaraman (2012). 54 Other states in the Census Northeast region are Maine (ME), Massachusetts (MA), New Hampshire (NH), New Jersey (NJ), New York (NY), Pennsylvania (PA), Rhode Island (RI), and Vermont (VT). 16

18 was lower than other Northeastern states by about 4 percentage points. After 2010, however, Medicaid coverage in Connecticut increased by 5 to 6 percentage points while other Northeastern states remained stable following their trends-levels. As a result, Medicaid coverage of low income childless adults in Connecticut became similar to (or little higher than) that of other Northeastern states. Table 7 presents the DD estimates of the effect of the Medicaid expansion on insurance coverage. Panel A presents the DD estimates of equation (7) using state year cell level data (9 states over 6 years). It shows an increase in Medicaid coverage by 5.9 percentage points (or 30% of the pre expansion rate of 19.7%) and in turn an increase in percent with public insurance by 5.6 percentage points. Notably, Column (3) shows a decrease in percent with ESI by 3.8 percentage points and a resulting decrease in percent with private insurance by 3.7 percentage points. Taken together, the crowd-out rate (i.e., the ratio of the decrease in private insurance to the increase in public insurance) is estimated to be 66%. Note that the magnitude of this crowd-out effect of Medicaid on private insurance is very similar to those in the recent study of Garthwaite, Gross, and Notowidigdo (2014) 55 and earlier studies (Cutler and Gruber, 1996; Gruber and Simon, 2008). Importantly, the resulting decrease in ESI coverage among low-income childless adults in Connecticut implies a corresponding decrease in their labor supply, which will be shown below. Note that Panel B of Table 7 shows that the DD estimates using individual data are the same as those using state year cell level data (Panel A) Effects on Labor Supply Panel B of Figure 6 shows employment rates of low income childless adults in Connecticut and in other Northeastern states over time. Prior to Connecticut s Medicaid expansion (from 2008 to 2010), employment rates of low income childless adults in both Connecticut and the other Northeastern states showed very similar trends with a higher employment rate in Connecticut by about 5 percentage points. After the Medicaid expansion (from 2011), the employment rate in Connecticut decreased by about 5 percentage points whereas there was no discernible change in the other Northeastern states, leading to virtually the same employment rates between Connecticut and the other Northeastern states. Interestingly, taken together Figure 6 shows that low income childless adults in Connecticut had a lower level of Medicaid coverage but a higher employment rate than those in the other Northeastern states prior to the Medicaid expansion, which then became almost the same as those in the other Northeastern states after the Medicaid expansion. In order to assure the common trends assumption, I extend the labor market outcomes back 55 Although Garthwaite, Gross, and Notowidigdo (2014) study the contraction of Medicaid in Tennessee and I examine the expansion of Medicaid in Connecticut, the crowd-out magnitudes are very similar. This implies the similar magnitudes of labor supply impacts with opposite directions. Indeed, the magnitudes of labor supply impacts (in absolute values) are very close as well (see the next Section). 17

19 to Figure 7 shows extremely parallel trends of labor market outcomes between low income childless adults in Connecticut (treatment group) and those in the other Northeastern states (control group) prior to the Medicaid expansion in Connecticut, providing strong graphical evidence of the common trends in labor market outcomes. Table 8 confirms this. Panel A of Table 8 shows the DD estimates using state year cell level data (9 states over 8 years). The estimates without, and with differential time trends (Panels A.1 and A.2, respectively) are virtually the same, which supports the common trends assumption. I focus on the estimates with differential time trends (Panel A.2) hereafter. As shown in Column (1), the DD estimate shows a decrease in the employment rate by 5.9 percentage points (or 18% of the pre expansion rate of 32.5%). Columns (2) and (3) of Table 8 show that most of the decreased employment come from those working less than 35 hours per week. In addition, Column (4) shows a decrease in hours of work (per week) by 1.2 hours (or 16% of the pre expansion level of 7.6 hours). 57 More importantly, in Column (5) of Table 8, I estimate a change in percent employed with ESI in order to examine how much of the overall labor supply responses resulted from the availability of Medicaid coverage as an alternative insurance to ESI (the substitution effect). The DD estimate shows a 3.8 percentage point decrease in employment with ESI. This estimate suggests that a 2.1 percentage point decrease in the employment rate (out of the overall 5.9 percentage point decrease in Column (1)) might be driven by the income effect i.e., by obtaining Medicaid coverage, low income childless workers without ESI effectively become wealthier, and may reduce their labor supply. Therefore, a 3.8 percentage point decrease would be a lower bound estimate of the labor supply impact of Connecticut s Medicaid expansion. This, along with the increase in Medicaid coverage by 5.9 percentage points (as shown in Panel A of Table 8), implies that 64 % of low income childless adults enrolled in the expanded Medicaid in Connecticut reduced their labor supply after obtaining Medicaid coverage. Interestingly, these estimates of the labor supply impacts of the Medicaid expansion are similar in magnitude to those of Garthwaite, Gross, and Notowidigdo (2014) s study on the Medicaid contraction in Tennessee. For example, they find that about 54 to 63% of people who lost their Medicaid coverage increased labor supply. In addition, Dague, DeLeire, and Leininger (2014) show 56 The ACS provides information on labor market activities even before 2008 unlike information on insurance coverage available only from 2008 onwards. I extended the labor market outcomes no earlier than 2006 because Group Quarters (both institutionalized and non-institutionalized) were included in the ACS for the first time from 2006 and this change makes the earnings of people and the poverty estimates from the 2006 ACS not comparable with those estimates from earlier years (Webster and Bishaw, 2007). 57 I show that the decrease in hours of work occurred along the extensive margin. Appendix Figure A5 shows the distribution of hours of work (per week) among low income childless workers with incomes up to 56% of the FPL before and after Connecticut s Medicaid expansion. As shown in Panel A of Figure A5, hours of work (per week) among those workers in Connecticut decreased between before and after the Medicaid expansion across all hours of work bins. Indeed, the DD estimate for hours of work (per week) conditional on working is statistically zero (coefficient: 0.09; t tario: 0.64). By contrast, Panel B shows an increase in hours of work (per week) among those workers in the other Northeastern states across hours of work bins. 18

20 similar results on the impact of the Medicaid expansion in Wisconsin. Their estimates imply that the Medicaid expansion to childless adults with incomes below 200% of the FPL in Wisconsin reduced the labor supply of enrollees by 2 to 24%. By contrast, in the study of the Oregon Health Insurance Experiment, Baicker et al. (2014) find a statistically insignificant labor market impact of the Medicaid expansion. This is likely the result of the lack of a crowd-out effect i.e., the Medicaid expansion did not affect private insurance coverage (Finkelstein et al., 2012). Indeed, one of the eligibility criteria requires lottery-selected individuals to have been uninsured for six months, and thus workers with ESI who would have most likely responded to the Medicaid expansion in their labor supply were not eligible for the Oregon Medicaid program. I then examine the labor supply responses of childless adults with incomes higher than 56% of the FPL (up to 200 % of the FPL) as a falsification check. If the reduced labor supply of eligible childless adults with incomes up to 56% was driven by Connecticut-specific factors other than the Medicaid expansion, one would see a similar pattern of (or at least a decrease in) labor supply among childless adults with incomes higher than 56% of the FPL in Connecticut, which is not the case as will be explained below. Panel A of Figure 8 shows Medicaid coverage of childless adults with incomes higher than 56% up to 200% of the FPL in Connecticut (solid line) and the other Northeastern states (dotted line) from 2008 to As would be expected given the ineligibility of these populations for the Medicaid expansion, there was no change in Medicaid coverage between pre and post expansion periods neither in Connecticut nor in the other Northeastern states. Column (1) of Table 9 presents the corresponding DD estimates, confirming no change in Medicaid coverage. Panels B through D of Figure 8 show the labor market outcomes of those ineligible for Connecticut s Medicaid expansion (i.e., childless adults with incomes higher than 56% up to 200% of the FPL) in Connecticut and the other Northeastern states from 2006 to The trends of labor market outcomes for those in Connecticut and in the other Northeastern states are very similar, and more importantly there seems to be no differential changes in the labor market outcomes from to between Connecticut and other Northeastern states. The DD estimates shown in Columns (2) through (4) of Table 9 are consistent with these graphical results: no impact of Medicaid expansion on labor supply of ineligible populations. One can also apply the DDD (triple differences) approach by utilizing the fact that only those childless adults with incomes at or below 56% of the FPL (as opposed to those with incomes higher than 56% up to 200 % of the FPL) are eligible for the Medicaid expansion in Connecticut. 58 Appendix Table A5 presents the DDD 58 The DDD regression takes the following form: Y gst = α + η g + λ s + δ t + µ gs + ν gt + ξ st + θ F P L56 g CT s P ost t + U gst (8) where η g area income group (up to or higher than 56% of the FPL) fixed effects; µ gs, ν gt, and ξ st are income group by state, income group by year, and state by year fixed effects, respectively; F P L56 g is an indicator variable for childless adults with incomes at or below 56% of the FPL; and other variables are the same as in equation (7). The 19