Finkelstein Etal Oregon 2012 PDF

THE
QUARTERLY JOURNAL
OF ECONOMICS
Vol. 127 August 2012 Issue 3
Downloaded from http://qje.oxfordjournals.org/ at MIT Libraries on September 24, 2012

THE OREGON HEALTH INSURANCE EXPERIMENT:
EVIDENCE FROM THE FIRST YEAR*
Amy Finkelstein
Sarah Taubman
Bill Wright
Mira Bernstein
Jonathan Gruber
Joseph P. Newhouse
Heidi Allen
Katherine Baicker
Oregon Health Study Group
In 2008, a group of uninsured low-income adults in Oregon was selected by

lottery to be given the chance to apply for Medicaid. This lottery provides an
opportunity to gauge the effects of expanding access to public health insurance
on the health care use, financial strain, and health of low-income adults using a
randomized controlled design. In the year after random assignment, the treat-
ment group selected by the lottery was about 25 percentage points more likely
to have insurance than the control group that was not selected. We find that in
this first year, the treatment group had substantively and statistically signifi-
cantly higher health care utilization (including primary and preventive care as
well as hospitalizations), lower out-of-pocket medical expenditures and medical
debt (including fewer bills sent to collection), and better self-reported physical
and mental health than the control group. JEL Codes: H51, H75, I1.
*We are grateful to Josh Angrist, Robert Avery, David Autor, Ethan
Cohen-Cole, Carlos Dobkin, Esther Duflo, Jack Fowler, Guido Imbens, Larry
Katz, Jeff Kling, John McConnell, Jon Levin, Richard Levin, Ben Olken, Alan
Zaslavsky, three anonymous referees, and numerous seminar participants for
helpful comments and advice; to Brandi Coates, Michael Gelman, John Graves,
Ahmed Jaber, Andrew Lai, Conrad Miller, Iuliana Pascu, Adam Sacarny,
Nivedhitha Subramanian, Zirui Song, James Wang, and Annetta Zhou for
expert research assistance; and to numerous Oregon state employees for help
acquiring the necessary data and for answering our many questions about the
! The Author(s) 2012. Published by Oxford University Press, on behalf of President and
Fellows of Harvard College. All rights reserved. For Permissions, please email: journals
.permissions@oup.com
The Quarterly Journal of Economics (2012), 10571106. doi:10.1093/qje/qjs020.
Advance Access publication on May 3, 2012.
1057
1058 QUARTERLY JOURNAL OF ECONOMICS
I. INTRODUCTION
In early 2008, Oregon opened a waiting list for a limited
number of spots in its Medicaid program for low-income adults,
which had previously been closed to new enrollment. The state
drew names by lottery from the 90,000 people who signed up. This
lottery presents an opportunity to study the effects of access to
public insurance using the framework of a randomized controlled
design.

Although the effects of health insurance on health and health
care use may seem intuitive, and there have been hundreds of
studies on the topic, research in this area has often been ham-
pered by the difficulty of controlling for unobserved differences
between the insured and uninsured (Levy and Meltzer 2008).
Random assignment of health insurance to some but not others
administration of state programs. We gratefully acknowledge funding from the

Assistant Secretary for Planning and Evaluation in the Department of Health
and Human Services, the California HealthCare Foundation, the John D. and
Catherine T. MacArthur Foundation, the National Institute on Aging
(P30AG012810, RC2AGO36631, and R01AG0345151), the Robert Wood Johnson
Foundation, the Sloan Foundation, the Smith Richardson Foundation, and the U.S.
Social Security Administration (through grant 5 RRC 08098400-03-00 to the
National Bureau of Economic Research as part of the SSA Retirement Research
Consortium). We also gratefully acknowledge Centers for Medicare and Medicaid
Services matching funds for this evaluation. The findings and conclusions ex-
pressed are solely those of the author(s) and do not represent the views of SSA,
the National Institute on Aging, the National Institutes of Health, any agency of the
federal government, any of our funders, or the NBER. In addition to the individuals
named as coauthors for this article, the Oregon Health Study Group includes Matt
Carlson (Portland State University), Tina Edlund (Deputy Directory, Oregon
Health Authority), Charles Gallia (Oregon DHS), Eric Schneider (RAND), and
Jeanene Smith (Office for Oregon Health Policy and Research). Author disclosures:
Finkelstein served on the CBOs Panel of Health Advisers through 2011. Wright is
employed by Providence Health & Services, a nonprofit integrated health care de-
livery system. Gruber was a paid technical consultant to the Obama administration
during the development of the Affordable Care Act and a paid consultant to the state
of Oregon for modeling health insurance expansion options, and serves on the
CBOs Panel of Health Advisers. Newhouse is a director of and holds equity in
Aetna, which sells Medicaid policies, and serves on the CBOs Panel of Health
Advisers. Allen is employed by Providence Health and Services, a nonprofit inte-
grated health care delivery system. She formerly served as director of the Medicaid
Advisory Committee and as staff to the Oregon Health Fund Board at the Office for
Oregon Health Policy and Research. Baicker is a MedPAC commissioner, serves on
the CBOs Panel of Health Advisers, is a director of Eli Lilly, has received honoraria
from several physician groups for speaking engagements, and previously served on
the Bush administrations Council of Economic Advisers.
OREGON HEALTH INSURANCE EXPERIMENT 1059
would avoid such confounding, but such opportunities are rarely

available.1
In this article we examine the effects of the Oregon Medicaid
lottery after approximately one year of insurance coverage. We
present comparisons of outcomes between the treatment group
(those selected by the lottery who had an opportunity to apply for
Medicaid) and the control group (those not selected and thus not
able to apply for Medicaid). We also present estimates of the
impact of insurance coverage, using the lottery as an instrument

for insurance coverage.
We organize our analysis around the potential costs and
benefits of health insurance. On the cost side, we examine the
impact of health insurance on increased health care utilization.
On the benefit side, we examine the impact of health insurance on
self-reported health, financial strain, and overall well-being. By
lowering the price of health care, health insurance is expected to
increase health care utilization. Ultimately additional health
care utilization may translate into improved health, although a
one-year window might be too short a time to observe health
improvements. Much less attention has been given in the litera-
ture to other potential benefits of health insurance. Because
risk-spreading is arguably the primary purpose of health insur-
ance (e.g., Zeckhauser 1970), we try to examine the impact of
health insurance on consumption smoothing, which we proxy
for with measures of financial strain. We also examine the
impact of health insurance on overall well-being, specifically
self-reported happiness; this may capture, among other things,
any benefits of health insurance from reductions in stress or
stigma.
The impact of Medicaid among a low-income population may
be lower than that of private insurance or insurance among
higher income individuals. The impact of Medicaid may be atte-
nuated (or potentially nonexistent) if public health clinics and
1. We know of only two other randomized health insurance experiments in the

United States. The RAND Health Insurance Experiment from the 1970s was de-
signed to investigate the marginal impact of varying insurance cost-sharing fea-
tures among approximately 6,000 insured individuals, not the effect of insurance
coverage itself (Manning et al. 1987; Newhouse and the Insurance Experiment
Group 1993). The more recent Accelerated Benefits Demonstration project was
designed to investigate the impact of health insurance for uninsured disabled
adults receiving Social Security Disability Insurance during the two-year waiting
period for Medicare (Michalopoulos et al. 2011).
uncompensated care allow low-income individuals to consume de

facto free medical care similar to that of the insured. Medicaids
impact would also be attenuated ifas is often claimed
Medicaid itself is not particularly good insurance in terms of
being able to access health care providers (e.g., Medicaid Access
Study Group 1994; GAO 2011; Rosenbaum 2011).
Our analysis draws on administrative data from hospital
discharge, credit report, and mortality records, as well as on re-
sponses to a large mail survey we conducted. The administrative

data are objectively measured and should not be biased by the
treatment and control groups differentially reporting outcomes,
but they only cover a relatively narrow set of outcomes. The
survey data allow examination of a much richer set of outcomes
than is feasible with administrative data alone, but with a 50%
effective response rate, are subject to potential nonresponse bias.
Our available evidence on this issue is limited but reasonably
reassuring.
Prior to looking at the data on outcomes for the treatment
group, virtually all of the analysis presented here was prespeci-
fied and publicly archived in a detailed analysis plan.2 Although
prespecification of hypotheses is the norm for randomized con-
trolled medical trials, is it rare in evaluation of social policy ex-
periments.3 Our prespecification was designed to minimize issues
of data and specification mining and to provide a record of the full
set of planned analyses.
About one year after enrollment, we find that those selected
by the lottery have substantial and statistically significantly
higher health care utilization, lower out-of-pocket medical
expenditures and medical debt, and better self-reported health
than the control group that was not given the opportunity to
apply for Medicaid. Being selected through the lottery is asso-
ciated with a 25 percentage point increase in the probability of
having insurance during our study period. This net increase in
2. Our analysis plan was archived on December 3, 2010, at http://www.nber

.org/sap/20101203/. Some of those analyses yielded little of interest, and therefore
we describe them briefly, presenting the full results only in appendixes. In the few
instances in which the results suggested the performance of additional analyses
that had not originally been planned, we have indicated this in the text and tables.
3. In economics, within the last few years, prespecification of hypotheses has
started to become more common in analyses of randomized experiments in develop-
ing countries (e.g., Alatas et al. 2010; Olken, Onishi, and Wong 2010; Schaner 2010;
Casey, Glennerster, and Miguel 2011).
insurance appears to come entirely through a gross increase in

Medicaid coverage, with little evidence of crowd-out of private
insurance. Using lottery selection as an instrument for insurance
coverage, we find that insurance coverage is associated with a 2.1
percentage point (30%) increase in the probability of having a
hospital admission, an 8.8 percentage point (15%) increase in
the probability of taking any prescription drugs, and a 21 per-
centage point (35%) increase in the probability of having an out-
patient visit. We are unable to reject the null of no change in

emergency room utilization, although the confidence intervals
do not allow us to rule out substantial effects in either direction.
In addition, insurance is associated with 0.3 standard devi-
ation increase in reported compliance with recommended pre-
ventive care such as mammograms and cholesterol monitoring.
Insurance also results in decreased exposure to medical liabilities
and out-of-pocket medical expenses, including a 6.4 percentage
point (25%) decline in the probability of having an unpaid medical
bill sent to a collections agency and a 20 percentage point (35%)
decline in having any out-of-pocket medical expenditures.
Because much medical debt is never paid, the financial incidence
of expanded coverage thus appears to be not only on the newly
insured but also on their medical providers (or whomever they
pass the costs on to).
Finally, we find that insurance is associated with improve-
ments across the board in measures of self-reported physical and
mental health, averaging 0.2 standard deviation improvement.
Two pieces of evidence suggest that the improvements in self-
reported health that we detect may at least partly reflect a gen-
eral sense of improved well-being. First, evidence from a separate
survey we conducted very shortly after random assignment
shows no impact of lottery selection on health care utilization
but improvements in self-reported health that are about two-
thirds the magnitude of our main survey results more than
a year later. Second, we find that one year later, Medicaid is
associated with about a 32% increase in self-reported overall
happiness, albeit reported in the context of a survey primarily
about health. Whether there are also improvements in objective,
physical health is more difficult to determine with the data
we now have available. More data on physical health, including
biometric measures such as blood pressure and blood sugar, will
be available from the in-person interviews and health exams
that we conducted about six months after the time frame in this
article.
Our estimates of the impact of public health insurance apply
to able-bodied uninsured adults below 100% of poverty who ex-
press interest in insurance coverage, a population of considerable
policy interest. In 2011, fewer than half of the states offered
Medicaid coverage to able-bodied adults with income up to
100% of poverty absent specific categorical requirements
(Kaiser Family Foundation 2011). As part of the 2010 Patient

Protection and Affordable Care Act, starting in 2014 all states
will be required to extend Medicaid eligibility to all adults up to
133% of the federal poverty level, with no financial penalties for
many individuals in this income range who do not take up cover-
age (Kaiser Family Foundation 2010a, 2010b; U.S. GPO 2010).
The rest of the article is structured as follows. Section II pro-
vides background on the Oregon Medicaid program and the lot-
tery design. Section III describes the primary data sources, and
Section IV presents our empirical framework. Section V presents
our main results. Section VI discusses interpretation and ex-
trapolation of our estimates. The Online Appendix provides add-
itional details.
II. OREGONS MEDICAID LOTTERY

The Oregon Health Plan (OHP)created by one of the first
federal waivers of traditional Medicaid rulescurrently consists
of two distinct programs: OHP Standard and OHP Plus. OHP
Plus serves the categorically eligible Medicaid population,
which includes (up to specific income thresholds) children and
pregnant women, the disabled, and families enrolled in
Temporary Assistance to Needy Families (TANF). OHP
Standard, which is the program that was lotteried, is a
Medicaid expansion program to cover low-income adults who
are not categorically eligible for OHP Plus. Specifically, it
covers adults ages 1964 not otherwise eligible for public insur-
ance who are Oregon residents, are U.S. citizens or legal immi-
grants, have been without health insurance for six months, have
income below the federal poverty level (FPL), and have assets
below $2,000 (Office for Oregon Health Policy and Research
2009).
OHP Standard provides relatively comprehensive benefits

with no consumer cost sharing. It covers physician services, pre-
scription drugs, all major hospital benefits, mental health and
chemical dependency services (including outpatient services),
hospice care, and some durable medical equipment. Vision is
not covered, nor are nonemergency dental services. Wallace
et al. (2008) estimate that in 20012004, average annual
Medicaid expenditures for an individual on OHP Standard were
about $3,000. Most care is provided through managed care organ-

izations. Monthly enrollee premiums range from $0 to $20 de-
pending on income, with those below 10% of the FPL paying $0.
At its peak in early 2002, about 110,000 people were enrolled
in OHP Standard, about one-third the size of OHP Plus enroll-
ment at that time. Due to budgetary shortfalls, OHP Standard
was closed to new enrollment in 2004. By early 2008, attrition had
reduced enrollment to about 19,000 and the state determined it
had the budget to enroll an additional 10,000 adults. Therefore, in
January 2008 the state reopened OHP Standard to new
enrollment.
Because the state (correctly) anticipated that the demand for
the program among eligible individuals would far exceed the
10,000 available slots, it applied for and received permission
from the Centers for Medicare and Medicaid Services to add the
new members through random lottery draws from a new reser-
vation list. From January 28 to February 29, 2008, anyone could
be added to the lottery list by telephone, by fax, in person sign-up,
by mail, or online. The state conducted an extensive public aware-
ness campaign about the lottery opportunity. To keep barriers to
sign-up low, the sign-up form (shown in Online Appendix Figure
A2) 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
sign-up for the lottery. A total of 89,824 individuals were placed
on the list during the five-week window it was open.
The state conducted eight lottery drawings from the list with
roughly equal numbers selected from each drawing; the drawings
were fairly evenly spaced from March through September 2008.
Selected individuals won the opportunityfor themselves and
any household member (whether listed or not)to apply for
OHP Standard coverage. Treatment thus occurred at the house-
hold level. In total, 35,169 individualsrepresenting 29,664
householdswere selected by lottery. If individuals in a selected
household submitted the appropriate paperwork within 45 days

after the state mailed them an application and demonstrated that
they met the eligibility requirements, they were enrolled in OHP
Standard.4 About 30% of selected individuals successfully en-
rolled. There were two main sources of slippage: only about 60%
of those selected sent back applications, and about half of those
who sent back applications were deemed ineligible, primarily due
to failure to meet the requirement of income in the last quarter
corresponding to annual income below the poverty level, which in

2008 was $10,400 for a single person and $21,200 for a family of
four (Allen et al. 2010). If they did successfully enroll in OHP
Standard, individuals could remain enrolled indefinitely, pro-
vided that they recertified their eligibility status every six
months.
III. DATA
We briefly describe each data source here. Additional details
can be found in Online Appendix 1.
III.A. Administrative Data on Outcomes: Hospital Discharges,

Credit Reports, and Mortality
We obtained standard individual-level hospital discharge
data for the entire state of Oregon from January 2008 through
September 2009 and probabilistically matched them to the lottery
list based on information provided at the time of lottery sign-up
on full name, ZIP code, and date of birth. The data include a
hospital identifier, dates of admission and discharge, source of
admission, detail on diagnoses and procedures, and discharge
destination. Similar discharge data have been used to study the
impact of health insurance in other contexts (see, e.g., Doyle
2005; Card, Dobkin, and Maestas 2008, 2009; Anderson,
Dobkin, and Gross 2010). Although inpatient admissions are rela-
tively rare (the annual admission rate for our controls is only
about 5%), they are expensive, accounting for about one-quarter
4. The state reviewed applications, first examining eligibility for OHP Plus
and then, if not eligible for Plus, examining eligibility for OHP Standard. Those who
did not apply during this window could not apply later (so unlike those categorically
eligible for Medicaid/OHP Plus, did not have conditional coverage if unenrolled).
of total medical expenditures for 1864-year-olds.5 We observe,

on average, about five months of prerandomization data.
We obtained detailed credit records from TransUnions
Consumer Credit Database. Credit bureaus like TransUnion
collect rich and detailed information on virtually all formal con-
sumer borrowing gleaned from public records, collections agen-
cies, and trade lines such as credit card balances (but do not
capture informal borrowing such as through relatives or pawn-
brokers). The analysis of such data is still relatively rare in the

economics literature and, to our knowledge, has never been done
before in a health insurance context.6 TransUnion used the full
name, full address, and date of birth reported at sign-up to match
68.5% of lottery participants to their prerandomization credit
report in February 2008. The credit bureau was able to track
over 97% of those found in the February 2008 file to their
September 2009 file. Our primary outcomes of financial strain
are measured in this 2009 file, which thus has an effective post-
randomization attrition rate of 3%. We also observe prerando-
mization outcomes measured in February 2008.
We obtained mortality data from Oregons Center of Health
Statistics for all deaths occurring in Oregon from January 1,
2008, through September 30, 2009, and probabilistically matched
our sample using full name, ZIP code, and date of birth.
III.B. Survey Data on Outcomes

We supplement the outcome measures available in the ad-
ministrative data with a mail survey that was sent out in seven
waves over July and August 2009 to virtually all individuals se-
lected by the lottery and a roughly equal number of unselected
individuals.7 The complete survey instrument is shown in Online
Appendix Figure A4. The basic protocol involved three mail
attempts. In addition, we designed a more intensive protocol,
which we conducted on approximately 30% of nonrespondents.
It included additional tracking efforts, mailings, and phone
5. Author calculations based on publicly available tables from the 2008

Medical Expenditure Panel Survey.
6. Avery, Calem and Canner (2003) provide an excellent, detailed discussion of
credit bureau data; much of our discussion of the data and our choice of analysis
variables is guided by their work.
7. The seven survey waves sent out do not map directly to lottery drawings (of
which there were eight). See Online Appendix Table A2 (on eight lottery drawings)
and Table A9 (on the seven survey waves) for more detail.
contacts. The response rate to the basic protocol was 36%; about
22% of those who did not respond to the basic protocol and who
received the intensive protocol responded. We calculate an effect-
ive response rate of 50%, with individuals who responded to the
intensive follow-up weighted by the inverse probability of being
included in the intensive follow-up subsample.
In Section V.C, we also briefly compare some of our estimates
from this main survey to those from two earlier, virtually identi-
cal surveys of the same population: an initial survey conducted

approximately one year earlier (i.e., shortly after random assign-
ment), and a six-month survey conducted about midway be-
tween the initial and main survey. The six-month survey was
conducted on a 20% subsample of the sample used in the other
two surveys. The earlier surveys used similar protocols but did
not have an intensive follow-up arm; the initial and six-month
surveys achieved response rates of 45% and 42%, respectively.
III.C. Other Data

We obtained prerandomization demographic information
that the participants provided at the time of lottery sign-up.
Online Appendix Figure A2 shows the sign-up form. We use
these data primarily to construct nine lottery list variables
that we use to examine treatment and control balance on preran-
domization demographics.8 We also obtained state administra-
tive records on the complete Medicaid enrollment history of
lottery list participants from prior to the lottery through
September 2009. We use these data as our primary measure of
the first-stage outcome (i.e., insurance coverage). Finally we ob-
tained state administrative records on the Food Stamp and TANF
benefit history of lottery list participants from prior to the lottery
through September 2009.
8. These nine lottery list variables are year of birth; sex; whether English is the
preferred language for receiving materials; whether the individuals signed them-
selves up for the lottery or were signed up by a household member; whether they
provided a phone number on sign-up; whether the individuals gave their address as
a PO box; whether they signed up the first day the lottery list was open; the median
household income in the 2000 census from their ZIP code; and whether the ZIP code
they gave is within a census-defined metropolitan statistical area.
III.D. Time Frame of the Study

In the administrative data we measure outcomes from the
date individuals were notified that they were selected (i.e., their
notification date) through the end of September 2009.9 This
observation period represents, on average, 16 months (std.
dev. = 2 months) after individuals are notified of their selection
and, on average 14 months (std. dev. = 3 months) after insurance
coverage is approved for those selected by the lottery who suc-

cessfully enrolled in OHP Standard. If an individual successfully
obtained insurance through the lottery, coverage was applied
retroactively to only a few days after the state mailed the appli-
cation to the individual, which was on average about one month
after the notification date and one month prior to the approval
date.10
In our survey most outcomes were asked with a six-month
look-back period (e.g., number of doctor visits in the last six
months) or based on current conditions (e.g., self-reported
health). There is variation across individuals in when surveys
were mailed and how long they took to respond, as well as their
lottery draw (and hence notification date). Our average survey
response occurs 15.3 months after notification date (std.
dev. = 2.7) months or 13.1 months after insurance approval (std.
dev. = 2.9 months).
III.E. Sample Definition and Descriptive Statistics

Of the 89,824 individuals who were on the lottery list, we
used prerandomization data to exclude individuals who were
not eligible for OHP Standard (because they gave an address out-
side of Oregon, were not in the right age range, or died prior to the
lottery), had institutional addresses, were signed up by third
parties, would have been eligible for Medicare by the end of our
study period, or were inadvertently included on the original list
multiple times by the state. These exclusions brought our study
9. We randomly assigned lottery draws to the control individuals as discussed

in more detail in Section IV.
10. We suspect, and focus group interviews with selected individuals suggest,
that selected individuals would have been unlikely to change their behavior while
their applications were being processed; however, the retroactive insurance cover-
age may affect the financial burden associated with health care utilization during
that time period.
1068
TABLE I
DEMOGRAPHIC CHARACTERISTICS OF STUDY POPULATION (CONTROL GROUP)
Variable Control mean Variable Control mean

Panel A: Full sample
Sex Language
% Female 0.557 % English preferred 0.922
Age ZIP codelevel variables
% 5064 0.267 % MSA 0.773
% 2050 0.733 ZIP code median household income $39,265
Panel B: Survey responders only
Lottery list variables
Sex Language
% Female 0.591 % English preferred 0.917
Age ZIP codelevel variables
% 5064 0.316 % MSA 0.751
% 2050 0.684 ZIP code median household income $39,225
12-month mail survey variables
QUARTERLY JOURNAL OF ECONOMICS
Race Health status

% White 0.820 Ever diagnosed with:
% Black 0.038 Diabetes 0.175
Asthma 0.276
High blood pressure 0.399
Ethnicity Emphysema or chronic bronchitis 0.129
% Spanish/Hispanic/Latino 0.123 Depression (screen positive) 0.557
(continued)

TABLE I
(CONTINUED)
Variable Control mean Variable Control mean

Education Income (% federal poverty line)
% Less than high school 0.177 <50% 0.406
% High school disploma or GED 0.491 5075% 0.138
% Vocational training or 2-year degree 0.220 75100% 0.140
% 4-year college degree or more 0.112 100150% 0.177
Above 150% 0.139
Employment
% dont currently work 0.551 Insurance coverage
% work <20 hours per week 0.090 Any insurance? 0.325
% work 2029 hours per week 0.099 OHP/Medicaid 0.117
% work 30+ hrs per week 0.259 Private insurance 0.128
Other 0.102
Average household income (2008) $ 13,035 # of months of last six with insurance 1.738
Notes. All statistics are reported for control individuals only. Panel A reports the control means for prerandomization demographics taken from the lottery list (from January
and February 2008) for the whole sample (N = 45,088 for controls). Age refers to age at the end of the study period. English as preferred language indicates whether the individual
OREGON HEALTH INSURANCE EXPERIMENT
did not check a box requesting materials in a language other than English. Panel B reports control means of lottery list prerandomization demographics and survey questions for
survey responders (N = 11,933 for controls), weighted using survey weights. Household income is gross household income (in $) for 2008 (before taxes and deductions but including
any cash assistance or unemployment assistance received); it is reported in bins and we assign individuals the income at the midpoint of their bin (see Online Appendix Figure A4 for
details). For the insurance questions, we code as yes if the respondent checked that insurance type box; because the survey allows one to check multiple boxes for types of
insurance, the subgroups (OHP/Mediciad, private, and other) wont necessarily add up to any. Private insurance includes employer and privately paid insurance; Other insurance
includes Medicare and other. We treat responses for insurance as missing if the responder checked I dont know or left all checkboxes blank. We construct income relative to the
federal poverty line based on self-reported income and self-reported (total) number of household members. See Online Appendix 3 for more details.
1069

population down to 74,922 individuals (representing 66,385

households). Of these, 29,834 individuals were selected by the
lottery and the remaining 45,088 individuals are controls. Of
these, we surveyed 29,589 treatment and 28,816 control individ-
uals. Online Appendix Figure A1 shows the overlap between the
full sample (which is used in the hospital discharge data and
mortality analysis), the credit report subsample, and the mail
survey subsample.
Table I provides some demographic summary statistics for

our control sample. Panel A, based on information provided prior
to randomization on the sign-up list, shows that our study popu-
lation is 56% female; about one-quarter are 50 to 64 years old at
the end of our study period (the average age is 41); 92% have a
preferred language of English; and about three-quarters live in a
metropolitan statistical area (MSA). Panel B reports additional
demographic characteristics of the control group from the survey.
These outcomes are only available for individuals who responded
to the mail survey and may therefore not be representative of the
full sample. The population is 4% black and 12% Hispanic. Almost
one-fifth has less than a high school education, and another half
has only a high school diploma or GED. Over half report that they
are not currently working. Most strikingly, they appear to be in
quite poor health: 18% report having ever been diagnosed with
diabetes, 28% with asthma, 40% with high blood pressure; 56%
screen positive for depression.11
Panel B also shows the distribution of household income
(relative to the federal poverty level) and insurance coverage.
Both are important for the first-stage impact of lottery selection
on insurance coverage. About 70% report incomes below the eli-
gibility cut-off of 100% of the federal poverty level; this is consist-
ent with our finding from analyzing application data that income
eligibility requirements disqualified a nontrivial share of selected
individuals (Allen et al. 2010). Finally, about 30% of the controls
report having insurance (which, if they had it at the time of the
11. By contrast among a general adult population, 7% report ever being diag-
nosed with diabetes, 14% with asthma, 24% with high blood pressure, and 28% with
depression. (These numbers are based on our calculation from the 20042009
Behavioral Risk Factor Social Surveillance Survey, which uses virtually identical
questions to our survey questions.)
lottery, would have also made them ineligible for OHP Standard);
13% report having private insurance.
IV. EMPIRICAL FRAMEWORK

IV.A. Intent to Treat Effects of Lottery
I. Intent-to-Treat Equation. We estimate the intent-to-treat

(ITT) effect of winning the lottery (i.e., the difference between
treatment and controls) by estimating the following ordinary
least squares (OLS) equation:
1 yihj 0 1 LOTTERYh Xih 2 Vih 3 "ihj ,
where i denotes an individual, h denotes a household, and j 2 J
denotes a domain of related outcomes (such as health or finan-
cial strain). For example, yij might be the self-reported health of
individual i, which is one of the health measures in the health
domain J. We define (sign) each outcome within a domain so
that higher values all have the same interpretation within a
domain (e.g., more health care use, more financial strain). As
we will discuss, we summarize the estimates within a domain
by the standardized treatment effect across outcomes in that
domain; we also report estimates for individual outcomes and
show p-values that are adjusted to account for the multiple out-
comes examined within the domain.
LOTTERY is an indicator variable for whether household h
was selected by the lottery. The coefficient on LOTTERY (b1) is
the main coefficient of interest and gives the average difference in
(adjusted) means between the treatment group (the lottery win-
ners) and the control group (those not selected by the lottery); it is
interpreted as the impact of being able to apply for OHP Standard
through the Oregon lottery.
We denote by Xih the set of covariates that are correlated
with treatment probability (and potentially with the outcome)
and therefore must be controlled for so that estimates of b1 give
an unbiased estimate of the relationship between winning the
lottery and the outcome. In all of our analyses, Xih includes indi-
cator variables for the number of individuals in the household
listed on the lottery sign-up form (hereafter household size);
as already noted, although the state randomly sampled from in-
dividuals on the list, the entire household of any selected
individual was considered selected and eligible to apply for insur-

ance. As a result, selected (treatment) individuals are dispropor-
tionately drawn from larger households.12 For outcomes in the
survey data, Xih also includes indicator variables for survey wave
(and the interaction of these indicator variables with household
size indicators) because the fraction of treatment individuals
varies across the seven survey waves.
We denote by Vih a second set of covariates that can be
included to potentially improve power by accounting for chance

differences in variables between treatment and control group but
that are not needed for b1 to give an unbiased estimate of the
relationship between winning the lottery and the outcome.
There are three potential sources of such variables: the lottery
list demographic variables, prerandomization measures of out-
comes in the credit report data and hospital discharge data, and
the lottery draw to which the individual is assigned.13 Our ana-
lysis of survey data will not control for any Vih covariates; our
analysis of administrative data will include lottery draw indica-
tors as well as the prerandomization measure for the outcome
analyzed in the hospital and credit report data. Online
Appendix Table A15 shows that our results are not sensitive to
other choices regarding the Vih covariates.
In all of our ITT estimates and in our subsequent instru-
mental variable estimates (see Section IV.C), we estimate linear
models even though a number of our outcomes are binary.
Because we are interested in the difference in conditional
means for the treatments and controls, linear probability
models pose no concerns in the absence of covariates or in fully
saturated models (Angrist 2001; Angrist and Pischke 2009). Our
models are not fully saturated, however, so it is possible that the
choice of models could make a difference, especially for outcomes
with very low or very high mean probability. In Online Appendix
Table A16 we show that the analysis of these tail outcomes is
12. The proportion of treated (respectively, control) individuals in household

size 1 is 66.5 (respectively, 83.6), in household size 2 is 33.1 (respectively, 16.4) and
in household size 3 is 0.5 (respectively, 0.04).
13. In the administrative data, we measure outcomes from the notification date.
For treatment individuals, notification date varies by lottery draw (which spans a
seven-month period). For control individuals, we randomly assigned a lottery draw
at the household level, stratified on household size, to match the distribution of
lottery draws among the treatments so that by construction, treatment probability
is uncorrelated with lottery draw within household size.
not sensitive to estimating nonlinear models instead (this ana-

lysis was not prespecified).
In all of our analyses we cluster the standard errors on the
household identifier because the treatment is at the household
level. Analyses of survey data are weighted to account for
the sampling design of the survey as described already (and
in Online Appendix 1.7); we show in Online Appendix
Table A15 that the results are similar if we do not use survey
weights or if we limit the analysis to the individuals who

responded to the basic protocol (this analysis was not
prespecified).
Handling many outcomes: standardized treatment effects and

multiple inference. We summarize multiple findings across
related outcomes within a domain J by the average standardized
treatment effect:14
X 1 1j
2 ,
j2J
J j
where j is the standard deviation of yj in the control group and 1j

is the coefficient of interest for outcome j. (Specifically, for the ITT
estimates in equation (1), the 1js correspond to the 1js). To ac-

count for covariance in the estimates of 1jj , we estimate
15
pooled OLS for all outcomes j 2 J. An important limitation of
standardized treatment effects is that they implicitly weight
each outcome within a domain equally, which may not be
desirable.
In addition to standardized treatment effects, in each domain
we also report the underlying estimates on the individual out-
comes (i.e., the 1js) due to their ease of interpretation and indi-
vidual interest. For the individual outcomes we report both per
comparison p-values and family-wise p-values adjusted to ac-
count for the multiple outcomes examined within the domain.16
14. The approach in this section draws heavily on Kling and Liebman (2004)
and Kling, Liebman, and Katz (2007).
15. Specifically, we stacked the data for the individual outcomes within a
domain and estimated a single regression equation that allowed the coefficients
on each covariate to vary flexibly across the outcomes and for correlation in the error
terms across outcomes.
16. The family-wise p-value corresponds to the probability of rejecting the null
hypothesis of no effect on a given outcome under the null family of hypotheses of no
The per comparison p-value may be appropriate if one is inter-

ested in the particular hypothesis of the impact of insurance on a
specific outcome (such as depression). The family-wise p-value is
more appropriate for considering the test of the specific outcome
as part of a set of tests on all the outcomes in the domain of that
standardized treatment effect (such as overall health). In prac-
tice, it is rare for one of our results to be statistically significant
(e.g., a p-value of less than .05) in the per comparison test and not
significant after adjusting for multiple inference.17

IV.B. Validity of the Experimental Design
Our causal inference rests on the twin assumptions that as-
signment of the ability to apply for OHP Standard was in fact
randomized in the way described, and that the treatment and
control individuals in the subsamples we use to analyze outcomes
are not differentially selected from the full sample. The lotterys
random selection process was performed by Oregons Department
of Human Services (DHS). We verified through independent com-
puter simulations that we could replicate their procedure to
within sampling error (Online Appendix 2.1 and Table A12 pro-
vide more detail); we also demonstrate that the procedure we
used to draw our survey sample produced balance of treatment
and control characteristics (see Online Appendix 2.2 and Table
A12).
Differences in attrition (match rates or response rates) or in
the prerandomization characteristics of the treatment and con-
trol analysis samples would raise concerns about the second key
assumption for causal inference. Table II therefore investigates
treatment-control balance for three different samples in columns
effect on any outcome in this domain. We calculate these family-wise error rate
adjusted p-values based on 10,000 iterations of the free step-down resampling
method of Westfall and Young (1993); see Kling and Liebman (2004) or Anderson
(2008) for more detailed discussions as well as applications.
17. In the archived analysis plan we proposed presenting standardized treat-
ment effects of related outcomes within a domain separately for both survey and
administrative data, as well as a third standardized treatment effect using the
outcomes from both survey and administrative data in a given domain. Given the
major substantive and methodological differences between the two types of data, in
this article we have opted for reporting only the standardized treatment effects
across outcomes within domains for the survey and administrative data separately.
In practice this makes a negligible difference to the adjusted p-values; results avail-
able on request.
TABLE II
TREATMENT: CONTROL BALANCE
Control mean (std. dev.) Difference between treatment and control

for full sample
Credit report Survey respondents
Full sample subsample subsample
(1) (2) (3) (4)
Panel A: Match/response rates
Matched in September 2009 credit data 0.663 0.0043
(0.473) (0.0037)
[0.247]
Responded to survey 0.506 0.016
(0.500) (0.0066)
[0.014]
Response time (in days) 53.0 1.638
(57.8) (1.088)
[0.132]
Panel B: Prerandomization characteristics
Lottery list variables
F-statistic 1.286 0.553 0.574
[p-value] [0.239] [0.836] [0.820]
Pre-randomization outcomes
F-statistic 0.543 0.921 1.266
[p-value] [0.844] [0.518] [0.281]
Both of the above
F-statistic 0.915 0.793 0.782
[p-value] [0.56] [0.726] [0.680]
N 74,922 49,980 23,741
Notes. Standard errors in parentheses; per comparison p-values in square brackets. In Panel A, we analyze match and response rates. The first column reports the mean and
standard deviation for the control sample of the outcome shown in the left column. Columns (3) and (4) report estimated differences between treatments and controls for the
dependent variable shown in the left column and the sample indicated in the column heading. Specifically they report the coefficient on LOTTERY based on estimating equation (1).
All regressions include household fixed effects and cluster on household. In addition, in column (4) we include survey wave fixed effects and the interaction of survey wave fixed
effects and household fixed effects and use survey weights. The full sample (i.e., the sample used in the hospital discharge and mortality data) is used in column (3). and the entire
survey sample is used in column (4). In Panel B we report the F-statistic and p-value from regressing multiple prerandomization characteristics on LOTTERY in equation (1).
Lottery list variables are common across all three samples and consist of nine demographic variables derived from information provided at the time of lottery sign up: year of
birth; sex; whether English is the preferred language for receiving materials; whether the individual signed him- or herself up for the lottery or were signed up by a household
member; whether the individual gave his or her address as a PO box; whether he or she signed up the first day the lottery list was open; the median household income in the 2000
census from their ZIP code; whether the ZIP code given is within a census-defined MSA; and whether he or she provided a phone number on sign up. Prerandomization outcomes
are specific to the sample (we look at the hospital outcomes that we subsequently analyze for column (2), the credit report outcomes we subsequently analyze for column (3), and a
1075
few measures from each that approximate survey questions we subsequently analyze for column (4)). More detail on the prerandomization outcomes, the exact regression speci-
fications, and the results for each variable analyzed in Panel B are presented in Online Appendix 3, Table A13.

(2) through (4), respectively: the sample universe (which is the

sample analyzed in the hospital discharge data and the mortality
data), the credit report subsample, and the survey respondents.
A priori we were most concerned about the potential for imbal-
ance between treatment and controls in the subsample of survey
respondents, given the 50% nonresponse rate (compared to an
effective match rate of over 97% in the credit report data).
Panel A shows the balance of match rates. In the credit report
data, the difference in (unconditional) match rates between treat-

ment and control groups is a statistically insignificant 0.4 per-
centage points. In the survey respondent sample, there is a
statistically significant 1.6 percentage point (std. err. = 0.7)
lower response rate for treated individuals, off of a 51% base.18
By way of comparison, our estimated difference in response rates
across treatment and control is much smaller than in the RAND
Health Insurance Experiment; there, the overall response rate
was higher (77%), but those randomized into more coverage had
systematically higher response rates, with a 24 percentage point
difference in response rate between the most and least compre-
hensive plan (87% versus 63%; Newhouse and the Insurance
Experiment Group 1993).
Among the matched or responding subsample, we investi-
gated the treatment-control balance for various prerandomiza-
tion characteristics. In total we examined treatment-control
balance for 31 different variables in up to three different samples
(54 total comparisons). In Panel B we report the summary F-stat-
istics and p-values on the treatment-control balance of groups of
variables. The first row (lottery list characteristics) shows the
F-statistics and p-values on the treatment-control balance of all
the nine lottery list demographics. The second row (prerandomi-
zation outcomes) shows the balance of prerandomization
outcomes that match ones that we subsequently analyze postran-
domization. The selection of these variables was prespecified and
is described in Online Appendix 2.2 (Appendix Table A13).
Finally, the third row (both of the above) examines balance on
the lottery list and prerandomization outcomes combined. In each
of the three samples we are unable to reject the null of treatment-
control balance on the lottery list variables, the prerandomization
measures, or the combined set of variables. All p-values are above
18. Conditional on response, response time between treatment and controls is

indistinguishable.
.23. For the survey respondent subsamplewhere a priori we

were most concerned about potential imbalancethe p-values
are all at least .28.19
The individual results for treatment-control balance for each
variable that contributes to the reported F-tests are shown in
Online Appendix Table A13. In only two of the comparisons
(one in the full sample and one in the credit report subsample)
did we obtain p-values below .05 (p = .039 and p = .014); this is
consistent with what we would expect by chance with 54

comparisons.
A separate question from the balance of treatment and con-
trols within a subsample is how the various subsamples compare
to each other. Differences in characteristics across subsamples do
not threaten causal inference but may be important for compar-
ing estimates across data sets as well as for extrapolating to other
contexts. Survey responders are on average almost two years
older and 3 percentage points more likely to be female than the
full sample; they look similar to the full sample in terms of their
language preference and urbanicity. There are no material
observable differences between the credit report subsample and
the full sample (see Online Appendix Table A1).
IV.C. Local Average Treatment Effect of Medicaid (LATE)

The ITT estimates from equation (1) provide an estimate of
the causal effect of winning the lottery (i.e., winning the ability to
apply for OHP Standard). This provides an estimate of the net
impact of expanding access to public health insurance. We are
also interested in the impact of insurance coverage. We model
this as follows:
3 yihj 0 1 INSURANCEih Xih 2 Vih 3 vihj ,

where INSURANCE is a measure of insurance coverage and all
other variables are as defined in equation (1). We estimate
19. We use the fairly conservative procedure of Lee (2009) to bound the potential
bias arising from differential response rates between treatments and controls. Our
ability to reject the null of no effect of health insurance on health care use or finan-
cial strain is generally robust to this bounding exercise (although naturally the
magnitudes are attenuated at the lower bound), but our ability to reject the null
of no impact of health insurance on self-reported health is generally not robust to
this bounding exercise. These results are presented in Online Appendix Table A14
and discussed in Online Appendix 2.2. (This analysis was not prespecified.)
equation (3) by two-stage least squares (2SLS), using the follow-

ing first-stage equation:
4 INSURANCEih 0 1 LOTTERYh Xih 2 Vih 3 ihj ,
in which the excluded instrument is the variable LOTTERY with

the first-stage coefficient of d1.20
We interpret the coefficient on insurance from instrumental
variable estimation of equation (3) as a local average treatment

effect (LATE) of insurance (Imbens and Angrist 1994). In other
words, our estimate of p1 identifies the causal impact of insurance
among the subset of individuals who obtain insurance on winning
the lottery and who would not obtain insurance without winning
the lottery (i.e., the compliers). Because the model is just identi-
fied, the LATE estimate of p1 is given by the ratio of the ITT
coefficient on LOTTERY (equation (1)) and the first-stage coeffi-
cient on LOTTERY (equation (4)), or 11 : Table III reports our
first-stage estimates based on estimation of equation (4). The
first row reports the results using the measure of insurance
that is used in all of our LATE estimates: whether the individual
was ever on Medicaid (which includes both OHP Standard and
OHP Plus) during our study period, as measured in the states
Medicaid enrollment files. The results indicate a first stage of 0.26
for both the full sample (column (2)) and the credit-report sub-
sample (column (4)) and a first stage of 0.29 for the survey re-
spondents (column (6)). All of these first stages have F-statistics
above 500. The first-stage coefficient is considerably less than 1,
primarily reflecting the 30% take-up already discussed; in add-
ition, a small percentage of the controls became eligible for OHP
Plus over our study period.21 Relative to our study population,
compliers are somewhat older, more likely white, in worse health,
and in lower socioeconomic status.22
20. When we report standardized treatment effects for LATE estimates, they
are calculated based on the formula in equation (2) and using pooled instrumental
variables (IV) estimates of equation (3) across outcomes.
21. As we discuss in more detail in Online Appendix 1.3, the way the state
identified lottery participants in the Medicaid enrollment files may cause it to
slightly underestimate enrollment among nonselected (control) individuals, and
thus cause us to overestimate our first stage by, we estimate, 2 percentage points
or less.
22. The relative likelihood of being a complier is given by the ratio of the first
stage in the subgroup to the overall first stage (Angrist and Pishke 2009, p. 171).
These results are reported in Online Appendix Table A26.
TABLE III
FIRST-STAGE ESTIMATES
Full sample Credit report subsample Survey respondents
Control mean Estimated FS Control mean Estimated FS Control mean Estimated FS

(1) (2) (3) (4) (5) (6)
(1) Ever on Medicaid 0.141 0.256 0.135 0.255 0.135 0.290
(0.0035) (0.0042) (0.0066)
(2) Ever on OHP Standard 0.027 0.264 0.028 0.264 0.026 0.302
(0.0029) (0.0036) (0.0055)
(3) # of months on Medicaid 1.408 3.355 1.352 3.366 1.509 3.943
(0.045) (0.055) (0.090)
(4) On Medicaid, end of study period 0.106 0.148 0.101 0.151 0.105 0.189
(0.0031) (0.0038) (0.0061)
(5) Currently have any insurance (self-report) 0.325 0.179
(0.0077)
(6) Currenty have private insurance (self-report) 0.128 0.0076
(0.0053)
(7) Currently on Medicaid (self-report) 0.117 0.197
(0.0063)
(8) Currently on Medicaid 0.105 0.191
(0.0060)
(9) Ever on TANF 0.031 0.0011 0.028 0.0021 0.023 0.0019
(0.0013) (0.0016) (0.0025)
(10) TANF benefits ($) 124 1.659 111 1.543 100 4.991
(5.813) (6.571) (10.884)
(11) Ever on food stamps 0.606 0.017 0.594 0.018 0.622 0.023
(0.0029) (0.0035) (0.0054)
(12) Food stamp benefits ($) 1776 61.3 1787 60.0 2202 122.4
(15.0) (18.8) (33.4)
N 74,922 49,980 23,741
Notes. Standard errors in parentheses. Even-numbered columns report the coefficient and standard error on LOTTERY from estimating the first-stage equation (4) with the
dependent variable INSURANCE defined in the left column; odd-numbered columns report the control mean for that measure of INSURANCE. Full sample is the sample analyzed
in the hospital discharge and mortality data. All regressions include dummies for household size and adjust standard errors for household clusters. The regressions in columns (2)
and (4) also include lottery draw dummies; the regressions in column (6) also include dummies for survey wave and survey wave interacted with household size dummies, and use
survey weights. The insurance measures are taken from the Medicaid enrollment administrative data except for those labeled self-report (rows (5) through (7)) which are taken
from the survey. In the survey, respondents could report various types of insurance; we define private insurance as employer or private insurance and any insurance as
Medicaid, Medicare, employer, private, or other insurance. In rows (1) and (2), ever refers to enrollment ever during the study period, as defined in the text. In row (3), # of
1079
months refers to number of months enrolled during the study period. In row (8), insurance is measured as being on Medicaid according to the state Medicaid enrollment data on the
day the survey was returned. All outcomes are measured for the individual except that in row (10) (12) where TANF (food stamp) benefits measure total household benefits received
over the study period. For purposes of measuring the first stage, the study period is defined as ending on September 30, 2009. For the second-stage outcomes that we analyze in the
administrative data, the study period in the first stage begins with the notification date (which varies by lottery draw). For the second stage outcomes that we analyze in the survey
data, the study period in the first stage begins on the first notification date (March 10, 2008).
The subsequent rows of Table III report first-stage estimates

for alternative definitions of insurance. We focus the discussion
on the results for the full sample (column (2)); the other samples
show very similar patterns. Not surprisingly, the results in row
(2) show that the estimated first stage is virtually the same if we
examine only coverage by OHP Standard (the program directly
affected by the lottery) versus coverage by either public program,
indicating that selection by the lottery is not associated with an
increase in coverage by OHP Plus.23 Row (3) shows that on aver-

age the lottery is associated with an increase of 3.4 months of
Medicaid coverage. Over time the difference in insurance cover-
age between treatment and controls attenuates as enrolled indi-
viduals had to recertify their eligibility every six months and
control individuals could find other insurance, particularly
OHP Plus (see Online Appendix Figure A3). As a result, when
the dependent variable is defined as on Medicaid at the end of
the study period rather than our primary measure of ever on
Medicaid, the estimated first stage declines from 0.26 (row (1)) to
0.15 (row (4)).
In our subsequent results, we use ever on Medicaid during
our study period for all of the reported LATE estimates of the
impact of Medicaid. However, these alternative definitions of in-
surance are also reasonable ways to scale the ITT estimates. The
number of months on Medicaid may be more appropriate than
ever on Medicaid where the effect of insurance on the outcome
is linear in the number of months insured. The measure on
Medicaid at the end of our study can be thought of as providing
a lower bound on the first stage (and hence an upper bound on our
estimate of the LATE of Medicaid), whereas our baseline measure
ever on Medicaid during our study period provides an upper
23. When insurance is defined as ever on OHP Standard we can probably be

comfortable interpreting the IV estimates of equation (3) as the treatment-
on-treated (ToT) rather than a LATE. In practice, there are two small violations
of this interpretation. First, if there were no way to get OHP Standard without
winning the lottery, there would be no always-takers in the terminology of
Angrist, Imbens, and Rubin (1996), but about 2% of our controls got onto OHP
standard through some limited mechanismsfor example, pregnant women who
are on OHP Plus can sometimes stay on OHP Standard after giving birth. Second,
it is possible that some compliers were put on OHP Plus rather than Standard,
because case workers are instructed to first check applicant eligibility for Plus; in
practice, this number is likely to be small since the estimated first stage is very
similar for ever on Medicaid (which includes Plus and Standard) and ever on
OHP Standard (see rows (1) and (2) of Table III).
bound on the first stage (and hence a lower bound on our estimate
of the LATE of Medicaid).
The survey data provide a broader measure of insurance
coverage than available in the Medicaid administrative data.
The results in row (5) indicate that the estimated increase in
self-reported any insurance coverage is very similar (within
two percentage points) to the estimated increase in self-reported
Medicaid coverage (row (7)). This suggests the lottery did not
crowd out other forms of insurance. Consistent with this, row

(6) indicates a statistically insignificant decline in self-reports of
private insurance coverage of only 0.8 percentage points (std.
err. = 0.5). The estimated increase in Medicaid coverage as re-
ported in the survey and as measured in the administrative
data are quite similar when measured over the same time horizon
(compare rows (7) and (8)).
The LATE interpretation requires the additional identifying
assumption that there is no effect on average on the outcomes
studied of winning the lottery that does not operate via the lot-
terys impact on insurance coverage. We believe this is a reason-
able approximation, but it may not be strictly true. There are (at
least) two possible types of violations. First, the event of winning
(or losing) the lottery may have direct effects on the outcomes we
study, although it seems unlikely to us that any such effects both
exist and persist a year after the lottery. Second, individuals who
apply for public health insurance may also be encouraged to apply
for other public programs for which they are eligible, such as food
stamps or cash welfare. In particular, if the individual applied for
OHP in person (rather than by mail), case workers were in-
structed to offer assistance to interested applicants in applying
for TANF (cash welfare) and the Supplemental Nutrition
Assistance Program (more commonly known as food stamps).
These other cash (or cash-equivalent) transfer programs could
have direct effects on the outcomes we study. This is not an idio-
syncratic feature of our setting but a more general feature of the
application process for public programs; as such, it may be a rele-
vant component of the impact of attempts to expand Medicaid
more generally. However, any direct impact of winning the lot-
tery on receipt of other benefits is a violation of the exclusion
restriction for the LATE interpretation of the impact of insurance
per se, as opposed to the effect of expanded access to Medicaid (the
ITT analysis).
Table III indicates that selection by the lottery is not asso-

ciated with any substantive or statistically significant change in
TANF receipt or benefits. However, lottery selection is associated
with a statistically significant but substantively trivial increase
in the probability of food stamp receipt (1.7 percentage points)
and in total food stamp benefits (about $60 over a 16-month
period, or less than 0.5% of annual income).24 Estimates of the
income elasticity of health care use range from a low end of about
0 to a high end of about 1.5 (Getzen 2000, table 1), suggesting that

the income effect of food stamp receipt on health care use would
be considerably less than 1%. The impact on health seems likely
to be negligible as well. Thus, we are comfortable interpreting our
IV estimates as the effect of insurance coverage, or more specif-
ically, the Oregon Medicaid program, on outcomes.
V. RESULTS
V.A. Health Care Utilization
1. Administrative Data. Table IV presents our primary esti-
mates using the hospital discharge data; all analyses exclude ad-
missions for childbirth.25 The most common admission is for
mental disorders (approximately 10% of admissions); skin infec-
tions, complications from diabetes, and alcohol-related disorders
are also common conditions (see Online Appendix Table A4).
Table IV, Panel A reports our estimates on admissions prob-
abilities. The LATE results suggest that insurance is associated
with an increase in the probability of any hospital admission of
2.1 percentage points (std. err. = 0.7), or about 30%. The increase
in hospital admissions appears to be disproportionately concen-
trated in the approximately 35% of admissions that do not
24. This is likely an upper estimate because the cash equivalent of food stamps
may be less than one (Hoynes and Schanzenbach 2009).
25. Regardless of lottery selection, many women in our sample would become
categorically eligible for OHP Plus for childbirth. However, Oregon does not have
presumptive Medicaid eligibility for pregnant women (Kaiser Family Foundation
2011), so it is possible that lottery selection could have an impact on
pregnancy-related hospital use. In Online Appendix Table A21, we show that our
results regarding hospital utilization are quite similar if we include admissions for
childbirth in the analysis or if we exclude women of childbearing age from the
sample (this analysis was not prespecified).
TABLE IV
HOSPITAL UTILIZATION
Control
mean ITT LATE p-values
(1) (2) (3) (4)
Panel A: Extensive margin
All hospital admissions 0.067 0.0054 0.021 [0.004]
(0.250) (0.0019) (0.0074)
Admissions through ER 0.048 0.0018 0.0070 [0.265]

(0.214) (0.0016) (0.0062)
Admissions not through ER 0.029 0.0041 0.016 [0.002]
(0.167) (0.0013) (0.0051)
Panel B: All hospital admissions
Days 0.498 0.026 0.101 [0.329]
(3.795) (0.027) (0.104) {0.328}
List charges 2,613 258 1,009 [0.077]
(19,942) (146) (569) {0.106}
Procedures 0.155 0.018 0.070 [0.031]
(1.08) (0.0083) (0.032) {0.059}
Standardized treatment effect 0.012 0.047 [0.073]
(0.0067) (0.026)
Panel C: Admissions through ER
Days 0.299 0.023 0.089 [0.183]
(2.326) (0.017) (0.067) {0.187}
List charges 1,502 163 636 [0.091]
(12,749) (96) (376) {0.171}
Procedures 0.081 0.0080 0.031 [0.135]
(0.694) (0.0054) (0.021) {0.187}
(0.0069) (0.027)
Panel D: Admissions not through ER
Days 0.199 0.0033 0.013 [0.841]
(2.38) (0.017) (0.065) {0.842}
List charges 1,110 98 384 [0.281]
(12,422) (91) (356) {0.383}
Procedures 0.075 0.010 0.038 [0.080]
(0.708) (0.0056) (0.022) {0.162}
(0.0068) (0.026)
Notes. Standard errors in parentheses; per comparison p-values in square brackets; family-wise
p-values in curly brackets. Table investigates nonchildbirth-related hospitalizations during the time
period from notification date to August 31, 2009. All outcomes are measured unconditionally (i.e., are
not conditional on admission). Column (2) reports the coefficient and standard error on LOTTERY from
estimating equation (1) by OLS. Column (3) reports the coefficient and standard error on INSURANCE
from estimating equation (3) by IV; for the IV estimates in column (3), the endogenous variable
INSURANCE is defined as ever on Medicaid during our study period and the first stage is given in
the first row of Table III. Column (4) reports the per comparison p-value and (where applicable) the
family-wise p-value across the three different measures of utilization used to create the standardized
treatment effect. Standardized treatment effect reports results based on equation (2). All regressions
include household size fixed effect, lottery draw fixed effects, and the analogous outcome measure for
the time period from January 1, 2008, through the notification date. All standard errors are clustered on
the household. Sample consists of entire sample universe (N = 74,922).
originate in the emergency room, suggesting that these admis-

sions may be more price sensitive.
Table IV, Panels BD examine three measures of total util-
ization commonly used in the literature (see Card, Dobkin, and
Maestas 2009): number of hospital days, total list charges, and
number of procedures performed.26 Although total utilization is
arguably of greater interest (particularly for estimating the
impact on total costs), not surprisinglygiven the skewed right
tail of total utilizationwe have less statistical precision here

than on the extensive margin. The LATE estimates in Panel B
show substantial increases in each of the three measures of
utilizationwith implied proportional increases of about 20%
for hospital days, 40% for list charges, and 45% for the number
of procedures, although only the result for procedures is statis-
tically significant at the 10% level. The standardized treat-
ment effect across all three measures indicates that insurance
is associated with a 0.047 standard deviation increase in overall
utilization (std. err. = 0.026); the relatively small standardized
effect partly reflects the large variance of the underlying
variables.
We undertook several additional investigations, which are
presented in detail in Online Appendix 3.2. First, we examined
hospital utilization for seven conditions of interest and of reason-
ably high prevalence in our population: heart disease, diabetes,
skin infections, mental disorders, alcohol and substance abuse,
back problems, and pneumonia. We found a statistically signifi-
cant increase in utilization (both extensive and total) only for
heart disease (see Online Appendix Table A18). We also explored
the impact of health insurance on a measure of the quality of
outpatient care (admissions for ambulatory caresensitive condi-
tions) and three measures of quality of care for inpatient care (not
having an adverse patient safety event, not being readmitted
within 30 days of discharge, and quality of hospital). We were
unable to reject the null of no effects on either outpatient or in-
patient quality, although our confidence intervals are sufficiently
wide that we cannot rule out quantitatively large effects (see
Online Appendix Table A19). Finally, we examined whether
26. List charges are accounting charges for rooms and procedures and do not
reflect transacted prices. They are perhaps best viewed as a price-weighted sum-
mary of treatment, albeit at somewhat artificial prices (Card, Dobkin, and Maestas
2009).
TABLE V
HEALTH CARE UTILIZATION (SURVEY DATA)
Extensive margin (any) Total utilization (number)
Control Control
mean ITT LATE p-values mean ITT LATE p-values
(1) (2) (3) (4) (5) (6) (7) (8)
Prescription drugs currently 0.637 0.025 0.088 [0.002] 2.318 0.100 0.347 [0.049]
(0.481) (0.0083) (0.029) {0.005} (2.878) (0.051) (0.176) {0.137}
Outpatient visits last six months 0.574 0.062 0.212 [<0.0001] 1.914 0.314 1.083 [<0.0001]
(0.494) (0.0074) (0.025) {<0.0001} (3.087) (0.054) (0.182) {<0.0001}
ER visits last six months 0.261 0.0065 0.022 [0.335] 0.47 0.0074 0.026 [0.645]
(0.439) (0.0067) (0.023) {0.547} (1.037) (0.016) (0.056) {0.643}
Inpatient hospital admissions last six months 0.072 0.0022 0.0077 [0.572] 0.097 0.0062 0.021 [0.311]
(0.259) (0.0040) (0.014) {0.570} (0.4) (0.0062) (0.021) {0.510}
Standardized treatment effect 0.050 0.173 [<0.0001] 0.040 0.137 [0.0003]
(0.011) (0.036) (0.011) (0.038)
Annual spendinga 3,156 226 778 [0.037]
(108) (371)
Notes. Standard errors in parentheses; per comparison p-values in square brackets; family-wise p-values in curly brackets. Hospital admissions exclude childbirth. Columns (2)
and (6) report the coefficient and standard error on LOTTERY from estimating equation (1) by OLS. Columns (3) and (7) report the coefficient and standard error on INSURANCE
from estimating equation (3) by IV; for the IV estimates in column (3), the endogenous variable INSURANCE is defined as ever on Medicaid during our study period and the first
stage is given in the first row of Table III. Columns (4) and (8) report the per-comparison p-value and the family-wise p-value across the four different measures of utilization used to
create the standardized treatment effect. Standardized treatment effect reports results based on equation (2). All regressions include household size fixed effects, survey wave fixed
effects, and the interaction between the two. All standard errors are clustered on the household and all regressions are weighted using survey weights. Sample consists of survey
responders (N = 23,741).
a
To calculate the implied spending effects associated with the estimated utilization effects, we use data from the 20022007 (pooled) Medical Expenditure Panel Survey (MEPS)
on expenditures of all nonelderly (age 1964) adults below 100% of poverty who are publicly insured. This gives us a total sample of over 7,500 individuals. We use their
expenditures (all inflated with the CPI-U to 2007 dollars) to calculate average expenditures per outpatient visit, average expenditures per ER visit, average expenditures per
inpatient visit (for visits not related to childbirth), and average semi-annual (six-month) spending on prescription drug. All spending is total expenditures (i.e., not just insured)
expenditures. The underlying costs are $150 per outpatient visit, $435 per ER visit, $7,523 per inpatient visit, and $156 six-month expenditure per current prescription drug; we
1085
multiply these all by two to get annual costs.

insurance was associated with a change in the proportion of pa-

tients going to public versus private hospitals and were unable to
detect any substantive or statistically significant differences (see
Online Appendix Table A20).
2. Survey Data. The survey data allow us to examine a

broader range of utilization outcomes. Table V shows the results.
Once again, we present results on both the extensive margin (left

panel) and on total utilization (right panel). On both margins
there are substantial and (mostly) statistically significant in-
creases in prescription drugs and outpatient use. For example,
the LATE estimates suggest that insurance is associated with a
0.35 (std. err. = 0.18) increase in the number of prescription drugs
currently taken (corresponding to an approximately 15% in-
crease) and a 1.08 (std. err. = 0.18) increase in the number of out-
patient visits (corresponding to an over 55% increase). The
responses on the extensive margin may account for a large
share of the increase in total utilization, although some of the
increase in outpatient utilizationand perhaps in total drug util-
izationlikely reflects increased use among existing users (i.e.,
on the intensive margin).
There is no discernible impact of insurance on emergency
room (ER) use on either margin. OHP Standard does not
impose financial penalties for individuals who go to the ER for
nonemergency reasons (Kaiser Family Foundation 2011), so it is
possible that our inability to detect an impact of the insurance on
ER use reflects the fact that people who are used to going to the
ER for nonemergency purposes continue to do so. Our finding
may also simply reflect a lack of power; for example, we cannot
reject the hypothesis that Medicaid is associated with an almost
10% decline in ER use or an up to 25% increase in ER use. There
is also no discernible impact on inpatient hospital use on either
margin. These estimates appear to reflect the time frame of the
survey measure; they are consistent with our findings in the ad-
ministrative data when we define hospital utilization with the
same look-back period as in the survey data.27
27. To mimic the survey measures in our administrative discharge data, we

constructed measures in the discharge data of any hospital visit in the last six
months and number of hospital visits in the last six months, each defined for the
six months prior to the individuals survey response date. For any visit in the last
six months, the reduced-form estimate in the discharge data is 0.0015 (std.
Overall, across the four utilization measures, we estimate

that insurance is associated with a statistically significant in-
crease in total utilization of 0.14 standard deviation and in any
utilization of 0.17 standard deviation. Because the four different
components of utilization have very different expected costs, in
the bottom row of Table V we make a back-of-the-envelope calcu-
lation of the increase in annual spending associated with insur-
ance by weighting each type of use by its average cost among
low-income publicly insured adults in the Medical Expenditure

Panel Survey (MEPS). This calculation suggests that insurance is
associated with a $778 (std. err. = $371) increase in annual spend-
ing, or about a 25% increase relative to the implied control mean
annual spending.
Although the longer run impact of health insurance on
health care utilization may differ from the one-year effects, we
suspect that our one-year estimates are not capturing an initial,
transitory surge of pent-up demand for health care among
the uninsured. In the survey, conducted about 13 months
after insurance coverage began, all of the questions ask about
current utilization or utilization over the past six months, not
about utilization right after insurance began. Moreover, the
evidence in Table XI from the survey conducted about six
months after insurance began (when a six-month look-back
period likely would have captured the initial effects of insurance)
shows no evidence of a larger initial utilization effect or
pent-up demand.
Table VI suggests that insurance is also associated with an
increase in compliance with recommended preventive care,
including blood cholesterol checks, blood tests for diabetes, mam-
mograms, and Pap tests. Overall, the results indicate a 0.3 stand-
ard deviation (std. err. = 0.04) increase in the probability of
getting recommended preventive care. This reflects statistically
significant increases in all four of the measures examined, includ-
ing a 20% increase in the probability of ever having ones blood
err. = 0.0023), compared to 0.0022 (std. err. = 0.0040) in the survey data; for
number of visits in the last six months, the reduced-form estimate in the dis-
charge data is 0.0025 (std. err. = 0.0034) compared to 0.0062 (std. err. = 0.0062) in
the survey data. Self-reports do tend to overstate inpatient hospital use on average.
The average for the controls of any visit in the last six months is 0.07 in the
survey compared to 0.03 in the discharge data; for number of visits in the last
six months these numbers are 0.10 and 0.03, respectively. There is not, however,
any difference in reporting error between treatments and controls.
TABLE VI
COMPLIANCE WITH RECOMMENDED PREVENTIVE CARE (SURVEY DATA)
Control
(1) (2) (3) (4)
Blood cholesterol checked (ever) 0.625 0.033 0.114 [<0.0001]
(0.484) (0.0074) (0.026) {<0.0001}
Blood tested for high blood sugar/diabetes (ever) 0.604 0.026 0.090 [0.0004]
(0.489) (0.0074) (0.026) {<0.0001}
Mammogram within last 12 months (women 40) 0.298 0.055 0.187 [<0.0001]

(0.457) (0.012) (0.04) {<0.0001}
Pap test within last 12 months (women) 0.406 0.051 0.183 [<0.0001]
(0.491) (0.01) (0.034) {<0.0001}
Standardized treatment effect 0.087 0.300 [<0.0001]
(0.012) (0.041)
p-values in curly brackets. Column (2) reports the coefficient and standard error on LOTTERY from
the first row of Table III. Column (4) reports the per comparison p-value and the family-wise p-value
across the four different preventive care measures used to create the standardized treatment effect.
Standardized treatment effect reports results based on equation (2). All regressions include household
size fixed effects, survey wave fixed effects, and the interaction between the two. All standard errors are
clustered on the household and all regressions are weighted using survey weights. Sample consists of
survey responders (N = 23,741).
cholesterol checked, a 15% increase in the probability of ever

having ones blood tested for high blood sugar or diabetes, a
60% increase in the probability of having a mammogram within
the past year (for women age 40 and over), and a 45% change
in the probability of having a Pap test within the last year (for
women).
V.B. Financial Strain
1. Administrative Data. Table VII, Panel A, analyzes five
measures of financial strain in the credit report data: whether
the individual has had a bankruptcy, lien, judgment, collection,
or delinquency (any credit account with a payment that is 30 days
or more late).28 Broadly speaking, all are measures of (reasonably
large) unpaid bills or outstanding obligations that are likely to
have a major negative impact on access to credit, at least in a
general population (Avery, Calem, and Canner 2003). Liens
28. Delinquencies are mechanically zero for the one quarter of our sample who
has no open credit over our study period.
TABLE VII
FINANCIAL STRAIN (ADMINISTRATIVE DATA)
Control
(1) (2) (3) (4)
Panel A: Overall
Any bankruptcy 0.014 0.0022 0.0086 [0.106]
(0.119) (0.0014) (0.0053) {0.358}
Any lien 0.021 0.0012 0.0047 [0.406]
(0.144) (0.0014) (0.0056) {0.698}

Any judgment 0.064 0.0014 0.0054 [0.573]
(0.244) (0.0024) (0.010) {0.698}
Any collection 0.500 0.012 0.048 [0.003]
(0.500) (0.0041) (0.016) {0.013}
Any delinquency (credit accounts) 0.366 0.0016 0.0063 [0.704]
(0.482) (0.0042) (0.017) {0.698}
(0.0049) (0.019)
Panel B: Medical debt
Any medical collection 0.281 0.016 0.064 [<0.0001]
(0.449) (0.0040) (0.016) {<0.0001}
Amount owed in medical collections 1,999 99 390 [0.028]
(6733) (45) (177) {0.025}
(0.0061) (0.024)
Panel C: Nonmedical debt
Any nonmedical collection 0.392 0.0046 0.018 [0.264]
(0.488) (0.0041) (0.016) {0.455}
Amount owed in nonmedical collections 2,740 20 79 [0.751]
(9,492) (63) (248) {0.752}
(0.0059) (0.023)
p-values in curly brackets. All outcomes are measured since notification date through September 2009.
Column (2) reports the coefficient and standard error on LOTTERY from estimating equation (1) by OLS.
Column (2) reports the coefficient and standard error on INSURANCE from estimating equation (3) by IV;
for the IV estimates in column (3), the endogenous variable INSURANCE is defined as ever on Medicaid
during our study period and the first stage is given in the first row of Table III. Column (4) reports the per
comparison p-value and the family-wise p-value across the different measures used to create the standar-
dized treatment effect. Standardized treatment effect reports results based on equation (2). All regressions
include household size fixed effects, lottery draw fixed effects, and the analogous outcome measure from
the February 2008 credit report data. All standard errors are clustered on the household. Sample consists
of all those matched to credit report data (N = 49,980).
refer to tax liens, generally taken out by governments for unpaid

taxes. Judgments for unpaid bills are sought by a variety of par-
ties, including medical providers, governments, utility compa-
nies, collections agencies, and creditors. Collections reflect
unpaid bills (mostly not related to revolving credit) that have
been sent to collections agencies for recovery attempts. These
measures capture only a subset of unpaid bills because not all
sources of borrowing are covered in the data, and some unpaid
bills may not be sent to collections agencies or have judgments or
liens against them. We return to this later. As the frequencies in

Table VII indicate, bankruptcies, judgments, and liens capture
relatively extreme right-tail events (ranging from 1% to 6% fre-
quency over our 16-month sample period), whereas collections
and delinquencies are much more common (about 50% and 37%
frequency, respectively).
The average standardized treatment effect suggests no evi-
dence of a decline in financial strain across all these measures;
the point estimate is of a statistically insignificant increase in

financial strain associated with health insurance of 0.009 stand-
ard deviations (std. err. = 0.019). Four of the five measures show
no significant change. However, health insurance is associated
with a decline in the probability of having any unpaid bills sent
to collection of 4.8 percentage points (std. err. = 0.016), or about
10% relative to the control mean. This result is highly statistic-
ally significant, even after adjusting for multiple tests. Two rea-
sons we may see a decline in collections but not in the other
measures are that a collection is less of a right tail event and
occurs with less of a lag following an unpaid bill than a judgment,
lien, or bankruptcy.
Another possibility is that there is more of an impact of
health insurance on collections because in both our population
and the general population, collections are disproportionately
medical (Avery, Calem, and Canner 2003). We are able to de-
compose the presence and size of collections into medical (Panel
B) and nonmedical (Panel C) components; this decomposition is
not feasible for the other measures. The decline in overall col-
lections shown in Panel A is primarily (or perhaps entirely)
driven by a decline in medical collections. We find declines in
both the existence and magnitude of medical collections. For
example, health insurance is associated with a decline in the
probability of having medical collections of 6.4 percentage
points (std. err. = 1.6), or about 25% relative to the control
mean, and a decline in the average (unconditional) amount
owed in medical collections of $390 (std. err. = $177), or about
20% relative to the control mean. The corresponding estimates
for nonmedical collections are substantially smaller and statis-
tically insignificant.
These results are subject to some potential limitations, dis-
cussed in more detail in the Online Appendix. First, not all col-
lections are reported to the credit bureaus, although our
investigations did not suggest any reason to suspect reporting
TABLE VIII
FINANCIAL STRAIN (SURVEY DATA)
Control
(1) (2) (3) (4)
Any out of pocket medical expenses, 0.555 0.058 0.200 [<0.0001]
last six months (0.497) (0.0077) (0.026) {<0.0001}
Owe money for medical expenses 0.597 0.052 0.180 [<0.0001]
currently (0.491) (0.0076) (0.026) {<0.0001}
Borrowed money or skipped other 0.364 0.045 0.154 [<0.0001]

bills to pay medical bills, last six (0.481) (0.0073) (0.025) {<0.0001}
months
Refused treatment because of med- 0.081 0.011 0.036 [0.01]
ical debt, last six months (0.273) (0.0041) (0.014) {0.01}
(0.010) (0.035)
p-values in curly brackets. Column (2) reports the coefficient and standard error on LOTTERY from
the first row of Table III. Column (4) reports the per comparison p-value and the family-wise p-value
across the four different measures of financial strain used to create the standardized treatment effect.
Standardized treatment effect reports results based on equation (2). All regressions include household size
fixed effects, survey wave fixed effects, and the interaction between the two. All standard errors are
clustered on the household and all regressions are weighted using survey weights. Sample consists of
survey responders (N = 23,741).
to be correlated with insurance status (see Online Appendix 1.6).

Second, in theory health insurance might affect access to revol-
ving credit itself, which could complicate interpretation of meas-
ures of financial strain based on late payments for revolving
credit (i.e., delinquencies); however, we found no evidence of ef-
fects on access to credit (Online Appendix 3.3). Third, many of the
measures capture only right-tail events and with a substantial
lag. Last, credit reports do not capture the use of informal or
nontraditional credit sourceswhich may be particularly im-
portant in a low-income population. Given these limitations, the
survey measures of financial strain are a useful complement to
the credit report measures.
2. Survey Data. Table VIII reports results for four measures of

financial strain: whether the respondent has any out-of-pocket
medical expenditures in the past six months, currently owes
money for medical expenses, had to borrow money (or skip
paying other bills or pay them late) to pay medical expenses in
the past six months, and has been refused medical treatment
because of medical debt in the past six months. We find a statis-

tically significant decline in all four survey measures of financial
strain, including, for example, a 20 percentage point (35%) de-
cline in the probability of having out-of-pocket expenses and a
15 percentage point (40%) decline in the probability of having
to borrow money or skip paying other bills to pay medical
expenses.29 The average standardized treatment effect indicates
that insurance is associated with a 0.3 standard deviation (std.
err. = 0.035) decline in these measures of financial strain.

The results suggest that financial benefits from Medicaid
coverage accrue not just to the newly insured. The declines in
out-of-pocket expenses and in reported difficulty paying nonme-
dical bills point to direct financial benefits to the newly insured.
At the same time, both the survey data and the credit report
data indicate a reduction in outstanding medical bills, the vast
majority of which are ultimately never paid. This suggests that
some of the financial benefits from Medicaid coverage accrue
to medical providers or to whomever they pass on the costs of
unpaid care.
For risk-averse consumers, the largest welfare gains from
any consumption-smoothing effects of insurance come from redu-
cing extreme negative shocks to consumption. Although we
cannot measure consumption directly, following Finkelstein and
McKnight (2008) we estimate quantile regression models of the
reduced-form equation (1) to examine the impact of health insur-
ance on the quantiles of the distribution of out-of-pocket medical
expenditures. As expected given the comprehensive nature of
Medicaid coverage, Figure I, Panels A and B indicate that selec-
tion by the lottery is associated with declines in out-of-pocket
29. These results imply that about 35% of those covered by OHP still have
out-of-pocket medical expenses. The control group reports, on average, $307 in
semi-annual out-of-pocket medical expenses; the LATE estimate of the impact of
insurance on these expenses is 122 (std. err. = 43), implying that those covered by
OHP average $185 in semi-annual out-of-pocket medical expenses. Our impression
from focus groups is that these reflect some combination of continued scheduled
payments on prior debts, reporting of monthly premiums as out-of-pocket medical
expenses, and perhaps including travel costs to the medical provider. It is also
possible that some individuals report out-of-pocket medical spending for other
family members (even though the question directed individuals to report only ex-
penditures on themselves) or for uncovered services, such as dental care (even
though the survey question explicitly said to exclude dental). Gross and
Notowidigdo (2011) similarly find evidence of reported out-of-pocket spending
from Medicaid recipients in the MEPS.

FIGURE I
Distribution of Out-of-Pocket Medical Expenses, Last Six Months (Survey Data)
Panel A shows the distribution of out-of-pocket medical spending for con-
trols, and the estimated distribution for treatments through the 95th quantile;
the estimated distribution for treatments is the control distribution added to
the beta on LOTTERY from the quantile estimation of the reduced-form equa-
tion (1). Panel B plots the quantile estimates from equation (1) (along with their
95% confidence interval) starting from the smallest quantile that is nonzero in
either the treatment or control distribution through the 95th quantile. The
confidence intervals are calculated based on 500 bootstraps clustered on house-
hold. Data are from the sample of survey responders (N = 24,012); all results
use survey weights. Quantile estimation of equation (1) includes household size
dummies, survey wave dummies, and the interaction of the two.
TABLE IX
HEALTH 1094
Control
(1) (2) (3) (4)
Panel A: Administrative data
Alive 0.992 0.00032 0.0013 [0.638]
(0.092) (0.00068) (0.0027)
Panel B: Survey data
Self-reported health good/very good/excellent (not fair or poor) 0.548 0.039 0.133 [<0.0001]
(0.498) (0.0076) (0.026) {<0.0001}
Self-reported health not poor (fair, good, very good, or 0.86 0.029 0.099 [<0.0001]
excellent) (0.347) (0.0051) (0.018) {<0.0001}
Health about the same or gotten better over last six months 0.714 0.033 0.113 [<0.0001]
(0.452) (0.0067) (0.023) {<0.0001}
# of days physical health good, past 30 days* 21.862 0.381 1.317 [0.019]
(10.384) (0.162) (0.563) {0.018}
# days poor physical or mental health did not impair usual 20.329 0.459 1.585 [0.009]
activity, past 30 days* (10.939) (0.175) (0.606) {0.015}
# of days mental health good, past 30 days* 18.738 0.603 2.082 [0.001]
(11.445) (0.184) (0.64) {0.003}
Did not screen positive for depression, last two weeks 0.671 0.023 0.078 [0.001]
(0.470) (0.0071) (0.025) {0.003}

(0.011) (0.039)
Notes. Standard errors in parentheses; per comparison p-values in square brackets; family-wise p-values in curly brackets. Column (2) reports the coefficient and standard error
on LOTTERY from estimating equation (1) by OLS. Column (3) reports the coefficient and standard error on INSURANCE from estimating equation (3) by IV; for the IV estimates in
column (3), the endogenous variable INSURANCE is defined as ever on Medicaid during our study period and the first stage is given in the first row of Table III. Column (4)
reports the per comparison p-value and the family-wise p-value across the different measures used to create the standardized treatment effect. Standardized treatment effect reports
results based on equation (2). All regressions include household size fixed effects and standard errors are clustered on the household. The regressions in Panel A include lottery draw
fixed effects, and the dependent variable alive is measured from the notification date through September 2009 (N = 74,922). The regressions in Panel B include survey wave fixed
effects and the interaction of survey wave fixed effects with household size fixed effects, and are weighted using the survey weights (N = 23,741).
*These questions were worded to ask about number of days health not good or impaired; we switched the sign for consistency with the other measures. See Online Appendix
Figure A4 for the exact survey wording.

spending at all the (nonzero) quantiles of the distribution. In the

Online Appendix we show similar effects for self-reported medical
liabilities and medical collections (Online Appendix Figures
A5aA6b).
V.C. Health
Table IX shows our estimates of the impact of health insur-
ance on health. We have one measure of health from administra-

tive data, mortality, which we measure from the notification date
through September 30, 2009. Mortalityalthough important and
objectively measuredis very low in our population; only
about 0.8% of the controls died over the 16-month study period.
Not surprisingly, Panel A shows that we do not detect any stat-
istically significant improvement in survival probability.
Panel B analyzes seven different measures of self-reported
health from the survey data. The first two use the question about
self-reported health (fair, poor, good, very good, or excellent) to
construct two binary measures: (1) self-reported health good, very
good, or excellent (55% of the population), and (2) to examine
tail behavior, self-reported health not poor (86% of the popula-
tion). The other measures are (3) whether self-reported health
status is about the same or has gotten better over past six
months (versus gotten worse), (4) the number of days in good
physical health in past month (030), (5) the number of days
not impaired by physical or mental health in the past month
(030), (6) the number of days in good mental health in the past
month (030), and (7) whether the respondent screened negative
for depression. Many of these measures capture both physical and
mental health; the last two capture only mental health.
The results in Table IX, Panel B indicate that insurance is
associated with statistically significant improvements in each of
the seven measures. On average, our results suggest that health
insurance is associated with a 0.2 standard deviation (std.
err. = 0.04) improvement in self-reported health. This includes,
among other things, an increase in the probability of screening
negative for depression of 7.8 percentage points (std. err. = 2.5) or
about 10% relative to the control mean and an increase in the
probability of reporting ones health as good, very good, or excel-
lent of 13 percentage points (std. err. = 2.6), or about 25% relative
to the control mean.
There is thus an overwhelming sense from the survey out-

comes that individuals feel better about their health and, as we
come to discuss shortly, their interactions with the health care
system. Given the subjective nature of the responses, however, it
is difficult to judge the extent to which these results reflect
improvements in physical health. For mental health, the
self-reported and subjective nature of the questions is less of an
issue, because diagnosis of depression, by its nature, relies on
such self-reports; the depression screen we use correlates

highly with clinical diagnoses of depression (Kroenke, Spitzer,
and Williams 2003). However, the self-reported physical health
measures could reflect a more general sense of improved
well-being rather than actual improvements in objective health.
(A priori, however, we were concerned that by increasing contact
with the health care system, health insurance would cause indi-
viduals to learn more about their health problems [e.g., a doctor
would tell a person who had not known it that they diabetes or
high blood pressure] and thus could cause them to report them-
selves to be in worse health; to the extent this happens, it does not
outweigh effects in the opposite direction.)
There is evidence of several mechanisms by which health
insurance could have improved objective, physical health.
Besides the previously documented increase in health care util-
ization and compliance with recommended preventive care, the
survey data also indicate that insurance is associated with stat-
istically significant increased self-reported access to care
(Table X, Panel A) and perceived quality of care (conditional on
receipt) (Table X, Panel B). However, there is also evidence that
a substantial part of the estimated improvements may reflect a
general sense of improved wellbeing. Table XI compares reduced-
form estimates from our main survey to reduced-form estimates
from the initial survey, which we fielded on average about
2.6 months after random assignment and about 1 month after
coverage was approved (this analysis was not prespecified). As
we would expect given this timing, there is no evidence of an in-
crease in health care utilization in this earlier survey. However,
there is evidence of an improvement in self-reported health of
about two-thirds the magnitude of our main survey estimates
from more than a year later; this reflects statistically significant
improvements in all of the individual health measures available in
the initial survey (all those in the main survey except the depres-
sion screen). Given the limited time after coverage approval and
TABLE X
POTENTIAL MECHANISMS FOR IMPROVED HEALTH (SURVEY DATA)
Control
(1) (2) (3) (4)
Panel A: Access to care
Have usual place of clinic-based care 0.499 0.099 0.339 [<0.0001]
(0.500) (0.0080) (0.027) {<0.0001}
Have personal doctor 0.490 0.081 0.280 [<0.0001]
(0.500) (0.0077) (0.026) {<0.0001}
Got all needed medical care, last six months 0.684 0.069 0.239 [<0.0001]
(0.465) (0.0063) (0.022) {<0.0001}
Got all needed drugs, last six months 0.765 0.056 0.195 [<0.0001]
(0.424) (0.0055) (0.019) {<0.0001}
Didnt use ER for nonemergency, last six months 0.916 0.0011 0.0037 [0.804]
(0.278) (0.0043) (0.015) {0.804}
(0.0084) (0.029)
Panel B: Quality of care
Quality of care received last six months good/very good/excellent (conditional on any) 0.708 0.043 0.142 [<0.0001]
(0.455) (0.0081) (0.027)
Panel C: Happiness
Very happy or pretty happy (vs. not too happy) 0.594 0.056 0.191 [<0.0001]
(0.491) (0.0074) (0.026)
Notes. Standard errors in parentheses; per comparison p-values in square brackets, family-wise p-values in curly brackets. Column (2) reports the coefficient and standard error
on LOTTERY from estimating equation (1) by OLS. Column (3) reports the coefficient and standard error on INSURANCE from estimating equation (3) by IV; for the IV estimates in
column (3), the endogenous variable INSURANCE is defined as ever on Medicaid during our study period and the first stage is given in the first row of Table III. Column (4)
reports the per comparison p-value and the family-wise p-value across the four different measures of financial strain used to create the standardized treatment effect. Standardized
treatment effect reports results based on equation (2). All regressions include household size fixed effects, survey wave fixed effects, and the interaction between the two. All
1097
standard errors are clustered on the household and all regressions are weighted using survey weights. Sample consists of survey responders (N = 23,741).

TABLE XI
ESTIMATED EFFECTS OF LOTTERY AT DIFFERENT TIMES
1098
ITT estimates from survey p-value of difference between
Initial Six month Main Initial and six month Six month and main Initial and main
(1) (2) (3) (4) (5) (6)
Utilization (extensive margin) 0.0038 0.047 0.050 0.035 0.867 <0.0001
(0.0084) (0.020) (0.011)
[0.656] [0.02] [<0.0001]
Utilization (total) 0.00023 0.027 0.040 0.188 0.556 0.001
(0.0086) (0.020) (0.011)
[0.978] [0.187] [0.0003]
Financial strain 0.035 0.099 0.089 0.002 0.613 <0.0001
(0.0089) (0.020) (0.010)
[<0.0001] [<0.0001] [<0.0001]
Health 0.042 0.097 0.061 0.014 0.121 0.112
(0.010) (0.023) (0.011)
[<0.0001] [<0.0001] [<0.0001]
Access 0.047 0.075 0.119 0.163 0.026 <0.0001
(0.0078) (0.019) (0.0086)
[<0.0001] [<0.0001] [<0.0001]
Notes. The analysis in this table was not prespecified. Standard errors in parentheses; per comparison p-values in square brackets. Table reports standardized treatment effects
from estimating equation (1) in three different surveys. The first column shows the results for our so-called initial survey, which occurred about 2.6 months after lottery notification;
the second column shows results for our six-month survey, which occured about 8 months after lottery notification (6 months after insurance coverage began); the third column
shows results for our main survey, which occured about 15 months after lottery notification; this is the survey that has been analyzed in prior tables. The surveys had similar
response rates and questionnaires; the six-month survey was conducted on an approximately 20percent sample of the other surveys (see Section III.D and Online Appendix 1 for
details). Columns (1), (2), and (3) report the coefficient, standard error, and p-value on LOTTERY from estimating equation (1) by OLS. Columns (3), (4), and (5) report the p-value of
the difference between various estimates. All regressions include household size fixed effects, survey wave fixed effects, and the interaction of survey wave and household size fixed
effects. Regressions based on the six-month and main surveys (columns (2) and (3)) are weighted using the survey weights for those surveys. Standard errors are clustered on the
household. N = 26,423 for the initial survey, N = 6,359 for the six-month survey, and N = 23,741 for the main survey. The individual components of the standardized treatment effects
are the same as in the earlier tables, except that health excludes the depression screen question and access excludes the personal doctor question (as these questions were not
asked in the initial survey). The reference period for the underlying questions is usually either currently or in the last six months (see prior tables for details).

the lack of any increase in health care utilization, it seems likely

that this immediate improvement does not reflect changes in ob-
jective physical health.30 Likewise, self-reported access to care also
shows a statistically significant improvement in the initial survey
(despite no evidence of an increase in utilization) of about 40%
the magnitude of the later survey, which again seems more
likely to reflect an improved outlook. It is not clear, however,
that the immediate effects are directly comparable to those
from one year later. Some of the immediate improvements may

reflect winning effects that are less likely to be picked up in the
estimates one year later, and the later estimates of self-reported
improvements in health may be biased downward by the impact
of health insurance on receiving health care and hence poor
health diagnoses, an effect that would not arise in the initial
survey because health care use does not appear to go up
immediately.
Consistent with an improved overall sense of well-being,
there is evidence in the later survey of a substantial (32%) in-
crease in self-reported overall happiness (Table X, Panel C). To
put this in perspective, if we compare our estimates to the litera-
ture on the impact of income on happiness, the impact of insur-
ance roughly equivalent to the impact of a doubling of income.31
Of course, it is difficult to know how much of the self-reported
happiness improvement reflects feeling better about ones health,
just as it is difficult to know how much of the self-reported health
improvement reflects more general improvement in a sense of
well-being. Overall, the evidence suggests that people feel
better off due to insurance, but with the current data it is difficult
to determine the fundamental drivers of this improvement.
30. There is also evidence of a decline in financial strain in the initial survey
that is about 40% the magnitude of the analogous measures in the later survey. This
is consistent with the fact that coverage was applied retroactively for lottery win-
ners to about one month prior to the approval date. Individuals in the initial survey
had therefore been covered retroactively for about two months, or about one-third of
the six-month look-back period. Thus, although they may not have changed their
health care utilization in the first month of coverage, the retroactive coverage for
two months had an impact on their finances.
31. Our IV or LATE estimates in Table X indicate that Medicaid increases hap-
piness by about 0.4 standard deviations. The literature on the relationship between
happiness and income tends to find that a log-point increase in income is associated
with an increase in happiness of about this magnitude (see, e.g., Sacks, Stevenson,
and Wolfers 2010, table 1).
VI. DISCUSSION: INTERPRETING AND EXTRAPOLATING THE RESULTS

VI.A. Comparison with Other Estimates
We compare our experimental estimates with those that
would be obtained by constructing observational estimates in
our own data (ignoring the role of the lottery in driving insurance
coverage) or in national survey data (see Online Appendix 3.6 and
Online Appendix Tables A27 and A28). Compared with our ex-
perimental estimates, observational analogs suggest larger im-

pacts of health insurance on health care utilization and
opposite-signed impacts of health insurance on health, with
health insurance appearing to worsen health in the observational
estimates. These differences suggest that at least within a
low-income population, individuals who select health insurance
coverage are in poorer health (and therefore demand more med-
ical care) than those who are uninsured, just as standard adverse
selection theory would predict.
We also compare our estimates to existing experimental es-
timates. Our estimates of the impact of moving low-income adults
from no insurance to Medicaid are not directly comparable to the
RAND experimental estimates of varying the extent of coverage
among the insured for a representative, nonelderly population.
Nevertheless, with this important caveat in mind, our estimates
seem slightly smaller than those found in RAND, whereas the
improvements in self-reported health are larger.32 The RAND
Experiment did not analyze the impact of health insurance on
financial risk exposure.33 Slightly more comparable to our
32. For example, RAND found that moving from the least comprehensive in-
surance planwhich still offered considerable insurance coverage to full insurance
was associated with a 45% increase in annual spending, whereas our
back-of-the-envelope calculation suggested that relative to being uninsured,
Medicaid was associated with a 25% increase in six-month spending. The same
insurance variation in RAND also produced about a 75% increase in the number
of annual outpatient visits, compared with the 55% increase we estimated
(Newhouse and the Insurance Experiment Group 1993, p. 41). RAND found no
evidence of an impact of insurance generosity on adult self-reported general
health or adult mental health (Newhouse and the Insurance Experimenta Group
1993, p. 209).
33. In the United States, we know of only three quasi-experimental studies of
the impact of health insurance on risk exposure. Like us, these studies find that
health insurance reduces the distribution of out-of-pocket medical expenditures
(Finkelstein and McKnight 2008; Englehardt and Gruber 2011) although, unlike
us, evidence that Medicaid expansions (at a slightly higher income level) are asso-
ciated with a decline in personal bankruptcies (Gross and Notowidigdo 2011).
estimates are the results from a contemporaneous randomized

trial of the one-year effects of extending public health insurance
coverage to uninsured nonelderly adults receiving Social Security
Disability Insurance (Michalopoulos et al. 2011). This population
is similar in age to the one in this study but had higher income
and was in much worse health (with not surprisingly substan-
tially higher health care use); it is also had a much higher fraction
nonwhite. Qualitatively, the findings are very similar; in both
experiments, insurance is associated with increased use of pre-

scription drugs and primary care (although no discernible impact
on ER or inpatient hospital use in survey data), decreased finan-
cial strain (such as declines in the right tail of out-of-pocket med-
ical expenditures), and improvements in self-reported health.
VI.B. Extrapolation to Other Contexts

Our results should be interpreted in light of the particular
characteristics of the study population and the nature and timing
of the policy intervention. Although it is a natural impulse to try
to generalize these experimental estimates to other contexts,
including the planned 2014 Medicaid expansions, it is important
to bear in mind several caveats.
First, our results speak only to the approximately one-year
impact of expanding Medicaid access. As discussed, our reading of
the available evidence is that the results do not reflect a transi-
torily large use response stemming from initial pent-up demand.
However, the increase in utilization may be larger than what we
would find if individuals were not at risk of losing their insurance
(and being unable to reapply for it) if they fail to continue to meet
the eligibility requirements. The effects of health insurance on
the outcomes studied may also change over longer time horizons
than we are able to analyze. For example, because health is a
stock rather than a flow, the long-run health impacts of health
insurance may exceed our one-year estimates. On the other hand,
there is evidence from the Moving to Opportunity experiment,
which provided housing vouchers to low-income families living
in public housing in high-poverty areas, suggesting that
short-run improvements in self-reported general health may
fade out after a few years, although self-reported improvements
in mental health outcomes may persist for longer (Katz, Kling,
and Liebman 2001; Kling, Liebman and Katz 2007).
Second, our findings speak to the partial equilibrium effects

of covering a small number of people, holding constant the rest of
the health care system; the effects of much larger health insur-
ance expansions might differ because of supply-side responses by
the health care sector (Finkelstein 2007).
Third, our results are specific to a particular population in a
particular health care environment at a particular time period.
Our population is not representative of the low-income uninsured
adults in the rest of the United States on a number of observable

(and presumably unobservable) dimensions. One striking differ-
ence is that our study population has more whites and fewer
African Americans (by about 15 percentage points each) than
the general low-income, uninsured, adult U.S. population. It is
also somewhat (four to five years) older and on some measures
appears to be in somewhat worse self-reported health (Allen et al.
2010). These differences are amplified when focusing on com-
pliers, who, relative to the overall lottery population, are some-
what older, more likely to be white, in worse health, and of lower
socioeconomic status (as proxied by education and having revol-
ving credit at the time of the lottery). Although we examined
heterogeneity in treatment effects by these and other observ-
ables, we lacked power to draw precise inferences (see Online
Appendix 3.5 and Online Table A26 for details).
In addition to observable demographic differences, our study
population voluntarily signed up for the lottery and enrolled in
health insurance. The impact of health insurance on, for example,
health care utilization may well be larger for those who select into
it than for those enrolling when coverage is mandatory; Einav
et al. (2011) provide evidence of such selection on moral
hazard in one specific context. Consistent with this idea, in our
setting we find some evidence that the approximately 10% of the
sample who signed up the first day the lottery list was open have
larger utilization responses than the rest of the lottery partici-
pants (see Online Appendix Table A26) (this analysis was not
prespecified). The impact of a public health insurance expansion
may also vary with the business cycle; our estimates come from a
time of substantial economic downturn.
Fourth, the impact of Medicaid may also vary with the spe-
cifics of the Medicaid program and the nature of the safety net
available to the uninsured. There is a great deal of heterogeneity
across states in these dimensions. Our reading of the data, how-
ever, suggests that Oregon is not particularly atypical on the
dimensions we can measure. For example, like Oregon, almost

all states operate Medicaid managed care programs. Medicaid
fees in Oregon are around the median of U.S. states. OHP
Standard is fairly generous in its lack of any copayments and
its coverage of nonmandated services, although some states
cover dental and vision, which OHP Standard does not
(Ramirez de Arellano and Wolfe 2007; Kaiser Family
Foundation 2011). It appears the safety net may be slightly
more generous in Oregon, although other measures of the

health care system like the share of admissions in public hos-
pitals and the physician-to-population ratio are quite similar to
the national average (Allen et al. 2010).
Finally, the insurance we study was free or heavily subsi-
dized, and as a result our estimates capture the combined effect
of insurance at actuarially fair prices and the wealth effect from
the large premium subsidy. Average annual OHP Standard ex-
pendituresand hence an actuarially fair premiumwere esti-
mated to be about $3,000 in 20012004 (Wallace et al. 2008),
which is quite high relative to the actual annual premium of $0
to $240. Presumably, however, most health insurance coverage
for this type of low-income population would also be heavily or
completely subsidized.
VI.C. Conclusion
Using a randomized controlled experiment design, we exam-
ined the approximately one-year impact of extending access to
Medicaid among a low-income, uninsured adult population. We
found evidence of increases in hospital, outpatient, and drug util-
ization; increases in compliance with recommended preventive
care; and declines in exposure to substantial out-of-pocket med-
ical expenses and medical debts. We also found evidence of im-
provement in self-reported mental and physical health measures,
perceived access to and quality of care, and overall well-being.
Our results suggest that Medicaid provides benefits to this popu-
lation above and beyond the non-Medicaid alternatives that exist
through various safety-net options. These results are important
inputs into a careful cost-benefit analysis of this expansion in
Medicaid, although such an analysis would require a number of
additional assumptions that go beyond the data that this experi-
ment can provide.
Supplementary Material
An Online Appendix for this article can be found at QJE
online (qje.oxfordjournals.org).
National Bureau of Economic Research and

Massachusetts Institute of Technology
National Bureau of Economic Research
Center for Outcomes Research and Education at

Providence Health and Services
National Bureau of Economic Research
Massachusetts Institute of Technology
National Bureau of Economic Research and Harvard
University
Center for Outcomes Research and Education at
Providence Health and Services
Harvard University
REFERENCES
Alatas, Vivi, Abhijit Banerjee, Rema Hanna, Benjamin A. Olken, and
Julia Tobias, Targeting the Poor: Evidence from a Field Experiment in
Indonesia, NBER Working Paper No. 15980, 2010.
Allen, Heidi, Katherine Baicker, Amy Finkelstein, Sarah Taubman, and
Bill Wright, and the Oregon Health Study Group, What the Oregon
Health Study Can Tell Us about Expanding Medicaid, Health Affairs, 29
(2010), 14981506.
Anderson, Michael, Multiple Inference and Gender Differences in the Effects of
Early Intervention: A Reevaluation of the Abecedarian, Perry Preschool, and
Early Training Projects, Journal of the American Statistical Association,
103 (2008), 14811495.
Anderson, Michael, Carlos Dobkin, and Tal Gross, The Effect of Health
Insurance Coverage on the Use of Medical Services, NBER Working Paper
No. 15823, 2010.
Angrist, Joshua, Estimation of Limited-Dependent Variable Models with
Dummy Endogenous Regressors: Simple Strategies for Empirical Practice,
Journal of Business Economics and Statistics, 19 (2001), 216.
Angrist, Joshua, Guido Imbens, and Donald Rubin, Identification and Causal
Effects Using Instrumental Variables, Journal of the American Statistical
Association, 91 (1996), 444472.
Angrist, Joshua, and Steve Pischke. Mostly Harmless Econometrics: An
Empiricists Companion (Princeton, NJ: Princeton University Press, 2009).
Avery, Robert, Paul Calem, and Glenn Canner, An Overview of Consumer Data
and Credit Reporting, Federal Reserve Bulletin, 89 (2003), 4778.
Card, David, Carlos Dobkin, and Nicole Maestas, The Impact of Nearly Universal
Coverage on Health Care Utilization: Evidence from Medicare, American
Economic Review, 98 (2008), 22422258.
, Does Medicare Save Lives? Quarterly Journal of Economics, 124 (2009),
597636.
Casey, Katherine, Rachel Glennerster, and Edward Miguel, Reshaping

Institutions: Evidence on Aid Impacts Using a Pre-Analysis Plan. Mimeo,
2011, available online at http://elsa.berkeley.edu/emiguel/pdfs/miguel_gbf
.pdf.
Doyle, Joseph, Health Insurance, Treatment, and Outcomes: Using Auto
Accidents as Health Shocks, Review of Economics and Statistics, 87
(2005), 256270.
Einav, Liran, Amy Finkelstein, Stephen Ryan, Paul Schrimpf, and Mark Cullen,
Selection on Moral Hazard in Health Insurance, NBER Working Paper No.
w16969, 2011.
Engelhardt, Gary V., and Jonathan Gruber, Medicare Part D and the Financial
Protection of the Elderly, American Economic Journal: Economic Policy, 3
(2011), 77102.

Finkelstein, Amy, The Aggregate Effects of Health Insurance: Evidence from the
Introduction of Medicare, Quarterly Journal of Economics, 122 (2007), 137.
Finkelstein, Amy, and Robin McKnight, What Did Medicare Do? The Initial
Impact of Medicare on Mortality and Out-of-pocket Medical Spending,
Journal of Public Economics, 92 (2008), 16441669.
General Accountability Office (GAO), Medicaid and CHIP: Most Physicians
Serve Covered Children but Have Difficulty Referring them to Specialists,
GAO-11-624, 2011; available online at http://www.gao.gov/new.items/
d11624.pdf.
Getzen, Thomas, Health Care Is an Individual Necessity and a National Luxury:
Applying Multilevel Decision Models to the Analysis of Health Care
Expenditures, Journal of Health Economics, 19 (2000), 259270.
Gross, Tal., and Matthew J. Notowidigdo, Health Insurance and the Consumer
Bankruptcy Decision: Evidence from the Expansion of Medicaid, Journal of
Public Economics, 95 (2011), 767778.
Hoynes, Hilary, and Diane Schanzenbach, Consumption Responses to In-Kind
Transfers: Evidence from the Introduction of Food Stamps, American
Economic Journal: Applied Economics, 1 (2009), 109139.
Imbens, Guido W., and Joshua Angrist, Identification and Estimation of Local
Average Treatment Effects, Econometrica, 62 (1994), 467475.
Kaiser Family Foundation, Implementation Timeline, (2010a); available online
at http://healthreform.kff.org/timeline.aspx (accessed October 27, 2010).
, Focus on Health Reform: Summary of New Health Reform Law,
Publication 8061, 2010b; available online at http://www.kff.org/healthre
form/8061.cfm (accessed February 22, 2011).
, State Health Facts, (2011); available online at http://www.statehealth
facts.org.
Katz, Lawrence F., Jeffrey R. Kling, and Jeffrey B. Liebman, Moving to
Opportunity in Boston: Early Results of a Randomized Mobility
Experiment, Quarterly Journal Of Economics, 116 (2001), 607654.
Kling, Jeffrey, and Jeffrey Liebman, Experimental Analysis of Neighborhood
Effects on Youth, Mimeo, 2004; available online at http://www.nber.org/
mtopublic/483.pdf.
Kling, Jeffrey, Jeffrey Liebman, and Lawrence Katz, Experimental Analysis of
Neighborhood Effects, Econometrica, 75 (2007), 83119.
Kroenke, K., R. L. Spitzer, and J. B. Williams, The Patient Health
Questionnaire-2: Validity of a Two-Item Depression Screener, Medical
Care, 41 (2003), 12841292.
Lee, David S., Training, Wages, and Sample Selection: Estimating Sharp Bounds
on Treatment Effects, Review of Economic Studies, 76 (2009), 10711102.
Levy, Helen, and David Meltzer, The Impact of Health Insurance on Health,
Annual Review of Public Health, 29 (2008), 399409.
Manning, Willard, Joseph Newhouse, Naihua Duan, Emmett Keeler,
Arleen Leibowitz, and Susan Marquis, Health Insurance and the Demand
for Medical Care: Evidence from a Randomized Experiment, American
Economic Review, 77 (1987), 251277.
Medicaid Access Study Group, Access of Medicaid Recipients to Outpatient
Care, New England Journal of Medicine, 330 (1994), 14261430.
Michalopoulos, Charles, David Wittenburg, Dina Israel, Jennifer Schore,

Anne Warren, Aparajita Zutshi, Stephen Freedman, and Lisa Schwartz,
The Accelerated Benefits Demonstration and Evaluation Project: Impacts
on Health and Employment at 12 months, MDRC, 2011; available online at
http://www.mdrc.org/publications/597/full.pdf.
Newhouse, Joseph P. and the Insurance Experiment Group. Free for All?: Lessons
from the RAND Health Insurance Experiment (Cambridge, MA: Harvard
University Press, 1993).
Olken, Benjamin A., Junko Onishi, and Susan Wong, Indonesias PNPM
Generasi Program: Interim Impact Evaluation Report, (World Bank, 2010).
Office for Oregon Health Policy and Research, Trends in Oregons HealthCare
Market and the Oregon Health Plan: A Report to the 75th Legislative
Assembly, (February 2009).

Ramirez de Arellano, Annette, and Sidney Wolfe, Unsettling Scores: A Ranking
of State Medicaid Programs, Public Citizen Health Research Group, Public
Citizen Press, 2007; available online at http://www.citizen.org/medicaid.
Rosenbaum, Sara, NEJM Perspective: Equal Access for Medicaid Beneficiaries
The Supreme Court and the Douglas Cases, New England Journal of
Medicine, 365 (2011), 22452247; available online at http://www.nejm.org/
doi/full/10.1056/NEJMp1111428.
Sacks, Daniel, Betsey Stevenson, and Justin Wolfers, Subjective Well-Being,
Income, Economic Development, and Growth, NBER Working Paper No.
w16441, 2010.
Schaner, Simone, Intrahousehold Preference Heterogeneity, Commitment and
Strategic Savings: Theory and Evidence from Kenya, MIT Working Paper,
2010; available online at http://econ-www.mit.edu/files/6221.
U.S. GPO, Patient Protection and Affordability Care Act (HR 3590), (2010);
available online at http://www.gpo.gov/fdsys/pkg/PLAW-111publ148/pdf/
PLAW-111publ148.pdf.
Wallace, Neal, John McConnell, Charles Gallia, and Jeanene Smith, How
Effective are Co-Payments for Reducing Expenditures for Low-Income
Adult Medicaid Beneficiaries? Experience from the Oregon Health Plan,
Health Services Research, 43 (2008), 515530.
Westfall, Peter H., and Stanley S. Young Resampling-Based Multiple Testing:
Examples and Methods for P-value Adjustment (New York: Wiley, 1993).
Zeckhauser, Richard, Medical Insurance: A Case Study of the Tradeoff between
Risk Spreading and Appropriate Incentives, Journal of Economic Theory, 2
(1970), 1026.

Finkelstein Etal Oregon 2012 PDF

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

Finkelstein Etal Oregon 2012 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Finkelstein Etal Oregon 2012 PDF

Uploaded by

Copyright:

Available Formats

THE

Downloaded from http://qje.oxfordjournals.org/ at MIT Libraries on September 24, 2012

In 2008, a group of uninsured low-income adults in Oregon was selected by

Downloaded from http://qje.oxfordjournals.org/ at MIT Libraries on September 24, 2012

administration of state programs. We gratefully acknowledge funding from the

would avoid such confounding, but such opportunities are rarely

Downloaded from http://qje.oxfordjournals.org/ at MIT Libraries on September 24, 2012

1. We know of only two other randomized health insurance experiments in the

uncompensated care allow low-income individuals to consume de

Downloaded from http://qje.oxfordjournals.org/ at MIT Libraries on September 24, 2012

2. Our analysis plan was archived on December 3, 2010, at http://www.nber

insurance appears to come entirely through a gross increase in

Downloaded from http://qje.oxfordjournals.org/ at MIT Libraries on September 24, 2012

Downloaded from http://qje.oxfordjournals.org/ at MIT Libraries on September 24, 2012

II. OREGONS MEDICAID LOTTERY

OHP Standard provides relatively comprehensive benefits

Downloaded from http://qje.oxfordjournals.org/ at MIT Libraries on September 24, 2012

household submitted the appropriate paperwork within 45 days

Downloaded from http://qje.oxfordjournals.org/ at MIT Libraries on September 24, 2012

III.A. Administrative Data on Outcomes: Hospital Discharges,

of total medical expenditures for 1864-year-olds.5 We observe,

Downloaded from http://qje.oxfordjournals.org/ at MIT Libraries on September 24, 2012

III.B. Survey Data on Outcomes

5. Author calculations based on publicly available tables from the 2008

Downloaded from http://qje.oxfordjournals.org/ at MIT Libraries on September 24, 2012

III.C. Other Data

III.D. Time Frame of the Study

Downloaded from http://qje.oxfordjournals.org/ at MIT Libraries on September 24, 2012

III.E. Sample Definition and Descriptive Statistics

9. We randomly assigned lottery draws to the control individuals as discussed

Variable Control mean Variable Control mean

Race Health status

Downloaded from http://qje.oxfordjournals.org/ at MIT Libraries on September 24, 2012

Variable Control mean Variable Control mean

Downloaded from http://qje.oxfordjournals.org/ at MIT Libraries on September 24, 2012

population down to 74,922 individuals (representing 66,385

Downloaded from http://qje.oxfordjournals.org/ at MIT Libraries on September 24, 2012

IV. EMPIRICAL FRAMEWORK

Downloaded from http://qje.oxfordjournals.org/ at MIT Libraries on September 24, 2012

individual was considered selected and eligible to apply for insur-

Downloaded from http://qje.oxfordjournals.org/ at MIT Libraries on September 24, 2012

12. The proportion of treated (respectively, control) individuals in household

not sensitive to estimating nonlinear models instead (this ana-

Downloaded from http://qje.oxfordjournals.org/ at MIT Libraries on September 24, 2012

Handling many outcomes: standardized treatment effects and

where j is the standard deviation of yj in the control group and 1j

The per comparison p-value may be appropriate if one is inter-

Downloaded from http://qje.oxfordjournals.org/ at MIT Libraries on September 24, 2012

Control mean (std. dev.) Difference between treatment and control

Downloaded from http://qje.oxfordjournals.org/ at MIT Libraries on September 24, 2012

(2) through (4), respectively: the sample universe (which is the

Downloaded from http://qje.oxfordjournals.org/ at MIT Libraries on September 24, 2012

18. Conditional on response, response time between treatment and controls is

.23. For the survey respondent subsamplewhere a priori we

Downloaded from http://qje.oxfordjournals.org/ at MIT Libraries on September 24, 2012

IV.C. Local Average Treatment Effect of Medicaid (LATE)

3 yihj 0 1 INSURANCEih Xih 2 Vih 3 vihj ,

equation (3) by two-stage least squares (2SLS), using the follow-

3 yihj 0 1 INSURANCEih Xih 2 Vih 3 vihj ,

4 INSURANCEih 0 1 LOTTERYh Xih 2 Vih 3 ihj ,