Subsidizing Health Insurance for Low-Income Adults: Evidence from
Subsidizing Health Insurance for Low-Income Adults: Evidence from Massachusetts Amy Finkelstein MIT and NBER Nathaniel Hendren Harvard University and NBER Mark Shepard Harvard Kennedy School and NBER November 2017 Motivation How much are low-income people willing to pay (WTP) for health insurance? What are the implications for insurance markets? Health insurance is by far the largest means-tested transfer in US In 2015: $550 billion on Medicaid vs. $70b or less on next biggest programs (food stamps, EITC, SSI, TANF) Key question: How much are recipients WTP for in-kind insurance? Little prior evidence until recently a non-traded good Health care reforms: Increasingly try to cover low-income uninsured via partially subsidized insurance ACA exchanges, state Medicaid reforms requiring premiums Will partial subsidies get to universal coverage? Depends on WTP Overview of This Paper Evidence from setting where can measure revealed preference WTP
(demand) and cost of health insurance for low-income population Setting: Subsidized exchange in Massachusetts (pre-ACA) Model for ACA: Similar design, low-income population choosing b/n heavily subsidized coverage vs. uninsurance Key feature: Subsidies vary by discrete income bin RD price variation E.g., 149% poverty person has $0 plan; 151% poverty pays $39/month Method: Use price variation to estimate WTP, cost of insurance Descriptive: How do lower subsidies affect take-up, average cost of enrollees in insurance market? Use simple model to map observed take-up, cost WTP, cost curves Summary of Results Descriptive findings: Insurance demand is highly price sensitive (2) Adverse selection (despite coverage mandate) Enrollee premium by $40 per month Take-up falls 25%, Average cost of enrollees higher by $10-50 (= 3-14%) (1) Implications of WTP, Cost Curve Estimates: Modest premiums are major deterrent to universal coverage E.g., subsidies covering 75% of insurer costs still leave >50% uninsured (2) Adverse selection exists but is not primary driver of low take-up WTP is far below not just average costs, but also marginal enrollees
costs imposed on insurer (for >70% of eligible pop.) (1) Discussion: Why is WTP so far below cost? And what are the normative implications? Key factor: Uncompensated care for uninsured (charity care, unpaid bills) Outline 1. Setting and Descriptive Evidence 2. Model and Estimates of WTP, Cost Curves 3. Discussion and Normative Implications 4. Conclusion 1. Setting and Descriptive Evidence Setting: Mass. Health Insurance Exchange Setting: Pre-ACA subsidized insurance exchange (CommCare) Introduced in 2006 Romneycare reform ACA-like eligible population: Low-income adults (<300% poverty) w/out access to employer insurance or other public insurance Relevant choice: CommCare plan vs. Uninsurance Focus on 135-300% poverty and simple market setting in 2011
Also show descriptive results for 2009-13 Benefits standardized, set to be quite generous Cover >94% of costs (like platinum plan in ACA) Each insurer offers a single plan, can set networks of doctors/hospitals Subsidies are also generous (cover >80% of insurer prices/costs) Set to make lowest enrollee premium affordable (0-5% of income) 400 Subsidy and Premium Discontinuities (2011) $ per month 300 200 Insurer Price Public Subsidies 100 $116
0 $39 $0 135 150 $77 Enrollee Premium 200 250 Income, % of Poverty Affordable Amt. (cheapest plan) 300 400 Subsidy and Premium Discontinuities (2011) $ per month 300 200
Insurer Price Public Subsidies ~$147 100 ~$106 ~$58 ~$11 0 Four Other Plans $39 $77 Enrollee Premium $116 Cheapest plan
$0 135 150 200 250 Income, % of Poverty 300 Data and Sample Construction 1. Plan enrollment (CommCare admin data) Start with individual-level data Collapse to plan enrollment by income (% of FPL) 2. Insurer medical costs (CommCare claims data) Linked to enrollment; Collapse to average cost by income (% of FPL) 3. Eligible population size (American Community Survey) Restrict to CommCare-eligible people based on observables (age, income, insurance status, citizenship) Graph Estimate smoothed eligible population size using this samples avg. insurance take-up rate (63%) and income distribution Do robustness checks on these assumptions
RD Analysis and Assumption RD Analysis: Use premium change at income thresholds to test for: Demand response: Does enrollment fall? Adverse Selection: Do average costs of enrolled population increase (suggesting healthy differentially leave)? Key assumption: No strategic income manipulation around cutoffs I.e., Eligible population is smooth through cutoffs so any enrollment change is driven by demand response No evidence of strategic manipulation 1. Institutional: Mapping from reported info to income measure used for subsidies (income as % of FPL) is not salient to applicants 2. No bunching in income distribution of eligible pop. in ACS ( Graphs) 3. No spikes/holes in enrollment distribution around cutoffs in 2011 RD = -1054 (318) % = -26% RD = -641 (157) % = -27% 3000
4000 5000 Enrollment Counts, by Income (2011) 0 1000 2000 RD = -326 (99) % = -25% 135 150 200 250 Income, % of Poverty 300 4500
Raw Data: Enrollment and Eligible Population (2011) Raw Count of CommCare Enrollment 0 1500 3000 (Smoothed) Estimate of Eligible Population Size 135 150 200 250 Income, % of Poverty 300 1 Share of Eligible Population Insured
340 380 420 Average Insurer Costs, by Income (2011) 133 150 200 250 Income, % of Poverty 300 Summary of Descriptive Evidence Low-income insurance demand is highly price sensitive Falls ~25% for each ~$40 increase in cheapest enrollee premium Modest premiums are a major deterrent to coverage for low-income population Cost RDs: Evidence of adverse selection Average cost rises by 2-15% ($6-50 / month) as covg. falls at RDs Enrollees who leave are lower cost than remaining people
How to translate into WTP/cost curves? Need a model (next step) 2. Model and Estimates of WTP, Cost Curves Translating Evidence into WTP, Cost Curves Goal: Use this evidence to measure implied WTP, cost curves Starting point: Einav, Finkelstein, Cullen (2010) framework Uses exogenous price variation to estimate (and graph) insurance demand and cost curves Extend framework to allow >2 options that are vertically ranked Focus on 2011 where market has useful vertical structure Four plans with broader networks set price w/in $3 of binding cap ($426) One limited-network plan (CeltiCare) set lower price ($405) Grouping: L plan (CeltiCare), H plan (all others), U (uninsured) Vertical model assumptions: 1. Vertical preferences: Everyone prefers H > L 2. Single index (s) of WTP heterogeneity (= WTP for generosity) Vertical Model and Identification WTP WH Identification:
PH PL WWHL Variation in PL + Observed Share(H+L) Identifies WL(s) Variation in PH PL + Observed Share(H) Identifies WHL(s) Calculate WH(s) = WL(s) + WHL(s) WL PL Buy H Buy L s*HL Buy U s*LU
40 WTP for H relative to L (WHL(s)) Adjusted Assumption WHL250% FPL WHL(s,y) = WHL(s + y, 150%) 30 (0.31, $31) Adjusted WHL WHL200% FPL (0.64, $19) (150% FPL) 10 $/month 20
(0.44, $29) WHL150% FPL 0 (0.80, $11) .2 .4 .6 s .8 1 160 Final WTP Curves for Insurance 120 WH(s)
WHL(s) $ per month 80 (0.36, $116) 40 (0.50, $77) (0.70, $39) WL(s) 0 (0.94, $0) 0 .2 .4 s
.6 .8 1 Estimating Cost of Marginal Consumers Efficiency in insurance markets depends on comparison b/n WTP (just estimated) vs. cost of marginal consumers Adverse selection: Marginal consumers are lower-cost than average Cost curve notation: Cj(s) = Insurer cost (in plan j) for consumer of type s = Cost of marginal enrollees if Pj = Wj(s) (in market with only plan j) ACj(s) = Insurers average cost if cover all types with WTP Wj(s) Use effect of price variation on ACH to estimate CH(s) Analogous to Einav, Finkelstein, Cullen (2010) Focus on CH but have alternate method for estimating CL(s) ( Details) Identifying Cost of H Plan CH(s) Cost Observed Avg. Cost
of H = ACH(s*HL) 0 Buy H Buy L s*HL s Uninsured 1 Identifying Cost of H Plan: Using Price Variation CH(s) Cost Observed Avg. Cost (PH2) Observed Avg. Cost (PH1) Infer CH(s) Premium Increase
0 Buy H s*HL(P2) s 1 s*HL(P1) Observed Average Costs (H Plan) 400 150% FPL Average Cost (H) $ per month 200 300 200% FPL 100
250% FPL 0 H and L Plan Cost RDs .2 .3 .4 .5 .6 .7 Fraction in H Plan .8 .9 1 Adjusted Average Cost Curve (H Plan) 400
s .7 .8 .9 1 400 Result #2: Little Take-up w/out Large Subsidies $ per month 200 300 ACH(s) CH(s) WH(s) Demand well below average cost 0
100 Premium = 25% of AC 49% take-up Premium = 10% of AC 79% take-up .2 .3 .4 .5 .6 s .7 .8 .9 1 400
Result #3: Adverse selection alone cannot explain low coverage $ per month 200 300 ACH(s) CH(s) WH(s) 0 100 Demand also well below enrollees own expected costs .2 .3 .4 .5 .6
s .7 .8 .9 1 Robustness Main result: Low-income WTP for insurance is far below own expected medical costs (for entire 70% of eligible pop. we observe) Also Adverse selection: Marginal enrollee costs < Average costs Robust to a variety of econometric, modeling choices ( Table) 1. Alternate RD specifications 2. Alternate take-up estimates from ACS 3. Accounting for mandate penalty (using normalized premiums) Low WTP not driven by inertia ( New enrollee RDs) Low WTP robust to relaxing vertical model ( Bounds method) 3. Discussion and Normative Implications
Discussion Two key questions: 1. Why is willingness to pay lower than individuals own costs? 2. What are the normative implications? Why is WTP Below Own Costs? Behavioral biases (inattention, inertia, information, misperception) Similar demand responsiveness for new enrollees argues against inattention / inertia as explanations But cannot rule out behavioral biases more broadly Moral hazard textbook reason for WTP < Cost But required magnitude not plausible Moral hazard would have to increase costs by ~200% to explain gap between WTP and Costs Oregon experiment moral hazard estimate: 25% Uncompensated care (charity care, unpaid bills) Important: Low-income uninsured pay just ~20% of their medical costs out of pocket (Finkelstein, Hendren, Luttmer 2015) Use this + moral hazard estimate to construct Net Cost Curve = Cost to insurer Savings to third parties on uncompensated care
400 WTP and Cost Curves, Adjusted for Uncomp. Care $ per month 200 300 ACH(s) Uncompensated Care Estimate CH(s) 100 Net Cost = CH(s)-CU(s) 0 WH(s) .2 .3
.4 .5 .6 .7 Fraction in H Plan (= s) .8 .9 1 Normative Implications Normative Question: Would a small subsidy increase be desirable? Note: Need to assume WTP curves are true metrics of consumer welfare Framework: Calculate Marginal Value of Public Funds (MVPF) spent on subsidy increase (following Hendren (2016)) MVPF Marginal WTP to Beneficiaries Marginal Cost to Govt.
Benefit-cost test: Is MVPF > relevant threshold for alternate use of funds? (e.g., 0.90 for EITC (Hendren 2016)) Main finding: Incidence of uncompensated care matters a lot 1. No value (pure waste) MVPF [0.28, 0.56] 2. Govt. cost savings / Benefit to low-income MVPF [0.80, 1.29] 3. Benefit to affluent (e.g., hospital CEOs) MVPF [0.67, 0.71] Conclusion Exploit discontinuities in premiums in Massachusetts to estimate WTP and cost of insurance among low-income population Main Finding: WTP is far below insurer costs Adverse selection exists but is not the key issue Enrollees not willing to pay own expected cost of coverage Uncompensated care provides a potential explanation Implications for economics of subsidizing health insurance Partial subsidies unlikely to get close to universal coverage Enrollees themselves may not be primary beneficiaries of subsidies Important topic for further work: Incidence and efficiency of uncompensated care vs. formal subsidized insurance Thank You!
Appendix Slides 0 5000 10000 15000 Estimate of Eligible Population from ACS (2011) 135 150 Go back 200 250 Income (% of Poverty) 5% FPL bins Regression Smoothed 300 0
5000 10000 15000 Estimate of Eligible Population from ACS (2009-2013) 135 150 Go back 200 250 Income (% of Poverty) 300 0 5000 10000 15000
Estimate of Eligible Population from ACS (2011) 135 150 Go back 200 250 Income (% of Poverty) 5% FPL bins Regression Smoothed 300 46 47 Average Age of Enrollees (2009-2013) 45 RD = 0.61 (0.25)
TIMED PAIR SHARE. On the last day he has to have a reformation meeting.. Does anyone know what this means? A reformation meeting is a meeting that Marcus has with someone from the Crime Prosecution Service and his Youth Worker...
It is divided into fundus, body and neck. branches of the ventral primary rami of spinal nerves S3-S4 elevates the pelvic floor anococcygeal raphe and the coccyx arcus tendineus levator ani and the ischial spine iliococcygeus deep branch of perineal...
y = sin x. y = cos x. y = tan x. And the reciprocal ratios: Period and Amplitude. ... Differentiating Trigonometric Functions: sin x. Consider the gradient of the curve y = sin x. The gradient function is actually...
Athena's symbols were the distaff and the aegis- a goat-like shield, to which the head of the Gorgon Medusa was fastened in order to terrify the opponents. Athena's sacred animal was the owl, the symbol of wisdom.
lnKeq = Slope = Intercept Van't Hoff plot Structure & Properties of Water Bent geometry, O-H bond length of 0.958Å Can form Hydrogen bonds Nonpolar/Polar Interactions and Structured Water A cage of water molecules (calatherate) surrounding the non-polar molecule. This...
Introduction to Physical Science Chapter 1, Section 1 What is Physical Science? Science is the way of learning about the natural world. Physical Science is the study of matter, energy, and the changes they undergo. Matter is anything that has...
Ready to download the document? Go ahead and hit continue!