230B: Public Economics In-Kind Transfers - Public health insurance

Transcription

1 230B: Public Economics In-Kind Transfers - Public health insurance Hilary Hoynes Berkeley 1

2 Thanks to others The slides I am using this term are the result of a collaboration across lots of other faculty teaching PE including: Emmanuel and Gabriel Raj Chetty Day Manoli John Friedman Nathan Hendren Owen Zidar Amy Finkelstein 3

3 OUTLINE 1) Overview / Facts 2) Research on Medicaid 2a) Oregon Health Insurance Experiment 2b) Medicaid expansions 3) Valuing Medicaid and public health insurance 4) Research on Medicare 4a) RD on age 65 (Card, Dobkin and Maestas) 2

4 U.S. Health Insurance Mix of public and private insurance 1) Government provided insurance Medicare for the elderly (65+) Medicaid (and CHIP) for low income (children, disabled, elderly, some nonelderly adults) Veterans benefits 2) Privately provided insurance: Employer provided health insurance (large) Direct private purchase (small) Obamacare exchanges 3) Uninsured: (1/6) of population (before Obamacare)

5

6 Medicare Started in 1965 as a universal health insurance system Federal program that provides health insurance to all people over age 65 and disabled Every citizen who has worked for 10 years (or their spouse) is eligible. Financed with an uncapped payroll tax totaling 2.9%; this funds about 50% of Medicare expenses, remainder is funded with general funds Physician reimbursement fairly generous (but not as high as private insurance)

7 Medicaid Provides health care for low income pop (means-tested) Financed from general revenues (federal/state) Targets low income kids, SSI or TANF recipients, and elderly (for non-medicare costs such as long-term care) 70% of recipients are mothers/kids but only 25% of costs (66% of expenditure goes to long-term care for elderly) Doctor reimbursement low some docs refuse Medicaid Big variation across states in Medicaid generosity Joint federal / state financing (higher fed match for low income states) Program eligibility criteria have been expanded over time (higher incomes allowed): Obamacare substantially expands Medicaid to reduce uninsured (but not all states implemented the expansion)

8

9 OUTLINE 1) Overview / Facts 2) Research on Medicaid 2a) Oregon Health Insurance Experiment 2b) Medicaid expansions 3) Valuing Medicaid and public health insurance 4) Research on Medicare 4a) RD on age 65 (Card, Dobkin and Maestas) 4

10 Medicaid Causal Research Designs Oregon Experiment, on older childless [RCT] Staggered expansions to pregnant women and children 1980s-1990s [state x year] Expansions in 1980s/1990s; covered children born after 9/30/83 [RD] ACA Medicaid expansions [DD] Massachusetts 2006 mandate [DD] Medicaid Rollout ( ) [DD]

11 The Oregon Medicaid Health Insurance Experiment The first use of a randomized controlled design to evaluate the impact of Medicaid in the United States In 2008, Oregon wanted to expand its Medicaid program to cover more adults (financially / income eligible, but not categorically eligible for Medicaid) They had a limited budget and used a lottery to select who would receive the Medicaid insurance coverage 90,000 signed up for this new insurance Had budget to enroll 10,000 30,000 lottery winners (treatment group) 55,000 lottery losers (control group) Lottery winners won ability to apply for Medicaid (needed 30K to get 10K enrolled, did not know eligibility ex ante)

12 Oregon Medicaid Health Insurance Experiment (cont) From January 28 to February 29, 2008, anyone could be added to the lottery list by telephone, by fax, in person signup, by mail, or online. The state conducted an extensive public awareness campaign about the lottery opportunity. To keep barriers to sign-up low, the sign-up form requested limited demographic information on the individual and any interested household member, and no attempt was made to verify the information or screen for program eligibility at signup for the lottery. Eligibility: 19-64, Oregon resident, U.S. citizens or legal immigrants, without health insurance for at least six months, incomes below poverty, less than $2,000 in assets

13 What do we expect to find? Uninsured decreased (first stage) Costs: Utilization/Access: increased health care utilization (moral hazard, natural property of demand curve!) costs increase More preventative care, less ER decrease costs? Benefits Improved health? Improved financial security (insurance!) Increase employment and earnings (due to health and financial security) or decrease (due to work disincentive to keep eligibility)

14 Importance of context Effects of Medicaid may be smaller than general insurance effects Social safety net (uncompensated care, public clinics) already providing some coverage Low reimbursement rate means limited access, wait times But, may be on steep portion of health production frontier Can only examine effects over 2 years (not long run)

15 Data Administrative: hospital, credit reports, earnings Mail surveys (25K respondents, 50% response rate) In person interviews (12K respondents)

16 Who signed up? Individuals in Oregon who joined the waiting list are somewhat older and appear to be sicker than the overall target population. For example, of those who signed up, about 30% report having been diagnosed with high blood pressure compared to 17% in the target population, 33% are compared to 9% in the target pop. Not everyone who won the lottery ended up on Medicaid About 60 percent of those who won filled out the paperwork Among this group, only about half met the eligibility requirements for Medicaid coverage

17 Methods and preliminaries Balance test pre-randomization covariates balanced between T and C Lottery experiment Intent to treat: compare those selected and those not selected program eval of effect of expanding access to Medicaid LATE /IV: use lottery receipt as instrument for Medicaid coverage effect of Medicaid coverage

18

19 Main first stage: ever on Medicaid

20

21 More ER, more hospitalization (QJE 2012, Science 2014)

22 Is it a surprise that ER use increased? Price decreased to 0 utilization should go up But price of doctor s visit also 0 cross price effects suggest that ER use should go down if substitutes

23 More preventative care (NEJM 2013)

24 More diagnosis chronic conditions (NEJM 2013)

25 More access (NEJM 2013)

26 Reduction in collections, self reported hardship (QJE 2012)

27 Self reported health & depression improves, insignificant on other health (NEJM 2013)

28 No effect on earnings (or employment) (AER P&P 2014)

29 The Oregon Experiment: Summary Key results: winning the Medicaid lottery leads to: 1. higher health care utilization (including primary and preventive care as well as hospitalizations) 2. lower out-of-pocket medical expenditures and medical debt (including fewer bills sent to collection) 3. better self-reported physical and mental health 4. more diagnoses of conditions, no evidence that the chronic conditions are more under control

30 Beyond Oregon: Impacts on Financial Strain & Debt Consistent evidence of improvement in financial strain and debt Massachusetts 2006 mandate (Mazumder and Miller 2016) Medicaid (Gross and Notowidigdo 2011) ACA Medicaid expansions: (Hu, Kaestner, Mazumder, Miller and Wong 2016 wp) Medicare (Barcellos & Jacobson 2015, Dobkin et al 2016)

31 Source: Mazumder and Miller (2016). Massachusetts 2006, County pre-expansion design.

32 Source: Hu, Kaestner, Mazumder, Miller & Wong, NBER WP

33 OUTLINE 1) Overview / Facts 2) Research on Medicaid 2a) Oregon Health Insurance Experiment 2b) Medicaid expansions 3) Valuing Medicaid and public health insurance 4) Research on Medicare 4a) RD on age 65 (Card, Dobkin and Maestas) 6

34 Medicaid expansions (for children) Goals: extend Medicaid eligibility beyond those eligible for AFDC ( welfare lock ). DRA 1984: Required coverage for children (born after ) if family is AFDC income eligible [1985: similar leg. for pregnant women] OBRA 1986: Permitted states to expand coverage for children and pregnant women with income below 100% poverty OBRA 1987: Permitted states to cover pregnant women and children<age 1 with family income below 185% poverty. OBRA 1989: Required states to cover pregnant women and children<6 with family income below 133% poverty OBRA 1990: Required states to coverage children<19 (born after ) with family incomes below 100% poverty 1997 Further expansion through CHIP Many states expanded beyond what was required; across a wider range of child ages and/or higher up the income distribution

35 Example of how changes vary across states (Gruber TPE 1997)

36 Major increase in health insurance among children, through expansions to Medicaid and CHIP in the 1980s and 1990s.

37 MEDICAID EXPANSIONS FOR CHILDREN AND PREGNANT MOTHERS Increased Eligibility Expanding In reality, crowd-out Medicaid turns eligibility out to be affects an important both the uninsured issue. and the insured. Step 1 Previously Uninsured Previously Privately Insured Step 2 Step 3 Step 4 Take-up Crowd-out Medicaid Coverage Access Medical Utilization Health Outcomes Program costs Cost-Effectiveness Even with crowd-out, however, Medicaid coverage goes up. This leads to better access to health care and greater utilization. Increased This is weighed utilization against should program improve costs to health outcomes. determine costeffectiveness.

38 Cutler-Gruber QJE 1996: Crowd-out Crowd Out: When Medicaid expands, by how much does health insurance coverage change? Does increase in Medicaid lead to reduction in private insurance? Important for knowing expected impacts on outcomes and for distributional implications. Data: CPS , women 15-44

39 Empirical methods (how do we get to SIMIV?) Naïve Cross-Sectional Estimator I: Regress outcome (y) on dummy for Medicaid participation (M). Yi X i M i i Take up of Medicaid is endogenous Naïve Cross-Sectional Estimator II: Regress outcome (y) on dummy for Medicaid eligibility (E). Y i = α + βx i + γe i + ε i Eligibility is a function of income, and that too is endogenous Simulated IV: Instrument for E with SIMELIG Build state eligibility calculator (function of child age, family income and year) Use national sample to calculate SIMELIG each state, year, [child age] Goal: create measure of Medicaid generosity as a function of policy (not family characteristics). We use this approach when you can t use a single parameter (e.g. state top up rate for EITC) as descriptor of program generosity.

40 In the weeds: SIMELIG varies by state, year, age of child (in child regressions), so should control for all 2-way interactions (state x year, age x year, state x age). They didn t and future papers showed results are not necessary robust. Crowdout (KIDS): For every 10 percentage point increase in eligibility, you get 2.35 percentage point increase in Medicaid (Take-up 24%) and 0.7 percentage point decrease in private insurance Crowd-out = 0.07/0.24=31% Problem: women crowdout > 100%

41 Issues with state x year variation Endogenous policy no pre-trend analysis. States that had low rates had large increases. Card and Shore-Shepard: Analyses using SIPP show much smaller effects. Show results using state*year effects, include other FE No placebo effects the estimates are applied to a sample of ALL education/income levels. Maybe should estimate model on a group not expected to respond to the policy change.

42 Other Medicaid Studies using State x Year Expansions Currie and Gruber QJE 1996 Examine impacts on health care utilization and child health using NHIS Significant positive effects on access, utilization Currie and Gruber, Saving Babies, JPE 1996: Examine expansions of Medicaid on pregnant women Significant increases in prenatal care (NLSY) and birth weight and reduction in infant mortality (vital statistics) Many later studies

43 Evidence of Medicaid Impacts using Birthdate RD Much of the 1990s Medicaid expansions compelled states to expand benefits to children in poor families; they defined the universe as children born after September 30, 1983 Regression discontinuity Initial work using this design examined impact on health insurance (Card and Shore-Sheppard 2004, RESTAT) Many later studies (covered next week) showing impacts on long run outcomes

44 Source: Wherry and Meyer (2015) Increase in eligibility by cohort

45 Source: Wherry and Meyer (2015) More on First Stage

46 Results from Medicaid Expansions 1. Crowd-out is important to measure: private insurance declines 20-50% of the public insurance increases 2. Utilization of health services increased: Early prenatal care visits and child regular visits rose by more than 50% (Currie and Gruber 1996) 3. Health care outcomes improved: Infant mortality declined (Currie and Gruber 1996) Child mortality declined (Wherry and Meyer 2015, Currie and Schwandt 2016) Highly cost-effective policy Contrast to evidence showing less substantial impacts on adults

47 Summary: Medicaid Public health insurance for adults leads to: Reductions in OOP spending Reductions in financial strain But, beneficiaries generally not willing to pay full cost Public health insurance for children leads to: Reductions in infant and child mortality Reductions in future medical costs and chronic conditions

48 OUTLINE 1) Overview Facts 2) Research on Medicaid 2a) Oregon Health Insurance Experiment 2b) Medicaid expansions 3) Valuing Medicaid and public health insurance 4) Research on Medicare 4a) RD on age 65 (Card, Dobkin and Maestas) 8

49 How health insurance is different from other inkind transfers Health care needs are highly variable, across people and over time (food is predictable, need it every period) Health insurance benefits are unlikely to free up a lot of cash period-to-period; not fungible (vs. SNAP. Highly fungible) Health insurance allows risk averse individuals to smooth MU of consumption in the face of unanticipated out of pocket medical expenses 9

50 Informal Welfare Analysis Media: Medicaid Makes Big Difference in Lives, Study Finds (NPR 2011) Spending on Medicaid Doesn t Actually Help the Poor (Washington Post 2013) Public policy centers: Oregon s lesson to the nation: Medicaid Works (Oregon Center for public policy 2013) Oregon Medicaid Study Shows Michigan Medicaid Expansion Not Worth the Cost (MacKinac Center for Public Policy 2013) 10

51 Observations on the value of health insurance Health insurance confers benefits to previously uninsured: reductions in out of pocket spending, reduced medical debt, improvements in health Uninsured pay small share of their medical expenses (prevalence of uncompensated care) Less than full take-up of public health insurance Recipients value formal health insurance at less than actuarial cost Growing evidence that substantial share of benefits of expanding insurance goes to firm side who would otherwise bear the costs of providing uncompensated care Distributional implications of public subsidies for health insurance depend on the economic incidence of these transfers (this is largely unknown) 11

52 Estimating Willingness to Pay Useful source for general discussion: Finkelstein, Mahoney and Notowidigdo (2017 nber wp, forthcoming Annual Review of Economics) Here we will present the basic modeling approach Discuss two methods to estimate the WTP: model based approach (ex-post outcomes) and revealed preference approach (demand curve) 12

53 Setup (common to both approaches) Individuals derive utility from health, h, and consumption of non-medical goods and services, c: u = u (c, h) Health h produced according to h = h(m; θ) Medical spending, m θ denotes underlying state variable (medical conditions, other factors affecting health, etc.) Assume each Medicaid recipient faces same distribution of θ 13

54 Setup (Cont) Goal: Estimate recipients value (γ) of in-kind transfer of (free) health insurance (e.g. Medicaid) Definition: γ is amount of money we could take away from insured that would leave them indifferent between having insurance (without γ) and being uninsured [compensating variation] 14

55 Defining recipient value of Medicaid Goal: Estimate recipients value (γ) of in-kind transfer of (free) health insurance (e.g. Medicaid) Notation: Presence of Medicaid denoted by q [0, 1] Define c(q; θ), h(q; θ) and m(q; θ) Welfare impact of Medicaid γ on recipients is the implicit solution to: E θ [u(c(0; θ), h(0; θ)] = E θ [u(c(1; θ) γ, h(1; θ))] 15

56 Two Approaches to Estimating γ Ex-post (outcomes-based) approach: Use estimated outcomes for individual with and without Medicaid to infer and compare utility with and without Medicaid Finkelstein, Hendren, Luttmer (2015) The Value of Medicaid: Interpreting Results from the Oregon Health Insurance Experiment Ex-ante (demand) approach: Estimate demand for Medicaid Demand curve reveals willingness to pay for Medicaid (γ) Finkestein, Hendren and Shepard (2017) Subsidizing health insurance for low-income adults: Evidence from Massachusetts 16

57 Outcomes Approach: FHL 2015 wp Oregon Health insurance experiment Specify utility function over c and h, implicit def of γ Use empirical estimates from Oregon to get distribution of h and c under T (q = 1) and C (q = 0) h measured several ways (self reported health, depression screen - each valued with quality of life years) c not measured in experiment proxy for consumption by the difference between average consumption for a low-income uninsured population and out-of-pocket medical expenditures reported by study participants Finding γ is (expressed as a share of costs) 17

58 Revealed Preference Approach: FHS 2017 wp Idea: If you could randomly vary the premium at which formal insurance is offered, then you could identify the wtp for those who take up as the price falls; demand curve shows willingness to pay for insurance Ideally: randomly vary price and observe number who buy at different price levels What they do: Use RD in prices faced for public health insurance Setting: 2006 Subsidized health insurance exchange in MA introduced in subsidies for low income non-elderly uninsured adults below 300% FPL 18

59 Finkelstein, Hendren and Shepard 2017

60 Finkelstein, Hendren and Shepard 2017

61 Finkelstein, Hendren and Shepard 2017

62 Finkelstein, Hendren and Shepard 2017 Cost = $420/month Median WTP = $100 γ/cost < 0.25

63 Key Assumption: Demand Curve Reveals Value why might it not? Economic frictions: liquidity constraints behavioral frictions: inattention, mis-information 20

64 Medicaid Value - General findings Ex-ante and ex post approaches yield similar result: value of Medicaid among poor adults is far below cost Implication: modest enrollee premiums will be deterrent to enrollment Why? Leading explanation is presence of implicit insurance Takes the form of uncompensated care from hospitals, free clinics, and uncollected medical debt (e.g., Garthwaite et al. 2015). Implies there are other beneficiaries of expansion of public health insurance Estimates imply that if there is no implicit insurance then value would be % of actuarial value More to be learned about possible role of behavioral biases

65 Open question: ultimate economic incidence What is ultimate incidence of implicit insurance Share holders (for profit hospitals) Buyers (patients, insurance companies) Suppliers (employees, drug/device manufacturers) Hospital competitors (community health centers) Local, state and federal governments

66 OUTLINE 1) Overview / Facts 2) Research on Medicaid 2a) Oregon Health Insurance Experiment 2b) Medicaid expansions 3) Valuing Medicaid and public health insurance 4) Research on Medicare 4a) RD on age 65 (Card, Dobkin and Maestas) 22

67 Medicare: Background and Program Details Basic Facts: Nearly universal coverage for elderly (65+) Benefits: Part A (Hospital), Part B (Physician s services) both with deductible and copayments, Part D (prescription drugs) History: Medicare enacted in July 1965; nearly universal implementation in July Single largest increase in health insurance coverage in the U.S.; increase in insurance coverage by 75 percentage points among elderly Eligibility: Can get at age 65 if you (or spouse) has worked for 40 quarters. Part A is free; Part B is optional and has low monthly premium; Part D is drug coverage.

68 Medicare Causal Research Designs Regression discontinuity compare those just over age 65 to those just under 65 Effects on insurance across groups, utilization, outcomes [Card, Dobkin and Maestas AER 2008] Effects on outcomes / mortality [Card, Dobkin and Maestas QJE 2009] Program Introduction compare outcomes before and after 1965

69 CDM AER 2008 Outcomes examined: Health insurance coverage (1st stage, CPS, NHIS) Health care utilization (BRFSS, NHIS) Health behaviors (BRFSS, NHIS) Hospital stays (Census of hospital discharges from CA, FL and NY, HCUP data) Measuring age: Limit to Data allows for measurement of age in quarters CDM QJE 2009 California Hospital Discharge File, 1/92-12/02, post-discharge mortality (to test if better/different services while in hospital) Select only hospital admissions that come through the emergency department (ED) for severe illnesses [= codes with same admit rate on weekdays and weekends] This gets out the elective surgery folks, such as knee replacement, that dominate the increase in hospital admissions

70 Identification? Standard checks for RD: No confounding factors that are also changing (discontinuously) near age 65 Other factors are (mostly) smoothly changing near age 65 (in contrast to discontinuous change due to Medicare) Possible confounders? Retirement, income, family structure, changes in medical guidelines at 65 Is the right experiment / what is the LATE here? The RD captures short term changes in health care utilization (and impacts on mortality) from shift from <65 to >65 Another question (not answerable in this framework) is how does access to Medicare change outcomes through older ages?

71 CDM Methodology y X h ( a) D ija ija j j a j ija y = first stage (insurance coverage) or outcome (health care utilization, mortality) a = age j = socioeconomic group (since we expect larger increases in coverage/outcomes for more disadvantaged groups) h(a) = smooth function of age D = dummy if 65+ π = treatment effect

72 First stage: sharp increase in coverage; more for disadvantaged Source: Card, Dobkin, and Maestas AER 2008

73 Bigger reduction at 62 Testing for confounders; estimate same model on employment: -- small employment discontinuity at age 65 (bigger at 62) -- they also look at marriage, poverty, mobility; no discontinuity

74 Increase in hospital admissions; driven by discretionary medical care, diagnostic heart treatments. (substitution over time?) Source: Card, Dobkin, and Maestas AER 2008

75 Nontrivial decrease in mortality (CDM QJE 2009) Source: Card, Dobkin, and Maestas QJE 2009

76 Other evidence on Medicare A second approach is to analyzing Medicare introduction Finkelstein & McKnight JPUBE 2008 Chay et al (working paper) Diff-Diff using cohort comparisons Both studies show dramatic decline in mortality, larger for more disadvantaged groups.