The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
The Impacts of Physician Payments on Patient Access, Use, and Health∗ Diane Alexander† Federal Reserve Bank of Chicago Molly Schnell‡ Northwestern University and NBER July 14, 2019 Abstract We examine how the amount a physician is paid influences who they are willing to see. Exploiting large, exogenous changes in Medicaid reimbursement rates, we find that increasing payments for new patient office visits reduces reports of providers turning away beneficiaries: closing the gap in payments between Medicaid and private insurers would reduce more than two-thirds of disparities in access among adults and would eliminate disparities among children. These improvements in access lead to more office visits, better self-reported health, and reduced school absenteeism. Our results demonstrate that financial incentives for physicians drive access to care and have important implications for patient health. JEL: H51, H75, I13, I14, I18, I24 Keywords: physician behavior, financial incentives, primary care rate increase, fee bump, Medicaid, Affordable Care Act, school absenteeism ∗ We thank Marguerite Burns, Janet Currie, Pinar Karaca-Mandic, Anthony Keats, Ilyana Kuziemko, Nicole Maestas, Melissa McInerney, Matthew Notowidigdo, Heather Sarsons, Hannes Schwandt, Jessica Van Parys, Laura Wherry, and participants in seminars at Stanford University, University of Illinois at Chicago, University of Zurich, the Ifo Institute for Economic Research, the 2017 CSWEP CeMENT workshop, the 2017 Society of Labor Economists Annual Meeting, the 2017 Midwest Health Economics Conference, the 2017 Barcelona GSE Summer Forum, the 2017 International Health Economics Association Congress, the 2018 NBER Program on Children Meeting, and the 2018 and 2019 American Society of Health Economists Annual Conferences. Leah Plachinski provided outstanding research assistance. Generous financial support from the Griswold Center for Economic Policy Studies and the Center for Health and Wellbeing at Princeton University is gratefully acknowledged. This paper was previously circulated under the title “Closing the Gap: The Impact of the Medicaid Primary Care Rate Increase on Access and Health.” Any views expressed in this paper do not necessarily reflect those of the Federal Reserve Bank of Chicago or the Federal Reserve System. All errors are our own. † dalexand@frbchi.org; 230 South LaSalle Street, Chicago, IL 60604 ‡ schnell@northwestern.edu; 2211 Campus Drive, Evanston, IL 60208 I Introduction Understanding how physicians respond to financial incentives is key for designing policies to control health care costs while maintaining quality of care. Prior work illustrates the impact of payment levels on treatment choices, decisions over treatment intensity, and adoption of new technologies, suggesting that higher fees lead providers to do more once a patient is through their door (Clemens and Gottlieb, 2014; Coey, 2015; Gruber et al., 1999; Rice, 1983; Yip, 1998). However, to what extent financial incentives influence extensive margin decisions governing who physicians are willing to see—and whether these decisions affect the health of patients—remains an open question. This question is particularly important in light of significant disparities in access to care between the publicly and privately insured: in 2009, office-based physicians were 35 percent less likely to accept new patients covered by Medicaid than those covered by private insurance (Decker, 2012, 2013; MACPAC, 2011). Since Medicaid historically pays physicians less than two-thirds of what Medicare and private insurers pay for the same services, these disparities in access could be driven by differences in payment generosity (Zuckerman and Goin, 2012). Alternatively, this preference for the privately insured could be driven by complex patient needs, payment delays, and high denial rates that are known to plague the Medicaid system (Cunningham and O’Malley, 2009; Gottlieb et al., 2018; Long, 2013; Niess et al., 2018; Sloan et al., 1978). Faced with little causal evidence that low payment levels are to blame for disparities in access to care, policy makers often lower Medicaid payments in response to economic downturns and budgetary shortfalls (MACPAC, 2015; Smith et al., 2004). In this paper, we exploit exogenous variation in reimbursement rates to estimate the effects of physician payment levels on patient access, use, and health. Most of our identifying variation comes from a federal mandate that required states to increase their Medicaid payments to match federally regulated Medicare levels for select primary care services in 2013 and 2014.1 As states traditionally had wide latitude in setting their Medicaid payments, 1 Designated in Section 1202 of the Affordable Care Act (ACA), the rate increase was federally funded 1 reimbursement rates varied dramatically across states before the primary care rate increase went into effect. While Medicaid payments for select primary care services increased by an average of 60 percent as a result of the mandate, rates more than doubled in eleven states and were unchanged in two. We find that increased physician reimbursement causes statistically and economically significant improvements in access to care. Combining a new database of state-level Medicaid reimbursement rates for new patient evaluation and management services from 2009 to 2014 with measures of access from the restricted-access National Health Interview Survey (NHIS) files, we estimate that a $10 increase in Medicaid payments reduces reports of doctors telling adult Medicaid beneficiaries that they are not accepting new patients or their insurance by 13 and 11 percent, respectively.2 Among children covered by Medicaid, a $10 increase in Medicaid payments leads to a 25 percent decrease in parents reporting having trouble finding a doctor for their children. Notably, we find little evidence that these improvements in access among Medicaid beneficiaries are offset by negative spillovers to the privately insured. Our results indicate that closing the gap in payments between private insurance and Medicaid— a $45 increase in Medicaid payments for the median state—would close over two-thirds of disparities in access for adults and would eliminate such disparities among children. If Medicaid beneficiaries eventually receive treatment despite difficulties accessing care, increased payments could reduce search costs but have no impact on the use of services or health among patients. However, we find that increased reimbursement rates lead to greater usage and improved health among beneficiaries. Again using data from the NHIS, we find that a $10 increase in Medicaid payments leads to a 1.4 percent increase in the probability that beneficiaries visited a doctor in the past two weeks and a 1.1 percent increase in the proband was intended to ease the absorption of new Medicaid enrollees entering through the ACA’s Medicaid expansions by encouraging primary care physicians to participate in Medicaid (Blumenthal and Collins, 2014). The primary care services covered by the mandate included evaluation and management services and vaccine administration provided by physicians in family medicine, general internal medicine, and pediatric medicine. 2 Compared to the average baseline payment of $76, these improvements in access imply payment elasticities of physician willingness to accept new adult Medicaid beneficiaries of 0.83 to 1.01. 2 ability that beneficiaries report being in very good or excellent health.3 Using self-reported data on school absences from the NHIS and administrative data on school attendance from the National Assessment of Educational Progress (NAEP), we further find that a $10 increase in Medicaid payments leads to a 14 percent decrease in chronic absenteeism due to illness or injury and a 2 percent decrease in chronic absenteeism overall. When the federally mandated rate increase expired at the end of 2014, 34 states chose to return to their previous payment levels (MACPAC, 2015). This provides us with a second round of changes in reimbursement to exploit.4 Using data from 2013 to 2015, we find that the reduction in reimbursement rates following the expiration of the federal mandate had effects of similar magnitudes—but opposite signs—as the primary care rate increase itself. This suggests that many of the improvements in access, use, and health that Medicaid beneficiaries experienced when payments increased were lost when payments returned to their previous levels, providing clear evidence of the importance of reimbursement rates in driving physician behavior. Of course, changes in Medicaid payments stemming from the primary care rate increase did not occur in isolation. The U.S. health care system in general, and Medicaid in particular, experienced many other changes over our sample period. Most relevant for our analysis, 27 states and the District of Columbia expanded their Medicaid programs in 2014 to include coverage for low-income, childless adults. Although the timing is similar, our results are not confounded by the 2014 Medicaid expansions for at least four reasons. First, we find similar effects of changing reimbursement rates in states that did and did not expand their Medicaid programs under the ACA. Second, we find the strongest effects on access and health among children, whose eligibility was unaffected by the Medicaid expansions.5 Third, we estimate 3 Among beneficiaries, these estimates imply an elasticity of office visits (self-reported good health) with respect to Medicaid payments of 0.11 (0.08). 4 The decision not to extend the augmented payments may have depended on a state’s experience during the federal mandate. However, we find that states that ultimately did and did not extend the higher payments experienced similar improvements in outcomes as a result of the primary care rate increase. 5 Recent work demonstrates that children’s use of preventive services increases when Medicaid eligibility is extended among adults (Venkataramani et al., 2017). While such spillovers could influence our estimates of child health, it is unlikely that an improvement in cost-sharing among adults would make it easier for 3 similar effects when we truncate the sample period to exclude the 2014 Medicaid expansions. Finally, as noted above, we estimate similar effects of reimbursement rates using variation in payments stemming from the expiration of the federal mandate in 2015, a year after the majority of Medicaid expansions had gone into effect. While economists, public health researchers, and policy makers have long been interested in the effects of program generosity on access, use, and health among beneficiaries, causal analyses have been hampered by two important data limitations. First, before the primary care rate increase, most states had not made large changes to their Medicaid reimbursement rates in the last decades, and those that had chose to do so voluntarily.6 Previous research on program generosity has therefore had to rely on cross-sectional associations that likely suffer from omitted variable bias, case studies of single fee changes that may be confounded by time trends, and difference-in-difference models in which treatment is potentially endogenous.7 In contrast, we exploit a federal mandate that induced large, exogenous changes in Medicaid reimbursement rates across the United States. Second, the rise of Medicaid managed care that began in the early 1990s has made it difficult to know how much physicians are actually reimbursed through Medicaid. In a feefor-service system, state Medicaid programs pay providers a fixed amount for each service they provide. Although time consuming, these payment rates can be collected by contacting each state (as we do in this study).8 Under managed care, in contrast, states typically pay managed care organizations (MCOs) a fixed amount per beneficiary to provide all covered parents to find physicians willing to see their children. 6 Physician reimbursement rates in Medicare offer even less variation, as changes are made to a single, nationwide fee schedule. Furthermore, Medicare reimbursement rates for physicians have remained essentially the same for the past decade and will remain largely unchanged until at least 2025 under the Medicare Access and CHIP Reauthorization Act of 2015. 7 Cross-sectional studies: Cohen (1993); Cohen and Cunningham (1995); Hadley (1979); Long et al. (1986); Mitchell (1991); Showalter (1997); Sloan et al. (1978). Case studies: Adams (1994); Coburn et al. (1999); Fanning and de Alteriis (1993); Fox et al. (1992); Gruber et al. (1997). Difference-in-difference models: Atherly and Mortensen (2014); Baker and Royalty (2000); Buchmueller et al. (2015); Callison and Nguyen (2017); Chen (2014); Decker (2007, 2009); Shen and Zuckerman (2005). 8 We note, however, that previous work relies almost exclusively on payment rates from infrequently published secondary sources such as the Urban Institute’s Medicaid-to-Medicare fee index or the American Academy of Pediatrics Medicaid Reimbursement Reports. 4 services, and MCOs then pay providers. While over 60 percent of Medicaid beneficiaries are enrolled in comprehensive risk-based managed care plans, states—and in turn researchers— have little knowledge about how or how much MCOs actually pay physicians for the services that they provide. Importantly, the primary care rate increase required states to raise their Medicaid payments to achieve parity with Medicare levels for both their fee-for-service and managed care programs.9 We therefore know exactly how much physicians were reimbursed under Medicaid in 2013 and 2014. To estimate payments in the pre- and post-periods, we combine quarterly fee-for-service rates collected directly from each state’s Medicaid office with state-level Medicaid managed care to fee-for-service reimbursement ratios and data on the fraction of beneficiaries enrolled in each system in each state. We are thus able to examine the effects of changing physician payments on the entire Medicaid system, rather than just on the rapidly shrinking fee-for-service portion. Our work contributes to an ongoing debate on the effects of the Medicaid primary care rate increase on access to care. An early audit study found that the federal mandate was associated with increases in appointment availability among Medicaid patients in ten states (Polsky et al., 2015).10 In contrast, recent work by Decker (2018) found that Medicaid acceptance rates in an annual survey of 1,500 physicians did not increase during the primary care rate increase. Using claims data on office visits among a convenience sample of 11 percent of primary care physicians, Mulcahy et al. (2018) also found no association between the rate increase and office visits among Medicaid beneficiaries. This previous work has relied on small, selected samples and included limited—if any—information on the size of the rate increase in given states.11 9 As outlined in Appendix A.3, the Centers for Medicare and Medicaid Services (CMS) required states to submit proposals outlining how the increased reimbursements would be incorporated into the state’s capitation payments made to MCOs in 2013 and 2014. 10 Candon et al. (2018) replicates the analysis following the end of the mandate in 2015 and finds that appointment availability declined in the sampled states that did not extend the increased payments. 11 The studies by Decker (2018) and Mulcahy et al. (2018) further incorporate implementation delays that were associated with the rate increase. Since increased payments were made retroactively in states that experienced such delays, we would expect the behavior of physicians—who are largely not credit constrained—to 5 In contrast, we use comprehensive data that cover every state and exploit continuous variation in the magnitude of payment increases across states. Notably, we demonstrate that the effects of the federal mandate scale linearly with the size of the payment increase. This highlights that simple before-after designs—which average treatment effects across states that experienced payment increases of 0 to over 200 percent as a result of the mandate— lead to estimates that mask the true relationship between reimbursement rates and access to care. Additionally, we look beyond access alone and find that improvements in access resulting from increased payments lead to increased use and improvements in health. More broadly, our work relates to the literature that studies the effects of insurance coverage itself on the use of medical services and health outcomes. Using both randomized controlled trials (Baicker et al., 2013; Finkelstein et al., 2012; Manning et al., 1987) and natural experiments (Card et al., 2008; Currie and Gruber, 1996a,b; Sommers et al., 2012), researchers have documented that having health insurance increases the use of health care services and improves health. Recent work further demonstrates the impacts of health insurance on educational attainment, labor market outcomes, and financial hardship (Baicker et al., 2013, 2014; Brown et al., 2015; Cohodes et al., 2016; Finkelstein et al., 2012; Gross and Notowidigdo, 2011). Our work complements these studies by demonstrating that the positive impacts of having health insurance will be mediated by program generosity. The rest of the paper proceeds as follows. We provide an overview of our data in Section II. Section III introduces our empirical strategy and examines the impacts of increased payments on access to care, use of services, and health. Section IV probes the robustness of these findings. Section V examines the effects of the reduction in reimbursement rates resulting from the end of the federal mandate. Section VI discusses mechanisms and concludes. respond when the augmented payments went into effect on January 1, 2013. In fact, even though some states took until May 2013 to release the increased payments (MACPAC, 2015), our event studies demonstrate that physician behavior responded equally in 2013 and 2014. Incorporating payment delays therefore biases the results towards zero because some of the “pre-period” in such specifications was actually treated. 6 II Data We use four main data sources to document how physician reimbursement rates affect access to primary care, frequency of office visits, and health among patients. To measure physician reimbursement, we construct a new dataset containing Medicaid payments for new patient evaluation and management services for all states from 2009 to 2015. To measure patient access, use, and health, we use the NHIS. Finally, to corroborate the NHIS outcomes related to schooling, we use data on school absences and test scores from the NAEP. These datasets are supplemented with information from the Health Resources and Services Administration’s (HRSA) Area Resource Files (ARF) to control for spatial and temporal differences in sociodemographics and health care resources. II.A Medicaid Reimbursement Rates Our primary explanatory variable is the amount that Medicaid pays physicians for new patient evaluation and management services across states and over time. Under a fee-for-service system, there are five Medicaid reimbursement rates for these services, each corresponding to a specific length and complexity of visit (current procedural terminology (CPT) codes 99201-99205).12 We obtained historical payment data for these five codes by contacting the Medicaid offices of all 50 states and the District of Columbia.13 Our main results use reimbursement rates associated with the most commonly billed new patient evaluation and management code over our sample period: new patient office visits of mid-level complexity (CPT code 99203).14 Given the strong correlation between Medicaid payments for CPT 12 Some states have modifier codes that allow for different payment rates depending on patient characteristics, such as age. Although ideally we would incorporate modifier codes for children, in practice this is difficult due to varying age group carve-outs across states (e.g., “pediatric only” or under age 15, 20, or 21) and restrictions on the level of aggregation allowed for merging variables within the NHIS. We therefore use base rates in all states but note that Medicaid payments for children will on average be slightly higher. 13 We have complete payment data for 44 states and the District of Columbia. Appendix A.2 outlines the methodologies used to impute payment rates for the six states with partial payment histories. 14 Of new patient visits billed to Medicare in 2009, the relative billing frequencies across CPT codes 9920199205 were 3 percent, 19 percent, 43 percent, 27 percent, and 8 percent, respectively (Levinson, 2012). Our results are robust to using a billing frequency–weighted average across the five reimbursement rates for new 7 codes 99201-99205 within states over time (see Figures A1 and A2), all of our results are robust to using payments for these alternative CPT codes. The amount physicians are paid under fee-for-service Medicaid does not tell the full story, however, as over half of Medicaid beneficiaries are enrolled in managed care. We take managed care into account by creating an expected Medicaid payment measure that combines the state-level fee-for-service data with state-level managed care to fee-for-service payment ratios and state-level Medicaid managed care enrollment shares.15 In particular, we first use Medicaid managed care to fee-for-service payment ratios from the Government Accountability Office (GAO) for evaluation and management services performed in doctors’ offices to calculate Medicaid managed care payments from the fee-for-service rates.16 Using data from CMS on the percentage of Medicaid beneficiaries enrolled in managed care annually in each state (shown in Figure A3), we then define expected Medicaid payments at the statequarter level as the enrollment-weighted average of Medicaid fee-for-service and managed care payments.17 Both the initial geographic variation in Medicaid payment rates and the changes over our sample period are substantial. Figure 1 plots our constructed measure of Medicaid patient visits. Unfortunately, analogous reports are not available for Medicaid. 15 As previously noted, the primary care rate increase applied to both fee-for-service and managed care Medicaid programs. Appendix A.3 provides an overview of how states incorporated the increased reimbursement rates into their payments to MCOs in 2013 and 2014. 16 These payment ratios come from a GAO report documenting the difference between managed care and fee-for-service payments under Medicaid at the state level in 2010 (GAO, 2014). The report provides payment ratios for two de-identified states and eighteen identified states (Arizona, California, Connecticut, Florida, Georgia, Indiana, Michigan, New Jersey, New Mexico, New York, Ohio, Pennsylvania, Rhode Island, South Carolina, Texas, Virginia, Washington, Wisconsin). We use the recorded ratio for the states in the report and the median of 5 percent more under managed care for the missing states. Our results are robust to only using the states in the GAO report and to imputing missing states with the mean. 17 FFS That is, letting Rsqy denote the Medicaid fee-for-service reimbursement rate in state s in quarter q of MC R year y, RF F S denote the managed care to fee-for-service payment ratio under Medicaid in state s in s,2010 MC 2010, and %Bsy denote the fraction of Medicaid beneficiaries enrolled in a managed care plan in state s in year y, the expected Medicaid reimbursement rate in each state-quarter before and after the primary care rate increase is approximated by MC R MC FFS MC FFS ˜ Rsqy = (1 − %Bsy ) · Rsqy + %Bsy · Rsqy · RF F S s,2010 8 payments at the state-quarter level from 2009 to 2015.18 In the first quarter of 2009, the expected Medicaid payment for a new patient office visit of mid-level complexity ranged from $37 in Minnesota to $160 in Alaska. Few states made meaningful changes to their reimbursement rates in the next three years: between 2009 and 2012, Medicaid payments for new patient office visits increased by an average of only $4.27 across states, with more than half of states making no changes to their payment schedules for evaluation and management services. This persistence of payments is one of the difficulties that previous work has faced in measuring how physicians respond to changes in program generosity. When the primary care rate increase went into effect in 2013, the range tightened, with states paying physicians between $101 (Alabama) to $171 (Alaska).19 Figure 2 provides maps of Medicaid payments across the United States in 2009, 2012, and 2013. The shading in the maps reflects state-level quintiles of Medicaid payments as defined in 2009. As shown in Figure 2, the primary care rate increase was sufficient to push all states into the top quintile of reimbursement rates from 2009. When the primary care rate increase was initially passed, it was unclear whether federal funding for the increased payments would extend beyond 2014. In the end, the funding was not extended, and in 2015, 34 states chose to return to their previous payment levels (see Figure A4). While this provides another large change in payment rates, states may have made this decision based on their experience during the primary care rate increase (MACPAC, 2015). Thus, in our main analysis we do not use variation in Medicaid payments stemming from the expiration of the federal mandate. Instead, we examine the effects of this reverse experiment on outcomes separately and explore the potential endogeneity concerns. Although the federal government mandated that states increase select Medicaid payments to primary care providers starting on January 1, 2013, many states experienced im18 Refer to Figure A2 for plots of quarterly fee-for-service Medicaid payments for CPT codes 99201-99205 from 2009 to 2015 for each state. 19 The remaining variation across states comes from two sources. First, Medicare payment levels vary across locations due to adjustments for geographic and market area differences. Second, Alaska and North Dakota maintained Medicaid payment rates that exceeded federally mandated Medicare levels over the sample period. 9 plementation delays (MACPAC, 2015). We do not incorporate state-level variation in the implementation of the primary care rate increase into our Medicaid payment variable; that is, we use the payment rates reported by the state as effective in each month and year. Because states with implementation delays were required to retroactively pay physicians the difference between the amount paid and the enhanced Medicaid rate, we believe that the behavior of physicians—who are largely not credit constrained—should respond at the start of the rate increase rather than when the higher payments were actually released. Finally, we can only expect physician behavior to respond to increased payments if providers are aware of changes in reimbursement. We note that the primary care rate increase was covered widely by news outlets: for example, The Washington Post published an article on December 21, 2012—before the federal mandate went into effect—titled “Obamacare is about to give Medicaid docs a 73 percent raise” (The Washington Post, 2012). As Medicaid payments more than doubled for primary care providers in some states, it is reasonable that physicians would take notice of the change, and thus there is scope for physician behavior to respond. II.B National Health Interview Survey The NHIS is the largest in-person household survey that tracks health care access, health care utilization, and health outcomes across the United States. While many surveys measure health patterns, the NHIS is well-suited for our study for two reasons. First, in contrast to most other surveys, the NHIS is large enough to be used for state-level estimates. This allows us to exploit state-level variation in Medicaid reimbursement rates over time. Second, while many surveys ask respondents whether they have health insurance, the NHIS collects information on insurance provider. This allows us to look separately at patients with private insurance and Medicaid beneficiaries. We use outcomes from three NHIS sample components in our analysis: the family file, the sample child file, and the sample adult file. The family component collects demographic 10 information and answers to basic questions (e.g., health status) for all members of a family. The sample child and sample adult components each sample one child and one adult in the family and ask a longer list of more detailed questions (e.g., days of school or work missed in the past year). Sample sizes are thus more limited when working with questions asked in the sample child or sample adult files relative to the full family sample. To measure access to health care services, we consider whether respondents report difficulty with doctors either not accepting new patients or not accepting their insurance.20 For children, we further consider indicators denoting whether parents report having difficulty finding a doctor to see their child and whether their child has a usual place of care. To measure use of health care services, we consider whether respondents report having had an office visit in the past two weeks. As seen in Table 1, Medicaid beneficiaries and the privately insured have a similar likelihood of visiting a doctor in the past two weeks. However, those covered by Medicaid are more than twice as likely to report difficulties finding physicians who are willing to accept them as new patients. Policies targeting health care access do so with the hope that improving access will improve health. To examine whether higher reimbursement rates lead to better health, we consider indicators denoting whether people rate their health as excellent/very good or fair/poor. We further consider the number of work days adults report having missed and the number of school days parents report their child having missed in the past year. Notably, the NHIS asks specifically about school absences due to illness or injury, allowing us to focus on the category of absences most likely to be sensitive to changes in access to primary care. We focus on chronic absenteeism when considering school absences, which we define as missing fourteen or more days of school in the past year.21 20 The exact survey questions used are outlined in Appendix A.1. All questions were asked throughout our full sample period except those asking whether children and adults had trouble finding a doctor, which started in 2011. 21 Chronic absenteeism is linked to low academic achievement, including test scores, test score growth, and on-time graduation rates (Buehler et al., 2012; Connolly and Olson, 2012; Schanzenbach et al., 2016; Spradlin et al., 2012; Utah Education Policy Center, 2012). Unlike average absenteeism, rates of chronic absenteeism vary widely across schools, and all but thirteen states use chronic absenteeism in their accountability system under the Every Student Succeeds Act (Bauer et al., 2018). 11 As shown in Table 1, baseline differences in health between Medicaid beneficiaries and the privately insured are large. Compared to respondents with private insurance, Medicaid beneficiaries are almost three times more likely to report being in fair or poor health, and children covered by Medicaid are twice as likely to be chronically absent. To account for differences in demographics and the availability of medical resources across locations and over time, we control for individual demographics from the NHIS and countylevel characteristics from the ARF. Table 2 reports summary statistics for individual and county-level characteristics by insurer. Relative to the privately insured, Medicaid beneficiaries have lower income and education levels, live in larger families, are less likely to be married, and are more likely to be black or Hispanic. Respondents covered by Medicaid also live in poorer, more densely populated areas. Although much of the NHIS data is publicly available, geographic identifiers for areas smaller than Census regions are restricted. In order to link our outcome measures to state-level variation in Medicaid reimbursement rates and county-level health resources, we obtained access to confidential state and county identifiers. All of our analyses using the NHIS are therefore conducted in a Census Research Data Center. II.C National Assessment of Educational Progress To examine whether increased payments to physicians lead to better educational outcomes among low-income children, we supplement self-reported days of missed school from the NHIS with administrative data from the NAEP. The NAEP is a congressionally mandated assessment that provides information on reading and mathematics performance in grades 4 and 8 every other year in all states. Not all schools are tested in each wave, although the schools and students participating in NAEP are selected to be representative of all schools nationally and of public schools at the district level. Importantly for our work, the NAEP reports the fraction of children who missed 0, 1–2, 3–4, 5–10, and 11 or more days of school (for any reason) in the month preceding their 12 national assessment exam.22 As in the NHIS, we focus on the fraction of chronically absent children. When considering absences over the period of a month (rather than over the course of a year, as in the NHIS), chronic absenteeism is commonly defined as three or more days of missed school (KewalRamani et al., 2007; Ginsburg et al., 2014; Schanzenbach et al., 2016). Figure A5 shows the distribution of absences averaged over math and reading assessments in grades 4 and 8. There are large differences by grade, with a larger fraction of students reporting zero absences in the past month in grade 4 than in grade 8. As was seen in the NHIS data, there are large differences in the number of absences by socioeconomic status. Although we do not observe whether children are covered by Medicaid in the NAEP data, we can identify children that are eligible to receive free school meals. Like Medicaid, free school lunch is a means-tested program; according to income eligibility limits for each program, all children who are eligible for free school meals are also eligible for Medicaid (but not vice versa).23 In grade 4, 53 percent of children ineligible for free lunch missed zero days in the past month compared to 44 percent among students eligible for free lunch. The discrepancy in school absences by free lunch eligibility is similar in grade 8, though fewer children report zero absences in both groups. In all grades and subjects, average test scores are monotonically decreasing in the number of school days missed in the past month (see Figure A6). Given the negative correlation between absences and test scores, it is possible that test scores could be affected by changes in access to primary care. We therefore also look at the effects of physician reimbursement on average state-level performance on national math and reading assessments. 22 While the NAEP data does not include information on absences due specifically to illness or injury (as in the NHIS), most school absences—particularly among young children—are attributable to either acute illnesses such as respiratory infections and gastroenteritis or chronic childhood diseases such as asthma (Ehrlich et al., 2014; Neuzil et al., 2002; Wiseman and Dawson, 2015). As these conditions are commonly treated in a primary care setting, overall school absenteeism may be responsive to changes in access to primary care. 23 Across the United States, children in households with income at or below 130 percent of the federal poverty level are eligible to receive free meals at school (FRAC, 2018). Medicaid eligibility requirements vary by state, although the federal minimum income limit for children’s health coverage is 138 percent of the federal poverty level with a median income limit of 149 percent across states (Kaiser Family Foundation, 2018). 13 III Physician Payments and Patient Access, Use, and Health The summary statistics in Table 1 demonstrate that those covered by Medicaid face greater difficulty accessing health care services and have worse health than the privately insured. To investigate whether differences in physician reimbursement contribute to these differences in outcomes, we examine the effects of changes in physician payments under Medicaid on access, use, and health among beneficiaries. Sections III.A through III.C consider a range of outcomes from the NHIS, while Section III.D turns to educational outcomes from the NAEP. We focus on the impacts of the increase in Medicaid payments stemming from the onset of the primary care rate increase in 2013 throughout Section III; Section V considers the effects of the reduction in Medicaid payments following the expiration of the federal mandate in 2015. III.A Raw Data We begin by examining patterns in the raw data. To do so, we divide states into deciles based on the size of the payment increase that they experienced under the Medicaid primary care rate increase. Figure 3 plots the average change in various outcomes in the two years after the payment increase (2013–2014) versus the two years before (2011–2012) against the average payment increase in each decile. We plot two lines for each outcome—one for Medicaid beneficiaries and one for privately insured patients—that depict the best fit line through these points. We adjust the outcomes such that higher values denote better outcomes; an increasing slope therefore indicates that larger payment increases are associated with larger improvements in a given outcome. Across a range of measures, we see that Medicaid beneficiaries in states with larger increases in Medicaid payments saw greater improvements in access, frequency of office visits, and health. For example, in the upper left subplot of Figure 3 we see that Medicaid 14 beneficiaries in states in the lowest decile of payment increases (average increase of $17.43) experienced little change in the probability of having an office visit in the past two weeks following the payment increase, whereas Medicaid beneficiaries in states in the highest decile of payment increases (average of $88.41) experienced an average increase of nearly 6 percent. Notably, across most outcomes there is no association between changes in Medicaid payments and changes in outcomes among privately insured patients; that is, the line is flat. III.B Event Studies To examine the timing of the effects and to control for differences across individuals and locations, we estimate event study specifications. In particular, letting ∆P ayments = P ayments,2013Q1 − P ayments,2012Q4 denote the change in Medicaid payments resulting from the primary care rate increase in state s, we estimate the following equation: Outcomeicsy = β0 + βy ∆P ayments ∗ λy + γXi + δZcy + λs + icsy (1) where Outcomeicsy denotes an outcome for Medicaid beneficiary i living in county c in state s in year y; Xi and Zcy are vectors of individual and county characteristics (listed in Table 2), respectively; and λs and λy are state and year fixed effects, respectively. By scaling the association between time and the outcome by the extent of the treatment, this specification exploits the full variation in Medicaid payments induced by the primary care rate increase. As in the raw data analysis, we adjust the outcomes such that higher values are indicative of better outcomes. We use the sample weights provided in the NHIS and cluster standard errors by state. Figure 4 plots the βy s from Equation (1). The coefficients before the primary care rate increase—β̂2009 through β̂2012 —are statistically indistinguishable from zero, indicating that the outcomes were stable before the federal mandate. Following the rate increase, however, there are persistent, significant increases in many of the outcomes. For example, the bottom 15 left subplot indicates that Medicaid beneficiaries saw improvements in physicians’ willingness to accept new patients when Medicaid reimbursement rates increased in 2013 and 2014. The effects are immediate for most outcomes, although there is some evidence that health effects—such as patients reporting their health as excellent or very good—accrue over time.24 III.C Regression Analysis Figure 4 demonstrates that increased Medicaid payments lead to improved outcomes among Medicaid beneficiaries. To quantify the effects of physician reimbursement on access, use, and health, we estimate the following specification: Outcomeicsqy = β0 + β1 P aymentsqy + γXi + δZcy + λs + λqy + icsqy (2) where Outcomeicsqy denotes an outcome for respondent i living in county c in state s in quarter q of year y, P aymentsqy is our primary measure of Medicaid payments in state s in quarter q of year y (see Section II.A), λqy are quarter-year fixed effects, and all other variables are defined as in Equation (1).25 We divide payments by $10 such that β1 represents the effect of a $10 increase in Medicaid payments.26 For the outcomes covering a retrospective time period of twelve months, the payment variable is the average Medicaid payment over the past four quarters; for all other outcomes we use the average payment in the quarter of the interview. Since we include state and quarter-year fixed effects, our main coefficient 24 Coefficients for 2013 may also be smaller because the reference periods for some questions include part of the pre-period. For example, when respondents are asked about difficulty finding a doctor over the past twelve months, respondents who were interviewed before the end of 2013 will include some of their experience from before the rate increase. Table A1 lists the reference window for each outcome. 25 Recall from Figure 1 that some states made minor adjustments to their Medicaid payments between 2009 and 2012 (over our sample window but before the federally mandated primary care rate increase). Most of our estimates of Equation (2) include these changes, although Figure 5 confirms that our results are robust to excluding variation in payments from before the federal mandate. 26 As shown in Figure 3, the relationship between increases in Medicaid payments and changes in our outcome variables is approximately linear. We therefore prefer a linear specification both because it is suggested by the data and because it allows for the coefficients to be easily interpreted as the effect of a $10 increase in payments. We can, however, use a specification in which we consider log(P aymentsqy ) on the right-hand side. We note that the elasticities implied from both specifications are quantitatively similar. 16 of interest, β1 , is identified by changes in Medicaid payments within states over time. As before, all regressions use the sampling weights provided by the NHIS, and standard errors are clustered by state. Results from estimation of Equation (2) are presented in Table 3. The first three columns of each panel show the effects of changes in Medicaid payments on survey respondents covered by Medicaid. Looking first to columns (1) and (2) of Panel B, we see that a $10 increase in Medicaid payments leads to a 0.54 percentage point decrease in the probability that parents report difficulty finding a doctor to see their child covered by Medicaid and a 0.36 percentage point decrease in the likelihood that the child has no usual place of care (reflecting decreases relative to the mean of 25 and 11 percent, respectively). Among adult Medicaid beneficiaries, a $10 increase in Medicaid payments causes both a 0.82 percentage point reduction in the probability of being told that a physician is not accepting new patients and a 0.89 percentage point reduction in the probability of being told that one’s insurance is not accepted (decreases of 13 and 11 percent of the mean, respectively; see columns (1) and (2) of Panel C). Notably, these improvements in access lead to more use: in the full sample, a $10 increase in Medicaid payments increases the probability that respondents covered by Medicaid had an office visit in the past two weeks by 0.28 percentage points (1.4 percent relative to the mean; column (1) of Panel A). In addition to improved access and use, increases in Medicaid payments lead to better health among the program’s beneficiaries. A $10 increase in physician reimbursement reduces the probability that beneficiaries report being in fair or poor health by 0.31 percentage points (1.8 percent of the mean; column (2) of Panel A) and increases the probability that beneficiaries report being in excellent or very good health by 0.62 percentage points (1.1 percent of the mean; column (3) of Panel A). Among young children covered by Medicaid, a $10 increase in Medicaid payments reduces the probability of being chronically absent due to illness or injury by 0.65 percentage points (a decrease of 14 percent of the mean; column 17 (3) of Panel B).27 There is no reduction in illness-related chronic absenteeism among older children covered by Medicaid (column (4) of Panel B); this is likely due to the fact that absences among older children are less closely tied to health (Ehrlich et al., 2014; Neuzil et al., 2002; Wiseman and Dawson, 2015). We further find no reduction in days of work missed among adult Medicaid beneficiaries (column (3) of Panel C). To get a sense of what these effects imply for the typical state under the primary care rate increase, we consider the effects of an increase in Medicaid payments of $35—the median increase in Medicaid payments across states from the fourth quarter of 2012 to the first quarter of 2013. Multiplying the point estimates in Table 3 by 3.5, we see that an increase of $35 in physician reimbursement under Medicaid leads to a 5.0 percent increase in the probability of having visited a doctor’s office in the past two weeks, a 6.2 percent decrease in the probability of being in fair or poor health, and a 3.9 percent increase in the probability of being in very good or excellent health among beneficiaries. Applying the same calculations to the access measures further indicates that the Medicaid primary care rate increase nearly eliminated parents having trouble finding doctors for their Medicaid-covered children and approximately halved these difficulties for adult beneficiaries in the median state. We can compute elasticities by comparing the effects of a $10 increase in Medicaid payments in percentage terms to the corresponding percent change in Medicaid payments implied by a $10 increase.28,29 As reported in column (4) of Table 4, our results imply elasticities with respect to Medicaid payments of physician willingness to accept new adult Medicaid patients of 0.83, office visits among beneficiaries of 0.11, and self-reported good health among beneficiaries of 0.08. The implied elasticities for access among children are even more pro27 We find similar results when we consider a continuous measure of school absences rather than an indicator for chronic absenteeism. As shown in Table A2, a $10 increase in Medicaid payments leads to an average reduction of 0.23 days of school missed due to illness or injury per year among young children covered by Medicaid, a 6.4 percent reduction relative to the mean. We find no effects of Medicaid payments on illness-related school absences for older children covered by Medicaid or for children with private insurance. 28 Compared to the average baseline Medicaid payment of $76 for a new patient office visit of mid-level complexity, a $10 increase in payments corresponds to a 13.2 percent increase. 29 Alternatively, we can calculate elasticities by including payments in logs instead of in levels in Equation (2). The elasticities from this alternative specification are very similar to those reported in Table 4. 18 nounced, suggesting that physicians are more elastic to payments for children. Although billing difficulties known to plague the Medicaid system should not depend on beneficiary age (Cunningham and O’Malley, 2009; Gottlieb et al., 2018), providers report that adult Medicaid beneficiaries have a wider breadth of needs, which makes managing their cases more difficult than those of children or patients with private insurance (Long, 2013; Niess et al., 2018).30 It is therefore reasonable that physician behavior would be more responsive to Medicaid payments for children. While we find strong evidence that increasing physician reimbursement under Medicaid improves access and health among the program’s beneficiaries, there is little evidence of spillovers to the privately insured. The last three columns of Table 3 present analogous estimates for privately insured respondents, who may be indirectly affected by Medicaid patients becoming relatively more attractive to physicians. However, we find no change in access, use, or health among the privately insured when Medicaid payments increase, with the exception of parents having slightly more trouble finding a doctor for their children (an increase of 0.13 percentage points, significant at the 10 percent level). Not only are the coefficients nearly all statistically insignificant despite large sample sizes, but across all outcomes the point estimates are much smaller than those observed among respondents covered by Medicaid. These effects have large implications for disparities in access to care between the publicly and privately insured. Column (3) of Table 4 reports baseline disparities in our outcome measures between Medicaid beneficiaries and patients with private insurance. Columns (6) through (8) show how much of this disparity is reduced by increasing Medicaid payments by $10, $35, and $45, respectively. As shown in column (7), increasing Medicaid payments by $35—the median increase under the primary care rate increase—reduces disparities in reports of doctors telling adult patients that they are not taking new patients or their insurance by 30 Niess et al. (2018) surveyed 806 physicians in Colorado to assess their beliefs and attitudes about adult Medicaid patients. Eighty-six percent of respondents had an unfavorable attitude toward adult Medicaid beneficiaries, with respondents most likely to agree that “socially complicated” and “medically complicated” described a typical adult Medicaid patient. 19 64 and 55 percent, respectively. Closing the gap in payments between private insurance and Medicaid—a $45 increase in Medicaid payments for the median state at baseline31 —closes over 80 percent of the gap in reports of doctors not taking new adult patients and over two-thirds of the gap in reports of doctors not taking an adult patient’s insurance. Because providers are more elastic to payments for children, it is easier to close gaps in access: as shown in column (9) of Table 4, it would take an increase in Medicaid payments of about $26 on average to eliminate disparities in access between children with private insurance and children with Medicaid.32 III.D Educational Outcomes in the NAEP All of the measures in the NHIS, including days of school missed, are self reported. To corroborate our finding that increased reimbursement rates reduce absenteeism among children— and to examine whether these reductions in absenteeism lead to improvements in test scores— we use administrative data from the NAEP. Here, we use a specification similar to Equation (2) but at the state-year level: Outcomesy = β0 + β1 P aymentsy + λs + λy + sy (3) where Outcomesy denotes an average schooling outcome in state s in year y, and λs and λy 31 We calculate the median difference in reimbursement rates between private insurance and Medicaid at baseline by combining private insurance to Medicaid payment ratios for evaluation and management services from the GAO with our data on Medicaid payments. The GAO data that we use documents the difference between private insurance and Medicaid payments for 32 states in 2010 (GAO, 2014). 32 As shown in column (7) of Table 4, the median increase in Medicaid payments of $35 under the federal mandate was sufficient to close more than the disparity in outcomes between children with private insurance and children with Medicaid; this suggests that children on Medicaid were more attractive to physicians than children with private insurance after the rate increase. As noted in footnote 12, state Medicaid programs often pay slightly more for children than for adults. The median payment increase of $35 will therefore close more of—or may even go beyond—the gap in payments between Medicaid and private insurance for children. Furthermore, we note that we do find some, albeit weak, evidence of negative spillovers to children with private insurance as a result of the Medicaid primary care rate increase: as shown in Table 3, a $10 increase in Medicaid payments increases reports of parents having trouble finding a doctor to see their child with private insurance by 0.13 percentage points, or 16 percent relative to the mean (significant at the 10 percent level). 20 are state and year fixed effects, respectively.33 As all state assessments take place between January and March, P aymentsy is the expected Medicaid payment in state s in the first quarter of year y. Standard errors are again clustered by state. Results from estimation of Equation (3) are presented in Table 5. Panel A shows the effects of changes in Medicaid payments on outcomes among students who qualify for free lunch, our proxy for Medicaid eligibility in the NAEP. For low-income children in the 4th grade, a $10 increase in Medicaid payments reduces the fraction of students who missed three or more days in the past month for any reason by 0.49 percentage points (2 percent relative to the mean; column (3)) and increases the fraction of students with zero absences by 0.46 percentage points (1 percent relative to the mean; column (1)). Columns (1) and (3) in Panel B show that no such pattern is present for children who do not qualify for free lunch (and are thus unlikely to be covered by Medicaid).34 Similar patterns are present for students in grade 8, though the point estimates are less precise. The larger effects in grade 4 relative to grade 8 likely reflect the fact that absences for younger children are more closely tied to health status, whereas absences for older children are more likely for non-health-care related reasons, such as truancy (Ehrlich et al., 2014; Neuzil et al., 2002; Wiseman and Dawson, 2015). As school absenteeism is strongly linked to test scores, it is possible that these reductions in absenteeism lead to improvements in student performance. However, we find no effects of increased physician reimbursement under Medicaid on average state-level scores on national math and reading assessments (Table A3). It is possible that more prolonged improvements in attendance are necessary to cause improvements in student performance. 33 There was little variation in most state-level characteristics over our sample period. As we include state fixed effects, our results are therefore quantitatively very similar if we include a range of time-varying, state-level controls. Our results are further robust to weighting by population. 34 Recall that the NAEP covers absences for any reason whereas the NHIS asks specifically about school absences due to illness or injury. It is therefore not surprising that we find larger effects in percentage terms when considering school absenteeism in the NHIS than in the NAEP. 21 IV IV.A Robustness 2014 Medicaid Expansions In 2014, 27 states and the District of Columbia expanded their Medicaid programs to extend coverage to low-income, childless adults. Although the timing of these expansions and the Medicaid primary care rate increase were similar, we are confident that our results are not confounded by the Medicaid expansions for several reasons. First, recall that states were required to raise their Medicaid payments to match Medicare levels for select primary care services beginning in January 2013, a year before most of the expansions. As shown in Figure 4, most of the effects of the rate increase are realized before 2014. Second, we find the largest effects on access and health among children, whose eligibility was unaffected by the expansions. Third, as shown in Section V below, we estimate similar effects of reimbursement rates on patient outcomes using variation in payments stemming only from the federal mandate expiring in 2015, a year after the majority of Medicaid expansions had gone into effect. Nevertheless, we conduct three additional analyses to verify that our results are not confounded by the Medicaid expansions. In particular, we estimate Equation (2) including only: (1) the years before the 2014 Medicaid expansions (2009–2013), (2) states that did not expand their Medicaid programs in 2014, and (3) families with children. The top rows of each subfigure in Figure 5 compare the estimates from our main sample with the results from these subsample analyses. Looking first to the results using data from 2009–2013 only, we see that our estimates are remarkably consistent when we exclude 2014. While some of our estimates lose precision when we only consider states that did not expand Medicaid, the general pattern of results is consistent with our main findings. Finally, we see that—if anything—our results are often stronger among households with children. Taken together, these results provide strong evidence that we are identifying changes in access, use, and health driven by changes in program generosity, not program eligibility. 22 IV.B Medicaid Payment Variable As outlined in Section II.A, we create expected Medicaid payment rates by combining: (1) state-level reimbursement rates under fee-for-service Medicaid collected directly from state Medicaid offices, (2) state-level Medicaid fee-for-service to managed care payment ratios from the GAO, and (3) state-level Medicaid managed care enrollment shares from CMS. While we have Medicaid fee-for-service rates and Medicaid managed care enrollment shares for all states, the GAO report only provides payment ratios for 20 states.35 In our main analysis, we use the median payment ratio among states in the GAO report (5 percent more under Medicaid managed care) for states that are not in the GAO data. To probe the robustness of our results to this imputation, we replicate our main results: (1) imputing states that are not in the GAO report with the mean payment ratio of 14 percent more under Medicaid managed care, (2) only using states in the GAO report, and (3) only using variation stemming from Medicaid fee-for-service payment rates. Results from these additional analyses are presented in Figure 5. Across all outcomes, imputing missing states with the mean payment ratio instead of the median has very little impact on the results. Narrowing the sample to only the 20 states in the GAO report tends to decrease the precision of our estimates, but the magnitudes of the effects are very similar to our primary specification. Finally, despite the fact that nearly 60 percent of Medicaid beneficiaries are enrolled in managed care, our results are very similar if we only consider fee-for-service reimbursement rates. This is not surprising given that most of the residual variation in our measure of Medicaid payments when controlling for state and time fixed effects comes from changes in fee-for-service payments within states over time.36 35 As previously noted, we have complete payment information for 44 states and the District of Columbia. Appendix A.2 outlines the methodologies used to impute payment rates for the six states with incomplete payment histories. Given that only a few imputations are required, our results are robust to only using non-imputed data and to using alternative imputation strategies. 36 Recall that variation in our Medicaid payment variable comes from three sources: (1) time-series variation in state-level fee-for-service payments, (2) cross-sectional variation in state-level Medicaid managed care to fee-for-service payment ratios, and (3) time-series variation in the fraction of Medicaid beneficiaries enrolled in managed care in each state. With the inclusion of state fixed effects, residual variation in payments comes only from (1) and (3). While the fraction of Medicaid beneficiaries enrolled in managed care varies 23 Recall that our primary payment variable includes all variation in state-level Medicaid payments for new patient office visits of mid-level complexity between 2009 and 2014. As previously noted, states that adjusted their reimbursement rates before the federal mandate chose to do so voluntarily, and thus the payment changes may be endogenous. The final row in each subfigure in Figure 5 replicates our main results using variation in Medicaid reimbursement rates stemming only from the federally mandated primary care rate increase. To do so, we impute state-level reimbursement rates from 2009 through the third quarter of 2012 with the relevant payment rate from the fourth quarter of 2012. As the overwhelming majority of variation in Medicaid reimbursement rates between 2009 and 2014 was driven by the federal mandate in 2013 (see Figure 1), our results are nearly identical if we ignore variation in payments between 2009 and 2012. IV.C Triple Difference Model In our preferred empirical specification, we conduct analyses separately among Medicaid beneficiaries and patients covered by private insurance. We look separately at these two groups, rather than using the privately insured as a control group, as changes in relative reimbursement rates could influence the treatment of individuals with private insurance. If, for example, increases in Medicaid payments lead physicians to see fewer patients with private insurance, then a triple difference strategy using patients with private insurance as a “control” group would overestimate improvements among Medicaid beneficiaries. However, as this strategy has been used previously when examining the impacts of changing reimbursement rates (see, for example, Atherly and Mortensen, 2014; Callison and Nguyen, 2017; and Shen and Zuckerman, 2005), we provide estimates from a triple within a state over time (see Figure A3), the vast majority of our identifying variation comes from changes in fee-for-service payments. 24 difference model for comparison. In particular, we estimate the following equation: Outcomeicsqy = β0 + β1 P aymentsqy + β2 I {M edicaidi } +β3 P aymentsqy ∗ I {M edicaidi } (4) +γXi + δZcy + λs + λqy + icsqy where I {M edicaidi } is an indicator denoting whether respondent i is a Medicaid beneficiary and all other variables are defined as in Equation (2). We only include patients with private insurance and Medicaid beneficiaries; that is, we exclude those without insurance and those covered by Medicare. Table 6 shows that the pattern of results from the triple difference model is very similar to that found using only Medicaid beneficiaries in a difference-in-difference framework. The similarity is not surprising given the minimal evidence of spillovers on the privately insured that we saw in Table 3. V Expiration of the Primary Care Rate Increase Both the federally mandated Medicaid primary care rate increase and the accompanying federal funding expired at the end of 2014. Beginning in 2015, states could therefore choose either to maintain the payments at higher levels—and pay for the higher payments themselves—or revert to their original payments. As shown in Figure A4, 34 states chose not to extend the increased payments; in these states, Medicaid reimbursement rates returned to their December 2012 levels in January 2015. While the reduction in Medicaid reimbursement rates resulting from the end of the federal mandate provides a second round of changes in reimbursement, the decision not to maintain the higher payments could be endogenous. Notably, states that experienced greater success under the federally mandated rate increase—that is, states in which the rate increase led to larger improvements in access, use, and health among Medicaid beneficiaries—may have 25 been more likely to extend the increased rates.37 To examine whether the primary care rate increase had smaller impacts in states that chose not to maintain the higher payments, we replicate our analysis using only the subset of states that did not extend the increased rates beyond 2014. Although we divide the sample by a decision made at the end of 2014, we only use variation in Medicaid payments stemming from the onset of the primary care rate increase in 2013; that is, we consider the effects of the primary care rate increase switching on in states that ultimately decided to switch it off. As shown in the top two rows of each subfigure in Figure 6, states that chose not to extend the higher payments saw improvements in outcomes during the federal mandate that were similar in magnitude to those experienced by the average state.38 While there is some evidence that states that did not extend the increased payments experienced smaller improvements in access, the point estimates for effects on use and health suggest that these states experienced similar—and if anything larger—improvements in downstream outcomes during the primary care rate increase. This suggests that states chose to return their payments to previous levels despite significant improvements in outcomes resulting from the increased payments. There are a number of reasons a state’s decision over whether to extend the payments may have been unrelated to its experience during the federally mandated rate increase. 37 Alternatively, states that did not extend the increased payments could have had less pronounced effects because providers in those states knew that the payments would be temporary (and physician-patient relationships tend to last for many years). However, it was difficult to forecast whether the rates would ultimately be extended. Notably, it was not announced until late 2014 that the increased payments would not receive federal funding in 2015; if the federal funding were to have persisted, it seems likely that all states would have maintained the higher rates. Furthermore, the states that chose to extend the increased payments show diversity in geography, demographics, and political affiliations (see Figure A4), so it is unlikely that providers could have predicted whether the payments would be extended based on state-level characteristics alone. 38 Ideally, we would estimate our primary specification separately among states that did and did not extend the payments. However, because only sixteen states opted to extend the payments, we lose precision and run into potential disclosure violations when we limit the sample to this subset of states. Alternatively, we can estimate an augmented version of Equation (2) that includes a state-level indicator denoting whether a state extended the increased payments beyond 2014 and an interaction between the payment variable and this indicator. As shown in Table A4, the interaction is not significant for any outcome other than days of work missed, demonstrating that states that did and did not ultimately extend the increased payments experienced similar effects from the mandated rate increase. 26 First, federal funding for the increased payments expired with the mandate. Budgetary considerations could therefore have led states to lower payments even if they were aware of the implications for the health care of Medicaid beneficiaries. Second, until this point, little comprehensive evidence has existed to demonstrate that the primary care rate increase had significant impacts on access, use, and health among Medicaid beneficiaries. Notably, a small survey of Medicaid officials, plan administrators, and provider organizations conducted by the Medicaid and CHIP Payment and Access Commission in the summer of 2014 suggests that states believed that the primary care rate increase had little impact on access to primary care (MACPAC, 2015). We therefore consider the effects of the primary care rate increase expiring in 2015 on access, use, and health among Medicaid beneficiaries. To do so, we estimate a specification analogous to Equation (2) that instead exploits variation in payments stemming only from the federal mandate expiring in 2015. Although the variation comes from payment decreases, the estimated coefficients again represent the effects of a $10 increase in Medicaid payments. As shown in the bottom row of each subfigure in Figure 6, the effects of reimbursement rates using variation stemming from the expiration of the primary care rate increase look similar, although slightly smaller, than the effects of reimbursement rates using variation stemming from the onset of the primary care rate increase. While we lose precision when considering a subset of states and years, comparing the point estimates between the final two rows in each subfigure—which hold the sample of states constant—suggests that the rate increase switching off had effects of similar magnitudes on access and use but smaller effects on health relative to the rate increase switching on. This could be because there is more persistence in health than physician and patient behavior. It is notable that we find similar effects of physician payments using variation stemming from a payment increase and a payment decrease. One could imagine that providers would adjust their practice in response to a payment increase in ways that would persist in the face of subsequent payment decreases. For example, providers might pay the fixed cost to 27 enroll as a Medicaid provider or invest in learning to deal with the complexities of Medicaid billing. Additionally, providers might establish relationships with Medicaid patients that would be difficult to abandon if payments were to subsequently decline. Although limited in precision, our results instead suggest that many of the improvements in access, use, and health that Medicaid beneficiaries experienced when payments increased were lost when payments returned to their previous levels. VI Discussion and Conclusion While it is known that financial incentives matter in health care, increasing reimbursement rates may not make physicians more willing to accept new patients for at least two reasons. First, factors other than low payments may lead providers to restrict access for certain patients. In the case of Medicaid, for example, payment delays, high denial rates, and complex patient needs may make treating beneficiaries unattractive regardless of relative payment levels (Cunningham and O’Malley, 2009; Gottlieb et al., 2018; Long, 2013; Niess et al., 2018; Sloan et al., 1978). Second, capacity constraints limit the number of patients that providers can see. With a fixed number of hours in the day, access to health care will necessarily be rationed when the supply of providers has not kept pace with growing demand. In contrast, we find that changes in physician reimbursement have meaningful effects on access to care for patients. Exploiting large, exogenous changes in physician reimbursement rates for primary care visits under Medicaid, we estimate that an increase in Medicaid payments of $35—the median increase across states over the federally mandated primary care rate increase—reduced the probability that adult Medicaid beneficiaries were told that a physician was not accepting their insurance by 3.1 percentage points, or 38 percent of the mean. Compared to the average Medicaid payment of $76 for a new patient office visit of mid-level complexity before the rate increase, our estimates imply an elasticity of physician willingness to accept adult Medicaid patients with respect to reimbursement of 0.83. 28 These improvements in access among Medicaid beneficiaries have large implications for disparities in access to care. Before the primary care rate increase, 8.2 percent of adult Medicaid beneficiaries reported being told that a provider was not accepting their insurance, as compared to only 2.5 percent among adults with private insurance. Our results demonstrate that increasing Medicaid reimbursement rates by $45—enough to close the median gap in payments between Medicaid and private insurers—would reduce disparities in access to care by at least 70 percent. Our results are even more pronounced among children, for whom we find that closing the gap in physician payment rates between Medicaid and private insurance would eliminate disparities in access. We further find that improving access leads to increased use and better health among Medicaid beneficiaries. Increasing Medicaid payments by $35 increases the probability that program beneficiaries had an office visit in the past two weeks by 5.0 percent and increases the probability that they report being in excellent or very good health by 3.9 percent. The implied elasticity of self-reported health with respect to outpatient care is consistent with the literature using exogenous variation in health insurance coverage: when Medicaid was extended to low-income adults using a lottery in Oregon, those who gained insurance saw a 50 percent increase in office visits and were 25 percent more likely to report being in excellent or very good health (Baicker et al., 2013; Finkelstein et al., 2012). We further find that increased access to primary care reduces school absenteeism among young children: a $35 increase in Medicaid payments leads to an average reduction of 0.79 days missed per year due to illness or injury, or 22 percent of the mean, and reduces illness-related chronic absenteeism by nearly 50 percent. An outstanding question is how physicians are able to absorb new patients when reimbursement rates increase. If physicians are capacity constrained, they could increase the number of Medicaid patients they see either by substituting away from patients with private insurance or by decreasing their appointment length per patient. We find little evidence that increasing Medicaid payments negatively impacts access among patients with private 29 insurance, suggesting that physicians do not respond to increased Medicaid payments by substituting away from the privately insured on the extensive margin. Furthermore, we find no health effects among the privately insured, whereas decreased appointment length may result in worse provision of care. If physicians are not capacity constrained, they could increase the number of Medicaid patients they see by increasing their total hours worked. Although we cannot look at physician labor supply directly in our data, we can divide counties by those with and without shortages of primary care providers as defined by the HRSA. If some physicians are capacity constrained, we would expect increased Medicaid payments to have smaller effects in areas where providers have little scope to take on new patients. However, as shown in Table A5, we find no evidence of differential effects between counties that are and are not designed as primary care shortage areas. This suggests that some providers in areas with and without an adequate supply of providers have scope to increase the number of patients they see. Understanding how physicians accommodate more patients when payments increase is an important area for future work. The difficulties that Medicaid patients face accessing care is commonly attributed to a combination of complex patient needs, billing complications, and low reimbursement rates. This has led policy makers, practitioners, and researchers alike to argue that increasing reimbursement rates alone will not be enough to improve the provision of care to Medicaid beneficiaries (Goroll, 2018). In contrast, we find that the majority of differences in access between Medicaid beneficiaries and privately insured patients are driven by differences in reimbursement. Not only does increasing Medicaid reimbursement rates improve access, but these improvements in access translate into meaningful improvements in self-reported health and school absenteeism among the program’s beneficiaries. While it is well-known that financial incentives matter in health care, they appear to matter even more than previously thought. 30 References Adams, Kathleen, “Effect of Increased Medicaid Fees on Physician Participation and Enrollee Service Utilization in Tennessee, 1985-1988,” Inquiry, 1994, 31 (2), 173–187. Atherly, Adam and Karoline Mortensen, “Medicaid Primary Care Physician Fees and the Use of Preventive Services Among Medicaid Enrollees,” Health Services Research, 2014, 49 (4), 1306–1328. Baicker, Katherine, Amy Finkelstein, Jae Song, and Sarah Taubman, “The Impact of Medicaid on Labor Market Activity and Program Participation: Evidence from the Oregon Health Insurance Experiment,” American Economic Review, 2014, 104 (5), 322– 328. , Sarah Taubman, Heidi Allen, Mira Bernstein, Jonathan Gruber, Joseph Newhouse, Eric Schneider, Bill Wright, Alan Zaslavsky, and Amy Finkelstein, “The Oregon Experiment — Effects of Medicaid on Clinical Outcomes,” New England Journal of Medicine, 2013, 368 (18), 1713–1722. Baker, Laurence and Anne Beeson Royalty, “Medicaid Policy, Physician Behavior, and Health Care for the Low-Income Population,” Journal of Human Resources, 2000, 35 (3), 480–502. Bauer, Lauren, Patrick Liu, Diane Whitmore Schanzenbach, and Jay Shambaugh, “Reducing Chronic Absenteeism under the Every Student Succeeds Act,” Technical Report, The Hamilton Project 2018. Blumenthal, David and Sara Collins, “Health Care Coverage under the Affordable Care Act - A Progress Report,” New England Journal of Medicine, 2014, 371, 275–281. Brown, David, Amanda Kowalski, and Ithai Lurie, “Medicaid as an Investment in Children: What is the Long-Term Impact on Tax Receipts?,” NBER Working Paper No. 20835, 2015. Buchmueller, Thomas, Sean Orzol, and Lara Shore-Sheppard, “The Effect of Medicaid Payment Rates on Access to Dental Care Among Children,” American Journal of Health Economics, 2015, 1 (2), 194–223. Buehler, Melanie H., John Tapogna, and Hedy N. Chang, “Why being in school matters: Chronic absenteeism in Oregon Public Schools,” Technical Report, Attendance Works 2012. Callison, Kevin and Binh Nguyen, “The Effect of Medicaid Physician Fee Increases on Health Care Access, Utilization, and Expenditures,” Health Services Research, 2017, 53 (2), 690–710. Candon, Molly, Stephen Zuckerman, Douglas Wissoker, Brendan Saloner, Genevieve Kenney, Karin Rhodes, and Daniel Polsky, “Declining Medicaid Fees 31 and Primary Care Appointment Availability for New Medicaid Patients,” JAMA Internal Medicine, 2018, 178 (1), 145–146. Card, David, Carlos Dobkin, and Nicole Maestas, “The Impact of Nearly Universal Insurance Coverage on Health Care Utilization: Evidence from Medicare,” American Economic Review, 2008, 98 (5), 2242–2258. Chen, Alice, “Do the Poor Benefit from More Generous Medicaid Physician Payments?,” SSRN Working Paper No. 2444286, 2014. Clemens, Jeffrey and Joshua Gottlieb, “Do Physicians’ Financial Incentives Affect Medical Treatment and Patient Health?,” American Economic Review, 2014, 104 (4), 1320–1349. Coburn, Andrew, Stephen Long, and M. Susan Marquis, “Effects of Changing Medicaid Fees on Physician Participation and Enrollee Access,” Inquiry, 1999, 36 (3), 265–279. Coey, Dominic, “Physicians’ Financial Incentives and Treatment Choices in Heart Attack Management,” Quantitative Economics, 2015, 6 (3), 703–748. Cohen, Joel, “Medicaid Physician Fees and Use of Physician and Hospital Services,” Inquiry, 1993, 30 (3), 281–292. and Peter Cunningham, “Medicaid Physician Fee Levels and Children’s Access to Care,” Health Affairs, 1995, 14 (1), 225–262. Cohodes, Sarah, Daniel Grossman, Samuel Kleiner, and Michael Lovenheim, “The Effect of Child Health Insurance Access on Schooling: Evidence from Public Insurance Expansions,” Journal of Human Resources, 2016, 51 (3), 727–759. Connolly, Faith and Linda Olson, “Early Elementary Performance and Attendance in Baltimore City Schools’ Pre-Kindergarten and Kindergarten,” Technical Report, Baltimore Education Research Consortium 2012. Cunningham, Peter and Ann O’Malley, “Do Reimbursement Delays Discourage Medicaid Participation by Physicians?,” Health Affairs, Web Exclusive, 2009, 28 (1), w17–w28. Currie, Janet and Jonathan Gruber, “Health Insurance Eligibility, Utilization of Medical Care, and Child Health,” Quarterly Journal of Economics, 1996, 111 (2), 431–466. and , “Saving Babies: The Efficacy and Cost of Recent Expansions of Medicaid Eligibility for Pregnant Women,” Journal of Political Economy, 1996, 104 (6), 1263–1296. Decker, Sandra, “Medicaid Physician Fees and the Quality of Medical Care of Medicaid Patients in the USA,” Review of Economics of the Household, 2007, 5 (1), 95–112. , “Changes in Medicaid Physician Fees and Patterns of Ambulatory Care,” Inquiry, 2009, 46, 291–304. 32 , “In 2011 Nearly One-Third of Physicians Said They Would Not Accept New Medicaid Patients, but Rising Fees May Help,” Health Affairs, 2012, 31 (8), 1673–1679. , “Two-Thirds of Primary Care Physicians Accepted New Medicaid Patients in 2011-2012: A Baseline to Measure Future Acceptance Rates,” Health Affairs, 2013, 32 (7). , “No Association Found Between the Medicaid Primary Care Fee Bump and PhysicianReported Participation in Medicaid,” Health Affairs, 2018, 37 (7). Ehrlich, Stacy B., Julia A. Gwynne, Amber Stitziel Pareja, and Elaine M. Allensworth, “Preschool Attendance in Chicago Public Schools: Relationships with Learning Outcomes and Reasons for Absences,” Technical Report, The University of Chicago Consortium on Chicago School Research 2014. Fanning, Thomas and Martin de Alteriis, “The Limits of Marginal Economic Incentives in the Medicaid Program: Concerns and Cautions,” Journal of Health Politics, Policy, and Law, 1993, 18 (1), 27–42. Finkelstein, Amy, Sarah Taubman, Bill Wright, Mira Bernstein, Jonathan Gruber, Joseph Newhouse, Heidi Allen, Katherine Baicker, and the Oregon Health Study Group, “The Oregon Health Insurance Experiment: Evidence from the First Year,” Quarterly Journal of Economics, 2012, 127 (3), 1057–1106. Food Research and Action Center, “School Meal Eligibility and Reimbursements,” http://frac.org/school-meal-eligibility-reimbursements. Accessed: December 2018. Fox, Michael, Jonathan Weiner, and Kai Phua, “Effect of Medicaid Payment Levels on Access to Obstetrical Care,” Health Affairs, 1992, 11 (4), 150–161. Ginsburg, Alan, Phyllis Jordan, and Hedy Chang, “Absences Add Up: How School Attendance Influences Student Success,” Attendance Works, 2014. Goroll, Allan, “Reforming Payment for Primary Care - It’s Not Just the Money, It’s the Payment System,” JAMA Internal Medicine, 2018, 178 (8), 1049–1050. Gottlieb, Joshua, Adam Hale Shapiro, and Abe Dunn, “The Complexity of Billing and Paying for Physician Care,” Health Affairs, 2018, 37 (4). Government Accountability Office, “Medicaid Payment: Comparisons of Selected Services under Fee-for-Service, Managed Care, and Private Insurance,” GAO-14-533, 2014. Gross, Tal and Matthew Notowidigdo, “Health Insurance and the Consumer Bankruptcy Decision: Evidence from Expansions of Medicaid,” Journal of Public Economics, 2011, 95 (7-8), 767–778. Gruber, Jonathan, John Kim, and Dina Mayzlin, “Physician Fees and Procedure Intensity: The Case of Cesarean Delivery,” Journal of Health Economics, 1999, 18 (4), 473–490. 33 , Kathleen Adams, and Joseph Newhouse, “Physician Fee Policy and Medicaid Program Costs,” Journal of Human Resources, 1997, 32 (4), 611–634. Hadley, Jack, “Physician Participation in Medicaid: Evidence from California,” Health Services Research, 1979, 14 (4), 266–280. Henry J. Kaiser Family Foundation, “Where are States Today? Medicaid and CHIP Eligibility Levels for Children, Pregnant Women, and Adults,” https://www.kff.org/medicaid/fact-sheet/where-are-states-today-medicaid-and-chip/ 2018. Accessed: December 2018. KewalRamani, Angelina, Lauren Gilbertson, Mary Ann Fox, and Stephen Provasnik, “Status and Trends in the Education of Racial and Ethnic Minorities,” National Center for Education Statistics, U.S. Department of Education, 2007. Leung, Pauline and Alexander Mas, “Employment Effects of the Affordable Care Act Medicaid Expansions,” Industrial Relations, 2018, 57 (2), 206–234. Levinson, Daniel, “Coding Trends of Medicare Evaluation and Management Services,” Department of Health and Human Services, Office of Inspector General, 2012. Long, Sharon, “Physicians May Need More than Higher Reimbursements to Expand Medicaid Participation: Findings from Washington State,” Health Affairs, 2013, 32 (9), 1560– 1567. Long, Stephen, Russell Settle, and Bruce Stuart, “Reimbursement and Access to Physicians’ Services Under Medicaid,” Journal of Health Economics, 1986, 5 (3), 236–251. Manning, Willard, Joseph Newhouse, Naihua Duan, Emmett Keeler, Arleen Leibowitz, and Susan Marquis, “Health Insurance and the Demand for Medical Care: Evidence from a Randomized Experiment,” American Economic Review, 1987, 77 (3), 251–277. Medicaid and CHIP Payment and Access Commission, “Examining Access to Care in Medicaid and CHIP,” in “Report to Congress on Medicaid and Chip” 2011. , “An Update on the Medicaid Primary Care Rate Increase,” in “Report to Congress on Medicaid and Chip” 2015. Mitchell, Janet, “Physician Participation in Medicaid Revisited,” Medical Care, 1991, 29 (7), 645–653. Mulcahy, Andrew, Tadeja Gracner, and Kenneth Finegold, “Associations Between the Patient Protection and Affordable Care Act Medicaid Primary Care Payment Increase and Physician Participation in Medicaid,” JAMA Internal Medicine, 2018, 178 (8), 1042– 1048. 34 Neuzil, Kathleen M., Cynthia Hohlbein, and Yuwei Zhu, “Illness Among Schoolchildren During Influenza Season: Effect on School Absenteeism, Parental Absenteeism From Work, and Secondary Illness in Families,” Archives of Pediatrics and Adolescent Medicine, 2002, 156 (10), 986–991. Niess, MA, IV Blair, A Furniss, and AJ Davidson, “Specialty Physician Attitudes and Beliefs about Medicaid Patients,” Journal of Family Medicine, 2018, 5 (3), 1–9. Polsky, Daniel, Michael Richards, Simon Basseyn, Douglas Wissoker, Genevieve Kenney, Stephen Zuckerman, and Karin Rhodes, “Appointment Availability After Increases in Medicaid Payments for Primary Care,” New England Journal of Medicine, 2015, 372, 537–545. Rice, Thomas, “The Impact of Changing Medicare Reimbursement Rates on PhysicianInduced Demand,” Medical Care, 1983, 21 (8), 803–815. Schanzenbach, Diane Whitmore, Lauren Bauer, and Megan Mumford, “Lessons for Broadening School Accountability Under the Every Student Succeeds Act,” The Hamilton Project 2016. Shen, Yu-Chu and Stephen Zuckerman, “The Effect of Medicaid Payment Generosity on Access and Use Among Beneficiaries,” Health Services Research, 2005, 40 (3), 723–744. Showalter, Mark, “Physicians’ Cost Shifting Behavior: Medicaid Versus Other Patients,” Contemporary Economic Policy, 1997, 15 (2), 74–84. Sloan, Frank, Janet Mitchell, and Jerry Cromwell, “Physician Participation in State Medicaid Programs,” Journal of Human Resources, 1978, 13, 211–245. Smith, Vernon, Kathleen Gifford, Eileen Ellis, Victoria Wachino, and Molly O’Malley, “States Respond to Fiscal Pressure: A 50-State Update on State Medicaid Spending Growth and Cost Containment Actions,” The Kaiser Commission on Medicaid and the Uninsured, 2004. Sommers, Benjamin D., Katherine Baicker, and Arnold M. Epstein, “Mortality and Access to Care among Adults after State Medicaid Expansions,” New England Journal of Medicine, 2012, 376, 1025–1034. Spradlin, Terry, Katherine Cierniak, Dingjing Shi, and Minge Chen., “Attendance and Chronic Absenteeism in Indiana: The Impact on Student Achievement,” Technical Report, Center for Evaluation and Education Policy 2012. The Washington Post, “Obamacare Is About to Give Medicaid Docs a 73 Percent Raise,” 2012. Utah Education Policy Center, “Research Brief: Chronic Absenteeism,” Technical Report, University of Utah 2012. 35 Venkataramani, Maya, Craig Evan Pollack, and Eric Roberts, “Spillover Effects of Adult Medicaid Expansions on Children’s Use of Preventive Services,” Pediatrics, 2017, 140 (6). Wiseman, Amy and Susan Dawson, “Why Do Students Miss School? The Central Texas Absence Reasons Study,” Technical Report 2015. Yip, Winnie, “Physician Responses to Medical Fee Reductions: Changes in the Volume and Intensity of Supply of Coronary Artery Bypass Graft (CABG) Surgeries in Medicare and Private Sectors,” Journal of Health Economics, 1998, 17 (6), 675–699. Zuckerman, Stephan and Dana Goin, “How Much Will Medicaid Physician Fees for Primary Care Rise in 2013? Evidence from a 2012 Survey of Medicaid Physician Fees,” The Kaiser Commission on Medicaid and the Uninsured, 2012. 36 VII Figures 100 0 50 Medicaid Payments 150 200 Figure 1: State-Level Medicaid Payments over Time 2009 2010 2011 2012 2013 2014 2015 Average across states Notes: The above figure depicts Medicaid payments at the state-quarter level from 2009 to 2015. As defined in footnote 17, the payments are beneficiary-weighted averages of Medicaid fee-for-service and managed care payments for new patient office visits of mid-level complexity; patterns are qualitatively robust to using Medicaid payment rates for alternative billing codes. The top two lines are Alaska (1) and North Dakota (2); the bottom two lines in 2009 are New Hampshire (50) and Minnesota (51). 37 Figure 2: Maps of State-Level Medicaid Payments 2009 2012 2013 Notes: The above maps depict Medicaid payments for each state in 2009 (first year in sample period), 2012 (year before rate increase), and 2013 (first year of rate increase). As defined in footnote 17, the payments are beneficiary-weighted averages of Medicaid fee-for-service and managed care payments for new patient office visits of mid-level complexity; patterns are qualitatively robust to using Medicaid payment rates for alternative billing codes. The colors reflect quintiles of reimbursement levels in 2009; following the primary care rate increase, all states had Medicaid payments that were in the highest 2009 quintile. 38 15 15 15 30 30 30 60 60 45 60 Accepting New Patients 45 No Trouble Finding MD 45 75 75 75 Office Visit in Past Two Weeks 90 90 90 15 15 15 45 60 75 45 60 75 30 45 60 75 Accepts Patient’s Insurance C. Adult Subsample 30 Usual Place of Care B. Child Subsample 30 Health: Not Poor or Fair A. Full Sample 90 90 90 Increase in Medicaid Payments 15 15 15 45 60 75 30 30 60 45 60 Work Days 45 75 75 <14 School Days Missed (Age ≤ 10) 30 Health: Excellent or Very Good Figure 3: Changes in Access, Use, and Health by Size of Payment Increase 90 90 90 Notes: The above figures plot the average change in each outcome between 2011–2012 and 2013–2014 across deciles of state-level Medicaid payment increases. The dots represent changes among Medicaid patients; the solid black line is the best fit line through these points. The dashed line depicts the best fit line for changes among patients with private insurance. Outcomes are adjusted such that higher values denote better outcomes. School days missed reflect absenteeism due to illness or injury over the previous year; the exact survey questions and the corresponding reference windows for all outcomes are outlined in Appendix A.1. Change Following Payment Increase .06 .04 .02 0 −.02 .01 .02 −.02 −.01 0 .1 .05 0 −.05 .04 .02 0 −.02 .02 0 −.02 .04 0 −.02 .02 .1 .05 0 −.05 .01 .02 .03 −.01 0 5 0 −5 −10 39 2011 2012 2013 2011 2012 2013 2010 2011 2012 2013 Accepting New Patients 2010 No Trouble Finding MD 2010 2014 2014 2014 2009 2009 2009 2011 2012 2013 2011 2012 2013 2010 2011 2012 2013 Accepts Patient’s Insurance C. Adult Subsample 2010 Usual Place of Care B. Child Subsample 2010 Health: Not Poor or Fair 2014 2014 2014 2009 2009 2009 2011 2012 2013 2010 2010 2012 2011 2012 Work Days 2011 2013 2013 <14 School Days Missed (Age ≤ 10) 2010 Health: Excellent or Very Good 2014 2014 2014 Notes: The above figures plot the coefficients and 95% confidence intervals on year indicators interacted with state-level changes in Medicaid payments induced by the primary care rate increase from estimation of Equation (1). Outcomes are adjusted such that higher values denote better outcomes. Outcomes with missing coefficients in 2009 and 2010 were only added to the survey in 2011. School days missed reflect absenteeism due to illness or injury over the previous year; the exact survey questions and the corresponding reference windows for all outcomes are outlined in Appendix A.1. 2009 2009 2009 Office Visit in Past Two Weeks A. Full Sample Figure 4: Event Study: Effects of Medicaid Payments on Access, Use, and Health Among Medicaid Beneficiaries Estimated Coefficient .01 .005 0 −.005 .01 .005 0 −.005 .01 0 −.01 .01 .005 0 −.01 −.005 .01 .005 0 −.01 −.005 .02 .01 0 −.01 .005 .01 .015 −.005 0 .02 .01 0 −.01 2 1 0 −1 −2 40 41 −.005 0 .005 .01 −.015 −.01 −.005 0 .005 0 .005 −.02 −.01 0 .01 Not Accepting Patient’s Insurance −.005 Not Accepting New Patients −.01 C. Adult Subsample −.015 Blank −.01 −.008−.006−.004−.002 0 No Usual Place of Care −.01 Trouble Finding MD .01 B. Child Subsample .005 Blank 0 Health: Poor or Fair Office Visit in Past Two Weeks −.005 A. Full Sample Blank 0 Blank .005 .01 .015 −2 −1 0 1 Work Days Missed Blank −.02 −.015 −.01 −.005 0 2 .005 14+ School Days Missed (Age ≤ 10) −.005 Health: Excellent or Very Good Blank Notes: Each dot in the above figures depicts the estimated effect of a $10 increase in Medicaid payments for the subsample listed on the y-axis from estimation of Equation (2). Each coefficient comes from a separate regression. The horizontal bars depict 95% confidence intervals for each coefficient. The dashed vertical line in each subplot displays the coefficient estimate from the main sample (as reported in Table 3), which includes years 2009–2014 and imputes missing payment ratios with the median across all states in the GAO report. School days missed reflect absenteeism due to illness or injury over the previous year; the exact survey questions and the corresponding reference windows for all outcomes are outlined in Appendix A.1. Main sample Truncated sample: 2009−2013 Non−expansion states Families with children Missing MMC ratios imputed with mean Only states in GAO report Only fee−for−service payments Only variation from federal mandate Blank Main sample Truncated sample: 2009−2013 Non−expansion states Missing MMC ratios imputed with mean Only states in GAO report Only fee−for−service payments Only variation from federal mandate Blank Main sample Truncated sample: 2009−2013 Non−expansion states Families with children Missing MMC ratios imputed with mean Only states in GAO report Only fee−for−service payments Only variation from federal mandate Blank Figure 5: Robustness: Effects of Medicaid Payments on Access, Use, and Health Among Medicaid Beneficiaries 42 −.015 −.01 −.005 0 .005 −.015 −.01 −.005 0 .005 Not Accepting Patient’s Insurance .002 .004 Not Accepting New Patients 0 C. Adult Subsample −.006 −.004 −.002 Blank 0 .005 No Usual Place of Care 0 Trouble Finding MD −.005 B. Child Subsample −.01 Blank .002 .004 .006 .008 −.008 −.006 −.004 −.002 0 Health: Poor or Fair Office Visit in Past Two Weeks −.002 A. Full Sample Blank 0 Blank .005 .01 .015 −2 −1 0 1 Work Days Missed Blank −.02 −.015 −.01 −.005 0 2 .005 14+ School Days Missed (Age ≤ 10) −.005 Health: Excellent or Very Good Blank Notes: Each dot in the above figures depicts the estimated effect of a $10 increase in Medicaid payments for the subsample listed on the y-axis from estimation of Equation (2). Although the third row in each subfigure exploits payment decreases, the estimated coefficients still represent the effects of a $10 increase in Medicaid payments. Each coefficient comes from a separate regression. The horizontal bars depict 95% confidence intervals for each coefficient. The dashed vertical line in each subplot displays the coefficient estimate from the main sample (as reported in Table 3), which includes years 2009–2014 and imputes missing payment ratios with the median across all states in the GAO report. School days missed reflect absenteeism due to illness or injury over the previous year; the exact survey questions and the corresponding reference windows for all outcomes are outlined in Appendix A.1. Turning off: 2013−2015 Turning on: 2012−2014 Turning on: 2012−2014 Blank Turning off: 2013−2015 Turning on: 2012−2014 Turning on: 2012−2014 Blank Turning off: 2013−2015 Turning on: 2012−2014 Turning on: 2012−2014 Blank Figure 6: Effects of Payment Increases versus Payment Decreases 43 0.055 0.075 4.929 0.021 0.031 3.433 4.428 0.044 0.064 0.196 0.174 0.566 0.016 0.022 3.730 0.008 0.021 2.887 3.221 0.022 0.032 0.174 0.061 0.729 0.062 0.082 5.010 0.022 0.034 3.516 4.745 0.046 0.070 0.197 0.176 0.562 0.017 0.025 3.711 0.008 0.022 2.933 3.302 0.023 0.034 0.175 0.062 0.726 0.049 0.069 4.798 0.020 0.026 3.292 3.895 0.042 0.053 0.194 0.169 0.573 Medicaid (5) 0.015 0.020 3.767 0.008 0.019 2.792 3.058 0.021 0.028 0.172 0.059 0.734 Private (6) Rate Increase (2013–2014) Notes: The reported statistics are weighted using the sample weights provided in the NHIS. School days missed reflect absenteeism due to illness or injury over the previous year; the exact survey questions and the corresponding reference windows for all outcomes are outlined in Appendix A.1. Not accepting new patients Not accepting patient’s insurance Work days missed C. Adult Subsample Trouble finding a doctor No usual place of care School days missed (age ≤ 10) School days missed (age > 10) 14+ school days missed (age ≤ 10) 14+ school days missed (age > 10) B. Child Subsample Office visit in past two weeks Health: poor or fair Health: excellent or very good A. Full Sample Private (4) Medicaid (3) Medicaid (1) Private (2) Baseline (2009–2012) Entire Sample (2009–2014) Table 1: Summary Statistics: NHIS Outcome Measures VIII Tables Table 2: Summary Statistics: Individual and County Controls All Medicaid Private Demographics Male Average age Black Hispanic U.S. citizen 0.489 0.373 0.132 0.167 0.927 0.439 0.242 0.252 0.296 0.936 0.489 0.384 0.097 0.101 0.959 Education < High school High school or GED Some college Associate’s degree Bachelor’s degree Master’s, professional, or Ph.D. 0.135 0.255 0.190 0.107 0.181 0.097 0.307 0.307 0.179 0.079 0.049 0.013 0.058 0.218 0.194 0.120 0.246 0.139 Family structure Married Live with partner No children 1 child 2 children 3 children 4 children 5+ children 0.582 0.055 0.479 0.176 0.191 0.099 0.036 0.019 0.400 0.049 0.229 0.193 0.243 0.185 0.090 0.059 0.666 0.045 0.503 0.179 0.197 0.086 0.025 0.010 Income and wealth Welfare Homeowner Income to poverty Income to poverty Income to poverty Income to poverty 0.127 0.334 0.138 0.166 0.250 0.299 0.483 0.645 0.475 0.285 0.109 0.025 0.035 0.223 0.036 0.097 0.286 0.436 1,126,919 2,010 53,749 0.083 286,546 0.093 234 876 401 3,254 1,284,943 3,087 50,031 0.087 362,920 0.118 265 969 447 3,750 1,050,948 1,834 55,408 0.081 255,757 0.096 222 838 386 3,037 603,074 96,128 338,174 Individual-level controls line: line: line: line: <1 1-1.99 2-3.99 4+ County-level controls Population Population density Median income Unemployment rate Medicaid eligibles Expansion state (2014) Number of pediatricians Number of primary care doctors Number of nurse practitioners Number of hospital beds Observations Notes: All statistics are in percentages unless otherwise specified; some categories do not sum to one due to missing responses. Individual-level statistics are weighted using the sample weights provided in the NHIS. County-level controls come from the HRSA’s Area Resource Files. 44 Table 3: Effects of Medicaid Payments on Access, Use, and Health A. Full Sample Medicaid Private (1) Office Visit (2 Weeks) (2) Health ≤ Fair (3) Health ≥ Very Good (4) Office Visit (2 Weeks) (5) Health ≤ Fair (6) Health ≥ Very Good 0.0028* (0.0017) -0.0031* (0.0017) 0.0062*** (0.0023) -0.0011 (0.0010) 0.0002 (0.0006) 0.0020 (0.0017) Observations R2 96,017 0.071 96,067 0.296 96,067 0.232 337,506 0.036 337,903 0.079 337,903 0.138 Baseline mean 0.197 0.176 0.562 0.175 0.062 0.726 Medicaid payments ($10s) B. Child Subsample Medicaid Private (1) Trouble Finding MD (2) No Usual Place of Care (3) (4) 14+ School 14+ School Absences Absences (Age≤10) (Age≥11) (5) Trouble Finding MD (6) No Usual Place of Care -0.0054*** (0.0015) -0.0036* (0.0018) -0.0065** (0.0030) 0.0019 (0.0038) 0.0013* (0.0007) 0.0000 (0.0010) 0.0012 (0.0028) 0.0003 (0.0026) Observations R2 16,752 0.016 21,221 0.022 6,665 0.036 6,766 0.047 26,277 0.007 33,994 0.029 10,079 0.025 14,337 0.021 Baseline mean 0.022 0.034 0.046 0.070 0.008 0.022 0.023 0.034 Medicaid payments ($10s) C. Adult Subsample Medicaid (7) (8) 14+ School 14+ School Absences Absences (Age≤10) (Age≥11) Private (1) Not Accepting New Patients (2) Not Accepting Patient’s Insurance (3) Work Days Missed (4) Not Accepting New Patients (5) Not Accepting Patient’s Insurance (6) Work Days Missed -0.0082*** (0.0015) -0.0089*** (0.0030) -0.1638 (0.3737) 0.0005 (0.0006) -0.0007 (0.0007) 0.1047 (0.1147) Observations R2 14,806 0.037 14,805 0.039 6,298 0.076 79,812 0.006 79,802 0.009 76,971 0.009 Baseline mean 0.062 0.082 5.010 0.017 0.025 3.711 Medicaid payments ($10s) Notes: Observations are at the individual level; standard errors are clustered by state. All regressions include state and quarter-year fixed effects and all individual and county-level controls listed in Table 2 (with age in five-year bins). Regressions are weighted using the sample weights provided in the NHIS. School days missed reflect absenteeism due to illness or injury over the previous year; the exact survey questions and the corresponding reference windows for all outcomes are outlined in Appendix A.1. Fewer children report days of missed school in the past year relative to other child outcomes because a child must be at least five years old to be asked this question. Similarly, only adults with employment histories are asked how many days of work they missed in the past year. 45 46 0.062 0.082 0.022 0.034 0.046 0.197 0.176 0.562 0.017 0.025 0.008 0.022 0.023 0.175 0.062 0.726 (2) (1) -0.045 -0.057 -0.014 -0.012 -0.023 -0.022 -0.114 0.164 (3) Disparity: (2) - (1) -0.0082 -0.0089 -0.0054 -0.0036 -0.0065 0.0028 -0.0031 0.0062 (4) -1.01 -0.83 -1.87 -0.81 -1.08 0.11 -0.13 0.08 (5) Effect of Implied $10 Increase Elasticity 0.18 0.16 0.39 0.30 0.28 NA 0.03 0.04 (6) $10 0.64 0.55 1.35 1.05 0.99 NA 0.10 0.13 (7) $35 0.82 0.70 1.74 1.35 1.27 NA 0.12 0.17 (8) $45 Share of Disparity Closed by $54.88 $64.04 $25.93 $33.33 $35.38 NA $367.74 $264.52 Increase to Erase Disparity (9) Notes: The above table displays the effects of changes in Medicaid payments on disparities in outcomes between Medicaid beneficiaries and privately insured patients. Columns (1) and (2) reproduce the averages from Table 1; column (3) is the difference in these means. Column (4) reproduces our main estimates from Table 3. Column (5) is the implied elasticity of each outcome with respect to physician payments; this is computed using a median baseline payment of $76 under Medicaid for an office visit of mid-level complexity ((5) = [(4)/(1)]/(10/76)]). Columns (6)–(8) show how much of the disparity from column (3) would be reduced by increasing Medicaid payments by $10, $35 (median payment increase from federal mandate), and $45 (median payment difference between Medicaid and private insurance), respectively. Column (9) shows the implied increase in Medicaid payments that would be required to eliminate the disparity in each outcome between Medicaid patients and patients with private insurance ((9) = [(3)/(4)]∗10). Office visits are excluded from columns (6)–(9) because Medicaid beneficiaries have more office visits than patients with private insurance. Not accepting new patients Not acc. patient’s insurance C. Adult Subsample Trouble finding MD No usual place of care 14+ school days missed B. Child Subsample Office visit (2 weeks) Health: poor or fair Health: excellent or very good A. Full Sample Private Medicaid Baseline Means Table 4: Effects of Medicaid Payments on Disparities between Publicly and Privately Insured Table 5: Effects of Medicaid Payments on School Absences: NAEP A. Free Lunch Eligible 0 days missed (%) 3+ days missed (%) (1) Grade 4 (2) Grade 8 (3) Grade 4 (4) Grade 8 0.0046*** (0.0012) 0.0041 (0.0025) -0.0049*** (0.0011) -0.0032* (0.0016) Observations R2 150 0.906 150 0.889 150 0.859 150 0.868 Baseline mean 0.444 0.381 0.243 0.267 Medicaid payments ($10s) B. Free Lunch Ineligible 0 days missed (%) 3+ days missed (%) (1) Grade 4 (2) Grade 8 (3) Grade 4 (4) Grade 8 0.0037 (0.0020) 0.0024 (0.0020) -0.0014 (0.0014) -0.0018 (0.0013) Observations R2 150 0.808 150 0.871 150 0.778 150 0.838 Baseline mean 0.536 0.468 0.162 0.168 Medicaid payments ($10s) Notes: Data are from the 2009, 2011, and 2013 National Assessment of Educational Progress and represent the percentage of students missing a given amount of school (for any reason) in the month preceding their national math and reading assessments. Observations are at the state-year level; standard errors clustered by state. All regressions include state and year fixed effects. Results are robust to weighting by state population. 47 Table 6: Effects of Medicaid Payments: Triple Difference Model A. Full Sample Medicaid payments ($10s) 1{Medicaid} Medicaid payments * 1{Medicaid} Observations R2 Baseline mean B. Child Subsample (1) Office Visit (2 Weeks) (2) Health: Poor or Fair (3) Health: Excellent or Very Good -0.0003 (0.0009) 0.0690*** (0.0107) -0.0007 (0.0008) 0.0001 (0.0005) 0.1137*** (0.0073) -0.0024*** (0.0007) 0.0022 (0.0015) -0.1359*** (0.0136) 0.0035** (0.0015) 430,800 0.040 431,244 0.128 431,244 0.163 0.179 0.082 0.698 (1) Trouble Finding MD (2) No Usual Place of Care (3) 14+ School Days Missed (Age≤10) (4) 14+ School Days Missed (Age≥11) -0.0006 (0.0008) 0.0267*** (0.0054) -0.0018*** (0.0005) -0.0011 (0.0008) 0.0043 (0.0065) -0.0009 (0.0007) -0.0008 (0.0023) 0.0376*** (0.0109) -0.0023** (0.0012) 0.0018 (0.0022) 0.0541*** (0.0160) -0.0032* (0.0016) Observations R2 42,541 0.010 54,602 0.021 16,544 0.023 21,473 0.026 Baseline mean 0.013 0.026 0.029 0.042 Medicaid payments ($10s) 1{Medicaid} Medicaid payments * 1{Medicaid} C. Adult Subsample (1) Not Accepting New Patients (2) Not Accepting Patient’s Insurance (3) Work Days Missed 0.0002 (0.0007) 0.0691*** (0.0123) -0.0042*** (0.0011) -0.0010 (0.0008) 0.0887*** (0.0118) -0.0046*** (0.0011) 0.0964 (0.1094) 1.9949* (1.1576) -0.1241 (0.1108) Observations R2 94,162 0.016 94,150 0.020 83,054 0.009 Baseline mean 0.022 0.031 3.785 Medicaid payments ($10s) 1{Medicaid} Medicaid payments * 1{Medicaid} Notes: Observations are at the individual level; standard errors are clustered by state. All regressions include state and quarter-year fixed effects and all individual and county-level controls listed in Table 2 (with age in five-year bins). Regressions are weighted using the sample weights provided in the NHIS. School days missed reflect absenteeism due to illness or injury over the previous year; the exact survey questions and the corresponding reference windows for all outcomes are outlined in Appendix A.1. Fewer children report days of missed school in the past year relative to other child outcomes because a child must be at least five years old to be asked this question. Similarly, only adults with employment histories are asked how many days of work they missed in the past year. 48 For Online Publication A Data Appendix A.1 Outcomes Table A1: Overview of Data Sources used for Outcome Measures Outcome Data Source Years Available Look Back Period Payment Variable Sample Office visit NHIS 2009-2015 Past 2 weeks Avg. rate in interview quarter Full sample Health: excellent or very good NHIS 2009-2015 Not specified Avg. rate in interview quarter Full sample Health: poor or fair NHIS 2009-2015 Not specified Avg. rate in interview quarter Full sample Trouble finding MD NHIS 2011-2015 Past 12 months Avg. rate over past 12 months Child subsample No usual place of care NHIS 2009-2015 Not specified Avg. rate over past 12 months Child subsample School days missed NHIS 2009-2015 Past 12 months Avg. rate over past 12 months Child subsample Not accepting new patients NHIS 2011-2015 Past 12 months Avg. rate over past 12 months Adult subsample Not acc. patient’s insurance NHIS 2011-2015 Past 12 months Avg. rate over past 12 months Adult subsample Work days missed NHIS 2009-2015 Past 12 months Avg. rate over past 12 months Adult subsample School absences NAEP 2009, 2011, 2013 30 days before test Avg. rate in first quarter 4th and 8th grade math and reading Test scores NAEP 2009, 2011, 2013 Testing occurs in Q1 Avg. rate in first quarter 4th and 8th grade math and reading 49 National Health Interview Survey questions • Full Sample (from Family File) – During the last two weeks, did {person} see a doctor or other health care professional at a doctor’s office, a clinic, an emergency room, or some other place? (Do not include times during an overnight hospital stay.) – Would you say {person’s} health in general is excellent, very good, good, fair, or poor? • Child Subsample – During the past 12 months, did you have any trouble finding a general doctor or provider who would see {sample child}? – Is there a place that {sample child} usually goes when {he/she} is sick or you need advice about {his/her} health? – During the past 12 months, that is, since {12-month ref. date}, about how many days did {sample child} miss school because of illness or injury? • Adult Subsample – During the past 12 months, were you told by a doctor’s office or clinic that they would not accept {sample adult} as a new patient? – During the past 12 months, were you told by a doctor’s office or clinic that they would not accept {sample adult}’s health care coverage? – During the past 12 months, about how many days did {sample adult} miss work? A.2 Medicaid Payments We collected data on fee-for-service reimbursement rates directly from state Medicaid offices. The data have two components: (1) standard fee-for-service rates applicable in 2009–2015 for 50 all providers and (2) augmented fee-for-service rates applicable in 2013–2014 (and 2015, depending on the state) for qualifying physicians in family medicine, general internal medicine, and pediatric medicine. In constructing our state-quarter panel of payments, we use standard rates in 2009–2012, augmented rates in 2013–2014, and either the standard or augmented rates in 2015 depending on whether a given state extended the primary care rate increase. 44 states and the District of Columbia provided us with complete, quarterly rate information used to construct this panel. For the remaining six states, we use the following procedures to impute missing rate information: • California: We only have the standard rates for 2009 and 2015. As the standard rates were the same in 2009 and 2015, we assume that they did not change over this period and pull forward the standard rates to 2012. • Hawaii: We only have the standard rates for 2009, 2012, and 2015. As the standard rates were the same in 2009 and 2012, we assume that they did not change over this period and pull forward the standard rates to 2011. • New Mexico: We are missing standard rates for January–November 2009. The rates changed over this period; we impute the missing months with the rate in the nearest month with non-missing rate information. • Utah: We are missing standard rates for January–May 2009 and July–December 12. We impute the missing months with the rate in the nearest month with non-missing rate information. • South Dakota: Standard rates are not archived, so we only have standard rates for 2015. We impute standard rates from 2009–2012 such that the change in reimbursement rates between each quarter and 2015 reflects the average change in reimbursement rates for neighboring states (MT, ND, MN, IA, NE, WY) over the same period. • Tennessee: We have no micro-data on reimbursement rates, as the state only uses 51 Medicaid managed care. However, we were told by the state that average reimbursements increased by 44 percent as a result of the primary care rate increase. We impute reimbursement rates for Tennessee in 2013 and 2014 by averaging the 2013 and 2014 augmented rates for neighboring states (MO, KY, VA, NC, GA, AL, MS, AR). We then apply the 44 percent increase from 2012 to 2013 to impute the rates for 2012. For 2009–2012 and 2015, we calculate the average change in physician payments across neighboring states in the relevant period and apply this rate change to Tennessee over the same window. A.3 Medicaid Managed Care The primary care rate increase applied to both Medicaid fee-for-service and Medicaid managed care programs. While states could just increase fee-for-service reimbursement rates for the covered services to comply with the mandated higher rates, determining how to increase reimbursement rates for physicians treating patients enrolled in Medicaid managed care was more complicated. To ensure that Medicaid managed care programs complied with the rate increase, each state’s Medicaid program was required to submit proposals to CMS that outlined methodologies for: 1. Identifying the proportion of the capitation payments made by the state to its contracted MCOs in 2009 that was spent on each of the applicable primary care services, as well as the per-unit cost of each of these services. These baseline costs were used to calculate the refunds that each state’s Medicaid program was eligible to receive from the federal government in 2013 and 2014. 2. Developing a “model” that incorporated the increased fees for primary care services into the state’s 2013 and 2014 capitation payments to MCOs. It was recommended that states implemented one of three types of models: • Model 1: “Full-risk prospective capitation” in which states incorporated increased 52 fees directly into their capitation payments to MCOs for 2013 and 2014 • Model 2: “Prospective capitation with risk-sharing that incorporates retrospective reconciliation” in which increased fees were built into states’ capitation payments for 2013 and 2014 (similar to Model 1), but capitation payments were to be adjusted at the end of an agreed-upon time period to reflect actual utilization and costs (states and MCOs engage in “retrospective reconciliation”) • Model 3: “Non-risk reconciled payments for enhanced rates” in which states’ initial capitation payments to MCOs for 2013 and 2014 did not incorporate increased fees. Instead, MCOs submitted encounter data to the state at the end of the quarter, year, etc., and the state sent an additional payment to the MCOs to cover the costs of the increased fees. CMS had to sign off on each state’s methodology for determining the 2009 rates and on its plan for implementing the rate increase for eligible physicians treating managed care enrollees. According to CMS, at least 21 states opted to receive the increased funding in lump-sum payments based on encounter data (Model 3). The rest of the states incorporated the increased fees directly into their capitation payments (Models 1 and 2); most of these states did not engage in any retrospective reconciliation based on actual utilization data. 53 B Supplementary Tables and Figures 0 Percent Increase from 2009 to 2013 75 150 225 Figure A1: Increases in Medicaid Payments from 2009 to 2013 Across New Patient Codes 0 75 150 Percent Increase from 2009 to 2013 (CPT 99203) CPT 99201 CPT 99202 CPT 99204 225 CPT 99205 Notes: The above figure displays state-level changes in Medicaid payments for CPT codes 99201, 99202, 99204, and 99205 from 2009 to 2013 versus state-level changes in Medicaid payments for CPT code 99203 (the CPT code used in the majority of our analyses) over the same period. The black line is the 45-degree line. The figure excludes CPT codes 99204 and 99204 for New Jersey; payments for these codes increased by 308 percent and 404 percent, respectively, while payment for CPT code 99203 increased by 169 percent. 54 2009 CPT 99201 2011 2013 Wisconsin West Virginia 2013 South Dakota South Carolina 2011 North Carolina Missouri Mississippi New York Kentucky Kansas 2009 Florida District of Columbia 2015 Alaska Alabama 2015 2009 2011 2015 2009 CPT 99202 2013 Wyoming Tennessee North Dakota Montana Louisiana Georgia Arizona 2011 2013 Texas Ohio Nebraska Maine Hawaii Arkansas 2015 2011 2013 CPT 99203 2009 Utah Oklahoma Nevada Maryland Idaho California 2015 2009 2011 2015 2009 2013 Virginia Pennsylvania New Jersey Michigan Indiana Connecticut 2011 CPT 99204 2013 Vermont Oregon New Hampshire Massachusetts Illinois Colorado Figure A2: Medicaid Payments Across New Patient Codes by State 2015 2009 2013 2015 CPT 99205 2011 Washington Rhode Island New Mexico Minnesota Iowa Delaware Notes: The above figure displays Medicaid payments by states for CPT codes 99201, 99202, 99203, 99204, and 99205 from 2009 to 2015. The payments for each CPT code in each state are normalized to their level in the first quarter of 2009. Percent Increase from First Quarter of 2009 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400 55 .8 .6 .4 .2 0 Medicaid Beneficiaries Enrolled in MMC (%) 1 Figure A3: State-Level Medicaid Managed Care Penetration over Time 2009 2010 2011 2012 2013 2014 2015 Average across states Notes: The above figure depicts the fraction of Medicaid beneficiaries enrolled in comprehensive risk-based managed care in each state from 2009 to 2015. The black line depicts the national average. Data for 2009 through 2014 come from CMS; data for 2015 comes from the Henry J. Kaiser Family Foundation. In 2014, eleven states had less than one percent of their Medicaid beneficiaries enrolled in managed care plans: Alabama, Alaska, Arkansas, Connecticut, Idaho, Maine, Montana, North Carolina, Oklahoma, South Dakota, and Wyoming. In the same year, nine states had more than 85 percent of their Medicaid beneficiaries enrolled in such plans: Arizona, Delaware, Hawaii, Kansas, Kentucky, New Hampshire, New Jersey, Tennessee, and Washington. 56 Figure A4: States that Extended the Medicaid Primary Care Rate Increase past 2014 Extended payment increase Did not extend payment increase Notes: The above map depicts whether states chose to maintain the primary care rate increase after the federal mandate expired in 2014: shaded states extended higher Medicaid payment rates into 2015. 57 Figure A5: Distribution of School Absences by Free Lunch Eligibility: NAEP 60 Grade 4 53.6 Percent 40 44.4 30.2 20 31.3 14.5 10.4 4.2 3.9 1.6 0 5.9 0 1−2 3−4 5−10 11+ Days missed Free lunch eligible Ineligible 40 50 Grade 8 46.8 38.1 36.4 20 Percent 30 35.3 16.7 10 11.5 6.9 3.1 1.3 0 3.9 0 1−2 3−4 5−10 11+ Days missed Free lunch eligible Ineligible Notes: The above figures display the average percentage of students who missed 0, 1–2, 3–4, 5–10, or 11+ days in the month preceding their national math and reading assessments in 2009, 2011, and 2013. Data come from the National Assessment of Educational Progress. 58 Figure A6: Average Test Scores by School Absences: NAEP 300 Math 288.7 284.4 273.9 270.1 254.9 244.5 240.4 234.9 233.5 0 Average test scores 100 200 218.3 0 1−2 3−4 5−10 11+ Days missed Grade 4 Grade 8 300 Reading 268.6 266.1 258.4 254.1 235.7 224.2 216.1 Average test scores 100 200 220.9 213.7 0 192.5 0 1−2 3−4 5−10 11+ Days missed Grade 4 Grade 8 Notes: The above figures display the average test scores of students who missed 0, 1–2, 3–4, 5–10, or 11+ days in the month preceding their national assessment in 2009, 2011, and 2013. Data come from the National Assessment of Educational Progress. 59 Figure A7: States that Expanded Medicaid Medicaid Expansion: None Before 2014 2014 Notes: The above map depicts whether states expanded their Medicaid programs: the dark-shaded states participated in the ACA Medicaid expansion in 2014, the light-shaded states expanded their Medicaid program prior to the ACA, and the remaining states did not participate in any type of Medicaid expansion by 2014. Data are from Leung and Mas (2018). Table A2: Effects of Medicaid Payments on School Days Missed (Continuous Measure) Child Subsample Medicaid Private (1) School Days Missed (Age≤10) (2) School Days Missed (Age≥11) (3) School Days Missed (Age≤10) (4) School Days Missed (Age≥11) -0.2256** (0.1022) -0.0702 (0.2251) 0.0014 (0.0463) 0.0649 (0.1018) Observations R2 6,687 0.043 6,794 0.062 10,168 0.033 15,046 0.029 Baseline mean 3.516 4.745 2.933 3.302 Medicaid payments ($10s) Notes: Observations are at the individual level; standard errors are clustered by state. All regressions include state and quarter-year fixed effects and all individual and county-level controls listed in Table 2 (with age in five-year bins). Regressions are weighted using the sample weights provided in the NHIS. School days missed reflect absenteeism due to illness or injury over the previous year. 60 Table A3: Effects of Medicaid Payments on Test Scores: NAEP A. Free Lunch Eligible Math Reading (1) Grade 4 (2) Grade 8 (3) Grade 4 (4) Grade 8 -0.109 (0.165) 0.353 (0.236) 0.141 (0.220) 0.0341 (0.207) Observations R2 150 0.925 150 0.937 150 0.903 150 0.893 Baseline mean 228.5 268.1 206.6 251.4 Medicaid payments ($10s) B. Free Lunch Ineligible Math Reading (1) Grade 4 (2) Grade 8 (3) Grade 4 (4) Grade 8 0.0114 (0.225) 0.448 (0.235) 0.116 (0.192) 0.143 (0.190) Observations R2 150 0.869 150 0.921 150 0.893 150 0.915 Baseline mean 250.6 294.1 232.8 273.7 Medicaid payments ($10s) Notes: Data is from the 2009, 2011, and 2013 National Assessment of Educational Progress. Observations are at the state-year level; standard errors clustered by state. All regressions include state and year fixed effects. Results are robust to weighting by state population. 61 Table A4: Effects of Medicaid Payments: States that Extended the Rate Increase A. Full Sample Medicaid Private (1) Office Visit (2 Weeks) (2) Health ≤ Fair (3) Health ≥ Very Good (4) Office Visit (2 Weeks) (5) Health ≤ Fair (6) Health ≥ Very Good 0.0026 (0.0017) -0.0991** (0.0374) -0.0018 (0.0020) -0.0028* (0.0017) -0.0254 (0.0395) 0.0027 (0.0023) 0.0063*** (0.0022) -0.2376*** (0.0538) 0.0008 (0.0033) -0.0009 (0.0010) -0.0487*** (0.0138) 0.0019** (0.0008) 0.0001 (0.0006) 0.0118 (0.0083) -0.0003 (0.0005) 0.0021 (0.0017) -0.0587** (0.0236) 0.0010 (0.0014) Observations R2 96,017 0.071 96,067 00.296 96,067 0.232 337,506 0.036 337,903 0.079 337,903 0.138 Baseline mean 0.197 0.176 0.562 0.175 0.062 0.726 MC payments ($10s) 1{Extended} Payments * 1{Extended} B. Child Subsample MC payments ($10s) 1{Extended} Payments * 1{Extended} Medicaid Private (1) Trouble Finding MD (2) No Usual Place of Care (3) (4) (5) 14+ School 14+ School Trouble Absences Absences Finding (Age≤10) (Age≥11) MD (6) No Usual Place of Care (7) (8) 14+ School 14+ School Absences Absences (Age≤10) (Age≥11) -0.0054*** (0.0015) 0.0324 (0.0209) 0.0002 (0.0011) -0.0034* (0.0019) -0.0199 (0.0327) 0.0016 (0.0017) -0.0074** (0.0032) 0.1609 (0.0974) -0.0074 (0.0053) 0.0010 (0.0040) -0.0234 (0.0802) -0.0082* (0.0045) 0.0000 (0.0010) 0.0666*** (0.0236) -0.0001 (0.0012) 0.0010 (0.0028) 0.0581 (0.0371) -0.0017 (0.0019) 0.0000 (0.0027) -0.0393 (0.0769) -0.0024 (0.0041) 0.0012* (0.0007) -0.0259*** (0.0080) -0.0002 (0.0004) Observations R2 16,752 0.016 21,221 0.022 6,665 0.036 6,766 0.047 26,277 0.007 33,994 0.029 10,079 0.025 14,337 0.021 Baseline mean 0.022 0.034 0.046 0.070 0.008 0.022 0.023 0.034 C. Adult Subsample Medicaid Private (1) Not Accepting New Patients (2) Not Accepting Patient’s Insurance (3) Work Days Missed (4) Not Accepting New Patients (5) Not Accepting Patient’s Insurance (6) Work Days Missed -0.0084*** (0.0014) 0.1212*** (0.0410) -0.0022 (0.0024) -0.0088*** (0.0030) 0.0322 (0.0699) 0.0008 (0.0037) -0.2808 (0.3704) 16.5395** (6.0360) -0.7897** (0.2961) 0.0005 (0.0006) 0.001 (0.0133) -0.0003 (0.0008) -0.0006 (0.0007) -0.0093 (0.0150) 0.0007 (0.0008) 0.1255 (0.1158) -4.4484** (1.8844) 0.1825* (0.0972) Observations R2 14,806 0.037 14,805 0.039 6,298 0.076 79,812 0.006 79,802 0.009 76,971 0.009 Baseline mean 0.062 0.082 5.010 0.017 0.025 3.711 MC payments ($10s) 1{Extended} Payments * 1{Extended} Notes: Observations are at the individual level; standard errors are clustered by state. All regressions include state and quarter-year fixed effects and all individual and county-level controls listed in Table 2 (with age in five-year bins). Regressions are weighted using the sample weights provided in the NHIS. School days missed reflect absenteeism due to illness or injury over the previous year; the exact survey questions and the corresponding reference windows for all outcomes are outlined in Appendix A.1. Fewer children report days of missed school in the past year relative to other child outcomes because a child must be at least five years old to be asked this question. Similarly, only adults with employment histories are asked how many days of work they missed in the past year. 62 Table A5: Effects of Medicaid Payments: Primary Care Shortage Areas A. Full Sample Medicaid Private (1) Office Visit (2 Weeks) (2) Health ≤ Fair (3) Health ≥ Very Good (4) Office Visit (2 Weeks) (5) Health ≤ Fair (6) Health ≥ Very Good 0.0025 (0.0018) -0.0056 (0.0127) 0.0008 (0.0017) -0.0030* (0.0017) 0.0148 (0.0113) -0.0007 (0.0014) 0.0060** (0.0027) -0.0242 (0.0183) 0.0011 (0.0022) -0.0008 (0.0010) 0.0028 (0.0046) -0.0005 (0.0006) 0.0006 (0.0006) 0.0087*** (0.0029) -0.0009** (0.0004) 0.0016 (0.0018) -0.0110* (0.0063) 0.0008 (0.0008) Observations R2 96,017 0.071 96,067 00.296 96,067 0.232 337,506 0.036 337,903 0.079 337,903 0.138 Baseline mean 0.197 0.176 0.562 0.175 0.062 0.726 MC payments ($10) 1{Shortage} Payments * 1{Shortage} B. Child Subsample Medicaid (1) Trouble Finding MD MC payments ($10) 1{Shortage} Payments * 1{Shortage} -0.0056*** (0.0017) -0.0081 (0.0141) 0.0006 (0.0015) Private (2) No Usual Place of Care (3) (4) (5) 14+ School 14+ School Trouble Absences Absences Finding (Age≤10) (Age≥11) MD (6) No Usual Place of Care (7) (8) 14+ School 14+ School Absences Absences (Age≤10) (Age≥11) -0.0036* (0.0020) -0.0001 (0.0115) 0.0000 (0.0012) -0.0052 (0.0034) 0.0065 (0.0266 ) -0.0024 (0.0028 ) 0.0005 (0.0047) -0.0187 (0.0309) 0.0026 (0.0037) 0.0012 (0.0008) 0.0010 (0.0052) 0.0002 (0.0006) 0.0002 (0.0012) 0.0028 (0.0059) -0.0003 (0.0007) 0.0015 (0.0027) 0.0095 (0.0118) -0.0005 (0.0014) 0.0000 (0.0028) -0.0086 (0.0110) 0.0007 (0.0013) Observations R2 16,752 0.016 21,221 0.022 6,665 0.036 6,766 0.047 26,277 0.007 33,994 0.029 10,079 0.026 14,337 0.021 Baseline mean 0.022 0.034 0.046 0.070 0.008 0.022 0.023 0.034 C. Adult Subsample Medicaid Private (1) Not Accepting New Patients (2) Not Accepting Patient’s Insurance (3) Work Days Missed (4) Not Accepting New Patients (5) Not Accepting Patient’s Insurance (6) Work Days Missed -0.0084*** (0.0017) -0.0212 (0.0128) 0.0008 (0.0012) -0.0077** (0.0030) -0.0002 (0.0168) -0.0020 (0.0016) -0.2740 (0.3170) -2.3331 (2.8151) 0.2592 (0.3144) 0.0004 (0.0007) -0.0026 (0.0051) 0.0002 (0.0006) -0.0005 (0.0008) 0.0048 (0.0075) -0.0003 (0.0008) 0.1710 (0.1273) 0.7920 (0.7211) -0.1233 (0.0841) Observations R2 14,806 0.038 14,805 0.040 6,298 0.076 79,812 0.006 79,802 0.009 76,971 0.010 Baseline mean 0.062 0.082 5.010 0.017 0.025 3.711 MC payments ($10) 1{Shortage} Payments * 1{Shortage} Notes: Observations are at the individual level; standard errors are clustered by state. All regressions include state and quarter-year fixed effects and all individual and county-level controls listed in Table 2 (with age in five-year bins). Regressions are weighted using the sample weights provided in the NHIS. School days missed reflect absenteeism due to illness or injury over the previous year; the exact survey questions and the corresponding reference windows for all outcomes are outlined in Appendix A.1. Fewer children report days of missed school in the past year relative to other child outcomes because a child must be at least five years old to be asked this question. Similarly, only adults with employment histories are asked how many days of work they missed in the past year. 63