Comments on Levy’s
description
Transcript of Comments on Levy’s
Comments on Levy’s“Health Insurance among Low-Skilled Adults
over the Business Cycle”
Rucker JohnsonUniv of California, Berkeley
Prepared for NPC “Working & Poor” Research Conference
June 9-10, 2005
Why Care about Loss of Health Insurance?Why Care about Loss of Health Insurance?(A little perspective on the paper)(A little perspective on the paper)
Some lose employer-sponsored HI & join public HI-- adds to budget strain
Uninsured may get less med treatment (Doyle 2001)
Uninsured may impose costs due to inefficient care mgmt (ER care was preventable if treated @ ofc visit)
Economic security & risk of bankruptcy in event of negative health shock
Despite strong economic growth, and expanding public Despite strong economic growth, and expanding public coverage, the rate of health insurance coverage fell during the coverage, the rate of health insurance coverage fell during the 1990s.1990s. 13.7% of non-elderly uninsured (1987)13.7% of non-elderly uninsured (1987) 15.8% of non-elderly uninsured (2000)15.8% of non-elderly uninsured (2000)
PuzzlePuzzle: Low-skilled had largest gains in employment : Low-skilled had largest gains in employment and largest largest declines in HIdeclines in HI
Goal of Levy paperGoal of Levy paper: Explain decline in HI coverage among : Explain decline in HI coverage among less-skilled in booming ’90sless-skilled in booming ’90s
Approach:Approach: Use CPS (’90-’03), relate HI coverage trends in late Use CPS (’90-’03), relate HI coverage trends in late ’90s boom, and ’01 downturn to changes in income and ’90s boom, and ’01 downturn to changes in income and employment by educ/gender– employment by educ/gender– How much can be explained? How much can be explained? Was decline driven by changes in public or private coverage?Was decline driven by changes in public or private coverage?
Factors that Affect HI Coverage
Labor market conditions Much of cyclical var in coverage related to change in
economic conditions (Cawley & Simon, 2005) Health care costs-- rising premiums
1% increase in premium drop of 300,000 (Lewin) Much of secular trend in HI coverage due to higher health
care costs (Chernew, Cutler, Keenan, 2002) Availability of public coverage
Medicaid/SCHIP expansions (Currie & Gruber, 1996) Structural changes of economy
Explain little of change in HI coverage over time (Glied & Stabile, 2000; Acs, 1995)
Demographic changes Changing value of alternatives (charity care) Regulatory changes Changing Load
Taxes
What about Changing Distribution of Coverage?
Decline not evenly distributed, concentrated among low-income adults
Prior research examines either effects of costs or economic conditions; But rarely both or distributional effects across
groups
• How does effect of macroeconomy on HI coverage differ for men, women, & children, by education and race?
• To what extent does public HI coverage compensate for secular and cyclical changes in private coverage, by gender, education, and race?
Overview of Trends in HI Coverage: Overview of Trends in HI Coverage: 1990-20031990-2003
Econometric SpecificationEconometric Specification
Identifying sources of variation: Exploit time and state variation in economic
conditions
Model: P(HI)= B1 * (state unemp rate) +
B2 * (own emp) + B3 * (spouse emp) +
B4 * (fam income) + B5 * (demographic vars)
+ B6 * (Medicaid generosity) +
(year dummies) + (state dummies)
Where HI = different health insurance outcomes
Main Concerns about Paper
Incomplete Characterization/Accounting Incomplete Characterization/Accounting
of Health Insurance Dynamics over periodof Health Insurance Dynamics over period
Decomposition Analysis—issues of interpretationDecomposition Analysis—issues of interpretation
Issues of Measurement that Raise Concern w/Analysis
Limitations of CPS Records whether covered by HI at any time in last 12 mos
Cannot use CPS to determine HI coverage in specific month matched w/macroeconomic cond’ns for that month
Multiple changes in survey question over time captured by year FE
Use of cross-sectional data Inability to remove unobserved time-invariant person-
specific heterogeneity Longitudinal data reveal much larger share of pop at risk
for being uninsured Short (‘04), using SIPP, finds ½ of persistently uninsured are missed
in svys that count only those uninsur for 12 consecutive months
Some Unresolved Issues
Does paper decompose relative roles of economic conditions, health care costs, Medicaid/SCHIP expansions, employment status, for health insurance coverage trends among low-skilled adults?
Does paper adequately acc’t for competing explanations over this time period?
Does paper help us to understand where risks of future gaps in coverage are likely to be greatest?
Decomposition Analysis: Trends in HI Coverage, 1990-2003
Total change can be decomposed as:
90
90909090
n
X
n
X
n
X
n
Xetotalchang i i
j
i ij
j
i ij
j
i jijj
Portion of Δs b/w yrs due to Δs in X
Portion of Δs in HI not explained by Δs in X: “residual” effect
• Cannot decompose residual component–- state dummies included
• State*yr dummies capture Δs in economic conditions, health care costs, public program generosity
• Sensitivity of decomposition estimates to inclusion of state unemp rate and proxy for state HI costs?
Relate time profile of residual effect with known trends of other factors that may have driven Δs in HI
Source: Brady & Lin, 2005
Source: Brady & Lin, 2005
Source: Brady & Lin, 2005
Health Insurance is Dynamic
Half of uninsured spells end within 5-6 months (Short, 2004)
Number uninsured part of a year number uninsured all year As many people lose or gain coverage as
remain uninsured Considerable turnover in uninsured
population
Timing matters in counting, characterizing, and covering the uninsured
Analyzing HI Dynamics using PSID (1997-2003)
Nationally-representative sampleHealth Insurance Info Individuals asked:
# of mos w/HI coverage in each yr b/w 1997-2002 Type of coverage in each yr b/w 1997-2002 HI premium costs Total HH medical care costs Out-of-pocket costs (hospital, Dr.office, Rx drugs)
Health status measures
Merged State-Level Data,1997-2002Merged State-Level Data,1997-2002
State Unemp RateState Unemp Rate BLSBLS
Health Insur CostsHealth Insur Costs Medicare hospital wage indexMedicare hospital wage index
Medicaid Medicaid CoverageCoverage
John Cawley & Kosali Simon: John Cawley & Kosali Simon: Similar to Cutler/Gruber MethodSimilar to Cutler/Gruber Method
% Unionized% Unionized Hirsch et al. (2001)Hirsch et al. (2001)
Special thanks to John Cawley and Kosali Simon for sharing their state-level data for Special thanks to John Cawley and Kosali Simon for sharing their state-level data for 1997-2002.1997-2002.
Merged data on:Merged data on: SCHIP/Medicaid EligibilitySCHIP/Medicaid Eligibility
Computed from running detailed simulation programs Computed from running detailed simulation programs (created by Cawley and Simon) on March CPS (created by Cawley and Simon) on March CPS respondents as in Cutler and Gruber (1996)respondents as in Cutler and Gruber (1996)
Take Take allall March CPS children in 1996, and calculate March CPS children in 1996, and calculate the weighted fraction of them that would be eligible the weighted fraction of them that would be eligible for Medicaid or SCHIP in a particular state in a for Medicaid or SCHIP in a particular state in a particular year. That fraction is used as measure of particular year. That fraction is used as measure of Medicaid/SCHIP generosity. The measure varies by Medicaid/SCHIP generosity. The measure varies by state and year.state and year.
Empirical Approach (using PSID) (similar to Cawley & Simon, ’05)
Separate models for men, women, children, by education (interactions w/race)—restrict to non-elderly
Estimate models w/dependent vars: Whether employer-sponsored HI full yr Whether uninsured all yr Govt-sponsored HI at any time (2-yr period) Medical care expenditures (2-yr period)
Explanatory var of interest: State unemp rate Include individual-specific & yr-specific fixed effects State-level controls for Medicaid generosity, HI costs, % unionized
Identification of effect of macroeconomic conditions on probability of HI coverage comes from variation w/in people over time in deviations from nat’l mean in that yr.
11.7
26.9
53.1
23.9
0
10
20
30
40
50
60
70
80
0Mos
0-12Mos
0-24Mos
0-36Mos
0-48Mos
0-60Mos
0-71Mos
72Mos
Whites
Minorities
Less-Educated Men Men (Especially Minorities)(Especially Minorities) Have High Uninsured Rates Have High Uninsured RatesPSID: #of Months w/PSID: #of Months w/Any HIAny HI,1997-2002,1997-2002
HS Dropouts
26.7
47.6
38.7
15.1
0
10
20
30
40
50
60
70
80
90
0Mos
0-12Mos
0-24Mos
0-36Mos
0-48Mos
0-60Mos
0-71Mos
72Mos
Whites
Minorities
Less-Educated Men Less-Educated Men (Especially Minorities)(Especially Minorities) Have High Uninsured Rates Have High Uninsured RatesPSID: #of Months w/PSID: #of Months w/ESI,ESI,1997-20021997-2002
HS Dropouts
10.5
21.7
54.2
34.6
0
10
20
30
40
50
60
70
0Mos
0-12Mos
0-24Mos
0-36Mos
0-48Mos
0-60Mos
0-71Mos
72Mos
Whites
Minorities
Less-Educated Women (Especially Minorities) Have High Uninsured RatesPSID: #of Months w/Any HI,1997-2002
HS Dropouts
33.5
51.6
33.9
18.2
0
10
20
30
40
50
60
70
80
90
0Mos
0-12Mos
0-24Mos
0-36Mos
0-48Mos
0-60Mos
0-71Mos
72Mos
Whites
Minorities
Less-Educated Women (Especially Minorities) Have High Uninsured RatesPSID: #of Months w/PSID: #of Months w/ESIESI,1997-2002,1997-2002
HS Dropouts
25
30
35
40
45
50
55
60
65
1997 1998 1999 2000 2001 2002
White Men
White Women
Minority Men
Minority Women
Less-Educated AdultsLess-Educated AdultsPSID: % with PSID: % with ESI Full YrESI Full Yr,1997-2002,1997-2002
HS Dropouts
10
15
20
25
1997-1998 1999-2000 2001-2002
White Women
Minority Women
Less-Educated WomenLess-Educated WomenPSID: % with PSID: % with Govt HI (at any time)Govt HI (at any time),,1997-2002
HS Dropouts
TABLE 1. TABLE 1. PSID men, HS dropout; whether HI as a fn of macroeconomic conditions
Linear probability model coefficients (Robust std errors)
Full-yr ESI Uninsured All-yr
Govt HI any time (2-yr per)
State unemp rate
State unemp rate*non-white -.0546***
(.0181)
State unemp rate*white -.0211
(.0206)Individual-specific fixed effects Yes
Yr-specific fixed effects Yes
State-level controls for: Medicaid generosity, HI costs, Unionization
Yes
Mean of dependent var .4609
# Observations 3,443
TABLE 1. PSID men, HS dropout; whether HI as a function of macroeconomic conditions
Linear probability model coefficients (Robust std errors)
Full-yr ESI Uninsured All-yr Govt HI any time (2-yr per)
State unemp rate
State unemp rate*non-white -.0546*** .0444**
(.0181) (.0227)
State unemp rate*white -.0211 -.0019
(.0206) (.0219)
Individual-specific fixed effects Yes Yes
Yr-specific fixed effects Yes Yes
State-level controls for: Medicaid generosity, HI costs, Unionization
Yes Yes
Mean of dependent var .4609 .3312
# Observations 3,443 3,443
Full-yr ESI Uninsured All-yr Govt HI any time (2-yr per)
State unemp rateState unemp rate -.0140-.0140
(.0113)(.0113)
State unemp rate*non-whiteState unemp rate*non-white -.0546***-.0546*** .0444**.0444**
(.0181)(.0181) (.0227)(.0227)
State unemp rate*whiteState unemp rate*white -.0211-.0211 -.0019-.0019
(.0206)(.0206) (.0219)(.0219)
Individual-specific fixed effectsIndividual-specific fixed effects YesYes YesYes YesYes
Yr-specific fixed effectsYr-specific fixed effects YesYes YesYes YesYes
State-level controls for: Medicaid State-level controls for: Medicaid generosity, HI costs, Unionizationgenerosity, HI costs, Unionization
YesYes YesYes YesYes
Mean of dependent varMean of dependent var .4609.4609 .3312.3312 .0981.0981
# Observations# Observations 3,4433,443 3,4433,443 1,7591,759
TABLE 1. PSID men, HS dropout; whether HI as a function of macroeconomic conditions
Linear probability model coefficients (Robust std errors)
TABLE 2. TABLE 2. PSID Women, HS dropout; whether HI as a fn of macroeconomic conditions
Linear probability model coefficients (Robust std errors)
Full-yr ESIFull-yr ESI Uninsured Uninsured All-yrAll-yr
Govt HI any Govt HI any time (2-yr per)time (2-yr per)
State unemp rateState unemp rate
State unemp rate*non-whiteState unemp rate*non-white -.0289*-.0289*
(.0162)(.0162)
State unemp rate*whiteState unemp rate*white -.0112-.0112
(.0229)(.0229)Individual-specific fixed effectsIndividual-specific fixed effects YesYes
Yr-specific fixed effectsYr-specific fixed effects YesYes
State-level controls for: Medicaid State-level controls for: Medicaid generosity, HI costs, Unionizationgenerosity, HI costs, Unionization
YesYes
Mean of dependent varMean of dependent var .3885.3885
# Observations# Observations 4,2224,222
Full-yr ESI Uninsured All-yr Govt HI any Govt HI any time (2-yr per)time (2-yr per)
State unemp rateState unemp rate
State unemp rate*non-whiteState unemp rate*non-white -.0289*-.0289* .0044.0044
(.0162)(.0162) (.0112)(.0112)
State unemp rate*whiteState unemp rate*white -.0112-.0112 -.0154-.0154
(.0229)(.0229) (.0201)(.0201)Individual-specific fixed effectsIndividual-specific fixed effects YesYes YesYes
Yr-specific fixed effectsYr-specific fixed effects YesYes YesYes
State-level controls for: Medicaid State-level controls for: Medicaid generosity, HI costs, Unionizationgenerosity, HI costs, Unionization
YesYes YesYes
Mean of dependent varMean of dependent var .3885.3885 .3161.3161
# Observations# Observations 4,2224,222 4,2224,222
TABLE 2. TABLE 2. PSID Women, HS dropout; whether HI as a fn of macroeconomic conditions
Linear probability model coefficients (Robust std errors)
Full-yr ESI Uninsured All-yr Govt HI any time (2-yr per)
State unemp rateState unemp rate -.0188-.0188
(.0176)(.0176)
State unemp rate*non-whiteState unemp rate*non-white -.0289*-.0289* .0044.0044
(.0162)(.0162) (.0112)(.0112)
State unemp rate*whiteState unemp rate*white -.0112-.0112 -.0154-.0154
(.0229)(.0229) (.0201)(.0201)Individual-specific fixed effectsIndividual-specific fixed effects YesYes YesYes YesYes
Yr-specific fixed effectsYr-specific fixed effects YesYes YesYes YesYes
State-level controls for: Medicaid State-level controls for: Medicaid generosity, HI costs, Unionizationgenerosity, HI costs, Unionization
YesYes YesYes YesYes
Mean of dependent varMean of dependent var .3885.3885 .3161.3161 .2149.2149
# Observations# Observations 4,2224,222 4,2224,222 2,1492,149
TABLE 2. TABLE 2. PSID Women, HS dropout; whether HI as a fn of macroeconomic conditions
Linear probability model coefficients (Robust std errors)
TABLE 3. PSID Children of HS dropout; whether HI as a fn of macroeconomic conditions
Linear probability model coefficients (Robust std errors)
Full-yr ESI Uninsured Uninsured All-yrAll-yr
Govt HI any Govt HI any time (2-yr per)time (2-yr per)
State unemp rateState unemp rate -.0402***-.0402***
(.0151)(.0151)
Individual-specific fixed effectsIndividual-specific fixed effects YesYes
Yr-specific fixed effectsYr-specific fixed effects YesYes
State-level controls for: Medicaid State-level controls for: Medicaid generosity, HI costs, Unionizationgenerosity, HI costs, Unionization
YesYes
Mean of dependent varMean of dependent var .2792.2792
# Observations# Observations 10,58710,587
Full-yr ESIFull-yr ESI Uninsured All-yrUninsured All-yr Govt HI any Govt HI any time (2-yr per)time (2-yr per)
State unemp rateState unemp rate -.0402***-.0402*** .0317**.0317**
(.0151)(.0151) (.0126)(.0126)
Individual-specific fixed effectsIndividual-specific fixed effects YesYes YesYes
Yr-specific fixed effectsYr-specific fixed effects YesYes YesYes
State-level controls for: Medicaid State-level controls for: Medicaid generosity, HI costs, Unionizationgenerosity, HI costs, Unionization
YesYes YesYes
Mean of dependent varMean of dependent var .2792.2792 .2715.2715
# Observations# Observations 10,58710,587 10,58710,587
TABLE 3. PSID Children of HS dropout; whether HI as a fn of macroeconomic conditions
Linear probability model coefficients (Robust std errors)
Full-yr ESI Uninsured All-yr Govt HI any time (2-yr per)
State unemp rate -.0402*** .0317** .0118
(.0151) (.0126) (.0180)
Individual-specific fixed effects Yes Yes Yes
Yr-specific fixed effects Yes Yes Yes
State-level controls for: Medicaid generosity, HI costs, Unionization
Yes Yes Yes
Mean of dependent var .2792 .2715 .3914
# Observations 10,587 10,587 5,234
TABLE 3. PSID Children of HS dropout; whether HI as a fn of macroeconomic conditions
Linear probability model coefficients (Robust std errors)