Mandate-Based Health Reform and the Labor Market: Evidence from the Massachusetts Reform

Transcription

1 Manate-Base Health Reform an the Labor Market: Evience from the Massachusetts Reform Jonathan T. Kolsta Wharton School, University of Pennsylvania an NBER Amana E. Kowalski Department of Economics, Yale University an NBER July 10, 2015 Abstract We moel the labor market impact of the key provisions of the national an Massachusetts manate-base health reforms: iniviual manates, employer manates, an subsiies. We characterize the compensating ifferential for employer-sponsore health insurance (ESHI) an the welfare impact of reform in terms of sufficient statistics. We compare welfare uner manate-base reform to welfare in a counterfactual worl where iniviuals o not value ESHI. Relying on the Massachusetts reform, we fin that jobs with ESHI pay $2,812 less annually, somewhat less than the cost of ESHI to employers. Accoringly, the eaweight loss of manatebase health reform was approximately 8 percent of its potential size. Keywors: Afforable Care Act, Massachusetts health reform, welfare effects We thank Iris Chan, Toby Chaiken, Erin Taylor, an especially Michael Punzalan an Aiti Sen for excellent research assistance. Amitabh Chanra, John Frieman, Jonathan Gruber, Ben Hanel, Samuel Kleiner, Lauren Nichols, Matt Notowiigo, an Hugh Gravelle provie helpful conference iscussions. This project has benefite greatly from comments from Joseph Altonji, Priyanka Anan, Bjoern Bruegemann, Tom Buchmueller, Marika Cabral, Joseph Doyle, Jesse Egerton, Jason Fletcher, Bill Gale, Alex Gelber, Michael Grossman, Martin Hackmann, Lisa Kahn, Charles Kolsta, Kory Kroft, Ilyana Kuziemko, Fabian Lange, Anup Malani, Amana Pallais, Mark Pauly, Vincent Pohl, Ebonya Washington, Heii Williams, an Cliffor Winston. We are also grateful for comments by seminar participants at Brookings, Brown, BU/Harvar/MIT Health Econ Workshop, the Bureau of Economic Analysis, FRB Richmon, Harvar, HEC Montreal, Olin, Penn, University of California Berkeley, the Urban Institute, University of Texas Austin, Stanfor, Yale School of Public Health, the ASSA meetings, the Duke Empirical Microeconomics Jamboree, the European Econometrics an Health Economics Workshop, ihea, the Miwestern Health Economics Conference, the National Tax Association annual meeting, the NBER Public Economics Program Meeting, the NBER Summer Institute, the New York Health Economics Workshop, an the Utah Winter Business Economics Conference. Funing from the National Institute on Aging grant #P30 AG012810, the Leonar Davis Institute for Health Economics, the Robert Woo Johnson Founation, the W.E. Upjohn Institute for Employment Research, an the Wharton Dean s Research Fun are gratefully acknowlege. During work on this project, Kowalski was the Okun Moel Early Career Fellow at the Brookings Institution. Contact information: jkolsta@wharton.upenn.eu, , amana.kowalski@yale.eu,

2 The Afforable Care Act (ACA) national health reform of 2010 an the Massachusetts state health reform of 2006 both focus on expaning health insurance coverage to near-universal levels. These manate-base" reforms rely on three key provisions to expan coverage: 1) a manate that iniviuals obtain coverage or pay a penalty, 2) a manate that employers offer coverage or pay a penalty, an 3) expansions in publicly subsiize coverage. While regulatory policy has long relie on manates (for example, comman an control regulation of technologies to reuce pollution), public policies that manate that iniviuals purchase privately supplie goos have little preceent. Such manates are sufficiently unpreceente, in fact, that uncertainty about whether the iniviual manate was constitutional at the national level was not resolve until the Supreme Court uphel it in June Despite the resolution of legal questions aroun manate-base policy, the question of economic efficiency remains. To inform the application of manate-base policy in health reform, as well as aitional applications in other settings, we evelop a simple moel of manate-base health reform. moel incorporates the three key features of the national an Massachusetts health reforms. Using this moel, we characterize the compensating ifferential for employee-sponsore health insurance (ESHI) the causal change in wages associate with gaining ESHI an we erive a set of sufficient statistics that capture the impact of the reforms on the labor market an on welfare. Although these sufficient statistics arise from potentially complex an ifficult-to-measure structural parameters that etermine iniviual health insurance an labor supply ecisions, we can recover them from easily measure changes in labor market outcomes. Our moel buils on the work of Summers (1989) who moels a full-compliance employer manate. 1 We apply the moel to current policy by allowing for a pay-or-play employer manate an aing a pay-or-play iniviual manate an expansions in subsiize coverage. The interaction between the employer an iniviual manates changes the preictions of the Summers moel. The central result that an employer manate reuces eaweight loss relative to a tax oes not hol if there is alreay an iniviual manate in place. This theoretical result is relevant for policy, given that as of this writing, the ACA employer manate has not yet been enforce. Base on the structure implie by our theory, we then estimate the relationship between ESHI an the labor market, allowing us to empirically assess the impact of health reform on welfare. Using variation inuce by the Massachusetts health reform which mirrors the national reform in all of the elements of our moel we estimate the empirical analog of our moel. We first estimate the compensating ifferential for health insurance. Our empirical strategy relies on exogenous shifts into an out of ESHI inuce by reform. Using longituinal ata from the Survey of Income an Program Participation on wages, employment, an hours worke, we stuy changes in labor market outcomes for iniviuals who switch to an from ESHI over the reform perio. We incorporate iniviual fixe effects to control for time-invariant attributes that etermine an iniviual s labor 1 The Summers moel has been use subsequently by Gruber an Krueger (1991), Gruber (1994), Buchmueller, DiNaro, an Valetta (2011), an Anan (2011), among others, but it has not been use, to our knowlege, to examine implications of the ACA. Krueger an Kuziemko (2013) consier the implications of the ACA on coverage for the uninsure, Pohl (2011) consiers the implications of the ACA for the labor supply of single mothers, an Heim an Lurie (2012) consier the implications of the Massachusetts reform for the self-employe. Our 2

3 market outcomes, an we incorporate variation between Massachusetts an other states to control for national trens. We also incorporate variation in firm size to allow some firms to be exempt from the employer manate an to control for variation in the Massachusetts labor market that is unrelate to the reform. Combining all of these sources of variation an the reform allows us to obtain causal estimates of the compensating ifferential associate with health insurance. Aing a small amount of structure to the estimate compensating ifferential for health insurance, we estimate the sufficient statistics that etermine the welfare impact of health reform. Specifically, our moel allows us to recover the cost of ESHI to employers, the unerlying worker valuation of ESHI, the labor supply an eman elasticities, an the magnitue of the behavioral responses to the key policy provisions of the Massachusetts reform (the iniviual an employer manates an subsiies). Once we emonstrate that these parameters are sufficient statistics for welfare analysis, we use our estimates to compute the eaweight loss associate with the manatebase reform in Massachusetts. We also compare our estimate eaweight loss to the eaweight loss of a counterfactual tax-base insurance expansion that woul involve levying a wage tax to pay for the provision of health insurance irectly. We fin a compensating ifferential for ESHI that is of the expecte theoretical sign though somewhat smaller in magnitue than the full cost of health insurance, suggesting moerately high average valuation of the benefit among the newly insure. Consistent with the large compensating ifferential, we fin a small hours ifferential between jobs with an without ESHI, also suggesting moerately high average valuation of the benefit among the newly insure. Translating our estimate compensating an hours ifferentials into sufficient statistics for welfare analysis, we fin that manate-base reform is a relatively efficient way to expan coverage. We estimate that manatebase coverage expansion in Massachusetts resulte in a eaweight loss ue to istortion of the labor market that was only 7.7 percent of the istortion associate with instea proviing health insurance through a tax on wages that workers o not link to receiving insurance. The relative efficiency of manate-base reform follows from the high estimate valuation of the newly insure; because people were willing to work for ESHI as well as wages, the istortion to the labor market of manating insurance was relatively small. We examine the robustness of our estimates to a variety of alternative specifications. Although our estimates vary, they always show that manate-base reform is substantially more efficient than tax-base reform because our fining that iniviuals value ESHI is very robust. Apart from our theoretical contributions, our finings contribute to the empirical literature on the incience of fringe benefits, with health insurance as the largest of those benefits. Typically, the enogeneity of fringe benefits an labor market outcomes leas researchers to fin wrong-signe compensating ifferentials for fringe benefits (see Gruber [2000] an Currie an Marian [1999] for reviews); most stuies fin that iniviuals who receive more fringe benefits also receive higher wages. Existing stuies that o not fin wrong-signe compensating ifferentials for health insurance rely on incremental changes in the cost of health insurance, such as premium increases ue to the aition of manate maternity benefits (Gruber 1994) or increasing malpractice costs (Baicker an Chanra 2005). By using variation from the Massachusetts reform, we fin a compensating 3

4 ifferential for the full cost of health insurance; iniviuals who receive ESHI receive wages that are lower by approximately the amount their employer spens on ESHI. While our focus is squarely on the application of the moel to health insurance an, specifically, the manate-base reforms in the ACA an Massachusetts, our approach has applications more broaly in consiering the value of fringe benefits. Our framework ties labor market outcomes to the market for fringe benefits as a function of a small set of parameters. We believe that our moel has potential for future applications both in evaluating the roll out of the ACA where each state will have ifferent unerlying conitions in the labor an insurance market -as well as in other markets (e.g., evaluating changes in pension offerings). Our work also complements recent work by Aizawa an Fang (2013). Their moel evaluates similar welfare questions on the interaction between health policy an the labor market but requires aitional structure. Thus, our approach provies a complementary set of results that evaluate similar policy questions with fewer assumptions require to both ientify the impacts of reform as well as to estimate welfare. The paper procees as follows: Section 1 iscusses the provisions of Massachusetts an national reforms that are likely to affect the labor market. Base on these provisions, Section 2 evelops a theory of manate-base health reform that we use to characterize the compensating ifferential for ESHI an the welfare impact of manate-base health reform relative to tax-base health reform; Section 3 iscusses ientification an estimation of the moel. Section 4 introuces the ata. Section 5 presents results an iscusses robustness, an Section 6 conclues. 1 Massachusetts Health Reform, the Afforable Care Act, an the Labor Market The Massachusetts health reform, passe in April 2006, an the feeral Patient Protection an Afforable Care Act (the ACA), passe in March 2010, contain a number of similar provisions that are likely to affect the labor market. We provie a sie-by-sie comparison in Appenix A. The cornerstone of both reforms is the iniviual manate to purchase health insurance. Unlike traitional full-compliance manates, the iniviual manate in both reforms is a pay-or-play manate that allows iniviuals to pay a penalty if they choose not to comply. The penalty in Massachusetts for those who were unable to emonstrate they ha coverage when they file their taxes was initially $219 per person per year, an it increase to 50 percent of the cost of the least generous ( Bronze ) plan available in the Massachusetts health insurance exchange ( the Connector ) in The penalty associate with the ACA iniviual manate, which will be phase in beginning in 2014, is the higher of $695 per uninsure member of the househol (up to three) or 2.5 percent of househol income. Compliance with the iniviual manate in Massachusetts has been high over 97 percent of tax filers submitte the tax form to comply with the iniviual manate in 2008, an less than 2 percent reporte any spell of uninsurance (Massachusetts Health Connector 2 Accoring to the Massachusetts Connector website in 2010, in the zip coe (Cambrige, MA), the cost of a Bronze plan for a family in Cambrige with two 40-year-ol parents was $11,000 annually. For a couple with two iniviuals age 35, the Bronze plan cost $6,600 annually. A 31-year-ol purchasing a Bronze woul expect to pay $2,868. 4

5 an Department of Revenue 2010). 3 Secon, both reforms inclue a pay-or-play employer manate, which requires employers to offer health insurance or pay a penalty. The Massachusetts reform requires employers with 11 or more full-time employees to offer their workers the option to purchase health insurance coverage. Health coverage options must inclue a plan that allows employees to purchase health insurance using pre-tax wages, an employers must contribute at least 33 percent of the value of the premium or they will be assesse a penalty of $295 per employee per year. The ACA incorporates a similar pay-or-play employer manate, but it efines large employers as those with 50 or more full-time employees (Kaiser Family Founation 2010b). The ACA also requires that coverage options be afforable, such that the insurance offere pays at least 60 percent of covere expenses an the employee is not require to pay more than 9.5 percent of family income for iniviual coverage (Burkhauser, Lyons, an Simon 2011). If the employer oes not offer coverage, the penalty is $2,000 per full-time employee, excluing the first 30 employees. If the employer oes not offer options that meet the efinition of afforable, an employees enroll in subsiize coverage through an exchange, the employer must pay a penalty of $3,000 per employee who obtains subsiize coverage, up to a maximum of $2,000 times the total number of employees minus 30 (Kaiser Family Founation 2010a). Despite the relatively low penalty, compliance in Massachusetts has been high. 4 The thir cornerstone of both reforms is the expansion of subsiize coverage: fully subsiize coverage through Meicai an partially subsiize coverage through new programs for low-income iniviuals who o not qualify for Meicai. The Massachusetts reform expane traitional Meicai (MassHealth) an new fully-subsiize CommCare plans to those earning less than 150 percent of the feeral poverty level (FPL), (150 percent of FPL for a family of three is $29,685 in 2014). 5 Iniviuals between 150 an 300 percent of the FPL can purchase CommCare plans with subsiies that ecline with income. Similarly, the ACA expans Meicai eligibility to all those with incomes below 133 percent of poverty. 6 The Supreme Court rule that states coul opt not to expan Meicai eligibility uner the ACA without losing existing Meicai funing, an about half of states have chosen not to expan at this writing. The ACA also extens subsiize coverage 3 To satisfy the manate, health insurance must meet or excee a specific value (calle minimum creitable coverage"). See Kaiser Family Founation (2009) an Raymon (2007). Iniviuals are automatically exempt from the iniviual manate penalty in Massachusetts if they have a gap in creitable coverage of three months or less in a given calenar year, if they claim a religious exemption, or if their annual income is uner 150 percent of the feeral poverty level (effectively because the lowest cost Connector plan woul be free for them). Other iniviuals can file for an exemption base on afforability using the Certificate of Exemption Application, which also provies etails on the efinition of minimum creitable coverage. (The application is available at org/portal/binary/com.epicentric.contentmanagement.servlet.contentdeliveryservlet/fininsurance/ Iniviual/Afforability2520Calculator/2011CertificateofExemption.pf, accessedecember1,2011.) 4 Only approximately 4.6 percent of employers large enough to be subject to the penalty (12 percent of all Massachusetts employers) were require to pay it in 2010 (Massachusetts DHCFP 2011b). In aition, employers are subject to a separate free rier surcharge" penalty if they o not offer a plan that allows employees to purchase health insurance using pre-tax wages an instea an employee receives care through the state s uncompensate care pool. The compliance cost for employers to avoi this penalty is minimal. Accoringly, zero employers were liable for the free rier surcharge in fiscal years 2008 an 2009 (Massachusetts DHCFP 2011a). 5 In the 48 contiguous states an the District of Columbia, the 2014 poverty level is $11,670 for a single iniviual, an it grows by $4,060 for each aitional family member (Feeral Register 2014). 6 Effectively, eligibility will be extene to 138 percent of poverty because there is a special euction of income uner 5 percent of poverty (Kaiser Family Founation 2010b). 5

6 higher up the income istribution to 400 percent of poverty ($79,160 for a family of three). Even though the national reform extens subsiies to families with higher incomes, the Massachusetts subsiies (Commonwealth Connector 2011a,b) are more generous than the national subsiies (Kaiser Family Founation 2010b) for almost all incomes an family sizes. 2 Moel of Manate-Base Health Reform an the Labor Market 2.1 The Moel We begin by consiering labor eman. resulting in the following labor eman function: A representative firm sets wages to maximize profits, L t D = L ESHI,t D (w + b)eshi t + L NoESHI,t D (w + t b)(1 ESHI t ). Willingness to eman hours of work L in perio t is a function L ESHI,t D or L NoESHI,t D of the monetary hourly wage w, an other arguments, epening on an inicator for whether the firm provies health insurance ESHI t at time t. 7 If the firm provies health insurance, labor eman epens solely on the cost of employing an iniviual in ollar terms wages an the ollar cost to the employer of a stanar health insurance benefit b. There are two perios: Before an After. The employer manate is not in place before reform, so Before =0, but it is in place after reform, so After =. If the firm oes not provie health insurance, labor eman epens on the wage an the per-worker penalty t b for not complying with the employer manate in place in perio t (since b is a fixe ollar amount, we express the employer penalty as a fraction of b instea of a fixe ollar amount without loss of generality). Figure 1 epicts labor eman graphically. Before reform, if the firm provies health insurance, labor eman is lower by b. 8 After reform, if the firm oes not provie health insurance, labor eman shifts own by the per-worker penalty for not complying with the employer manate. Next consier labor supply. A representative iniviual chooses how many hours to work to maximize utility, resulting in the following labor supply function: L t S = L ESHI,t S (w + b + t b µ xt b)eshi t + L NoESHI,t S (w)(1 ESHI t ). Willingness to supply hours of work L in perio t is a function L ESHI,t S or L NoESHI,t S of the hourly wage w. For an iniviual with ESHI, given by the inicator ESHI t, which is exogenous for now, labor supply is also a function of factors beyon the wage. As shown, it is a function of the cost to the 7 We evelop the moel relying on the simplifying assumption that we can measure L in hours of work incluing nonworkers with zero hours, ignoring the potential ifference between the extensive an intensive margin of employment. When we estimate the moel, we relax this assumption in a series of specification checks that allow us to compare the intensive an extensive margin impacts in Massachusetts. We also note that we o not measure L as the probability of employment because only employe workers can have ESHI. We o not measure L as the number of employees because the goal of our moel is to make preictions about labor market outcomes for iniviual workers that we observe in multiple perios. 8 As epicte, L ESHI,t D an L NoESHI,t D have the same linear functional form (the linear functional form is an approximation to a general nonlinear functional form). 6

7 w w NoESHI, Before w NoESHI, After Figure 1: Graphical Moel B A NoESHI L S b, t ESHI, Before L S ( x )b ESHI, After L S w ESHI, Before w ESHI, After D F L ESHI, Before L ESHI, After L NoESHI, After L NoESHI, Before b b NoESHI, Before L D NoESHI, After L D ESHI, t L D L employer for a stanar health insurance benefit b, scale by the amount that an iniviual values a ollar of ESHI relative to a ollar of wages,, an policy parameters in place at time t: theiniviual penalty for not having health insurance t, an the subsiy µ xt available on the iniviual health insurance market, which varies in generosity base on income group x. The iniviual manate an the subsiies are not in place before reform, so Before = Before =µ x,before =0, but they are in place after reform, so After =,anµ x,after = µ x. Figure 1 epicts labor supply graphically. Before the reform, if an iniviual moves from not having ESHI to having ESHI, labor supply shifts ownwar by b because the iniviual is willing to work for lower wages in a job that provies ESHI. This shift results because ESHI is not merely a cost to the employer; it also has value to the employee. In the iniviual s choice problem, several factors can affect the magnitue of the unerlying valuation of ESHI relative to a ollar of wages:. For example, canonical insurance theory emonstrates that willingness to pay for insurance is etermine by an iniviual s wealth, health risk, risk preferences, an the available insurance contract (see, e.g., Arrow [1963] an Rothschil an Stiglitz [1976]). Furthermore, there is a tax preference for ESHI, so we expect the tax preference to increase as a function of the iniviual s marginal tax rate. Rather than moeling these factors iniviually, we moel only, which we will emonstrate to be a sufficient statistic for welfare analysis in the spirit of Chetty (2009). After the reform, labor supply also reflects the penalty associate with the iniviual manate an any subsiy available to that iniviual for health insurance outsie of employment. That is, the iniviual penalty augments the iniviual s unerlying valuation of ESHI, shifting his labor supply curve further ownwar for jobs offering ESHI even if the iniviual oes not value health insurance on its own merits, he will value it at least as much as the penalty that he must pay 7

8 for not having it. 9 A subsiy available outsie of employment also affects the iniviual s labor supply if he obtains health insurance through his employer because the outsie coverage option has change. He is more willing to work for ESHI instea of wages in the face of a penalty, an he is less willing to work for ESHI instea of wages in the face of a subsiy for health insurance outsie of employment. After the reform, if the iniviual moves from not having ESHI to having ESHI, his labor supply shifts ownwar by + µ x, which we call the penalty-an-subsiy-inclusive valuation of ESHI, multiplie by the cost of health insurance b. The penalty-an-subsiy-inclusive valuation incorporates unerlying preferences for health insurance an the key policy features of manate-base reform in a simple measure: the shift in labor supply. We will show that the penalty-an-subsiy-inclusive valuation is an important sufficient statistic for our welfare analysis. Putting the two sies of the market together yiels a labor market equilibrium (w, L) in perio t that reflects the unerlying parameters that etermine labor supply an eman. As shown in Figure 1, there are two potential equilibria in each perio t, conitional on whether the iniviual receives health insurance through the employer: D an A are the equilibria for iniviuals with an without ESHI before the reform, respectively; F an B are the equilibria for iniviuals with an without ESHI after the reform, respectively. Our remaining theoretical an empirical analysis relies on the istances between points A, B, D, anf. The remainer of this section focuses on translating these istances into parameters of interest that ultimately allow us to analyze welfare. 2.2 Characterization of the Compensating Differential for ESHI We begin by emonstrating how the moel allows us to ientify the compensating ifferential for health insurance, efine as the causal ifference in wages between jobs with ESHI an jobs without ESHI. We can also characterize the corresponing hours ifferential using hours in lieu of wages. To obtain compensating an hours ifferentials, we simply compare wages an hours in the equilibria with ESHI (equilibria D an F ) to wages an hours in the equilibria without ESHI (equilibria A an B). The first column of Table 1 shows the four possible comparisons of equilibrium wages an hours that we can use to measure compensating an hours ifferentials. As shown in Figure 1, because 9 We o not expect the iniviual penalty to increase the total valuation of health insurance for an iniviual who alreay values it fully. Therefore, we specify that the magnitue of is affecte by the unerlying valuation an the statutory penalty as follows: 8 <, for apple 1 = 1, for 1 apple apple 1 : 0, for >1 In the first case, is small, so takes on its statutory value, an the penalty-inclusive valuation, which we efine as +,islessthan1.intheseconcase, is large enough to augment until the penalty-inclusive valuation is full. In the thir case, the iniviual s unerlying valuation of health insurance is higher than the cost to the employer. Such a case coul arise if an iniviual cannot access health insurance outsie of employment, perhaps because of preexisting conitions that are exclue on the iniviual market. Such a case coul also arise if health insurance through the employer is cheaper than other insurance, which is likely because of the tax-preference for employersponsore health insurance an because employers have more negotiating power than iniviuals. In this case, the penalty-inclusive valuation of health insurance is his unerlying valuation, an the penalty has no impact. We efine the subsiy similarly so that it cannot reuce an iniviual s penalty-an-subsiy-inclusive valuation beyon zero. 8

9 health reform shifts labor supply an labor eman, the compensating an hours ifferentials are ifferent epening on which equilibria we compare. Table 1: Compensating an Hours Differentials Compensating Di erential Su cient Statistics Coe cients s w D w A b s 8 [+ 8e ] (1 )s ( + µ w F w x ) B b s [+ 1e + 8e ] (1 )s w D w B b s 8 11 [+ 8e ] s [ + µ w F w x ] A b s [+ 1e + 8e ] Hours Di erential Su cient Statistics Coe cients 1 L D L A b s 8 [+ 8e ] 1 ( + µ L F L x ) B b s [+ 1e + 8e ] 1 L D L B b s 8 11 [+ 8e ] 1 ( + µ L F L x ) A b s [+ 1e + 8e ] We can also use the compensating an hours ifferentials to learn about the valuation of ESHI an the other moel parameters. In the secon column of Table 1, we express each compensating or hours ifferential in terms of the sufficient statistics of the moel. These expressions follow irectly from the geometry of Figure 1. We represent the slope of the labor supply curve with s an the slope of the labor eman curve with (these slopes are elasticities if we specify w as the logarithm of wages an h as the logarithm of hours). In our empirical implementation, we will be particularly intereste in the compensating an hours ifferentials for iniviuals who switch from not having ESHI before the reform (equilibrium A) to having it after the reform (equilibrium F ). For these iniviuals, as shown in the expression in the last row of the each panel of Table 1, if the penalty-an-subsiy-inclusive valuation is full ( + µ x =1), then the absolute value of the compensating ifferential is equal to the cost of ESHI (ESHI ecreases wages by b), an the hours ifferential is zero (ESHI oes not istort hours worke). Therefore, if the compensating ifferential is equal to the cost of the benefit b an the hours ifferential is zero, then we can infer that the penalty-an-subsiy-inclusive valuation is full Characterization of the Welfare Impact of Manate-Base Health Reform Using Sufficient Statistics To this point, we have evelope a simple moel that allows us to express all of the key parameters of manate-base reform as well as the set of preferences that etermine an iniviual s valuation 10 Previous stuies base on the Summers moel have stoppe at relate inferences because they only have enough variation to ientify the valuation if it is full. If the compensating ifferential is less than the cost of the benefit, an the hours ifferential is nonzero, then they cannot infer the magnitue of the penalty-an-subsiy-inclusive valuation beyon stating that it is not full. However, as we iscuss in Section 3, the aitional sources of variation in our moel enrich the empirical content of the Summers moel, allowing us to ientify the penalty-an-subsiy-inclusive valuation, regarless of the true magnitue. 9

10 of ESHI in terms of labor market equilibria. We are also intereste in how the Massachusetts an national manate-base reform affect welfare an how these policies compare to alternative approaches that coul be taken to expan health insurance coverage. Our moel allows us to conuct welfare analysis simply with sufficient statistics, builing on our estimate compensating ifferential. w NoESHI, Before w NoESHI, After w ESHI, Before w ESHI, After Figure 2: Graphical Moel with Deaweight Loss w T NoESHI, t L D S B ESHI, F A b L S ( x )b F ESHI, B L S D T D F L ESHI, Before L ESHI, After L NoESHI, After L NoESHI, Before b b Before After NoESHI, Before L D NoESHI, After L D ESHI, t L D L Manate-base policy can reuce welfare in two ways. First, if it istorts the labor market such that workers are willing to work for wages lower than the market wage an employers are willing to hire workers for more than the market wage, but the transaction oes not occur. Secon, workers at firms not offering ESHI face a penalty cost. Graphically, the welfare reuction correspons to a weighte combination of two overlapping triangles shown in Figure 2. The eaweight loss is the triangle given by F 0 AF 00 if the representative iniviual has ESHI after the reform, an it is equal to the triangle given by B 0 AB if he oes not. 11 We can also express these triangles in terms of the key parameters of the moel. The combine eaweight loss of the policies of manate-base health reform (enote by the subscript m) is as follows: DWL m = b 2 2(s ) ((1 ( + µ x)) 2 ESHI After + 2 (1 ESHI After )). (1) If we know the values for all of the terms in this equation, we can calculate the welfare impact of manate-base health reform on the labor market. Thus, these terms are sufficient statistics for welfare analysis. For example, if the employer penalty is equal to zero an the penalty-ansubsiy-inclusive valuation is full ( + manate-base health reform. µ x =1), then there is no eaweight loss associate with 11 As shown, the eaweight loss for iniviuals with ESHI is smaller, but the relative magnitues of the triangles can reverse, epening on the magnitues of the policy parameters. 10

11 2.4 Characterization of the Welfare Impact of Manate-Base to Tax-Base Health Reform While the welfare impact of manate-base reform is clearly of interest, we are also intereste in comparing the welfare impact of manate-base reform to the welfare impact of a counterfactual reform to inform policy ecisions an to give us a sense of whether the welfare impacts that we fin are large or small. Using Equation (1), we can compare the eaweight loss of manate-base reform to the eaweight loss of an alternative policy. If we can express the key policy elements in terms of labor market equilibria, then we can compare ifferent policies simply by taking the ratio of the eaweight losses. In the traition of Summers (1989), we compare the welfare impact of manate-base health reform to the welfare impact of an alternative tax-base health reform. Before consiering the counterfactual policy of the Summers moel, we note that the Summers moel is a special case of our moel. The Summers moel incorporates the establishment of a fullcompliance employer manate starting from a regime where no firms provie health insurance. We exten the Summers moel to incorporate the establishment of the three elements of manate-base reform starting from a regime where some firms provie health insurance an others o not. From a moeling perspective, the Summers moel is a special case of our moel with a single equilibrium in each perio an a ifferent policy intervention. In our moel, before the policy intervention, there are two equilibria the equilibrium at A for jobs that o not inclue ESHI an the equilibrium at D for jobs that inclue ESHI. In the Summers moel, before the policy intervention, there is only one equilibrium at A no jobs inclue ESHI. In our moel, after the policy intervention, there are two new equilibria because of the imposition of the employer manate, the iniviual manate, an the employer manate. One occurs at F for firms that provie ESHI, an the other occurs at B for firms that o not provie ESHI an instea pay the penalty. In the Summers moel, there is only one equilibrium after the policy intervention. The policy intervention consists of a manate that all employers must provie health insurance, an it is not a pay-or-play manate, so there is full compliance with the manate. The single equilibrium after the full compliance manate occurs at D, which is equivalent to the ESHI equilibrium before the reform in our moel. As an alternative to the full-compliance manate, Summers (1989) consiers a single counterfactual policy, which he refers to as a benefit tax. Uner this counterfactual policy, again there is a single equilibrium before its implementation at A no jobs inclue ESHI. Upon the implementation of the policy, the government levies a tax on employers to provie health insurance. Suppose for now that the tax is equal to the cost of proviing a stanar health insurance benefit b. As shown in Figure 1, labor eman shifts ownwar by b, an holing labor supply constant, the new labor market equilibrium is at point T. The eaweight loss of the tax-base reform is given by the triangle T 0 AT :. DWL = 2 2(s ). The key Summers (1989) assumption about tax-base reform is that it oes not inuce a shift in labor supply. An equivalent way of stating this assumption is that workers value insurance, 11

12 but because insurance is not linke to work uner the tax-base reform, this valuation oes not show up in the labor supply curve. This assumption implies that iniviuals woul place no value on health insurance offere by their employers (ESHI) because they alreay have health insurance from a tax-finance government source. Yet another equivalent way of stating this assumption is that tax-base policy makes the link between taxes an benefits less salient workers can recognize the link between ESHI an wages uner manate-base health reform, but they o not recognize the link between tax-finance health insurance an the tax that employers pay on wages uner tax-base reform. We compare the welfare impact of manate-base reform to the welfare impact of the same tax-base reform as Summers (1989). Taking the ratio of the eaweight loss of manate-base reform to the eaweight loss of tax-base reform, allowing b 6= gives DWL m DWL = b2 2 ((1 ( + µ x)) 2 ESHI After + 2 (1 ESHI After )). (2) This equation characterizes the welfare of the combine features of manate-base reform relative to a tax-base reform in terms of a small number of sufficient statistics: the cost b that employers pay for ESHI compare to the necessary tax revenue for the same benefit; the penalty-an-subsiyinclusive valuation, + µ x, for iniviuals who have ESHI after reform; the employer penalty for iniviuals who o not have ESHI after reform; an the fraction of iniviuals with ESHI after reform, ESHI After. Since the same iniviuals woul be covere by both manate-base an tax-base reform, unerlying health risk is invariant to the plan. Thus the ratio of b to is just the relative loaing cost of ESHI an government-provie health insurance. We show in Section 3.3 that two of the key sufficient statistics the cost of the benefit b an the penalty-an-subsiyinclusive valuation + to after the reform. µ x are functions of the compensating an hours ifferentials from before Our welfare ratio generalizes the ratio implie by Summers (1989) an offers some insights that are perhaps counter-intuitive. To see this, first consier the ratio implie by Summers (1989): the ratio of the eaweight loss of a full-compliance employer manate given by the triangle D 0 AD 00 to the eaweight loss of tax-base reform given by triangle T 0 AT. Using Equation (2), all agents have ESHI after reform (ESHI After =1), an there are no penalties or subsiies ( = µ x =0). Assuming that the tax is equal to the cost of health insurance (t = b), the ratio of the eaweight loss of the full-compliance employer manate to the ratio of the tax-base reform simplifies to (1 ) 2. This special case yiels the theoretical contribution of Summers (1989): an employer manate reuces eaweight loss relative to a tax. However, our moel emonstrates that an employer manate oes not always reuce eaweight loss relative to a tax. If there is alreay a pay-or-play iniviual manate in place, then the aition of a full-compliance or a pay-or-play employer manate weakly ecreases welfare relative to a tax. Consier the case where there is alreay an iniviual pay-or-play manate in place, an some firms provie ESHI while others o not, but there is no employer manate of any kin. If the tax is equal to the cost of health insurance (t = b), then the eaweight loss of the iniviual manate 12

13 relative to a tax is given by Equation (2) with µ x = =0, which simplifies to (1 ) 2 ESHI After. We can see from this expression that if there is no employer penalty, then there is no istortion to the labor market for firms that o not provie ESHI. Aing a full-compliance employer manate weakly increases the eaweight loss ratio because after its imposition, all firms must provie ESHI (ESHI After =1); zero istortion without ESHI is no longer possible. Likewise, aing a pay-or-play manate weakly increases the eaweight loss ratio, which becomes (1 ) 2 ESHI After + 2 (1 ESHI After ) because there is now a eaweight loss triangle for firms without ESHI. Intuitively, the iniviual manate has a smaller eaweight loss than a tax because it makes iniviuals willing to work for lower wages if they receive ESHI. When the iniviual manate is alreay in place, the employer manate results in aitional eaweight loss for iniviuals without ESHI. Another relate implication of our welfare moel is that as long as some iniviuals receive ESHI from their employers, holing existing subsiies an employer manates constant, the aition of a pay-or-play iniviual manate or an increase in the size of the penalty associate with the iniviual manate can reuce eaweight loss in the labor market. Algebraically, an increase in in Equation (1) results in ecrease eaweight loss. This implication seems counterintuitive if we generally expect that new or expane policies increase istortions Ientifying an Estimating the Moel In this section, we evelop the empirical analog of our theoretical moel. We have shown that we can express the compensating ifferential for ESHI an the welfare impact of health reform in terms of ifferences between the four labor market equilibria. Thus, to estimate the moel we must ientify wages an hours at each equilibrium. To o so, we rely on the variation inuce by the reform in Massachusetts. The simplest approach woul require only eight pieces of ata to estimate the four labor market equilibria in Figure 1: average wages an hours for jobs with an without ESHI before an after reform within Massachusetts. We coul then calculate the compensating ifferential for ESHI an the sufficient statistics for the welfare impact of health reform. However, we nee to incorporate aitional sources of variation to account for factors outsie of the moel that woul bias our estimates were we to merely compare means in Massachusetts over time across groups. In practice, we also calibrate the sufficient statistics that are the least well-ientifie by our empirical variation. 3.1 The Estimating Equation To estimate all of the relevant ifferences between labor market equilibria, the compensating an hours ifferentials an the welfare impact of health reform, we specify an estimate wage an hours 12 We o note that the aition of an iniviual manate can increase istortion in the market for goos an services by requiring iniviuals to allocate more ollars to health insurance an fewer ollars to other goos. However, we focus on a specific set of istortions to the labor market, assuming that the government has alreay ecie to expan health insurance coverage as in the case of the Massachusetts reform an the ACA an is looking for the most efficient way to o so. For a given level of coverage, istortions to the market for goos an services are the same uner manate-base reform an tax-base reform. By taking the ratio of the istortion uner manate-base reform to the istortion uner tax-base reform, we can focus on the welfare impact of health reform on the labor market. 13

14 equations of the following form: Y it =[ 1(MA ESHI After Large) it + 8 (MA ESHI Large) it + 11 (MA After Large) it + 12(ESHI After Large) it + 19 (ESHI Large) it + 22 (After Large) it + 23(Large) it +( s Large) it +] 1[e](MA ESHI After) it + 8[e] (MA ESHI) it + 11[e] (MA After) it + 12[e](ESHI After) it + 19[e] (ESHI) it + 22[e] (After) it + s + i + " it, (3) where Y it measures wages w or hours L for iniviual i in state s at time t. We specify wages an hours in levels. The level specification allows us to capture the impact of the reform on the intensive margin of how many hours to work an the extensive margin of whether to work because we can inclue unemploye workers in the sample, specifying that they have wages an hours of zero. The level specification also allows health insurance to have a realistic aitive rather than multiplicative effect on wages, but we also investigate robustness to specifying wages an hours in logarithmic form. MA is an inicator for the state of Massachusetts relative to other states, ESHI is an inicator for ESHI relative to the absence of ESHI, After is an inicator for the perio after the reform relative to the perio before the reform, an Large is an inicator for large firms relative to firms of known small size that are exempt from the employer manate. We represent the coefficients of the wage equation with subscripte coefficients of the hours equation with subscripte coefficients, an we represent the corresponing coefficients. The numbers of the coefficients convey that they are a subset of the coefficients of the full equation that we use to separately ientify ifferent values of µ x, which we present in the Online Appenix. We inclue state fixe effects s with a state other than Massachusetts omitte to control for ifferences in wages across states, an we inclue iniviual fixe effects i to control for time-invariant ifferences across iniviuals, allowing for iniviual-specific shocks at time t, " it. We inclue a time fixe effect, After, to control for changes in the labor market over time. 13 We begin with a baseline specification that exclues all brackete terms. This specification exclues variation between large an small (exempt) firms. We subsequently inclue the brackete terms in our full specification. Our approach incorporates three key sources of variation in aition to the changes in Massachusetts over time in labor market outcomes by ESHI status. First, we rely on variation within iniviuals over time by incluing iniviual fixe effects. The iniviual fixe effects are essential because they allow us to control for a myria of worker characteristics that shift labor supply an eman for a given iniviual for reasons that are correlate with having ESHI. That is, iniviuals who have ESHI are likely to iffer from those iniviuals who o not, an those ifferences also manifest in labor market outcomes. Unobserve ifferences between iniviuals with an without ESHI is the stanar concern that has plague the literature on the compensating ifferential for health insurance. A more subtle but critical reason to incorporate iniviual fixe effects is the nee to aress compositional change among those with ESHI in Massachusetts from before to after the 13 In all specifications, we also allow for a uring implementation perio that is separate from the before an after perios. In our results tables, we represent the coefficients on uring perio terms with corresponing superscripts. 14

15 reform. If manate-base reform ifferentially increases ESHI rates among iniviuals with lower wages an/or work hours, without iniviual fixe effects, we coul spuriously estimate a negative relationship between ESHI an wages after the reform. Secon, we incorporate variation between Massachusetts an other states to control for factors that shift labor supply an eman nationally for reasons that are unrelate to Massachusetts health reform. Incorporating this variation allows us to control for any aggregate trens in the relationship between ESHI an labor market outcomes. Finally, in our full specification, we incorporate variation between small an large firms. 14 This aitional source of variation allows us to better ientify the impact of the employer penalty by comparing firms that qualifie relative to firms that were exempt. In contrast, because it oes not inclue variation by firm size, our baseline specification assumes that all Massachusetts firms are subject to the employer penalty after reform. Incorporating variation by firm size also helps to control for Massachusetts-specific factors unrelate to health reform that coul shift labor supply an eman. 15 Our estimating equations are relatively straightforwar, allowing us to estimate them with orinary least squares. The simplicity of the estimating equations is an avantage of our moel relative to alternative structural moels because robustness analysis is easier to implement, an the results are more transparent. Furthermore, because the functional form of these equations is relatively simple, we can interpret the coefficients irectly as well as the combinations of coefficients that make up the sufficient statistics. 3.2 Estimating Wages, Hours, an the Compensating Differential for ESHI To ientify the wage an hours associate with ifferent equilibria, we focus on the linear combinations of coefficients that correspon to wage an hours at each equilibrium A, B, D an F,as oppose to focusing on a single coefficient as in a traitional ifference-in-ifferences moel. To see the istinction between our estimating equation an a ifference-in-ifferences moel, consier that in the baseline specification that omits brackete terms, the first coefficient 1 is the traitional ifference-in-ifference-in-ifferences coefficient. It gives the impact on wages for iniviuals in Massachusetts relative to iniviuals in other states, when they have ESHI relative to when they o not have ESHI, after the reform relative to before the reform. In this specification, iniviuals in Massachusetts with ESHI after reform are the only group treate by reform, an all other iniviuals are in control groups. However, our moel preicts that iniviuals in Massachusetts without ESHI after reform have also been treate by the reform their wages an hours shoul be lower because of the employer penalty so we o not want to use them as a control group in our 14 We recognize that firm size can be enogenous in the sense that iniviuals can choose to work at small or large firms in response to health reform or firms that are near the firm size cutoff may enogenously change their size to avoi penalties. However, we want to allow for such behavior to capture the broaest possible impact of reform. 15 We exten the moel to incorporate variation in subsiy amounts in the Online Appenix. This variation allows us to ientify separate equilibria for iniviuals for ifferent subsiy amounts. With these equilibria, we can separately ientify from µ x, an we can ientify a ifferent value of µ x for each income eligibility group x. However,because using this variation requires us to ivie the ata into small groups base on income eligibility threshols, we o not use this variation in our full specification. 15

16 preferre estimate of the compensating ifferential. Similarly, our full specification that inclues brackete terms resembles a ifference-in-ifferences specification, but we o not want to examine only the resulting 1 coefficient because oing so woul incorporate control groups that shoul not be incorporate accoring to our moel. 16 Accounting for ifferences with relevant control groups, we express the wages associate with each equilibrium in Table 2. The hours associate with each equilibrium are equivalent with in place of. To ease interpretation, we normalize w A =0an L A =0so that all equilibria are relative to the equilibrium without ESHI before reform. The erivation of these expressions is straightforwar. For example, the ifference in wages between equilibrium B an A (the equilibrium without ESHI after the reform relative to the equilibrium without ESHI before the reform) is 11, the change in wages from after the reform to before the reform for iniviuals who remain without ESHI in Massachusetts, relative to iniviuals in other states who remain without ESHI. 17 Table 2: Wages in Terms of Coefficients w A NoESHI, Before 0 w B NoESHI, After 11 w D ESHI, Before 8 [+ 8e ] w F ESHI, After [+ 1e + 8e ] Using the expressions for the labor market equilibria in Table 2, we can then express the compensating an hours ifferentials in terms of regression coefficients in the last column of Table 1. Our preferre measure of the compensating ifferential, w F w A, is the sum of several coefficients: [+ 1e + 8e ]. These coefficients reflect the change in wages observe for iniviuals who switch from not having ESHI before the reform to having it after the reform, relative to iniviuals who have the same switch in ESHI status from before to after reform in other states. Because the most convincing ientification comes from changes in ESHI status for a given iniviual inuce by the reform, we focus on this comparison for our preferre estimates of the compensating an hours ifferentials. In contrast, the first two ifferentials in Table 1 rely on changes in ESHI status for a given iniviual within the perio either before or after reform. The changes in ESHI status that ientify these compensating ifferentials coul be enogenous, even after incluing iniviual fixe effects, if iniviuals gain ESHI when they get a better job that inclues health insurance. 16 As in the baseline specification, we o not want to inclue iniviuals in Massachusetts without ESHI after reform as a control group because they are treate by the employer penalty. Even more importantly, we o not want to inclue iniviuals in small firms as controls for iniviuals in large firms in all instances because our moel preicts that iniviuals at firms of both sizes shoul experience eclines in wages when they shift into ESHI. Only iniviuals in small firms without ESHI are a vali control group because they are not subject to the employer penalty. 17 In the full specification, which inclues the brackete terms in Equation (3), 11 also reflects the ifference between iniviuals in large firms an iniviuals in small exempt firms, thus controlling for Massachusetts-specific factors after reform. 16

17 3.3 Estimating the Welfare Impact of Manate-Base Health Reform To estimate the welfare impact of health reform given in Equations (1) an (2), we first estimate the unerlying sufficient statistics. We can express most of the sufficient statistics in terms of ifferences in wages an hours between the four labor market equilibria epicte in Figure 1. Our erivation follows irectly from the geometry of the figure. In the first two rows of Table 3, we express the supply an eman curve slopes in terms of wages an hours ifferences between equilibria. The last column gives equivalent expressions in terms of coefficients. In the subsequent rows of the table, we express other sufficient statistics in terms of the slope of the labor supply an eman curves as well as ifferences between other equilibria. Table 3: Sufficient Statistics Su cient Statistics Wages an Hours Coe cients w s B w A 11 L B L A 11 w F w D [+ 1e ] L F L D [ 1e ] b (L F L A ) (w F w A ) ( [ 1e + 8e ]) ( [+ 1e + 8e ]) (L B L A ) (w B w A ) ( 11 ) ( 11 ) b b s(l D L A ) (w D w A ) s( 8 [+ 8e ]) ( 8 [+ 8e ]) b b s(l µ F L D ) (w F w D ) s( [+ 1e ]) ( [+ 1e ]) x b b s( [+ 1e + 8e ]) ( [+ 1e + 8e ]) b + µ x s(l F L A ) (w F w A ) b As iscusse above, the ifferences between labor market equilibria that are ientifie by changes resulting from the Massachusetts reform are arguably best ientifie. Therefore, some sufficient statistics are ientifie more convincingly than others. Fortunately, these sufficient statistics are the most important for welfare analysis: the penalty-an-subsiy-inclusive valuation ( + µ x ) an the cost of ESHI to employers (b). The other sufficient statistics are ientifie in principle, but not as convincingly because they o not epen on changes in ESHI status inuce by the reform. 18 In practice, we estimate values 18 We can ientify the slope of the eman curve by comparing iniviuals with ESHI before an after the reform; we can ientify the slope of the supply curve s by comparing iniviuals without ESHI before an after the reform; an we can also ientify the employer penalty by comparing iniviuals without ESHI before an after reform, using a value for. We can also ientify,, an µ x separately from their sum, which woul allow us to analyze the welfare impact of the separate components of health reform inepenently. As shown in Table 3, ientification of requires a value for s an the comparison of people with an without ESHI before reform. The inclusion of iniviual fixe effects shoul help to ientify because we control for time-invariant factors that affect wages an benefits. However, any changes over time that affect both simultaneously will lea to bias. For example, if an iniviual without health insurance gets promote to a job with higher wages an ESHI, we will estimate a negative value for, eveniftheiniviualvaluesthebenefitsuchthatthetruevalueof is positive. Such bias is precisely the problem that has hinere previous efforts to ientify compensating ifferentials, which we overcome in ientifying the penalty-an-subsiy-inclusive valuation but not in ientifying separately. We can also attempt to separately ientify an µ x.asshownintable3,ientificationoftheifference µ x requires a value for s an the comparison of people with ESHI before an after reform. To separately ientify µ x from, an to ientify ifferent values of 17

18 for these parameters that o not accor well with values that we expect base on the literature an the empirical magnitue of the employer penalty. Given that these parameters are not ientifie by the Massachusetts reform an that their misspecification can affect the estimates of all the other sufficient statistics through the s an terms in their erivations, we iscar the empirical estimates an calibrate them. Reviewing the literature suggests that reasonable magnitues for labor supply an eman elasticities are 0.1 an 0.2, respectively (Blunell an MaCury 1999, Hamermesh 1996). 19 We also calibrate the employer penalty such that the ollar value of the employer penalty b is equal to the statutory penalty of $295 per year. Given that we calibrate some sufficient statistics, one might be tempte to calibrate most of our moel using the statutory values of the policy parameters, rather than estimating any sufficient statistics. However, it is important to estimate the sufficient statistics for two main reasons. First, the iniviual s unerlying valuation oes not have a statutory value. Secon, the behavioral response to the policy parameters might be smaller or larger than the statutory policy parameters because of interactions between them an the iniviual s unerlying valuation (see footnote 9), or if iniviuals over respon if they are averse to paying penalties on moral grouns. 20 Therefore, we only calibrate values when we have reason to believe that ientification is not convincing an the empirical results are not consistent with the moel. 4 Data an Summary Statistics 4.1 The Survey of Income an Program Participation For our main analysis, we use the Survey of Income an Program Participation (SIPP), a nationally representative longituinal survey covering househols in the civilian noninstitutionalize population. As we iscuss in etail, the longituinal nature of the SIPP is critical for ientification. Iniviuals selecte into the SIPP sample are interviewe once every four months over a four-year panel. Each interview covers information about the previous four-month perio, resulting in personmonth-level ata. Interview months iffer across iniviuals in the sample. Previous research has shown evience of seam bias" in the SIPP, whereby iniviuals ten to give the same responses uring one interview for all four months associate with the interview perio, but they o change responses from one interview to the next (see Chetty [2008]). To aress seam bias, we restrict our ata to the interview month in our regression specifications. We use weights in all summary statistics an regressions to account for the SIPP sampling an response unit esign. 21 µ x for people eligible for ifferent subsiy amounts, we can incorporate variation in subsiy amounts across income eligibility threshols as we iscuss in the Online Appenix. 19 Because our specification is in levels (not logarithms), we convert these into slopes at the mean wage an hours. 20 The CBO consiere a behavioral response to an iniviual manate in their estimates of the impact of the ACA on coverage (Auerbach et al. 2010). They highlight the nee to unerstan responses to the iniviual manate in more etail, an our methoology coul prove useful. 21 We use ata from the core content of the SIPP. We construct our ata by appening the 12 iniviual-wave files from the 2004 panel an merging longituinal weights onto the full file by iniviual person ientifiers. Longituinal panel weights account for people who were in the sample in wave 1 of the panel an for whom ata were obtaine (either reporte or impute) for every month of the panel. There are four panel weights associate with the 2004 SIPP panel; the first covers people present in waves 1 4, the secon covers people in waves 1 7, the thir covers 18

19 We use the full 2004 SIPP panel, which covers October 2003 to December A potential limitation of this SIPP panel is that it oes not exten for a long time after reform was fully implemente, an it will not be extene further because an entirely new SIPP cohort began in Despite this potential limitation, we believe the SIPP ata are the best-suite to moeling the labor market impact of the Massachusetts reform for a number of reasons. First an foremost, the SIPP is the largest longituinal ata set that we are aware of that inclues labor market outcomes an insurance information. The Meical Expeniture Panel Survey (MEPS) is longituinal, but it only extens for two an a half years, an the sample size is only approximately 15 percent of the size of the SIPP, with 160 iniviuals in Massachusetts. 22 Current Population Survey (CPS) an the American Community Survey (ACS) o not inclue longituinal ientifiers for health insurance. Although aministrative ata from Social Security are longituinal, they o not inclue information on hours worke or insurance. Secon, while we cannot compare wage an hours trens conitional on iniviual fixe effects across ata sets because of ata availability, raw wage an hours trens are very similar in the SIPP to those in other ata sets insie an outsie of Massachusetts. We present these comparisons in Appenix B. Thir, although we observe a relatively short perio of responses after the iniviual manate went into effect on July 1, 2007, we also observe a full year of responses uring the implementation of the reform. Thus, we are able to observe the response in wages an hours of all iniviuals who change ESHI status in Massachusetts at any time after the reform was passe. This is particularly valuable because open-enrollment perios for ESHI are generally in November, with new coverage starting in January. Thus, to satisfy the iniviual manate in July 2007 by taking up ESHI, many iniviuals woul have to start coverage in January 2007, well before our ata en at the en of December Using ata from the CPS, we fin that of the eventual increase in coverage among those in Massachusetts by 2010, 87 percent ha occurre by the en of 2007 (measure in the March 2008 CPS). The share of the eventual increase in those covere by ESHI by 2010 is even higher at 91 percent, suggesting that our sample perio covers the time in which much of the expansion ue to the reform ha occurre espite our relatively short post-reform perio. 24 people in waves 1 10, an the fourth covers people who have ata for the whole sample (waves 1 12). The panel weighting scheme oes not assign weights to people who enter the sample universe after wave 1 (panel weight = 0 if the iniviual was not in the sample in wave 1, if they have missing ata for one or more month(s), or both). In choosing the appropriate weights, there is a traeoff between length of iniviual ata an reuctions in sample size associate with attrition. Our specification oes not use panel weights an instea uses iniviual weights, therefore maximizing the number of responents. In results not reporte, we reestimate our regressions using each panel weight. Reassuringly, the coefficients of interest are relatively robust to these weight changes. Using weights 3 or 4 oes lea to substantial loss of precision as the sample size falls when moving from longituinal weight 1 to We have run our regressions in the restricte-access MEPS with state ientifiers, but the sample size is not large enough for us to obtain reliable results. 23 We also note that even if we ha ata beyon December 2007, we woul be reticent to rely on it to estimate our moel because a recession began at that time. Dubay et al. (2012) present insurance coverage an employment measures in Massachusetts, a set of northeastern control states, an the remainer of the country over time. They show the impact of the recession in both Massachusetts an the various control groups beginning almost immeiately in With 2008 ata, we woul thus be concerne that the recession coul have ha a ifferential impact on Massachusetts relative to other states. 24 We o, however, note that even though the CPS asks about coverage in the previous year, it is well known that The 19

20 Despite all of the avantages of the 2004 SIPP panel, one limitation is that the sample size ecreases over time, primarily because of interview reuctions but also because of attrition. Our group of interest is the population between the ages of 18 an 64. In 2004, there are 72,057 unique iniviuals in this sample across states, of which 2,047 unique iniviuals are in Massachusetts. In 2007, there are 28,661 unique iniviuals in the sample, of which 685 unique iniviuals are in Massachusetts. Using the SIPP, we construct our main epenent variables: hourly wages w an hours worke per week L. We convert all wages into 2006 ollars using the Consumer Price Inex for all urban consumers (CPI-U) to ajust for inflation. The SIPP allows responents to report wages an hours for up to two jobs, but our estimates rely on income an hours worke only in the primary job. Because the SIPP ata only inclue monthly income, not monthly wages, we must ivie income by a measure of hours worke to obtain monthly wages. However, because our moel relies on separate movements in w an L, it woul be problematic for both measures to reflect contemporaneous movements in L. To get aroun this issue, which is relate to the ivision bias problem in the labor economics literature, we ivie income by the average hours reporte in the first four interviews (representing a 16-month perio). Our regression estimates are robust to alternative wage measures because hours vary less than wages. 4.2 Summary Statistics Before proceeing to our regression results, we assess the empirical valiity of comparing Massachusetts to other states by comparing labor market, health insurance, an emographic variables. From Appenix Figures B1 an B2, it is clear that aggregate labor market trens are similar in Massachusetts an other states before reform across a variety of ata sets. We also observe similar aggregate labor market trens in Massachusetts an other states after reform, which is not surprising given our moel. Our moel preicts ifferent movements in wages an hours base on ESHI status before an after reform. The overall impact on the labor market, however, is ambiguous an unlikely to be substantial unless a large share of the population change ESHI status. As shown in Figure 1, we expect iniviuals who switch from not having ESHI before the reform (equilibrium A) to having ESHI after the reform (equilibrium F ) to receive weakly lower wages an to work weakly lower hours. While these iniviuals are an important group from a policy perspective, they are only a small fraction of the aggregate population in the labor market. 25 We know from prior research that, though the reform resulte in a significant ecline in the percentage of people responses also reflect current coverage. Thus, the responses from March of 2008 may reflect some share of people who gaine coverage in Nevertheless, because the survey was very early in 2008, we woul expect much of the change in coverage to have occurre in our sample perio in For iniviuals who continue to have ESHI (equilibrium D to equilibrium F ), our moel preicts an ambiguous change in hours an wages, epening on the relative magnitues of the iniviual penalty an the subsiy µ x. For the small group of other iniviuals who switche from having ESHI before reform (equilibrium D) to not having it after reform (equilibrium B) (perhaps because subsiies became available),the moel also preicts an ambiguous change in wages an hours, epening on the relative magnitues of the unerlying valuation an the employer penalty. Therefore, while our moel makes clear preictions about changes in wages an hours for iniviuals base on changes in ESHI status, the impact of reform on wages an hours in the aggregate labor market is an empirical question. 20