投影片 1

Health Insurance and the Demand for
Medical Care: Evidence from a
Randomized Experiment
Willard G. Manning et al. (1987)
June 1, 2007
Willard G.
Question: Demand response
The size of demand response
 Demand response and income
 Demand response and age group
 Demand response and different services
 Demand response and HMO
 Health use and health status

Problems

People who face lower price use more health care
→ demand response

But in reality Insurance (e.g. choices of different copayment rate) is endogenous



People who face lower price use more health insurance
People who need more use more health insurance
Bias in the estimation
Solutions: random experiment

People cannot choose what they want. They are
randomly assigned to different insurance programs

Thus



No self-selection
No income effect
Pure moral hazard problem
Experiment Design

Time:



Observations:



Nov, 1974-Feb, 1977
70% for 3 years and 30% for 4 years
5809 persons from six cities (FFS)
1982 (HMO)
Insurance:




Coinsurance rate: 0, 25, 50, 95
Upper limit: 5,10,15 (income)
Annual Maximum Out-pocket expense (MDE): 1000
Individual deductible – 95% coinsurance rate for
outpatient, and free for inpatient care (has upper limit)
Statistical Methods

Three characteristics of distribution of medical
expenses:




A large proportion use no medical services.
Medical expense is highly skewed.
Distribution is different between outpatient and inpatient
use
Methods


ANOVA
Four part equations
Four part equations
Results (I)

Large medical increase when co-payment rate moves
from 25% to 0%; but a much smaller increase when
moving from 95% to 50% or 50% to 25%

Total Exp (0%)~1.5 Total Exp (95%)
Results (II)

EXP(In) does not change with coinsurance rate

EXP(Out) increases significantly when coinsurance
rate moves from 0 to 25%
Results (III)

No large demand response on inpatient service

U shape expenditure: probably due to the upper
limit
Results (IV)
Children are less responsive for inpatient care.
 Adults have significant lower use of inpatient
services on the family-pay plans than on the free
plan.
 For other subgroups, for example, health status
(healthy v.s. sick), there is no evidence to show the
differential response to health insurance coverage
between these two subgroups

Results (IV)
Health status outcome: Patients with relatively
prevalent chronic problems (high blood pressure,
myopia) have specific gains in use of free FFS rather
in use plans with cost sharing.
 HMO results (table 7):



Same rate of using service among the plans.
The participants with one or more hospital admissions
differs with plans.
Random experiment
Is it worth it?
 Cost: 80 millions
 Benefit?

Was it worth it?
Rand Health Insurance Experiment cost $80 million
 Initial results published in 1981





In the next 2 years, # of insurance companies with firstdollar coinsurance for hospital care increased from 30%
to 63%
# of insurance companies with annual deductible of $200
per person increased from 4% to 21%
Estimated cost saving from the demand for medical care
= $7 billion
Government sponsored studies often yield
important knowledge for business