Male, Age 40+ Estimated Welfare Loss from the Obesity Externality

Who Pays for
Obesity?
Jay Bhattacharya
Stanford University
(with Kate Bundorf
and Neeraj Sood)
February 2008
Motivation

Obese individuals have more chronic diseases


Medical expenditures are greater on the obese



Higher rates of diabetes, heart disease,
hypercholesterolemia, hypertension, and stroke.
$31 billion (in year 2000 $) spent during 1996 for adult
overweight/obesity-related cardiovascular disease
treatments alone.
Among the overweight, per capita lifetime medical costs
can be reduced by $2,200 - $5,300 following a 10
percent reduction in body weight.
Should we care?
Rates of Obesity in the U.S.
Proportion Obese (BMI>=30) - Adults 20-55
0.35
0.30
0.25
0.20
Women
Men
0.15
0.10
0.05
0.00
1971-74
1976-80
1988-94
1999-2000
Source: Anderson, Butcher, and Levine, 2003. Authors’ calculations from the NHANES
The Incremental Annual Medical
Spending Attributable to Obesity
800
700
600
500
$ 400
300
200
100
0
732
423
247
143
All
Privately Insured
Overweight
Obese
Source: Finkelstein et. al. “National Medical Spending Attributable to Overweight and Obesity:
How Much, and Who’s Paying?”, Health Affairs – Web Exclusive, 14 May 2003.
Should We Care?

Obesity can have severe personal medical
and social consequences.


These consequences should (and often do) play
an important role in private decisions about body
weight.
But is obesity a “public” health crisis?

To what extent are the costs of obesity external?
Obesity Externalities

Private health insurance


Since medical costs are higher for the obese and
premiums do not depend on weight, lighter
people in the same pool pay for the
food/exercise decisions of the obese.
Government social programs

The negative health effects of obesity may
decrease the ability of the obese pay for and
increase the use of government social programs.


Disability insurance, Medicaid, Medicare
(Calculating the direction of the transfer is
complicated in the case of Medicare)
Aims of the Talk



Develop an economic framework for thinking
about health insurance/obesity externalities.
Estimate the external costs of obesity
assuming complete risk pooling.
Examine the extent of risk pooling among the
obese and non-obese in employer-provided
health insurance.
An Economic Framework for
Thinking about the Health
Insurance/Obesity Externality
Model Summary


Each consumer starts with an initial genetic
endowment of weight and maximizes
expected utility.
Losing weight (exercising, avoiding donuts)




Increases income (by reducing uninsurable
disability).
Decreases the probability of falling sick, which
in turn increases expected medical care costs.
Causes some disutility directly.
Consumers purchase insurance to insure
against health shocks
Two Regimes


Health insurance premium underwriting
depends on body weight.
Pooled health insurance.
Regime 1: Underwriting
Allowed



Insurance premiums equal expected
medical costs, given weight.
Under full insurance, consumption is the
same regardless of health state.
Two marginal benefits from weight loss:



Weight loss increases income
Weight loss lowers insurance premiums
One marginal cost from weight loss:

Direct disutility of dieting, exercising.
Regime 1: Welfare
Implications


No moral hazard problem as premiums are
dependent on weight
Consumers choose to lose weight even when
fully insured



They face the full costs of their weight choice
through the health insurance premium.
Weight loss is at socially optimal level
Full insurance is optimal when premiums are
actuarially fair
Regime 2: Pooled Insurance



Insurance premiums depend on the
distribution of weight within the pool.
Premiums are set at the expected level of
medical expenditures for the whole
insurance pool.
If the pool size is large, consumers pick
their weight without taking into account the
effect of their choices on premiums.

External costs of weight loss decisions are
imposed on other pool members.
Regime 2: First Order
Conditions

One marginal benefit from weight loss:


One marginal cost from weight loss:


Weight loss increases income
Direct disutility of dieting, exercising.
Unlike Regime 1, there is no incentive for
weight loss through decreased premiums.
Regime 2: Welfare
Implications

Weight loss lowers premiums for
everybody in insurance pool but
consumers ignore this when making
individual decisions



Weight loss creates a positive externality, and
hence is underprovided
Full insurance is not socially optimal
Consumer heterogeneity is not a
necessary condition for this result
Welfare Loss Due to the
Externality
 
 


DWL  EU **  EU *  U  . P W0  * 

Dead weight loss (DWL) under
pooling due to the obesity externality
is proportional to:


The effect of weight loss on expected
medical expenditures (P′).
The effect of insurance on body weight
decisions (Δω).
The State of the Literature

There is a large literature documenting the
difference in medical expenditures between
obese and non-obese populations.


A related literature documents that public and
private insurance pay for a high proportion of
obesity related expenditures.
Almost no work examines the effect of pooled
insurance on body weight decisions.
Policy Implications of the
Framework

If pooled health insurance does not cause
obesity (Δω = 0)


No social harm from obesity (through the
health insurance mechanism) even with full
insurance.
 Insurance induces transfers but no welfare
loss.
If pooled insurance changes body weight:

Potentially large social harm from not
underwriting premiums based upon body
weight.
The Welfare Loss from Pooled
Insurance: A Simulation
Exercise
Simulation Setup

Utility function:



U(c, ) = ln c -  2
Consumers can choose from one of three
weight categories – normal, overweight,
obese
Parameters




Probability distribution of initial weight
Disutility from weight loss: 
Co-Insurance: 
Estimated assuming pooling
Calibration

For a given set of parameters




Estimate weight distribution under each regime
Estimate expected medical expenditure
Estimate welfare loss from not allowing weightbased underwriting (CV)
Choose utility function parameter to match
weight distribution in data.
Modeling Health Care
Expenditures


Use standard two-part model of medical care
expenditures as a function age, sex, race, and
indicators of obesity and overweight.
Use parameter estimate to approximate the true
distribution by discrete distribution with 6 points of
support:
  0,50,100,1000,10000,50000
 k   k 1 
  k   k 1
Pr    k   Pr 
m

2
2


Medical Expenditures:
Female, Age 25-39
normal
overweight
obese
.6
.5
Probability
.4
.3
.2
.1
0
0
50
100
1000
10000
Medical Expenditure Category
50000
Medical Expenditures:
Male, Age 25-39
normal
overweight
obese
.6
.5
Probability
.4
.3
.2
.1
0
0
50
100
1000
10000
Medical Expenditure Category
50000
Medical Expenditures:
Female, Age 40+
normal
overweight
obese
.6
.5
Probability
.4
.3
.2
.1
0
0
50
100
1000
10000
Medical Expenditure Category
50000
Medical Expenditures:
Male, Age 40+
normal
overweight
obese
.6
.5
Probability
.4
.3
.2
.1
0
0
50
100
1000
10000
Medical Expenditure Category
50000
Estimated Welfare Loss from
the Obesity Externality
Group
Age 25-39
Welfare Loss
Change in Distribution of Weight from Obesity
Due to Pooled Premiums
Externality (Y)
Normal Overweight
Obese
Males
Females
Age 40+
-5%
0%
-9%
-16%
14%
16%
$7
$78
Males
Females
All Groups
0%
-7%
-3%
-19%
-14%
-15%
19%
21%
19%
$80
$304
$149
Measuring the Elasticity of Body
Weight with Respect to Insurance
Coverage: Three Studies
RAND Health Insurance
Experiment



Use data from the RAND health insurance
experiment to measure the insurance
elasticity of body weight
Take advantage of randomized insurance
assignment
Surprisingly, this elasticity is never reported in
the voluminous literature on the HIE.
Change in BMI per year
(1)
(2)
(3)
0.000
0.000
0.001
(0.23)
(0.60)
(0.85)
0.001
0.000
-0.000
(0.66)
(0.07)
(0.14)
0.022
-0.019
-0.042
(0.34)
(0.29)
(0.59)
-0.000
-0.000
0.000
(1.11)
(0.38)
(0.32)
0.148
0.294
0.341
(6.34)**
(3.00)**
(3.30)**
Observations
2628
2540
2540
Demographics
No
Yes
Yes
Site & Part.Incentive
No
No
Yes
R-squared
0.00
0.00
0.01
Outpatient Copay
Inpatient Copay
Deductible
Maximum OOP
Constant
Absolute value of t statistics in parentheses
* significant at 5%; ** significant at 1%
Probability of Turning Obese
(1)
(2)
(3)
0.000
0.000
0.000
(0.80)
(0.90)
(1.04)
-0.000
-0.000
-0.000
(0.25)
(0.53)
(0.73)
-0.011
-0.014
-0.019
(0.68)
(0.81)
(1.03)
-0.000
-0.000
0.000
(0.90)
(0.34)
(0.28)
0.010
0.086
0.112
(0.94)
(3.09)**
(3.68)**
Observations
3480
2969
2969
Demographics
No
Yes
Yes
Site & Part. Incentive
No
No
Yes
R-squared
0.00
0.01
0.02
Outpatient Copay
Inpatient Copay
Deductible
Maximum OOP
Constant
Absolute value of t statistics in parentheses
* significant at 5%; ** significant at 1%
Preliminary Conclusions


Consistent with zero insurance elasticity of
body weight
Limitations


Does not include zero insurance branch
The data are dated (thought there’s no reason to
think that the elasticity has changed)
An Empirical Examination of
Pooling in Employer-Provided
Health Insurance
Obesity and Wages in the
Labor Market



The wages of the obese are lower than similar
normal weight workers.
For men, these differences are explained by job
and occupation choice.
For women, these differences are less easily
explained, raising the concern that they are
attributable to invidious discrimination.
Study Design


Compare wages of obese and non-obese
workers with employer-sponsored health
insurance
Use the difference between the wages of the
obese and the non-obese workers without
employer-sponsored health insurance as a
control.
Data—NLSY


Nationally representative sample of people 1422 in 1979.
Survey years 1989-1998.


Health insurance status available after 1988.
Sample:


Full-time workers (usually worked 7+ hours per day at
full time job), excluding pregnant women.
Primary sample includes workers with employersponsored insurance and uninsured (35,750/24,805
worker-years)
NLSY Study Sample
1989 1990 1992 1993 1994 1996 1998
Overweight
31% 33% 34% 35% 37% 39% 39%
Obese
11% 12% 16% 16% 18% 21% 23%
Uninsured
23% 21% 24% 23% 21% 21% 19%
Age
29
30
32
33
34
36
38
Unadjusted Estimate of the Wage
Offset for Obesity
-$1.70
-$0.40
Insured
Uninsured
16
14
12
10
Hourly
8
Wage $
6
4
2
0
Obese
Non-obese
Difference-in-difference estimate: $-1.30 (p<=0.05)
Wages of Workers with Employer
Provided Health Insurance
Wages ($/hour)
19
17
15
13
11
9
7
5
1989 1990 1992 1993 1994 1996 1998
Obese
Non-Obese
Wages of Workers without
Employer Provided Health
Insurance
Wages ($/hour)
19
17
15
13
11
9
7
5
1989 1990 1992 1993 1994 1996 1998
Obese
Non-Obese
Difference in Difference Estimates
of the Wage Offset for Health
Insurance by Year
2
1
0
$
-1
-2
All
89
90
92
93
94
96
98
** *
***
** **
-3
-4
**
***
-5
Unadjusted
Adjusted
Effects of Other Fringe Benefits
on Wages
Table 4: Difference in difference estimates of the effect of incidence of other benefits on wages
Unadjusted
Adjusted
Fringe Benefit
n
Coefficient
SEs
n
Coefficient
SEs
Life Insurance
32643
-0.079
0.465
22914
0.111
0.499
Dental Insurance
32915
-0.518
0.492
23122
-0.838
0.543
Maternity Benefits
30801
-0.305
0.599
21405
-0.862
0.733
Retirement
32518
-0.121
0.532
22809
-0.414
0.618
Profit Sharing
32637
-0.602
0.596
22911
-0.382
0.682
Training/Education
32506
-0.300
0.487
22841
-0.183
0.556
Childcare
32292
0.888
1.520
22657
1.577
1.987
Flexible Working Hours
*** 1%, ** 5%, * 10% stat sig
32985
-0.638
0.497
23187
-0.125
0.580
Note: Standard errors adjusted for clustering within individual. We estimate these models on the sample of
workers employed full-time in each year either with employer sponsored coverage or uninsured and present both
unadjusted and adjusted estimates. The table entries show the coefficients and standard errors from the interaction
terms between obesity and fringe benefits offered from employers. Each table entry represents a different
regression. Full regression results are available in Appendix A4.
Can Lower Wages Of The Obese
Be Attributed To Higher Medical
Care Costs?
Table 5: The Effect of Obesity on Wages
Sample: Full-time workers either with current employer-sponsored coverage in their own name or uninsured
Pooled Sample
Men
Women
(1)
(2)
(3)
(1)
(2)
(3)
(1)
(2)
(3)
Obese
-0.87
-0.89
0.04
-0.68
-0.70
-0.64
-1.38
-1.38
0.85
(0.30)*** (0.30)***
(0.64)
(0.40)*
(0.39)*
(0.50)
(0.47)*** (0.47)***
(1.41)
Employer Coverage
1.81
2.01
2.10
2.12
1.27
1.78
(0.29)*** (0.29)***
(0.38)*** (0.42)***
(0.50)** (0.37)***
Obese*Employer Coverage
-1.20
-0.08
-2.89
(0.70)*
(0.69)
(1.40)**
Constant
7.37
7.63
7.43
5.61
6.11
6.10
9.79
9.68
9.16
(2.09)*** (1.99)*** (1.98)***
(2.49)** (2.39)** (2.39)**
(3.31)*** (3.21)*** (3.17)***
Observations
24085
24085
24085
14203
14203
14203
9882
9882
9882
R-squared
0.10
0.11
0.11
0.11
0.11
0.11
0.09
0.09
0.09
* significant at 10%; ** significant at 5%; *** significant at 1%
Note: Standard errors in parentheses. Standard errors are adjusted for repeated observations of individuals. Estimates
include controls for marital status, urban residence, age, firm size, job tenure, education, sex, race, year, AFQT score,
Data Sources

1998 Linked Medical Expenditures Panel
Survey (MEPS) and the National Health
Interview Survey (NHIS)


Medical expenditures and other control variables
from MEPS
Height and body weight from NHIS
Incremental Medical Care Costs Of
Obesity Relative To Normal Weight
By Sex
Table 6: Average Health Expenditures by Sex and Weight
Male
Normal
Weight
Obese
Sample
Difference
Aged 18-64
$1,721
$2,271
$551 *
Aged 20-50
$1,106
$1,061
-$45
Privately Insured and Aged 20-50
$1,086
$1,011
-$76
Data Source: 1998 Medical Expenditure Panel Survey
* significant at 10%; ** significant at 5%; *** significant at 1%
Female
Normal
Weight
$2,294
$1,536
$1,521
Obese
$3,277
$2,284
$2,190
Difference
$983 ***
$748 ***
$669 **
Sources of Expenditure
Differences Between Obese and
Non-obese
Table 7: Incremental Medical Expenditures Associated with
Obesity by Type of Service and Source of Payment
Sample: Privately Insured Individuals age 20-50
Total expenditure
By Type of Expenditure
Inpatient
Outpatient
Emergency
Prescription Drugs
Male
-$76
-$31
-$45
-$1
$74 *
Female
$669 **
$234 *
$435 **
$24
$103 **
By Source of Payment
Self-pay
-$16
-$17
Insured
-$60
$686 ***
Data Source: 1998 Medical Expenditure Panel Survey
* significant at 10%; ** significant at 5%; *** significant at 1%
Reconciling the Estimates


Incremental annual medical care costs of obesity for
women are approximately $700.
The annual wage offset for health insurance for
obese women is almost $6,000.


2.89 * 2,041 hours annually
Explanations for the difference



Loading of health insurance?
Residual discrimination concentrated in high end jobs that
provide health insurance?
Noise in the point estimates?
Conclusions

The welfare loss from the obesity externality
requires:




Pooled health insurance that induces a transfer from nonobese to obese individuals in the pool.
Increased body weight as a result of this transfer.
Obesity related wage offsets “undoes” pooling for
employer provided health insurance.
Two ways to limit the welfare loss from the obesity
externality in public insurance:


This externality arises because weight based underwriting
of health insurance premiums is not permitted.
Modest copayment can also limit these external effects.
MEPS Expenditure Difference
by Disease and Sex
Table 8: Expenditure and Prevalence Differences by Condition
Women
Disease Prevalence Differences
Thin
Obese Difference
Diabetes
1.15%
4.64%
3.49%
Asthma
9.29%
14.58%
5.30%
Hypertension
6.18%
22.14%
15.96%
Coronary Artery Disease
0.13%
0.68%
0.56%
Angina
0.18%
0.46%
0.29%
Myocardial Infarction
0.22%
0.69%
0.48%
Other Heart Disease
3.26%
4.46%
1.21%
Stroke
0.44%
0.62%
0.17%
Emphysema
0.10%
0.24%
0.14%
Joint Pain
22.53%
35.57%
13.04%
Arthritis
8.07%
17.96%
9.89%
se
0.49%
1.07%
1.02%
0.18%
0.17%
0.20%
0.64%
0.24%
0.13%
1.51%
1.06%
t
7.19
4.97
15.62
3.11
1.65
2.38
1.88
0.72
1.10
8.66
9.36
se
0.58%
0.98%
1.25%
0.30%
0.23%
0.30%
0.54%
0.18%
0.13%
1.63%
1.00%
9.30
-1.35
13.62
2.15
1.98
1.72
1.32
2.21
0.65
4.04
6.05
Expenditures Conditional on Disease Differences
Thin
Obese Difference se
t
*** $4,246
$5,769
$1,522
$1,261
1.21
*** $3,805
$4,147
$342
$635
0.54
*** $3,834
$4,278
$444
$596
0.75
*** $19,274
$6,641
-$12,633 $12,367
-1.02
* $2,637
$8,574
$5,937
$5,635
1.05
** $6,709
$8,240
$1,531
$5,301
0.29
* $4,333
$3,900
-$433
$1,592
-0.27
$10,728
$7,969
-$2,760
$3,285
-0.84
$13,712
$8,851
-$4,861
$7,415
-0.66
*** $3,740
$4,726
$987
$818
1.21
*** $4,141
$6,097
$1,956
$764
2.56
**
Thin
$5,425
$2,043
*** $3,276
** $12,618
** $7,766
* $11,812
$2,440
** $5,635
$1,781
*** $4,514
*** $2,926
*
Men
Diabetes
Asthma
Hypertension
Coronary Artery Disease
Angina
Myocardial Infarction
Other Heart Disease
Stroke
Emphysema
Joint Pain
Arthritis
Thin
1.23%
7.99%
9.76%
0.50%
0.28%
0.54%
2.03%
0.15%
0.11%
24.93%
6.54%
Obese
6.60%
6.66%
26.84%
1.15%
0.74%
1.06%
2.75%
0.56%
0.20%
31.53%
12.60%
Difference
5.38%
-1.33%
17.08%
0.64%
0.46%
0.52%
0.72%
0.41%
0.09%
6.59%
6.06%
***
Obese Difference se
$4,623
-$802
$1,350
$2,533
$490
$705
$2,996
-$280
$565
$6,959
-$5,658
$5,229
$9,610
$1,844
$4,383
$6,123
-$5,690
$5,324
$4,014
$1,574
$1,292
$12,730
$7,095
$6,693
$106
-$1,675
$1,007
$3,215
-$1,298
$2,718
$4,150
$1,224
$691
-0.59
0.70
-0.50
-1.08
0.42
-1.07
1.22
1.06
-1.66
-0.48
1.77