ESSAYS IN PUBLIC AND HEALTH ECONOMICS
By
Ehren Schuttringer

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
Economics – Doctor of Philosophy
2014

ABSTRACT
ESSAYS IN PUBLIC AND HEALTH ECONOMICS
By
Ehren Schuttringer
This dissertation consists of three empirical studies in public and health economics. In
the first chapter I use inpatient level hospital data to examine the effect of state tort reforms on
physician behavior. For the second chapter, I evaluate the impact of a large expansion in public
health insurance for children on the labor supply of single mothers. In the third chapter I
describe trends in health insurance coverage during the Great Recession, and investigate the
degree to which business cycle variation accounts for the large decline in insurance coverage
among adults during the recession.
The first dissertation chapter explores the relationship between medical malpractice laws
and the behavior of medical providers. Physician incentives in the current medical malpractice
system may encourage socially wasteful behavior if doctors, out of liability concerns, provide
medical services to patients of little benefit relative to the cost. Tort reform is often touted as a
way to reduce liability induced provision of costly and unnecessary medical services. Empirical
studies that assess the effect of tort reform on physician behavior offer mixed evidence of a
behavioral response. Using a source of inpatient data not previously used in this literature, I
investigate potential explanations for this lack of consensus. Results indicate some evidence that
tort reforms influence physician behavior in a sample of heart attack patients, but not in a sample
of stroke patients. These results are not robust to different specifications, however. I conclude
that there is little evidence of a relationship between tort reform and hospital treatment for AMI
or stroke.

In the second chapter, I evaluate the implementation of the State Children’s Health
Insurance Program (SCHIP) and its impact on the labor supply of single mothers. SCHIP,
established in 1997, is an important source of health care access for children in near poor
families. As with other means tested government programs, a worry is that program eligibility
rules distort parental labor supply decisions. Because states have a large amount of flexibility in
the design and administration of their SCHIP programs, there is considerable heterogeneity in
state eligibility rules. Consequently, reduced form methods that associate SCHIP eligibility
policies with labor supply must average outcomes over a diverse set of households with varying
incentives for insurance coverage and employment. With data from the March Current
Population Survey (CPS), I evaluate the program’s effect on the labor supply using an
instrumental variables estimation strategy that relates child insurance coverage with program
eligibility rules. Results suggest that SCHIP led to an increase in public coverage and private
insurance crowd-out, with no effects on maternal work behavior.
For the third chapter, I investigate adult health insurance coverage during the Great
Recession (2007-2009). The Great Recession is associated with large reductions in employment
and health insurance coverage. However, it is not clear that the decline in health insurance
coverage can be completely attributed to the recession. Using data from the CPS, I relate health
insurance coverage for working age adults with state level measures of employment. Estimates
from this model are used to generate out of sample predictions of health insurance coverage
during the Great Recession. Results show that the proportion of adults without insurance
coverage during the recession is greater than the level predicted by the regression model. These
results are sensitive to different specifications of the model, however.

To my parents, for their constant support and encouragement.
To Kristen, whom I love.

iv

ACKNOWLEDGEMENTS

I sincerely thank John Goddeeris and Steven Haider for their invaluable feedback and
guidance throughout my graduate career. Their comments were instrumental in my development
as an economist. I would also like to thank Stacy Dickert-Conlin and Zhehui Luo for their
important assistance and for serving on my dissertation committee. Additionally, I am grateful
for the support and friendship from many of the current and former graduate students in the
Economics Department at Michigan State University. Their diverse personalities and
perspectives inspired me to become a better student and researcher.

v

TABLE OF CONTENTS

LIST OF TABLES ....................................................................................................................... viii
LIST OF FIGURES ....................................................................................................................... xi
CHAPTER 1
TORT REFORM, MEDICAL EXPENDITURES, AND PATIENT OUTCOMES .......................1
I. Introduction ......................................................................................................................1
II. Background Information .................................................................................................3
II.A. Types of tort reforms .......................................................................................3
II.B. The medical malpractice system and predicted effects of tort reform ............4
II.C. Relevant literature............................................................................................7
III. Empirical Methodology .................................................................................................8
IV. Data ..............................................................................................................................10
V. Results ...........................................................................................................................13
VI. Conclusion ...................................................................................................................20
APPENDIX ....................................................................................................................................22
WORKS CITED ............................................................................................................................35
CHAPTER 2
THE STATE CHILDREN’S HEALTH INSURANCE PROGRAM AND MATERNAL LABOR
SUPPLY INCENTIVES ................................................................................................................38
I. Introduction ....................................................................................................................38
II. Background Information ...............................................................................................39
II.A. The State Children’s Health Insurance Program ...........................................39
II.B. Relevant literature..........................................................................................42
III. Methods........................................................................................................................44
III.A. Predictions from a static labor supply model ...............................................44
III.B. Data ..............................................................................................................47
III.C. Empirical methodology ................................................................................52
IV. SCHIP Eligibility and Children’s Public Insurance Coverage ....................................55
V. The Effect of SCHIP Coverage on Maternal Labor Supply .........................................60
VI. Conclusion ...................................................................................................................62
APPENDICES ...............................................................................................................................65
Appendix A: Analysis with Additional Medicaid and SCHIP Policy Variables ...............66
Appendix B: Additional Tables .........................................................................................74
WORKS CITED ............................................................................................................................82
CHAPTER 3
HEALTH INSURANCE COVERAGE DURING THE GREAT RECESSION ..........................87
I. Introduction ....................................................................................................................87
vi

II. Methods .........................................................................................................................88
II.A. Data ...............................................................................................................88
II.B. Empirical methodology .................................................................................92
III. Results ..............................................................................................................................
IV. Conclusion .................................................................................................................100
APPENDIX ..................................................................................................................................104
WORKS CITED ..........................................................................................................................111

vii

93

LIST OF TABLES

Table 1.1: Means of key variables, specific reforms .....................................................................12
Table 1.2: Changes in state tort laws .............................................................................................13
Table 1.3: AMI results ...................................................................................................................14
Table 1.4: Stroke results ................................................................................................................15
Table 1.5: AMI results, no state time trends ..................................................................................16
Table 1.6: Stroke results, no state time trends ...............................................................................17
Table 1.7: AMI lag and lead results, direct and indirect reform specification ..............................18
Table 1.8: AMI lags and leads results, individual reform specification ........................................19
Table 1.9: Stroke lags and leads results, direct and indirect reform specification.........................23
Table 1.10: Stroke lags and leads results, individual reform specification ...................................24
Table 1.12: Stroke comparison results ...........................................................................................26
Table 1.13: AMI results, direct/indirect specification ...................................................................27
Table 1.14: AMI results, individual reform specification..............................................................28
Table 1.15: Stroke results, direct/indirect specification.................................................................29
Table 1.16: Stroke results, individual reform specification ...........................................................30
Table 1.17: AMI results, all cap definitions ..................................................................................31
Table 1.18: Stroke results, all cap definitions ................................................................................32
Table 1.19: AMI, patient compensation fund and contingency fee ...............................................33
Table 1.20: Stroke, patient compensation fund and contingency fee ............................................34
Table 2.1: Means of key variables .................................................................................................50
Table 2.2: Health insurance coverage, equation (2) ......................................................................56

viii

Table 2.3: Group and non-group private coverage, equation (2) ...................................................59
Table 2.4: Labor supply, equation (3) ............................................................................................61
Table 2.5: Labor supply, equation (1) ............................................................................................61
Table 2.6: Means of additional policy variables ............................................................................67
Table 2.7: First stage (maternal sample), equation (2) ..................................................................68
Table 2.8: Labor supply, equation (3) ............................................................................................69
Table 2.9: Labor supply, equation (3) ............................................................................................70
Table 2.10: Labor supply, equation (3) ..........................................................................................71
Table 2.11: Labor supply, equation (3) ..........................................................................................72
Table 2.12: Labor supply, equation (1) ..........................................................................................73
Table 2.13: Health insurance coverage (maternal sample), equation (2), 1996-2002 sample years
only .............................................................................................................................75
Table 2.14: Group and non-group private coverage (maternal sample), 1996-2002 sample years
only .............................................................................................................................75
Table 2.15: Labor supply, equation (3), 1996-2002 sample years only ........................................76
Table 2.16: Labor supply 2SLS only, equation (1), 1996-2002 sample years only ......................76
Table 2.17: Health insurance coverage (maternal sample), equation (2), no state-year
interactions .................................................................................................................77
Table 2.18: Group and non-group private coverage (maternal sample), no state-year
interactions .................................................................................................................77
Table 2.19: Labor supply, equation (3), no state-year interactions ...............................................78
Table 2.20: Labor supply 2SLS only, equation (1), no state-year interactions .............................78
Table 2.21: Health insurance coverage (maternal sample), equation (2), results with overlap
category ......................................................................................................................79
Table 2.22: Health insurance coverage (maternal sample), equation (2), no weights ...................79
Table 2.23: Group and non-group private coverage (maternal sample), no weights .....................80
ix

Table 2.24: Labor supply, equation (3), no weights ......................................................................80
Table 2.25: Labor supply 2SLS only, equation (1), no weights ....................................................81
Table 3.1: Means of key variables (1987-2012) ............................................................................90
Table 3.2: Insurance coverage trends during the Great Recession, 19-64 adults ..........................91
Table 3.3: Great Recession and insurance coverage (BLS LAUS employment variables) .........101
Table 3.4: Great Recession and insurance coverage (1988-2012; 19-64 year old employment
variables) ....................................................................................................................101
Table 3.5: Great Recession, proportion of adults unemployed at least 1 week in year ...............105
Table 3.6: Trends in main employment variables over past decade ............................................105
Table 3.7: Great Recession, parental insurance coverage trends .................................................106
Table 3.8: 19-64 Adults results from equation (1) used for prediction, uninsured (1987-2006) .107
Table 3.9: Great Recession and insurance coverage (full set of results from Table 3.3) ............108

x

LIST OF FIGURES

Figure 2.1: Effect of SCHIP on labor supply (intensive and extensive margin) ...........................45
Figure 2.2: Child Income Eligibility Thresholds for Medicaid and SCHIP ..................................51
Figure 2.3: Child Health Insurance Coverage Trends, by Income Status ......................................52
Figure 3.1: Proportion uninsured, unemployment rate (BLS LAUS) ............................................94
Figure 3.2: Proportion uninsured, EP ratio (BLS LAUS) ..............................................................94
Figure 3.3: Private coverage, unemployment rate (BLS LAUS) ...................................................96
Figure 3.4: Private coverage, EP ratio (BLS LAUS) .....................................................................96
Figure 3.5: 1988-2012 Proportion uninsured, unemployment rate (19-64 year old employment
variable) ......................................................................................................................97
Figure 3.6: 1988-2012 Proportion uninsured, EP ratio (19-64 year old employment variable) ....97
Figure 3.7: Proportion uninsured trend comparison, unemployment rate (BLS LAUS) ...............99
Figure 3.8: Proportion uninsured trend comparison, EP ratio (BLS LAUS) .................................99
Figure 3.9: Public coverage, unemployment rate (BLS LAUS) ..................................................110
Figure 3.10: Public coverage, EP ratio (BLS LAUS) ..................................................................110

xi

CHAPTER 1
TORT REFORM, MEDICAL EXPENDITURES, AND PATIENT OUTCOMES
I. Introduction
In the United States, rising medical costs and large government budget deficits have
renewed focus on reducing expenses associated with medical care provision. While there are
many methods that may achieve this goal, it is often argued that the medical malpractice system
is a large source of the cost growth and should be reformed. Expenses directly related to the
medical malpractice industry (such as malpractice payments, liability insurance premiums, and
legal expenses) are relatively small: Mello et al. (2010) report that existing research shows these
direct costs to be less than 2% of total health care spending. Nevertheless, it is possible that
indirect expenses associated with medical malpractice may potentially be much larger.1 Indirect
expenses of malpractice litigation include defensive medicine, a phenomenon that occurs when
the threat of litigation causes doctors to prescribe medical services of relatively little benefit to
patients (Kessler and McClellan 1996).
States adopt tort reforms in an effort to control expenses related to medical malpractice.
Tort reform may lessen an individual’s incentive to file a malpractice claim, primarily by
reducing the amount of damages they can pursue. In response, physician perceptions of
malpractice risk may decline, lowering the provision of malpractice induced medical services.
This is a frequent justification for tort reform. Former United States Senator Judd Gregg argues,
“We cannot possibly reduce the cost of health care in this country without reducing defensive
medicine” (Langel, 2009). However, it is important to note that the liability system is intended
to induce physicians to take appropriate precaution when treating patients. Studdert et al. (2004)
state that “lawsuits deter physicians [from unsafe practice] by reminding those who wish to avoid
1

Generally, indirect costs are the result of the malpractice system’s effect on physician behavior.

1

the emotional and financial costs of litigation that they must take care” (p.283). Limits to
liability may reduce malpractice induced behavior and health care costs, but may also harm
patient outcomes if such services have some positive benefit. The critical question is whether
costs from such behavior exceed the benefits.
In order to address this question, numerous studies have been conducted to evaluate the
effect of tort reform on medical expenses, procedure choice, and patient outcomes. This
literature has not reached a consensus regarding these relationships. There are two primary
explanations for the lack of consensus. Researchers have analyzed different populations of
patients. This inhibits comparison across studies if the effect of malpractice pressure on medical
treatment varies across patient groups. In addition, the specification of the tort reform variable
varies across studies. Some authors (Kessler and McClellan 1996, 2002; Sloan and Shadle 2009)
group reforms into direct and indirect categories. Others (Currie and MacLeod, 2008) consider
separately the effect of each individual reform. Critically, both tort variable specification and the
population of interest vary across studies. This makes it difficult to isolate either factor as an
explanation of the literature’s mixed results.
This paper seeks to add to the literature in two important ways. First, I evaluate the effect
of state tort reform laws on patient outcomes and measures of medical resource use.2 Two
samples, patients hospitalized with a primary diagnosis of either heart attack (acute myocardial
infarction, or AMI) or stroke, are derived from a nationally representative data set of hospital
inpatient stays unique to the literature. As opposed to Kessler and McClellan (1996 and 2002)
and Sloan and Shadle (2009), who focus only on Medicare patients with an initial
hospitalization, this is an all-payer data set that is representative of all hospital inpatient
2

I use two measures of medical resource use: total charges and length of stay. There is good reason to think that
charges are not well correlated with actual medical costs (Reinhardt, 2006). Consequently, I use length of stay as an
additional measure of resource use and medical service intensity.

2

admissions in the country. Secondly, I seek to determine if either the specification of the tort
variable or the patient population is important in accounting for the literature’s results. I test for
the existence of malpractice induced behavior using two specifications of the tort variable across
similar disease groups – one similar to the indirect and direct classifications used by Kessler and
McClellan (1996 and 2002) and Sloan and Shadle (2009), and another similar to the approach of
Currie and MacLeod (2008) that considers individual tort reforms. The paper is thus an attempt
to reconcile the different methods used to estimate the effect of tort reform on medical practice.
Results show that several reforms are associated with at least one measure of resource use
in the AMI sample. Similar associations are found across both tort reform specifications,
suggesting that estimates of tort reform are not sensitive to the specification of that variable.
However, these results are not consistent across different measures of medical resource use and
are not robust to standard sensitivity checks. In addition, I find almost no evidence of a
statistically significant relationship between reform adoption and the treatment of stroke.
In section II, I discuss some of the important institutional details associated with tort
reform and malpractice litigation, relevant academic literature, and predictions of the effect of
tort reform on medical practice. The empirical strategy is explained in Section III, and the data
are discussed in Section IV. Results are presented in Section V. Section VI concludes.

II. Background Information
II.A. Types of tort reforms
States adopted various tort reforms over the past several decades. Kessler and McClellan
(1996) identify eight common reforms, of which I focus on the most popular between 1988 and
2008: caps on award size, collateral source rule reform, joint and several liability reform, and

3

mandatory periodic payments. Caps on award size are any tort reform that limits the size of a
malpractice award, whether punitive or compensatory. These include limits on non-economic
damages, punitive damages, or total damages. Collateral source rule reform (CSR) reduces
payments to a plaintiff by the dollar amount of payments from other sources, such as insurance.
Reforms to joint and several liability (JSL) rules limit the ability of a plaintiff to seek damages
from multiple defendants. Finally, mandatory periodic payments (PPA) require malpractice
awards to be paid out in fixed amounts over time.
Direct tort reforms, as defined by Kessler and McClellan (1996), are reforms that directly
reduce payment size and claim frequency. These include damage caps and CSR reforms.
Indirect tort reforms are reforms which do not directly affect payment size and claim frequency.
This category includes PPA and JSL reforms.

II.B. The medical malpractice system and predicted effects of tort reform
The medical malpractice system serves two purposes: provide compensation to patients
who suffer losses as the result of negligent medical care and to discourage doctors from behaving
carelessly (Congressional Budget Office [CBO], 2006). Medical malpractice cases are judged
using the negligence rule, a common law legal standard. The rule holds that physicians are liable
for malpractice if the plaintiff demonstrates they suffered an injury, the injury was caused by the
physician, and the physician’s level of care deviated from what is considered by the court to be
due or customary care. Due care is associated with the practice of physicians in good standing;
injuries and other adverse medical outcomes that are the result of due care are not considered
negligent. It is often not clear exactly what customary practice should be in any given case. This
standard must be established in every case, contributing to variance in legal outcomes (Danzon,

4

2000). Assuming physicians are risk averse, the ambiguity of the negligence rule may create
incentives for doctors to behave defensively, since they cannot be sure of exactly what
constitutes due care and whether a given level of care will protect them from a lawsuit.
It is not obvious that malpractice litigation should affect physician behavior, however.
Physicians are insured from malpractice award and settlement payments, and their insurance
plans generally do not contain experience rated premiums or cost-sharing. Instead, premiums
vary by physician specialty and geographic location (Danzon, 2000). Consequently, physicians
are well insulated from any of the pecuniary costs of a successful tort claim. Physicians do face
significant lawsuit associated costs, however. These include mental, time, and reputational costs.
To illustrate, Danzon (2000) cites a 1993 survey of physicians that finds physicians with a claims
history are significantly more likely to discuss medical risks with patients and order additional
procedures. These physicians indicate they incurred significant financial and non-financial costs
of being sued, despite having insurance.
Litigation related costs imply that the malpractice system should affect physician
behavior. Along with the uncertain nature of the negligence rule, these costs provide physicians
with incentive to prescribe an inefficient amount of medical services in order to limit their
exposure to litigation. Argues Danzon (2000), “Because liability is all-or-nothing, by incurring a
small additional cost [the physician’s] probability of a large penalty may be significantly
reduced” (p.1348). Tort reform may affect this calculation in several ways. Before a liability
reducing reform, a physician’s private incentives are such that the marginal benefit of additional
treatment is high relative to the costs. The introduction of a tort reform may reduce physician
liability, decreasing the benefit of malpractice induced behavior to the physician.3

3

A key assumption throughout the literature is that changes in tort law affect physician perceptions about the risk of
malpractice litigation, and therefore affect practice behavior (Danzon, 2000).

5

Since physicians are fully insured against the size of an award, the primary way a tort
reform may reduce physician liability is by lowering claim frequency. Direct reforms, including
award caps, limit the maximum size of the malpractice award and reduce a potential plaintiff’s
incentive to file a malpractice claim. The relationship between indirect reforms and liability is
less clear. JSL reform reduces the ability of a plaintiff to hold more than one defendant liable for
damages, potentially lowering the expected award size. In states that also enacted award caps,
however, the effect of JSL reform on claim frequency may be limited. PPA reform does not
directly alter the award size, and will only reduce claim frequency if potential plaintiffs prefer
lump sum payments.
A “first-stage” literature, meant to evaluate the link between tort reform and malpractice
liability, consistently estimates a negative relationship between direct reforms and measures of
malpractice liability. Dependent variables include the size of a malpractice claim payment, the
likelihood of a malpractice claim being filed, and premium payments for liability insurance. In a
survey of this literature, Mello (2006) finds that the strongest studies show a negative
relationship between damage caps and malpractice claim payout size, as well as liability
insurance premiums. She finds only mixed evidence for CSR reform, with equal number of
studies indicating either no effect or a negative relationship with claim frequency and payout.
Additional studies, including Kessler and McClellan (2002) and Avraham (2007), offer evidence
that direct reforms reduce claim frequency and payment size.
Indirect reforms, however, are not consistently associated with measures of physician
liability. Mello (2006) finds little evidence in the literature that JSL or PPA reforms are related
to claim payouts, claim frequency, or liability premiums. However, Avarham (2007) estimates
that both JSL and PPA reforms have a negative effect on claim payment size. Kessler and

6

McClellan (2002) find that indirect reforms, which include JSL and PPA, reduce claim
frequency but have a positive effect on other forms of malpractice pressure.

II.C. Relevant literature
Authors of several previous studies use variation in tort reform adoption to test for a
relationship between malpractice pressure and physician behavior. Kessler and McClellan
(1996, 2002) estimate the effect of state level tort reforms on medical expenditures and outcomes
one year after initial diagnosis for a sample of Medicare patients with AMI or ischemic heart
disease. Their 1996 paper covers a period between 1984 and 1990, while the later paper extends
this time frame to 1994. Kessler and McClellan (1996) argue that defensive medicine is present
if tort reforms meant to reduce liability risk lower medical expenditures without harming patient
outcomes. The authors generally find that direct tort reforms are associated with a decrease in
expenditures, and indirect reforms have a small positive relationship with expenditures.
Coefficients for medical outcomes are rarely significant and when significant are very small.
Results from the CBO (2006) mirror these results. They estimate the effect of state level tort
reform adoption on aggregate measures of Medicare expenditures using two decades of data,
starting in 1980. Authors find that individual direct reforms and the direct reform variable
decrease expenditures. JSL reforms increase expenditures, though the authors find no evidence
of a relationship between the indirect reform variable and expenditures.
Currie and MacLeod (2008) also find an effect between tort reform and medical practice.
Employing data from the National Vital Statistics birth dataset for 1989 to 2001, the authors use
variation in state tort reform adoption to estimate the effect of malpractice pressure on usage
rates of different birth procedures, as well as health outcomes for mother and baby. Using a

7

theoretical model where the effect of JSL reform on physician liability is the opposite of award
caps, the authors estimate the effect of each individual reform. They find that damage caps
increase the incidence of a Caesarean section by 5%, whereas JSL reform decreases the
probability of a Caesarean section 7%. In addition, JSL reforms reduce the likelihood of
preventable labor complications by 13%, while caps on noneconomic damages increase these
complications by 6%. These results show that reductions in physician liability due to tort reform
may exacerbate socially wasteful behavior while worsening patient outcomes.
Finally, one study does not find evidence of a relationship between tort reform and
physician behavior, despite using a methodology similar to previous studies. Sloan and Shadle
(2009) extend Kessler and McClellan’s analysis by examining all expenditures (from inpatient as
well as outpatient services) and mortality outcomes one year after any hospital diagnosis, as well
as separately for diagnoses of AMI, stroke, breast cancer, and diabetes. Using 1985 to 2000 data
from the National Long Term Care Survey matched with Medicare information, they fail to
replicate any of the results found by Kessler and McClellan (1996). One explanation for these
results is that outpatient expenditures are not sensitive to changes in tort law. Given the results
of Kessler and McClellan (1996) on inpatient expenditures for AMI treatment, this suggests that
tort reform has a heterogeneous effect across inpatient and outpatient services. However, the
authors do not investigate whether tort reform has a differential effect across these categories.

III. Empirical Methodology
To examine the relationship between tort reform and physician behavior, the following
empirical specification is used (where i denotes hospital admission, s state, and t year):

Yist = αs + δt + λs∙t + γTortst + Xistβ + εist
8

This specification is derived from natural experiment methods found in Kessler and McClellan
(1996) and Currie and MacLeod (2008).4 Here Yist is the outcome measure, indicating the natural
log of real total charges, length of inpatient stay, or mortality. Total charges and length of stay
model resource use per inpatient stay in a hospital, and mortality is an indicator that equals one if
the patient died while hospitalized.5 State and year indicator variables are indicated with αs and
δt, respectively. The specification also includes state specific time trends, indicated with λs∙t. Xist
is a vector of patient controls. This vector includes indicators for whether the patient is an
elective admission, female, white, expected primary payer, age and age squared, and an indicator
for whether the hospital associated with the inpatient record is a teaching hospital. Finally,
standard errors are clustered by state to account for within state correlation.
Tortst is an indicator for the presence of a reform in state s during year t. For tort reforms
with an effective date on or after July 1st of year t, the law change is coded as Tortst = 1 in t+1.
This variable is modeled in two ways: as a vector including indicator variables for each tort
reform or as a vector including an indicator for direct and indirect reforms. This first
specification is similar to the empirical strategy employed by Currie and MacLeod (2008).6 The
second specification is similar to the method first outlined in Kessler and McClellan (1996).
There are two main threats to the validity of this empirical approach. It is possible that
tort reform adoption influences the likelihood of admission. If so, this will affect the mix of
patients, so that any estimate of the effect of tort reform would reflect both the law change and
the change in admission behavior. One way to address this issue is to select a subpopulation of

4

There are some differences. Kessler and McClellan (1996) include a vector of state legal and political
characteristics but do not estimate a time trend, and Currie and MacLeod (2008) use county fixed effects.
5
Real total charges are measured in 1988 dollars. Length of stay is measured in days.
6
I combine all damage caps – punitive, non-economic, and total – into one variable in this specification. An
alternate specification is run with separate variables for each type of cap. Results are similar, though the result for
AMI JSL on stay length is of greater magnitude and significance. These are found in Tables 1.17 and 1.18.

9

patients with limited discretion over admission. For this analysis, patients with a hospitalization
for one of two medical emergencies are selected. These are patients with a primary diagnosis for
AMI or stroke.
Unobserved trends correlated with reform adoption also threaten the above empirical
strategy. Identification requires that trends in the dependent variable for both adopting states and
non-adopting states are similar, in the absence of a treatment. To investigate this, leads and lags
of the tort reform variables are estimated in an alternate specification to examine whether there
are unobserved trends correlated with reform adoption. Estimated in this specification are one
and two year leads, a year of adoption dummy, one and two year lags, and a lag variable for three
years or longer after the policy change. Significant leads may indicate a differential trend for
adopting states that is correlated with reform adoption, calling into question the causality of
results in the main analysis. If, for instance, leads for the effect of direct reforms on charges are
negative, then this may indicate that expenditure trends in direct reform adopting states began to
decline before the law change. Any estimated effect would be biased, since it includes both the
effect of the tort reform and the pre-adoption trend. Lags are included to investigate how quickly
physicians respond to tort reform.

IV. Data
I use data from the 1988-2008 Healthcare Cost and Utilization Project (HCUP)
Nationwide Inpatient Sample (NIS). The HCUP NIS is an all payer database of hospital
inpatient stays from nearly 1,000 U.S. hospitals. This database uses a stratified sampling
strategy to approximate a 20% sample of all U.S. community hospitals. Forty-two states are
included in the 2008 release of the NIS, while only 8 states are included in the original 1988

10

release.7 The NIS data are at the discharge level. The data include information on diagnoses,
procedures, charges, outcomes, and other patient and hospital information related to the inpatient
admission. Primary diagnosis information is used to create the two patient samples used in the
analysis. A patient discharge record is included in the sample by selecting the relevant ICD-9CM codes for either AMI or stroke.
Table 1.1 includes a summary of means for the key variables used in the analysis,
organized by disease group and tort reform type. The first column indicates means for all states.
For AMI patients, means for length of stay and mortality are 6.15 and 0.088, respectively. These
means tend to be consistent across tort reform status. The mean total charge for an inpatient stay
in the first column is $20,803.45. This is consistent across tort reform status as well, with the
exception of CSR and PPA reforms which are lower and higher, respectively, than the overall
average. These trends are generally true of stroke patients as well. Across disease samples, AMI
treatment incurs greater charges but lower stay length than stroke. The incidence of mortality is
greater in the stroke sample. Stroke patients are more likely to be female, and are several years
older, than patients in the AMI sample.
Information on state tort laws is from data compiled by Ronen Avraham (2010).
Included in this analysis are award caps, PPA mandates, reforms to JSL, and CSR reform. I
include two other types of tort reforms, contingency fee caps and patient compensation fund
requirements, as indirect reforms but not in the individual reform specification.8 Only two states
enacted these reforms, and these states contribute a small amount of observations. Table 1.2
7

The data is unbalanced in states, potentially confounding estimates of tort reform if states are not randomly
missing. To account for this, the regression model used in this analysis includes state fixed effects. Estimates using
a 1988-2008 balanced panel, which includes only 8 states, are similar to main results from this paper but have larger
standard errors.
8
In an alternative specification, I estimate the effect of both contingency fee (adopted by Nevada) and patient
compensation funds (adopted by West Virginia). Controlling for these laws does not affect the results. These
additional estimates are presented in Tables 1.19 and 1.20.

11

Variable
Total charges
Length of Stay
Mortality
Age
Female
White
Medicare
Medicaid
Private Insurance
Elective Admission
Teaching Hospital
Observations
Total charges
Length of Stay
Mortality
Age
Female
White
Medicare
Medicaid
Private Insurance
Elective Admission
Teaching Hospital
Observations

Table 1.1: Means of key variables, specific reforms
All states No tort reform Awards cap
JSL
CSR
PPA
Acute Myocardial Infarction (Heart Attack)
20,803.450
20,560.580
20,803.610 19,153.640 17,722.820 23,478.880
6.150
5.959
6.299
6.368
6.650
5.861
0.088
0.083
0.092
0.092
0.094
0.079
67.695
67.638
67.786
67.724
68.029
66.685
0.398
0.403
0.397
0.412
0.407
0.397
0.580
0.714
0.498
0.388
0.590
0.412
0.581
0.581
0.583
0.585
0.597
0.543
0.042
0.052
0.033
0.038
0.034
0.044
0.300
0.301
0.299
0.306
0.296
0.314
0.075
0.079
0.073
0.059
0.052
0.086
0.420
0.479
0.381
0.517
0.547
0.433
2,551,355
991,060
1,500,431
557,094
303,614
450,630
Stroke
15,821.300
16,376.290
15,440.140 14,317.100 13,610.620 17,518.840
7.869
7.868
7.925
8.043
8.321
7.419
0.116
0.120
0.115
0.114
0.119
0.107
71.014
70.941
71.113
71.171
71.988
70.068
0.532
0.540
0.527
0.535
0.531
0.537
0.526
0.660
0.444
0.328
0.533
0.363
0.680
0.678
0.683
0.682
0.696
0.635
0.053
0.068
0.042
0.047
0.035
0.051
0.200
0.200
0.199
0.198
0.182
0.210
0.085
0.079
0.089
0.078
0.083
0.083
0.414
0.474
0.373
0.498
0.525
0.442
1,791,518
682,579
1,072,249
397,541
200,662
305,773

indicates year of both tort reform adoption and reforms turning off in relevant states. Several of
the adopting states in the sample enacted multiple reforms at the same time. For instance,
Wisconsin adopted damage caps, CSR, and JSL reform in the same year. This complicates the
analysis, since it is harder to identify the effects of any one tort reform if states adopt several
reforms simultaneously.

12

Table 1.2: Changes in state tort laws
Any cap
JSL
CSR

Year
1988
1989
CO, WA*
1990
WI*
1991
1992
1993
1994
1995
IL, WI
IL, WI
1996
NJ
1997
PA, IL*
IL*
1998
1999
OR*
2000
2001
2002
PA
2003
NV, OH
OH
2004
FL, TX
2005
GA
2006
IL, SC, MO
SC
2007
2008
Law change only included if state was in HCUP that year
*Indicates that law turned off, usually due to court ruling

PPA
CO

AZ*
WI

PA
WV

PA
OH
TX
GA

V. Results
Table 1.3 presents main results for the AMI sample. Estimates from this sample are
consistent across specifications. Columns 1 and 2 indicate results on total hospital charges for
individual and grouped reform categories, respectively. Award caps and CSR reform, as well as
the direct reform category, do not have a statistically significant effect on hospital charges. In
addition, results from columns 5 and 6 show no evidence that these reforms influence patient
mortality. Indirect reforms, including JSL and PPA reform, appear to influence hospital charges.
Coefficients for these reforms are positive and statistically significant for both PPA reform and
the overall indirect reform category. These point estimates imply that PPA reform increases
13

Variable
Award cap
CSR reform

Table 1.3: AMI results
Natural log charges
LOS
1
-0.013
(0.025)
0.020
(0.040)

Direct reforms
JSL reform
Periodic payment
Indirect reforms

2

3
-0.262**
(0.098)
-0.153
(0.145)

-0.021
(0.024)
0.026
(0.034)
0.047**
(0.022)

Mortality
4

5
0.000
(0.001)
-0.002
(0.002)

-0.251**
(0.105)
0.269*
(0.151)
0.128
(0.107)

0.052**
(0.023)

0.001
(0.001)
-0.002
(0.002)
-0.001
(0.002)

0.230**
(0.113)

Observations

6

-0.002
(0.002)

2,551,355

Cluster robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Not indicated are the following control variables: elective admission, female, white, Medicare, Medicaid,
private insurance, age, age squared, teaching hospital, state and year dummies, state linear time trends

charges 4.8% and the indirect category raises total charges 5.3%.9 As with direct reforms,
neither PPA reform nor the indirect category has any effect on patient mortality. Overall, AMI
results for direct reforms diverge from those of Kessler and McClellan (1996), who find a
negative relationship between the direct reform category and expenditures. Indirect reform
results for AMI are consistent with Kessler and McClellan (1996), however, implying that
indirect reforms have a positive relationship with resource use but no relationship with outcomes.
Columns 3 and 4 of Table 1.3 present results on hospital stay length, which is meant as
an additional measure of resource use. As with total hospital charges in columns 1 and 2, the
length of stay results are consistent across both tort variable specifications. However, the types

9

I exponentiate the point estimates here. For instance, PPA reform: 100*[e^(0.047)-1] = 4.81.

14

Variable

Table 1.4: Stroke results
Natural log charges
LOS
1

Award cap
CSR reform

0.014
(0.042)
0.134**
(0.057)

Direct reforms
JSL reform
Periodic payment
Indirect reforms

2

Mortality

3

4

-0.518*
(0.264)
0.092
(0.531)
0.002
(0.044)

-0.030
(0.052)
-0.025
(0.045)

5
-0.005
(0.003)
0.008
(0.013)

-0.376
(0.261)
0.472
(0.444)
-0.660
(0.395)

0.012
(0.052)

-0.004**
(0.002)
0.008
(0.005)
-0.014*
(0.007)

-0.035
(0.334)

Observations

6

0.003
(0.005)

1,791,518

Cluster robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Not indicated are the following control variables: elective admission, female, white, Medicare, Medicaid,
private insurance, age, age squared, teaching hospital, state and year dummies, state linear time trends

of reforms estimated to influence stay length are different than for total charges. Award caps
have a negative effect on stay length with a coefficient of -0.262, or a 4.3% reduction relative to
mean stay length. Direct reforms have a similar relationship with stay length, with the
coefficient implying a 4.1% reduction relative to the mean. JSL reform, which does not have a
significant effect on charges in column 1, has a positive effect on stay length in column 3. The
opposite is true of PPA reform, which is not associated with length of stay. The only variable
that is consistent across either measure of resource use is the overall indirect reform category in
column 4, estimated to increase stay length 3.7%. These results suggest that tort reform
estimates are sensitive to the choice of medical resource use variable.

15

Results are also sensitive to the use of state specific time trends. Table 1.5 presents AMI
estimates without these trends. None of the tort reform variables have a statistically significant
effect on either measure of resource use. The only significant relationship between tort reform

Variable
Award cap
CSR reform
Direct reforms
JSL reform
Periodic payment
Indirect reforms

Table 1.5: AMI results, no state time trends
Natural log charges
LOS
1
2
3
4
0.099
-0.154
(0.081)
(0.272)
-0.038
-0.154
(0.049)
(0.195)
0.106
-0.145
(0.081)
(0.274)
-0.040
0.150
(0.064)
(0.257)
-0.024
0.078
(0.053)
(0.176)
-0.068
0.086
(0.061)
(0.232)

Mortality
5
-0.003*
(0.002)
0.001
(0.002)

6

-0.003*
(0.002)
0.001
(0.002)
0.000
(0.001)
0.002
(0.002)

Observations
2,551,355
Cluster robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Not indicated are the following control variables: elective admission, female, white, Medicare, Medicaid,
private insurance, age, age squared, teaching hospital, state and year dummies, state linear time trends

and the dependent variables is with patient mortality. Award caps and the overall direct reform
category reduce the likelihood of mortality 0.3% points. The results from specifications with and
without state specific time trends imply that states have differential underlying trends in the
dependent variables. Finally, lead and lag results in Tables 1.7 and 1.8 present additional
evidence that the main results from Table 1.3 are sensitive to unobservable trends. Both tables
show that direct and indirect reform categories, as well as award caps, have significant and
positive 1 year leads for total charges. This implies that hospital charges in reform adopting
states were increasing relative to the state specific trend prior to the law change.
16

Results for the stroke sample show almost no relationship between tort reform and
medical resource, regardless of whether state time trends are included in the model. Table 1.4
and 1.6 present estimates for the stroke sample. The only significant effects on resource use are

Variable
Award cap
CSR reform
Direct reforms
JSL reform
Periodic payment
Indirect reforms

Table 1.6: Stroke results, no state time trends
Natural log charges
LOS
1
2
3
4
0.114
-0.329
(0.095)
(0.743)
0.072
0.694
(0.068)
(0.570)
0.111
-0.372
(0.095)
(0.763)
-0.093
-0.096
(0.075)
(0.687)
-0.094
-0.548
(0.067)
(0.576)
-0.087
-0.079
(0.076)
(0.687)

Mortality
5
-0.007*
(0.004)
0.015***
(0.004)

6

-0.009**
(0.003)
0.004
(0.004)
-0.004
(0.004)
0.008*
(0.004)

Observations
1,791,518
Cluster robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Not indicated are the following control variables: elective admission, female, white, Medicare, Medicaid,
private insurance, age, age squared, teaching hospital, state and year dummies, state linear time trends

associated with individual reforms. In Table 1.4, CSR reform and award caps affect hospital
charges and stay length, respectively. CSR reform is associated with a large positive expenditure
increase of 14.3%. For length of stay, the coefficient on award caps is -0.518, or a 6.6%
reduction in hospital stay. Despite no evidence of a relationship with either measure of resource
use, the direct reform category and PPA reform lower the likelihood of mortality during the
inpatient stay. Tort reform is also associated with hospital mortality in specifications without
state time trends. In Table 1.6, both the direct and indirect reform categories, as well as CSR and
award caps, have a statistically significant relationship with mortality. This is despite no
17

evidence of a relationship between resource use and any of the reform variables in Table 1.6.
Estimates from an additional specification, found in Table 1.12, show that results are similar
even when different definitions of the stroke sample are considered.10
Table 1.7: AMI lag and lead results, direct and indirect reform specification
Variables
Natural log charges
LOS
Mortality
Direct 2 year lead
-0.016
-0.068
-0.002
(0.029)
(0.144)
(0.002)
Direct 1 year lead
0.075***
0.121
0.002
(0.025)
(0.140)
(0.002)
Direct year of adoption
0.037*
-0.096
0.002
(0.022)
(0.144)
(0.002)
Direct 1 year lag
-0.042
-0.271
-0.001
(0.035)
(0.170)
(0.001)
Direct 2 year lag
0.026
-0.247
-0.002
(0.043)
(0.195)
(0.002)
Direct 3 year plus lag
0.014
-0.394*
-0.001
(0.048)
(0.211)
(0.002)
Indirect 2 year lead
0.052
0.243
-0.001
(0.034)
(0.148)
(0.002)
Indirect 1 year lead
0.045*
0.014
-0.002
(0.023)
(0.111)
(0.002)
Indirect year of adoption
0.071***
0.260**
-0.003**
(0.025)
(0.113)
(0.001)
Indirect 1 year lag
0.140***
0.288*
-0.002
(0.046)
(0.168)
(0.002)
Indirect 2 year lag
0.065
0.131
-0.003
(0.046)
(0.180)
(0.004)
Indirect 3 year plus lag
0.125**
0.209
-0.002
(0.052)
(0.212)
(0.003)
Observations
2,551,355
Cluster robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Not indicated are the following control variables: elective admission, female, white, Medicare, Medicaid,
private insurance, age, age squared, teaching hospital, state and year dummies, state linear time trends

10

Specifically, two additional samples are created. These include a “wide” sample, which consists of those patients
indicated as having some type of cerebrovascular disease. A “narrow” sample consists of only those patients
reported to have a cerebral hemorrhage or infarction. The sample used in the main analysis is an intermediate form
of these definitions, including all cerebral hemorrhages and infarctions as well as diagnoses where it is not indicated
if the patient suffered an infarction.

18

Table 1.8: AMI lags and leads results, individual reform specification
Variables
Natural log charges
LOS
Award cap 2 year lead
0.000
-0.065
(0.027)
(0.132)
Award cap 1 year lead
0.064***
0.026
(0.023)
(0.137)
Award cap year adoption
0.043*
-0.149
(0.022)
(0.146)
Award cap 1 year lag
-0.029
-0.305*
(0.029)
(0.164)
Award cap 2 year lag
0.032
-0.356*
(0.044)
(0.190)
Award cap 3 year plus lag
0.023
-0.449**
(0.046)
(0.215)
CSR 2 year lead
-0.015
0.449**
(0.039)
(0.217)
CSR 1 year lead
0.011
0.564**
(0.069)
(0.229)
CSR year of adoption
-0.034
0.424**
(0.047)
(0.201)
CSR 1 year lag
-0.028
0.199
(0.055)
(0.344)
CSR 2 year lag
-0.025
0.748*
(0.104)
(0.377)
CSR 3 year plus lag
-0.040
0.365
(0.065)
(0.427)
JSL 2 year lead
0.024
0.004
(0.037)
(0.195)
JSL 1 year lead
0.015
-0.187
(0.041)
(0.189)
JSL year of adoption
0.055
0.122
(0.043)
(0.168)
JSL 1 year lag
0.093**
0.070
(0.039)
(0.178)
JSL 2 year lag
0.043
-0.170
(0.062)
(0.293)
JSL 3 year plus lag
0.088*
-0.036
(0.046)
(0.250)
PPA 2 year lead
0.042
-0.063
(0.025)
(0.111)
PPA 1 year lead
0.065
-0.245
(0.044)
(0.149)

19

Mortality
-0.002
(0.001)
0.000
(0.001)
0.001
(0.002)
0.000
(0.002)
-0.002
(0.002)
-0.001
(0.002)
0.001
(0.002)
0.005
(0.003)
0.004
(0.002)
0.000
(0.003)
-0.002
(0.004)
-0.006*
(0.004)
-0.004*
(0.002)
-0.004
(0.003)
-0.005**
(0.002)
-0.004**
(0.002)
0.002
(0.002)
0.002
(0.002)
0.001
(0.002)
-0.003
(0.003)

PPA year of adoption
PPA 1 year lag
PPA 2 year lag
PPA 3 year plus lag

Table 1.8 (cont’d)
0.086*
(0.049)
0.167***
(0.057)
0.092
(0.068)
0.184**
(0.076)

-0.116
(0.194)
-0.045
(0.278)
-0.156
(0.312)
-0.109
(0.363)

-0.004
(0.003)
-0.005
(0.004)
-0.012**
(0.005)
-0.006
(0.005)

Observations
2,551,355
Cluster robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Not indicated are the following control variables: elective admission, female, white, Medicare, Medicaid,
private insurance, age, age squared, teaching hospital, state and year dummies, state linear time trends

VI. Conclusion
This paper presents inconclusive results regarding the relationship between physician
behavior and state tort reform adoption. For the treatment of AMI, aggregating individual law
changes into direct and indirect categories of tort reform does not seem to be an important part of
the story behind mixed evidence in the literature. Results on stay length, as well as the
relationship between indirect reforms and hospital charges, are generally consistent with the
empirical results of Kessler and McClellan (1996). They are also consistent with theoretical
predictions of Currie and MacLeod (2008), who argue that the effect of direct reforms and JSL
reform on physician behavior should be oppositely signed. However, the AMI results are
sensitive to the choice of variable meant to measure medical resource use and the use of state
time trends. Results from an event history analysis indicate significant leads in hospital charges
prior to tort reform adoption, suggesting that results from the main analysis in Table 1.3 do not
represent the true causal effect of the law change.

20

There is little evidence of a behavioral response in the stroke sample. The few significant
associations between tort reform and resource use in Table 1.4 – the relationship between CSR
reform and charges, as well as award caps and stay length – are sensitive to the use of state
specific time trends. Because there is little evidence of a relationship between physician
behavior and tort reform in stroke treatment, and the main results in the AMI sample are
sensitive to several robustness checks, it is difficult to make any conclusions regarding the role
of patient population in explaining the lack of consensus in the tort reform literature.
Finally, results from this paper do little to identify the presence of defensive medicine in
medical practice. Authors from the tort reform literature analyze a more comprehensive set of
patient outcome variables than presented in this paper. Notably, Kessler and McClellan (1996)
examine both mortality and health status one year after initial diagnosis, and Currie and
MacLeod (2008) explore the effect of tort reform on birth complications and APGAR scores.
Information on outcomes beyond mortality is limited in the HCUP NIS, and observation of
patient outcomes ends upon hospital discharge. Without better information on patient health
status after treatment, it is impossible to obtain a good picture of the benefits, if any, of liability
induced care. If incentives are appropriate in the current malpractice system, then reductions in
liability may actually harm patient outcomes if physicians no longer take sufficient measures to
avoid negligent injury. Future investigations of the effect of tort reform on physician behavior
must take into account both benefits and costs generated by the malpractice system. Given
mixed results on outcomes from the literature, whether tort reforms induce efficient or inefficient
behavior is a critical point of clarification in the evaluation of these laws.

21

APPENDIX

22

Table 1.9: Stroke lags and leads results, direct and indirect reform specification
Variables
Natural log charges
LOS
Mortality
Direct 2 year lead
0.010
0.155
0.004
(0.038)
(0.170)
(0.003)
Direct 1 year lead
0.062*
0.076
0.012***
(0.037)
(0.278)
(0.004)
Direct year of adoption
0.033
-0.082
0.007
(0.048)
(0.332)
(0.004)
Direct 1 year lag
-0.001
-0.319
0.000
(0.065)
(0.379)
(0.006)
Direct 2 year lag
0.057
-0.377
-0.001
(0.068)
(0.396)
(0.006)
Direct 3 year plus lag
-0.016
-0.711
0.000
(0.076)
(0.576)
(0.007)
Indirect 2 year lead
0.018
0.269
0.001
(0.038)
(0.227)
(0.004)
Indirect 1 year lead
0.099***
0.086
0.001*
(0.034)
(0.281)
(0.005)
Indirect year of adoption
0.082**
0.257
0.005
(0.037)
(0.317)
(0.006)
Indirect 1 year lag
0.128**
0.121
0.009*
(0.061)
(0.424)
(0.004)
Indirect 2 year lag
0.072
0.135
0.014*
(0.055)
(0.427)
(0.007)
Indirect 3 year plus lag
0.147**
0.355
0.014
(0.072)
(0.627)
(0.009)
Observations

1,791,518

Cluster robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Not indicated are the following control variables: elective admission, female, white, Medicare, Medicaid,
private insurance, age, age squared, teaching hospital, state and year dummies, state linear time trends

23

Table 1.10: Stroke lags and leads results, individual reform specification
Variables
Natural log charges
LOS
Award cap 2 year lead
0.017
0.044
(0.040)
(0.164)
Award cap 1 year lead
0.069*
-0.102
(0.040)
(0.252)
Award cap year adoption
0.040
-0.298
(0.052)
(0.267)
Award cap 1 year lag
0.005
-0.601*
(0.062)
(0.337)
Award cap 2 year lag
0.047
-0.668*
(0.074)
(0.361)
Award cap 3 year plus lag
-0.020
-0.907*
(0.077)
(0.515)
CSR 2 year lead
0.008
0.657***
(0.045)
(0.211)
CSR 1 year lead
0.045
1.364***
(0.066)
(0.343)
CSR year of adoption
0.098
1.446***
(0.060)
(0.304)
CSR 1 year lag
0.167**
1.466***
(0.082)
(0.517)
CSR 2 year lag
0.117
1.411**
(0.095)
(0.568)
CSR 3 year plus lag
0.305***
1.854**
(0.105)
(0.906)
JSL 2 year lead
0.031
0.060
(0.041)
(0.306)
JSL 1 year lead
0.080**
-0.317
(0.034)
(0.349)
JSL year of adoption
0.018
-0.027
(0.028)
(0.214)
JSL 1 year lag
0.028
-0.330
(0.033)
(0.350)
JSL 2 year lag
0.003
-0.212
(0.053)
(0.214)
JSL 3 year plus lag
0.019
-0.016
(0.038)
(0.233)
PPA 2 year lead
-0.051
-0.345*
(0.043)
(0.196)
PPA 1 year lead
0.015
-0.969***
(0.059)
(0.303)

24

Mortality
0.004
(0.003)
0.013***
(0.004)
0.006
(0.004)
0.000
(0.005)
-0.002
(0.005)
-0.001
(0.006)
0.002
(0.004)
-0.001
(0.009)
0.000
(0.011)
0.015
(0.010)
0.022*
(0.011)
0.022
(0.015)
0.001
(0.005)
0.009
(0.005)
0.004
(0.005)
0.006**
(0.003)
0.015***
(0.004)
0.007**
(0.003)
-0.006
(0.008)
0.000
(0.011)

PPA year of adoption
PPA 1 year lag
PPA 2 year lag
PPA 3 year plus lag

Table 1.10 (cont’d)
-0.021
-1.252***
(0.073)
(0.442)
0.034
-1.372**
(0.088)
(0.656)
-0.031
-1.911***
(0.098)
(0.659)
-0.041
-2.563**
(0.139)
(1.033)

Observations

-0.007
(0.011)
-0.012
(0.011)
-0.018
(0.013)
-0.018
(0.018)

1,791,518

Cluster robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Not indicated are the following control variables: elective admission, female, white, Medicare, Medicaid,
private insurance, age, age squared, teaching hospital, state and year dummies, state linear time trends

25

Variables
Award cap
CSR reform
JSL reform
PPA

Direct ref.
Indirect ref.

Observations

Narrow
Nat. log
charges

Main results
Nat. log
charges

0.024
(0.045)
0.142**
(0.053)
-0.050
(0.052)
-0.027
(0.040)

0.014
(0.042)
0.134**
(0.057)
-0.030
(0.052)
-0.025
(0.045)

0.008
(0.049)
-0.003
(0.054)

0.002
(0.044)
0.012
(0.052)

1,450,969

1,791,518

Table 1.12: Stroke comparison results
Wide
Narrow
Main results
Wide
Nat. log
charges
LOS
LOS
LOS
Individual reform specification
0.005
-0.295
-0.518*
-0.562**
(0.034)
(0.214)
(0.264)
(0.277)
0.140***
0.271
0.092
0.067
(0.049)
(0.496)
(0.531)
(0.509)
-0.024
0.297
0.472
0.586
(0.038)
(0.380)
(0.444)
(0.423)
-0.028
-0.619*
-0.660
-0.637
(0.048)
(0.339)
(0.395)
(0.436)
Direct and indirect reform specification
-0.008
-0.188
-0.376
-0.416
(0.038)
(0.206)
(0.261)
(0.282)
0.019
-0.092
-0.035
0.065
(0.049)
(0.289)
(0.334)
(0.352)
2,247,711

1,450,969

1,791,518

2,247,711

Cluster robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Not indicated are the following control variables: elective admission, female, white, Medicare, Medicaid,
private insurance, age, age squared, teaching hospital, state and year dummies, state linear time trends

26

Narrow

Main results

Wide

Mortality

Mortality

Mortality

-0.004
(0.003)
0.010
(0.013)
0.005
(0.006)
-0.016**
(0.007)

-0.005
(0.003)
0.008
(0.013)
0.008
(0.005)
-0.014*
(0.007)

-0.005**
(0.002)
0.000
(0.010)
0.008*
(0.004)
-0.012*
(0.006)

-0.004*
(0.002)
0.001
(0.006)

-0.004**
(0.002)
0.003
(0.005)

-0.004**
(0.002)
0.001
(0.004)

1,450,969

1,791,518

2,247,711

Variable
Direct ref.
Indirect ref.
Female
White
Age
Age squared
Medicare
Medicaid
Pri. insurance
Elective
Teaching hos.

State, year FE
State time
trends

Table 1.13: AMI results, direct/indirect specification
Natural log
Natural log
charges
charges
LOS
LOS
0.106
-0.021
-0.145
-0.251**
(0.081)
(0.024)
(0.274)
(0.105)
-0.068
0.052**
0.086
0.230**
(0.061)
(0.023)
(0.232)
(0.113)
-0.043***
-0.043***
0.260***
0.257***
(0.004)
(0.004)
(0.020)
(0.020)
-0.047**
-0.037**
-0.209**
-0.327***
(0.023)
(0.015)
(0.081)
(0.062)
0.050***
0.050***
0.141***
0.141***
(0.002)
(0.002)
(0.008)
(0.008)
-0.000***
-0.000***
-0.001***
-0.001***
(0.000)
(0.000)
(0.000)
(0.000)
0.057***
0.064***
0.581***
0.594***
(0.014)
(0.013)
(0.089)
(0.086)
0.035***
0.042***
1.064***
1.054***
(0.013)
(0.010)
(0.102)
(0.090)
0.049***
0.053***
-0.222***
-0.206***
(0.013)
(0.012)
(0.052)
(0.050)
0.040**
0.035**
-0.041
-0.057
(0.015)
(0.014)
(0.099)
(0.091)
0.473***
0.479***
1.537***
1.533***
(0.042)
(0.041)
(0.122)
(0.121)

Mortality
-0.003*
(0.002)
0.002
(0.002)
0.007***
(0.001)
-0.004***
(0.001)
-0.006***
(0.000)
0.000***
(0.000)
0.000
(0.001)
0.011***
(0.002)
-0.013***
(0.001)
-0.019***
(0.002)
0.003***
(0.001)

Mortality
0.001
(0.001)
-0.002
(0.002)
0.007***
(0.001)
-0.004***
(0.001)
-0.006***
(0.000)
0.000***
(0.000)
0.000
(0.001)
0.011***
(0.002)
-0.014***
(0.001)
-0.019***
(0.002)
0.003**
(0.001)

Yes

Yes

Yes

Yes

Yes

Yes

No

Yes

No

Yes

No

Yes

Observations

2,551,355

Cluster robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Not indicated are the following control variables: elective admission, female, white, Medicare, Medicaid,
private insurance, age, age squared, teaching hospital, state and year dummies, state linear time trends

27

Variable
Award cap
CSR reform
JSL reform
PPA
Female
White
Age
Age squared
Medicare
Medicaid
Pri. insurance
Elective
Teaching hos.

State, year FE
State time
trends

Table 1.14: AMI results, individual reform specification
Natural log
Natural log
charges
charges
LOS
LOS
Mortality
0.099
-0.013
-0.154
-0.262**
-0.003*
(0.081)
(0.025)
(0.272)
(0.098)
(0.002)
-0.038
0.020
-0.154
-0.153
0.001
(0.049)
(0.040)
(0.195)
(0.145)
(0.002)
-0.040
0.026
0.150
0.269*
0.001
(0.064)
(0.034)
(0.257)
(0.151)
(0.002)
-0.024
0.047**
0.078
0.128
0.000
(0.053)
(0.022)
(0.176)
(0.107)
(0.001)
-0.043***
-0.043***
0.260***
0.257***
0.007***
(0.004)
(0.004)
(0.020)
(0.020)
(0.001)
-0.046*
-0.038**
-0.207**
-0.330***
-0.004***
(0.023)
(0.015)
(0.079)
(0.061)
(0.001)
0.050***
0.050***
0.141***
0.141***
-0.006***
(0.002)
(0.002)
(0.008)
(0.008)
(0.000)
-0.000***
-0.000***
-0.001***
-0.001***
0.000***
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
0.057***
0.064***
0.581***
0.593***
0.000
(0.014)
(0.013)
(0.089)
(0.086)
(0.001)
0.036***
0.043***
1.064***
1.053***
0.011***
(0.013)
(0.010)
(0.102)
(0.090)
(0.002)
0.049***
0.054***
-0.222***
-0.207***
-0.013***
(0.013)
(0.012)
(0.052)
(0.050)
(0.001)
0.040**
0.035**
-0.0411
-0.057
-0.019***
(0.015)
(0.014)
(0.099)
(0.091)
(0.002)
0.473***
0.479***
1.537***
1.533***
0.003***
(0.042)
(0.041)
(0.122)
(0.121)
(0.001)

Mortality
0.000
(0.001)
-0.002
(0.002)
-0.002
(0.002)
-0.001
(0.002)
0.007***
(0.001)
-0.004***
(0.001)
-0.006***
(0.000)
0.000***
(0.000)
0.000
(0.001)
0.011***
(0.002)
-0.014***
(0.001)
-0.019***
(0.002)
0.003**
(0.001)

Yes

Yes

Yes

Yes

Yes

Yes

No

Yes

No

Yes

No

Yes

Observations

2,551,355

Cluster robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Not indicated are the following control variables: elective admission, female, white, Medicare, Medicaid,
private insurance, age, age squared, teaching hospital, state and year dummies, state linear time trends

28

Variable
Direct ref.
Indirect ref.
Female
White
Age
Age squared
Medicare
Medicaid
Pri. insurance
Elective
Teaching hos.

State, year FE
State time
trends

Table 1.15: Stroke results, direct/indirect specification
Natural log
Natural log
charges
charges
LOS
LOS
Mortality
0.111
0.002
-0.372
-0.376
-0.009**
(0.095)
(0.044)
(0.763)
(0.261)
(0.003)
-0.087
0.012
-0.079
-0.035
0.008*
(0.076)
(0.052)
(0.687)
(0.334)
(0.004)
0.000
0.001
0.211***
0.199***
0.002***
(0.003)
(0.003)
(0.049)
(0.046)
(0.001)
-0.109***
-0.098***
-0.715***
-0.985***
0.000
(0.025)
(0.017)
(0.183)
(0.183)
(0.002)
-0.004***
-0.004***
-0.173
-0.174
-0.004***
(0.001)
(0.001)
(0.122)
(0.122)
(0.000)
-0.000***
-0.000***
0.001
0.001
0.000***
(0.000)
(0.000)
(0.001)
(0.001)
(0.000)
0.066***
0.074***
-0.124
-0.067
-0.046***
(0.015)
(0.013)
(0.470)
(0.458)
(0.006)
0.175***
0.183***
2.498***
2.456***
-0.025***
(0.016)
(0.012)
(0.312)
(0.291)
(0.005)
0.043**
0.049***
-0.840***
-0.774**
-0.039***
(0.017)
(0.017)
(0.296)
(0.290)
(0.005)
-0.151***
-0.150***
-1.437***
-1.399**
-0.057***
(0.021)
(0.021)
(0.522)
(0.532)
(0.006)
0.311***
0.314***
1.417***
1.380***
0.019***
(0.033)
(0.033)
(0.105)
(0.103)
(0.003)

Mortality
-0.004**
(0.002)
0.003
(0.005)
0.002***
(0.001)
0.000
(0.002)
-0.004***
(0.000)
0.000***
(0.000)
-0.047***
(0.006)
-0.026***
(0.005)
-0.040***
(0.004)
-0.056***
(0.006)
0.019***
(0.003)

Yes

Yes

Yes

Yes

Yes

Yes

No

Yes

No

Yes

No

Yes

Observations

1,791,518

Cluster robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Not indicated are the following control variables: elective admission, female, white, Medicare, Medicaid,
private insurance, age, age squared, teaching hospital, state and year dummies, state linear time trends

29

Variable
Award cap
CSR reform
JSL reform
PPA
Female
White
Age
Age squared
Medicare
Medicaid
Pri. insurance
Elective
Teaching hos.

State, year FE
State time
trends

Table 1.16: Stroke results, individual reform specification
Natural log
Natural log
charges
charges
LOS
LOS
Mortality
0.114
0.014
-0.329
-0.518*
-0.007*
(0.095)
(0.042)
(0.743)
(0.264)
(0.004)
0.072
0.134**
0.694
0.092
0.015***
(0.068)
(0.057)
(0.570)
(0.531)
(0.004)
-0.093
-0.030
-0.096
0.472
0.004
(0.075)
(0.052)
(0.687)
(0.444)
(0.004)
-0.094
-0.025
-0.548
-0.660
-0.004
(0.067)
(0.045)
(0.576)
(0.395)
(0.004)
0.000
0.001
0.210***
0.199***
0.002***
(0.003)
(0.003)
(0.049)
(0.046)
(0.001)
-0.110***
-0.098***
-0.724***
-0.988***
0.000
(0.024)
(0.017)
(0.183)
(0.182)
(0.002)
-0.004***
-0.004***
-0.173
-0.174
-0.004***
(0.001)
(0.001)
(0.122)
(0.122)
(0.000)
-0.000***
-0.000***
0.001
0.001
0.000***
(0.000)
(0.000)
(0.001)
(0.001)
(0.000)
0.067***
0.075***
-0.125
-0.072
-0.046***
(0.015)
(0.013)
(0.467)
(0.460)
(0.006)
0.175***
0.185***
2.497***
2.451***
-0.025***
(0.016)
(0.012)
(0.310)
(0.287)
(0.005)
0.043**
0.050***
-0.840***
-0.779**
-0.039***
(0.017)
(0.017)
(0.296)
(0.291)
(0.005)
-0.151***
-0.150***
-1.434***
-1.399**
-0.057***
(0.021)
(0.021)
(0.521)
(0.532)
(0.006)
0.311***
0.314***
1.417***
1.378***
0.019***
(0.033)
(0.033)
(0.105)
(0.103)
(0.003)

Mortality
-0.005
(0.003)
0.008
(0.013)
0.008
(0.005)
-0.014*
(0.007)
0.002***
(0.001)
0.000
(0.002)
-0.004***
(0.000)
0.000***
(0.000)
-0.047***
(0.006)
-0.026***
(0.005)
-0.040***
(0.004)
-0.056***
(0.006)
0.019***
(0.003)

Yes

Yes

Yes

Yes

Yes

Yes

No

Yes

No

Yes

No

Yes

Observations

1,791,518

Cluster robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Not indicated are the following control variables: elective admission, female, white, Medicare, Medicaid,
private insurance, age, age squared, teaching hospital, state and year dummies, state linear time trends

30

Variable
NE damage cap
Punitive cap
Total damage cap
CSR reform
JSL reform
Periodic payment

Table 1.17: AMI results, all cap definitions
Natural log charges
LOS
Individual reform specification
-0.027
-0.319***
(0.028)
(0.111)
0.020
-0.101
(0.020)
(0.143)
-0.182***
-0.725***
(0.035)
(0.183)
0.010
-0.186
(0.036)
(0.145)
0.052
0.371**
(0.031)
(0.158)
0.046**
0.121
(0.021)
(0.115)

Observations

Mortality
-0.001
(0.001)
0.002
(0.002)
0.000
(0.002)
-0.002
(0.002)
-0.001
(0.002)
-0.001
(0.002)

2,551,355

Cluster robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Not indicated are the following control variables: elective admission, female, white, Medicare, Medicaid,
private insurance, age, age squared, teaching hospital, state and year dummies, state linear time trends

31

Variable
NE damage cap
Punitive cap
Total damage cap
CSR reform
JSL reform
Periodic payment

Table 1.18: Stroke results, all cap definitions
Natural log charges
LOS
Individual reform specification
0.007
-0.535
(0.041)
(0.344)
0.036
-0.380
(0.061)
(0.361)
-0.162***
-0.489
(0.049)
(0.501)
0.124**
0.078
(0.055)
(0.539)
-0.008
0.529
(0.053)
(0.532)
-0.027
-0.670*
(0.046)
(0.393)

Observations

Mortality
-0.001
(0.004)
-0.012*
(0.007)
0.004
(0.004)
0.008
(0.012)
0.004
(0.006)
-0.015**
(0.007)

1,791,518

Cluster robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Not indicated are the following control variables: elective admission, female, white, Medicare, Medicaid,
private insurance, age, age squared, teaching hospital, state and year dummies, state linear time trends

32

Table 1.19: AMI, patient compensation fund and contingency fee
Variable
Natural log total charges
LOS
Award cap
-0.013
-0.262**
(0.025)
(0.098)
CSR reform
0.022
-0.155
(0.040)
(0.145)
JSL reform
0.026
0.270*
(0.034)
(0.151)
Periodic payment
0.046**
0.129
(0.022)
(0.107)
Patient comp
-0.104***
0.057
(0.026)
(0.082)
Contingency fee
-0.160***
0.014
(0.030)
(0.089)
Observations

Mortality
0.001
(0.001)
-0.001
(0.002)
-0.002
(0.002)
-0.002
(0.002)
-0.012***
(0.001)
0.015***
(0.002)

2,551,355

Cluster robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Not indicated are the following control variables: elective admission, female, white, Medicare, Medicaid,
private insurance, age, age squared, teaching hospital, state and year dummies, state linear time trends

33

Table 1.20: Stroke, patient compensation fund and contingency fee
Variable
Natural log total charges
LOS
Award cap
0.013
-0.520*
(0.042)
(0.264)
CSR reform
0.137**
0.117
(0.056)
(0.528)
JSL reform
-0.030
0.468
(0.052)
(0.443)
Periodic payment
-0.026
-0.669*
(0.045)
(0.395)
Patient comp
-0.116***
-1.042***
(0.032)
(0.218)
Contingency fee
-0.286***
-1.074***
(0.037)
(0.163)
Observations

Mortality
-0.005
(0.003)
0.009
(0.012)
0.007
(0.005)
-0.015**
(0.007)
-0.041***
(0.006)
-0.036***
(0.004)

1,791,518

Cluster robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Not indicated are the following control variables: elective admission, female, white, Medicare, Medicaid,
private insurance, age, age squared, teaching hospital, state and year dummies, state linear time trends

34

WORKS CITED

35

WORKS CITED

Avraham, R. (2010, April). Database of State Tort Law Reforms (DSTLR 3rd). Retrieved
September 30, 2010, from http://www.utexas.edu/law/faculty/ravraham/dstlr.html
Avraham, R. (2007, June). An Empirical Study of the Impact of Tort Reforms on Medical
Malpractice Settlement Payments. The Journal of Legal Studies, 36(S2), S183-S229.
Congressional Budget Office. (2006). Medical Malpractice Tort Limits and Health Care
Spending. Washington, DC: Congress of the United States.
Currie, J., & MacLeod, W. (2008). First Do No Harm? Tort Reform and Birth Outcomes. The
Quarterly Journal of Economics, 123(2), 795-830.
Danzon, P. M. (2000). Liability for Medical Malpractice, Chapter 26. In Handbook of Health
Economics (pp. 1339-1404). N.p.: Elsevier.
HCUP Databases. Healthcare Cost and Utilization Project (HCUP). 1988-2008. Agency for
Healthcare Research and Quality, Rockville, MD. www.hcup-us.ahrq.gov/databases.jsp.
Kessler, D. P. (2011). Evaluating the Medical Malpractice System and Options for Reform.
Journal of Economic Perspectives, 25(2), 93-110.
Kessler, D., & McClellan, M. (1996). Do Doctors Practice Defensive Medicine? The Quarterly
Journal of Economics, 111(2), 353-390.
Kessler, D., & McClellan, M. (2002). How Liability Law Affects Medical Productivity. Journal
of Health Economics, 21(6), 931-955.
Kim, B. (2007, June). The Impact of Malpractice Risk on the Use of Obstetrics Procedures.
Journal of Legal Studies, 36(S2), S120-S79.
Langel, S. (2009, July 6). GOP Pushing Malpractice Reform. Roll Call. Retrieved March 22,
2012, from http://www.rollcall.com/issues/55_1/-36429-1.html
Mello, M. M. (2006, May). Medical Malpractice: Impact of the Crisis and Effect of State Tort
Reforms. Research Synthesis Report 10. Retrieved April 1, 2012, from
http://www.rwjf.org/files/research/15168.medmalpracticeimpact.report.pdf
Mello, M. M., Chandra, A., Gawande, A. A., & Studdert, D. M. (2010). National Costs Of The
Medical Liability System. Health Affairs, 29(9), 1569-1577.
Reinhardt, U. E. (2006, January/February). The Pricing Of U.S. Hospital Services: Chaos Behind
A Veil Of Secrecy. Health Affairs, 25(1), 57-69.
36

Sloan, F. A., & Shadle, J. H. (2009). Is There Empirical Evidence for "Defensive Medicine"? A
Reassessment. Journal of Health Economics, 28(2), 481-491.
Studdert, D. M., Mello, M. M., & Brennan, T. A. (2004). Medical Malpractice. The New
England Journal of Medicine, 350(3), 283-291.

37

CHAPTER 2
THE STATE CHILDREN’S HEALTH INSURANCE PROGRAM AND MATERNAL LABOR
SUPPLY INCENTIVES
I. Introduction
In recent decades, there have been numerous reforms to the public health insurance
system in the United States. One of the largest is the State Children’s Health Insurance Program
(SCHIP), adopted in 1997. This program provides health insurance to children less than 19 years
of age in families with incomes above existing Medicaid eligibility thresholds. SCHIP quickly
became an important source of health care access for near poor children. Between 1999 and
2006 total enrollment grew from two million children to well over six and a half million
(Congressional Budget Office [CBO], 2007).
Despite the benefits of health insurance, means-tested programs like SCHIP and
Medicaid are often criticized for distorting labor supply behavior. Recent research focuses on
the separation of Medicaid and AFDC eligibility rules during the late 1980s and early 1990s, and
Medicaid’s initial implementation period (Ham and Shore-Sheppard, 2005a; Strumpf, 2011).
These authors find no evidence of a Medicaid labor supply effect among mothers.11
The effect of Medicaid on labor supply need not be comparable to the SCHIP program,
however. The SCHIP expansion provides insurance coverage to children in families with higher
incomes and, frequently, access to employer sponsored health plans. In addition, the SCHIP
program is associated with greater variation in state level eligibility rules than traditional
Medicaid. The federal government gives individual states a large degree of flexibility in the
design and operation of their SCHIP programs. Some states established income eligibility
thresholds well beyond the federal poverty limit (FPL). New Jersey, for instance, covered
11

In an early paper, Yelowitz (1995) estimates reduced form relationships between Medicaid eligibility rules and
labor supply behavior. He finds that reforms to Medicaid eligibility rules for children increase maternal labor force
participation. These findings were questioned in later research (Ham and Shore-Sheppard, 2005a).

38

children living in families with incomes less than or equal to 350% of the federal poverty limit in
2007, whereas thresholds in North Dakota and Oregon were 140% and 185% FPL respectively.
Because it is likely that access to private or employer sponsored insurance varies with these
thresholds, the incentives to participate in SCHIP and to work could vary substantially. While
such heterogeneity is useful from an empirical perspective, it complicates the interpretation of
the program’s relationship with labor supply.
In this paper, I study the role of SCHIP on maternal labor supply behavior in two steps.
First, I estimate the effect of the eligibility expansion on health insurance coverage for a sample
of children belonging to single mothers, similar to the insurance crowd-out literature. Second, I
examine the effect of changes in income eligibility thresholds on the work effort of single
mothers. Results indicate that SCHIP is associated with public insurance take-up among
children, suggesting the program was successful in enrolling uninsured children. Corresponding
estimates on private and any insurance coverage, however, imply that a large fraction of the takeup occurs among children who were previously privately insured. Finally, I find no evidence of
a relationship between SCHIP and work behavior in the labor supply analysis.
In section II of this paper, I discuss important institutional details about SCHIP and
review related academic literature. Theoretical labor supply predictions, data, and empirical
methods are outlined in section III. Health insurance coverage results are found in section IV,
and labor supply results are discussed in section V. Finally, section VI concludes.

II. Background Information
II.A. The State Children’s Health Insurance Program

39

The federal government established the State Children’s Health Insurance Program as
part of the Balanced Budget Act of 1997. The program is intended to expand health insurance
access to children less than 19 years in age in families earning too much income to qualify for
Medicaid coverage. By the end of 2000, all states finished initial implementation of their SCHIP
programs. Matching funds are provided to states from the federal government, but total funding
is capped. Additionally, the law allows states considerable flexibility in the design and
maintenance of their programs. To satisfy SCHIP requirements, states can implement an
expansion of their existing Medicaid programs, create a stand-alone program, or enact some
combination of the two options. As a consequence, eligibility and program rules vary
considerably across states.
The discussion of SCHIP eligibility rules regarding family income can be separated into
two parts – thresholds and disregards. Children are generally eligible for SCHIP coverage if
their family’s income is below the income limit, or threshold, for eligibility. Income disregards
raise eligibility cutoffs by allowing families to earn an income in excess of a given threshold.
This is done by exempting specific amounts of family income in order to reflect work status or
child care expenses, as well as child support. As with eligibility thresholds, there is considerable
variation in income disregards. These vary with expense type, such that some states only exempt
work expenses and others discount all three types. The amount of income disregarded varies as
well. In 2008, for instance, Alabama allowed families to disregard $90 of income per month for
work expenses. The same disregard in Kansas was $200 monthly (Ross et al., 2008a). The size
of child care disregards often vary with child age, with younger children eligible for larger

40

income exemptions. All income eligibility threshold information in this analysis is calculated
using work and child care disregards.12
SCHIP includes a number of additional provisions regarding cost-sharing and private
insurance crowd-out. Under traditional Medicaid, families were not required to pay premiums or
cost-sharing. The SCHIP expansion, however, does allow states to institute these policies.
Generally states can charge both premiums and cost sharing if family income exceeds 100%
FPL. Families earning below 100% FPL do not pay premiums, but can face cost sharing charges
for non-preventative medical services (CBO, 2007). In 2006, 35 states charged premiums and
roughly a third required cost-sharing payments (Ross et al., 2007). These payments vary across
states and income level, but are usually low.13 In addition, the total sum of premium and costsharing payments is capped at 5% of family income (CBO, 2007).
In addition, most states use enrollment waiting periods to inhibit private insurance crowdout. Waiting periods require children in families be uninsured for a specific length of time
before receiving coverage, though some states have exceptions for involuntary loss of private
coverage (CBO, 2007). As of 2006, 35 states had some sort of waiting period, almost all of
which are 6 months in length or less (Ross et al., 2007). For privately insured families, these
policies raise the cost of leaving existing coverage for public insurance. However, this may also
have the effect of reducing take-up among uninsured children.
The SCHIP program is also associated with a limited expansion of parental eligibility for
public insurance. The Congressional Budget Office reports that 13 states used SCHIP funding to
12

I do not apply disregards for child support expenses or income. March CPS data does not include information on
child support expenses, though it does report information on child support income.
13
Ross et al. (2007) report that for a family of three with two children and an income of 200% FPL in 2006,
effective annual premium payments exceeded $500 in 9 states. For similar families earning 150% FPL, only 4 states
exceeded this amount. At 100% FPL, no states charged more than $500 yearly. This is well below average yearly
private insurance premium expenditures for families of 3 persons or more, which were $2,846 in 2006 (Bernard and
Banthin, 2009).

41

expand parental eligibility by 2007 (CBO, 2007). In general, income eligibility is much lower
for parents than children in either the Medicaid or SCHIP programs. Only 8 states had parental
income eligibility thresholds for Medicaid or SCHIP above 200% FPL in 2006, with thresholds
in a majority of states less than or equal to the poverty level (Ross et al., 2007). Large
differences between child and parental eligibility for SCHIP may encourage public insurance
take-up among children without changing labor supply among working mothers.

II.B. Relevant literature
Authors of recent papers find little evidence of a relationship between maternal work
behavior and public health insurance programs for children.14 Strumpf (2011) examines the
initial implementation of Medicaid in the late 1960’s and its effect on labor supply behavior.
She uses both a difference-in-difference and triple difference specification to evaluate the effect
of the program’s introduction on labor force participation of single mothers. No evidence of a
relationship between Medicaid eligibility and labor supply behavior is found. Ham and ShoreSheppard (2005a) use variation in income eligibility thresholds to determine an effect on AFDC
and labor force participation for a sample of single mothers during the middle 1980s to early
1990s. They estimate specifications with AFDC and Medicaid thresholds entered separately,
finding no evidence that Medicaid reforms in this period increase labor force participation. The
authors do find, however, that AFDC thresholds are consistently associated with both of the
dependent variables.

14

Older papers in the literature offer mixed evidence of a labor supply effect from Medicaid. Yelowitz (1995) finds
evidence of an association between Medicaid and maternal labor supply. Others (Winkler, 1991; Moffitt and Wolfe,
1992; Montgomery and Navin, 2000; Meyer and Rosenbaum, 2001) conclude that the program has no effect on
maternal labor supply.

42

There is a sizable literature devoted to evaluating the effects of public insurance
expansions on both take-up and private insurance crowd-out. Many authors from this literature
analyze the late 1980s to early 1990s Medicaid expansions. Cutler and Gruber (1996) were the
first to assess the relationship between Medicaid eligibility and insurance coverage. Using
Current Population Survey (CPS) data and a simulated eligibility measure to instrument for
program eligibility, they estimate crowd out rates of 31-50%. Ham and Shore-Sheppard (2005b)
also conduct a Cutler and Gruber style analysis, but with Survey of Income and Program
Participation (SIPP) data. They find no evidence of crowd-out.
Researchers have also used SCHIP to investigate questions of public coverage take-up
and private insurance crowd-out. Authors estimate high levels of crowd-out due to the SCHIP
expansion. These researchers (Lo Sasso and Buchmueller, 2004; Hudson et al., 2005; Gruber
and Simon, 2008) estimate crowd-out rates between 50% and 60%, though their estimates are
sensitive to specification. Results are robust to the choice of data set as authors use the CPS,
Medical Expenditure Panel Survey (MEPS), and SIPP. Estimates of SCHIP take-up among
eligible children are low. Gruber and Simon (2008) and Lo Sasso and Buchmueller (2004)
estimate only 5 – 16% of eligible children receive SCHIP coverage, though the latter pair of
authors argue that the take-up rate for previously uninsured children is over 24%. Additionally,
SCHIP legislation included several provisions designed to reduce crowd-out. Lo Sasso and
Buchmueller (2004) find that waiting periods reduce both take-up of public insurance and
crowd-out. Gruber and Simon (2008) conclude differently, arguing that waiting periods and cost
sharing associated with SCHIP exacerbate crowd-out by limiting take-up of the uninsured more
than they limit movement away from private insurance.

43

III. Methods
III.A. Predictions from a static labor supply model
Standard static labor supply models imply specific predictions for both the extensive and
intensive margin labor supply responses, depicted in Figure 2.1 below. For single mothers that
value children’s health insurance, SCHIP effectively extends children’s insurance eligibility
beyond existing coverage from the Medicaid program. This implies that the traditional Medicaid
budget constraint notch exists at higher income levels under SCHIP.15 Working mothers above
this notch, with or without insurance for their children, may decrease earnings to become eligible
for the program. These mothers are indicated with an A in Figure 2.1. Employed mothers at B,
constrained to Medicaid coverage at the old notch, may increase their labor supply in response to
eligibility expansion. Finally, those located between these two points – in the SCHIP expansion
region of the budget constraint – may reduce labor supply because of the income effect generated
by the program. This group includes mothers with employer sponsored insurance as well as
those with uninsured children.16
The overall intensive margin response, then, is ambiguous. Families with existing public
coverage through Medicaid may increase labor supply if the SCHIP expansion allows mothers to
increase work effort and reach a higher indifference curve. Mothers of children without public
insurance coverage, either uninsured or privately covered, face negative intensive margin
incentives. For these families, the income effect generated by SCHIP allows them to reduce
labor supply and move to a more desirable mix of income and leisure. The size of the income
15

This is a simplification, since few states expanded SCHIP eligibility to parents. In states without expanded
parental eligibility, there may be an additional notch between Medicaid and SCHIP. This generates additional
discontinuities in the mother’s budget set.
16
These predictions do not account for any differences in the set of insurance coverage options for the mother as a
result of changes in her labor supply behavior. A change in labor supply that results in the loss of employer
sponsored coverage, or even a change in medical providers, may reduce the mother’s incentive to alter employment
behavior in response to SCHIP.

44

Figure 2.1: Effect of SCHIP on labor supply (intensive and extensive margin)

effect generated by SCHIP for working mothers will depend on how much they value public
insurance coverage. Mothers will value public coverage more if they are financing the employer
plan with premium contributions or if their employer plan does not provide a level of benefits
comparable to SCHIP. Investigating employer-sponsored insurance trends between 2000 and
2008, Vistnes et al. (2010) find both an increase in premium contributions paid by employees
and a decline in insurance offers from some employers. Both trends may encourage families to
take up public coverage.
Figure 2.1 also depicts the extensive margin labor supply response. The static labor
supply model predicts an extensive margin response only from unemployed single mothers with
Medicaid coverage for their children. These mothers, depicted at point C on the budget

45

constraint, may be induced to join if SCHIP allows them to work and retain insurance coverage
for their children. This may be especially relevant for individuals with limited discretion over
hours of work. It is not expected that SCHIP would result in a negative extensive margin
response among working mothers with employer sponsored insurance or without coverage for
their children. These families were already free to leave the labor force and obtain Medicaid
coverage before the program was in effect.
Despite the extensive margin predictions, SCHIP may have a role in reducing labor force
participation. This is because the program is associated with efforts to increase enrollment
among low-income families eligible for traditional Medicaid. Selden et al. (2004) argue that,
“concerns about Medicaid and SCHIP enrollment have led to unprecedented efforts to improve
outreach, reduce stigma, simplify enrollment, and retain eligible enrollees since 1996” (p.40).
They also note that SCHIP legislation requires applicants be screened for Medicaid eligibility,
potentially increasing Medicaid enrollment (Selden et al., 2004). Consequently, SCHIP
implementation may also increase public coverage among families already eligible for Medicaid.
This can reduce labor force participation among low income mothers working to finance health
care with employer coverage or their wages. A similar response can also occur among relatively
high income eligible families with limited exposure to, or knowledge of, the public health
insurance system. A final possibility for a negative participation effect is among families not
previously eligible for traditional Medicaid even if they left work, possibly due to high levels of
non-work related income.17

17

Child support payments are an important source of income for some single mothers. Approximately one-third of
the mothers in this analysis receive child support payments. Conditional on positive values, the median annual
payment is just over $2,990 (in 1996 dollars). For comparison, the federal poverty limit in 1996 for a family of
three was $12,980.

46

III.B. Data
This analysis uses data from the IPUMS CPS, which is an integrated version of the
March Current Population Survey (CPS) managed by the University of Minnesota, survey years
1997 through 2007.18 I chose this time period to provide sufficient policy variation before and
after the implementation of SCHIP, and to avoid the Great Recession era and the introduction of
the Affordable Care Act. I lower child and maternal age by one year to reflect the age
concurrent with recorded labor market information. I draw two extracts from this data set. The
first consists of 19-65 year old single mothers whose youngest child is less than or equal to 18
years of age. Attached to each observation is information regarding all related children living
with the mother. The sample does not contain information about children greater than 18 years
of age, or from children designated in the data as household heads. The second extract consists
of 0-18 year old children belonging to the single mothers identified in the first sample. In both
samples, I drop observations from Tennessee. Throughout much of the sample period,
Tennessee’s TennCare program extended Medicaid coverage to uninsured children not eligible
for traditional Medicaid. Importantly, TennCare did not restrict eligibility based on income,
implying that children did not face an income eligibility cutoff for coverage.
Four measures of labor supply are collected in the maternal sample: labor force
participation, full-time work participation, usual weekly work hours, and annual weeks worked.
Labor force participation is a binary indicator equal to one if the mother spends at least one week
at work during the year. Full-time employment is a binary variable equal to one if the mother is
both employed and works thirty-five or more hours a week, based on usual weekly work hours.
Usual weekly hours is a self-reported measure of work hours the mother experiences in a typical

18

Because the March CPS contains data on the previous year, this survey period reflects information from years
1996 through 2006.

47

week. Finally, the weeks worked variable counts the number of weeks the mother is employed
over the course of the year.
One feature of the IPUMS CPS is the summary health insurance variables constructed by
the State Health Access Data Assistance Center (SHADAC) at the University of Minnesota.
These insurance variables are created to be consistent across survey years, and are distinct from
health insurance variables reported in the March CPS supplement. IPUMS CPS documentation
notes that health insurance information from the March supplement is vulnerable to changes in
the CPS survey, primarily the introduction of an insurance coverage verification question and
changes in editing procedures. To account for these changes, SHADAC releases an enhanced
version of the March variables meant to be consistent across time.19
Policy data come from a variety of sources. Income eligibility thresholds for Medicaid
and SCHIP, both before and after the introduction of SCHIP, are primarily from Rosenbach et al.
(2001) and annual surveys of state SCHIP policies by the Kaiser Family Foundation.20 For this
analysis, the relevant eligibility threshold for each state and child age group is the Medicaid or
SCHIP income cutoff during that year. Dates of state level SCHIP program implementation
come from Rosenbach et al. (2001). Medicaid and SCHIP income disregard information are
from Ku et al. (1999) and Ross et al. (2008b), as well as the Urban Institute’s TRIM 3 database.21
Ku et al. (1999) describe state level disregard information for Medicaid and SCHIP as of October
1998, and Ross et al. (2008b) define the same information as of January 2008. These are used as
endpoints; if disregards are unchanged between the two data sources then that disregard
19

In addition, the documentation also advises the use of SHADAC summary health insurance weights as opposed to
standard CPS weights. These weights remove observations from survey respondents who do not answer questions
from the March supplement. Responses for these individuals are normally imputed. Consequently, sample sizes
using SHADAC weights are smaller than with standard weights.
20
See works cited page for complete list of surveys.
21
For my analysis, the relevant disregard information (Medicaid or SCHIP disregards) is the one that applies to the
highest eligibility threshold in that state.

48

information is applied to all intervening years. When this is not the case, the disregard change is
coded as occurring during the year of the SCHIP policy adoption. Disregard information for
years before 1998 are from TRIM 3.
Sample means of key variables from both samples are found on Table 2.1. These means
are calculated using SHADAC weights. The top panel presents mean and standard deviation
information for policy and labor supply variables. The mean SCHIP or Medicaid income
threshold in the sample is just under 2.3, implying the average eligibility cutoff for children is
230% of FPL. Three-quarters of all children in the child sample are imputed to be program
eligible based on SCHIP and Medicaid eligibility rules. Maternal labor supply characteristics are
also depicted in the top panel. Most mothers, 80.6%, are in the labor force. Those working are
on the job nearly 40 hours a week and over 45 weeks out of the year, on average. The third
panel indicates health insurance information in both samples. A majority of the children have
health insurance, roughly 86% to 87% in either sample. Private insurance is more prevalent than
public for these children. In the fourth panel, a majority of mothers are separated or divorced
from their spouse, and roughly a quarter have at least a two year college degree.
Trends in income eligibility and insurance coverage rates are displayed in Figures 2.2 and
2.3. Figure 2.2 illustrates average income eligibility thresholds across years, by age group.22
These trends demonstrate the variation present in the key policy variable used in this analysis.
Before SCHIP was fully implemented, there was considerable age variation in eligibility rules.
As SCHIP moved towards complete implementation, differences in age group eligibility

22

Age cutoffs for the age groups presented on these figures are based on eligibility rules for traditional Medicaid.
Infants (<1 years), on average, experience the highest levels of income eligibility. Federal law mandates that
children less than six years in age (1-5 years) have income eligibility of at least 133% FPL. Children born after
September 30, 1983, and who are six years of age or greater, have eligibility requirements of at least 100% FPL (616 year olds by the end of the SCHIP implementation period in 2000). The last group, 17 to 18 year olds, is never
subject to increased eligibility under traditional Medicaid before the SCHIP expansion.

49

Table 2.1: Means of key variables
Variable
Child sample
Mean
SD
Policy and labor supply variables
Medicaid/SCHIP income eligibility threshold
2.282
0.609
Imputed eligibility
0.751
0.432
Imputed eligibility (share of children)
Imputed eligibility (all children)
Labor force participation (LFP)
Full-time work (FT)
Hours worked, positive values~
Weeks worked, positive values~
Health insurance coverage
Public
0.398
0.489
Private
0.467
0.499
Group private
0.386
0.487
Non-group private
0.081
0.272
Any
0.865
0.342
Public (share)
Private (share)
Group private (share)
Non-group private (share)
Any (share)
Public (all)
Private (all)
Group private (all)
Non-group private (all)
Any (all)
Sample characteristics
Female
0.500
0.500
Foreign born
0.039
0.193
Share of family w/ some college
0.009
0.056
Share of family employed
0.050
0.128
Family size
3.770
1.569
White
0.608
0.488
Age
8.961
5.273
Child less than 5
Cohabitation
Separated or divorced
High school
Some college
Two year degree
Four year degree

50

Maternal sample
Mean
SD
2.276

0.593

0.704
0.694
0.806
0.650
38.050
45.356

0.451
0.461
0.396
0.477
9.588
13.010

0.356
0.504
0.417
0.086
0.860
0.340
0.488
0.397
0.075
0.845

0.471
0.492
0.483
0.271
0.337
0.474
0.500
0.489
0.264
0.362

3.279
0.649
34.933
0.306
0.033
0.553
0.356
0.234
0.096
0.097

1.374
0.477
9.009
0.461
0.179
0.497
0.479
0.423
0.295
0.296

Table 2.1 (cont’d)
Graduate degree
Observations
117,131
~There are 54,513 observations for the conditional work variables

0.036

0.186
67,016

disappeared. This implies older age groups experienced relatively large increases in income
eligibility. Figure 2.3 shows rates of insurance coverage, by type, across income. Income is
measured in units of the FPL. At 200% FPL, roughly where SCHIP is designed to operate, a
majority of children have private insurance coverage. Less than one fifth of children have public
coverage at this income level. These insurance trends show that SCHIP extends public insurance
eligibility to children living in families that are much more likely to have private coverage,
suggesting the program’s potential for private insurance crowd-out.
Figure 2.2: Child Income Eligibility Thresholds for Medicaid and SCHIP

51

Figure 2.3: Child Health Insurance Coverage Trends, by Income Status

III.C. Empirical methodology
To organize my analysis, consider the following labor supply equation (where i denotes
mother, a youngest child age, s state, and t year):
LSiast = αs + δt + μa + λst + γ PublicCoverageiast + Xiβ + εiast

(1)

LSiast is a maternal labor supply variable, meant to capture extensive margin (indicators for labor
force participation and full-time work) or intensive margin effects (usual hours worked per week
and number of weeks worked during the year). Specifications with an intensive margin variable
are conditional on positive values. PublicCoverageiast reflects the public insurance coverage
status of children attached to the mother’s observation. I use two definitions of this variable in
the analysis. The first is a share variable meant to indicate the fraction of children with public

52

insurance. The second definition is an indicator that equals one if all children have coverage.23
Xi is a vector with information on race, age, family size and composition, cohabitation status,
separated or divorced status, and education level. Standard errors are clustered at the state level,
and observations are weighted using SHADAC summary health insurance weights.
To control for unobserved effects, I include state and year indicators in the model. These
allow the model to capture unobservable time and state specific variation that may be correlated
with eligibility rules, and are denoted with αs and δt. I also include youngest child age fixed
effects, μa, to account for time invariant age group specific behavior. In an effort to control for
year differences that affect certain states more or less than others, I include a fully interacted set
of state and year dummies in all specifications. These are denoted with λst. Including these
interactions imply that equation (1) achieves identification from changes in income eligibility
thresholds across youngest child age groups within state and year combinations. This would
include comparisons of similar families where some of those families, before the implementation
of SCHIP, were ineligible for Medicaid because their children were older and faced a lower
income threshold for coverage. SCHIP generally removes differences in income eligibility based
on child age, which existed under traditional Medicaid.
Because individuals select into coverage, PublicCoverageiast should be regarded as
endogenous in this specification. As a solution, I estimate equation (1) using an income
eligibility threshold variable as an instrument for PublicCoverageiast. Variation in the threshold
variable is due solely to state, year, and child age group differences in eligibility policies for state
Medicaid or SCHIP programs. This approach follows methods from the Medicaid labor supply

23

Other definitions of insurance coverage were considered. These include an indicator for any (at least one) child in
the family with coverage, and whether the youngest child in the family has coverage. Estimating these variables
does not change the results substantially.

53

literature (Yelowitz, 1995; Ham and Shore-Sheppard, 2005a; Dave et al. 2013), where authors
relate variation in program eligibility rules with labor supply.
A benefit of equation (1) is that it connects two related literatures. When estimating
equation (1) by instrumental variables, the first stage is similar to health insurance models from
the health insurance crowd-out and take-up literature:
PublicCoverageiast = αs + δt + μa + λst + γEligibilityThresholdiast + Xiβ + εiast

(2)

Fundamentally, this literature seeks to understand the relationship between public health
insurance eligibility and insurance coverage behavior among eligible individuals and families.
Estimates of private insurance crowd-out would imply that privately insured families move to
public coverage in response to SCHIP. The labor supply analysis helps determine whether
insurance crowd-out is the result of a change in maternal labor supply behavior or decisions to
enroll with employer sponsored coverage.
Equation (1) is also related to the Medicaid labor supply literature. The reduced form
associated with estimating equation (1) by instrumental variables updates methods from
Yelowitz (1995) and Ham and Shore-Sheppard (2005a), who associate labor supply with
measures of program eligibility rules. The reduced form relationship between maternal labor
supply and SCHIP eligibility is:
LSiast = αs + δt + μa + λst + γEligibilityThresholdiast + Xiβ + εiast

(3)

Dave et al. (2013) use a similar reduced form approach, relating the work behavior of pregnant
women with a simulated measure of Medicaid eligibility. A disadvantage of reduced form
models like equation (3) is that they estimate the average effect of eligibility across households
with varying incentives for program take-up and labor supply. As discussed in section III.A,

54

SCHIP labor supply predictions can be either positive or negative depending on a child’s initial
insurance coverage status.
Instrumental variable estimation of equation (1) also shares this disadvantage. The
instrumental variables strategy assumes that the threshold variable affects labor supply though
only the public insurance take-up decision. Mothers with children covered under traditional
Medicaid, however, may increase work effort with no change to their public insurance status
(point B mothers in Figure 2.1). In this case, estimating equation (1) by instrumental variables
will not deliver the causal labor supply effect of gaining public coverage. If the assumption
necessary for instrumental variables estimation is true, however, instrumental variable estimation
of equation (1) will reflect the causal effect on maternal labor supply of children gaining public
insurance coverage as a result of the SCHIP eligibility expansion.

IV. SCHIP Eligibility and Children’s Public Insurance Coverage
Equation (2) investigates the effect of program eligibility rules on public insurance takeup among children. This equation is estimated in both the child and maternal level samples. The
purpose of the child level estimation is to benchmark results on insurance coverage to the crowdout literature. The maternal level analysis serves as the first stage for instrumental variable
estimation of equation (1). Insurance coverage variables are defined differently for each sample,
with a binary coverage indicator in the child sample and either a share variable or indicator for
all children with coverage in the maternal sample.24
Estimates of the relationship between SCHIP and public insurance coverage are
presented in the first column of Table 2.2. In the child sample, the eligibility expansion is
24

Other specification differences between the child and maternal level analyses are: different variables in the vector
of controls, Xi; child age dummies are used in the child level sample, but are replaced with youngest child age
dummies in the maternal level sample.

55

Variable
SCHIP threshold

Mean value

Table 2.2: Health insurance coverage, equation (2)
Public
Private
Child sample
0.130**
-0.055
(0.044)
(0.040)
[2.955]
[1.375]
0.398

0.467

Observations

SCHIP threshold

Mean value

SCHIP threshold

Mean value

Any
0.075**
(0.016)
[4.688]
0.865

117,131
Maternal sample: share of children
0.080*
-0.029
(0.031)
(0.032)
[2.580]
[0.906]
0.356

0.504

Maternal sample: all children
0.070*
-0.037
(0.030)
(0.033)
[2.333]
[1.121]
0.340

0.488

0.051**
(0.015)
[3.400]
0.860

0.040*
(0.016)
[2.500]
0.845

Observations
67,016
Cluster robust standard errors in parentheses, absolute value of t-statistic in brackets
** p<0.01, * p<0.05
Not indicated are estimates from the following control variables: family size, white racial status, black
racial status, female, foreign born, share of family with some college education, share of family in labor
force, age, age squared, child age dummies, year dummies, and state dummies.

associated with public insurance take-up. For a one unit increase in the eligibility threshold (one
unit equals 100% FPL), the eligibility expansion is associated with a 13 percentage point
increase in the likelihood of public insurance coverage. This represents an increase in the mean
number of children with public insurance of almost 33%. Maternal level estimates, found in the
middle and bottom panels, are similar. The threshold variable is associated with positive public

56

insurance take-up, the middle panel showing an 8 percentage point increase in the share of
children with coverage from a one unit increase in the eligibility threshold. For the bottom
panel, the same threshold change is associated with a 7 percentage point increase in the
likelihood of all children having public insurance coverage.
To better understand the public insurance take-up decision among children, dependent
variables reflecting private and any insurance coverage are estimated in addition to the
PublicCoverageist variable. Insurance coverage variables are defined to be mutually exclusive;
summing across public and private coverage equals the any insurance category.25 Private
coverage estimates are reported in the middle column of Table 2.2. In both the child and
maternal samples, coefficients for private coverage are negative but never significant. Results
for any insurance coverage are found in the last column of Table 2.2. In the child sample, the
coefficient on any insurance coverage indicates a statistically significant 7.5 percentage point
decline in the probability of a child being uninsured. Maternal sample estimates are similar,
suggesting that the SCHIP expansion increased the likelihood of having any insurance coverage.
In both the child and maternal samples, the significant effect of the threshold variable on
public insurance take-up is associated with a less than equivalent rise in the likelihood of any
insurance coverage. This suggests that some fraction of public insurance take-up associated with
SCHIP is due to children leaving private coverage. Public and any insurance coverage estimates
imply a crowd-out rate of just over 42% in the child sample and roughly 36% in the maternal

25

Individuals reporting both private and public insurance in a given year are recorded as having private insurance.
Additionally, the all children insurance coverage variables in the maternal sample do not sum to the overall coverage
category. Within some families, different children have different sources of coverage. For these families, the all
insurance coverage definition records a 0 for any specific coverage type, but a 1 for having any insurance.

57

sample.26 These are smaller than similar estimates from Lo Sasso and Buchmueller (2004) or
Gruber and Simon (2008), whose preferred estimates range from nearly 50% to 60%.
The effect of the threshold variable on private insurance coverage is statistically
insignificant in both samples. This result, however, masks variation across group and non-group
private insurance coverage. Table 2.3 reports the effect of SCHIP on private insurance
categories for the child and maternal samples. The effect of the eligibility expansion on group
coverage is large – a one unit increase in the threshold is associated with a 10.9 percentage point
decline in the probability of group coverage for the child sample. The same threshold change
yields group coverage declines of 8.1 to 9.4 percentage points for the maternal sample. The
effect of the threshold variable on non-group private coverage is smaller in magnitude, but show
an increase of 3.8 to 5.2 percentage points in the maternal sample. Lo Sasso and Buchmueller
(2004) find similar results on group and non-group private coverage. They argue this is evidence
that parents are incorrectly reporting SCHIP coverage as private coverage, and that this is
reflected in positive estimates on private non-group coverage and negative estimates on group
insurance (Lo Sasso and Buchmueller, 2004). If such an assumption is true about non-group
private coverage, then insurance coverage estimates from Table 2.2 understate both the true
effect of SCHIP on public coverage and the extent of private insurance crowd-out.27
Finally, public insurance coverage results from the maternal level sample suggest that the
eligibility threshold is a weak instrumental variable. From maternal sample results in Table 2.2,
the t-statistic associated with the share of children coefficient is only 2.58. For the all children

26

The crowd-out rate is the difference between the increase in public and any coverage, divided by the increase in
public coverage. I use the following formula: 1 – (coefficient for any/coefficient for public).
27
This can be seen from the expression used to calculate crowd-out, 1 – (coefficient for any/coefficient for public).
If the increase in non-group private coverage actually reflects public coverage take-up, then the coefficient for
public coverage is too small. The coefficient for any coverage is not affected. This decreases the size of the crowdout expression above.

58

Table 2.3: Group and non-group private coverage, equation (2)
Variable
Group Private
Non-Group Private
Child sample
SCHIP threshold
-0.109**
0.054**
(0.031)
(0.014)
[3.516]
[3.857]
Mean value

0.386

Observations

SCHIP threshold

Mean value

SCHIP threshold

Mean value

0.081
117,131

Maternal sample: share of children
-0.081**
(0.024)
[3.375]
0.417
Maternal sample: all children
-0.094**
(0.028)
[3.357]
0.397

0.052**
(0.014)
[3.714]
0.086

0.038*
(0.015)
[2.533]
0.075

Observations
67,016
Cluster robust standard errors in parentheses, absolute value of t-statistic in brackets
** p<0.01, * p<0.05
Not indicated are estimates from the following control variables: family size, white racial status, black
racial status, female, foreign born, share of family with some college education, share of family in labor
force, age, age squared, child age dummies, year dummies, and state dummies.

public coverage coefficient, the t-statistic is 2.33. This shows that the threshold variable is only
weakly correlated with public coverage, which is the key explanatory variable in the labor supply
model.28 A weak relationship between these variables may mean that instrumental variables

28

Tables 2.6 through 2.12 present results from specifications including additional policy variables (parental
eligibility thresholds for public insurance, waiting periods, premium payments, and cost-sharing payments) as first
stage instruments. These policy variables do not strengthen the first stage relationship.

59

estimation of equation (1) will yield large standard errors and point estimates (Bound, Jaeger,
and Baker, 1995). Consequently these estimates, which follow, should be interpreted cautiously.

V. The Effect of SCHIP Coverage on Maternal Labor Supply
Labor supply estimates from equation (3), which provide the direct effect of SCHIP on
maternal work behavior, are reported in Table 2.4. This specification is meant to be similar to
reduced form methods used in the literature, and serves as the starting point for the labor supply
analysis. As with the analysis of Ham and Shore-Sheppard (2005a), estimates from Table 2.4
indicate almost no evidence that SCHIP influences the labor supply of single mothers.
Coefficients generally show a negative extensive and intensive margin labor supply effect,
though the coefficient on the threshold variable for hours is positive. These estimates are
insignificant at the 5% level, with the exception of the threshold variable coefficient for annual
weeks worked. This estimate, in the last column of Table 2.4, implies a reduction in work of
over a week and a half. Relative to the sample mean, this represents a decline of nearly 3.5%. It
is possible that the negative coefficient on weeks reflects labor force exits, since the variable
includes mothers who work a fraction of the year before leaving work.
Estimates from equation (3) do not provide any evidence of a SCHIP labor supply effect.
This leaves equation (1), which focuses on only the public insurance take-up decision, to indicate
a relationship between the program and maternal work behavior. Table 2.5 reports ordinary least
squares and two-stage least squares (2SLS) estimates of equation (1). The first four columns of
Table 2.5 report results for both the labor force and full-time work participation variables. OLS
estimates for both variables are negative, reflecting the fact that public coverage status is
associated with lower rates of labor force attachment. 2SLS estimates are negative and

60

Policy variable
SCHIP threshold

Mean value

Table 2.4: Labor supply, equation (3)
LFP
FT
Hours~
-0.026
-0.030
0.086
(0.019)
(0.020)
(0.402)
[1.368]
[1.500]
[0.214]

Weeks~
-1.566*
(0.752)
[2.082]

0.806

45.356

0.650

38.050

Observations
67,016
54,513
Cluster robust standard errors in parentheses, absolute value of t-statistic in brackets
** p<0.01, * p<0.05
~Estimates conditional on positive values
Not indicated are estimates from the following control variables: family size, child less than five years in
age, white racial status, cohabitate with partner but not married, separated or divorced, high school
education, some college education, two year college degree, four year college degree, graduate degree,
age, age squared, youngest child age dummies, year dummies, and state dummies.

Public
coverage

Table 2.5: Labor supply, equation (1)
LFP
FT
Hours
Weeks
(1) OLS
(2) 2SLS
(3) OLS
(4) 2SLS
(5) OLS
(6) 2SLS
(7) OLS
(8) 2SLS
Share of children (first stage t-statistic: 2.580)
-0.219**
-0.322
-0.302**
-0.381
-3.479**
1.346
-6.962** -24.49**
(0.014)
(0.202)
(0.011)
(0.199)
(0.153)
(6.391)
(0.232)
(8.365)

Public
coverage

-0.214**
(0.013)

Variable

-0.370
(0.225)

All children (first stage t-statistic: 2.333)
-0.291**
-0.437
-3.350**
1.554
(0.010)
(0.223)
(0.144)
(7.414)

-6.780**
(0.231)

-28.27**
(10.590)

Observations
67,016
67,016
54,513
54,513
Cluster robust standard errors in parentheses
** p<0.01, * p<0.05
Not indicated are estimates from the following control variables: family size, child less than five years in
age, white racial status, cohabitate with partner but not married, separated or divorced, high school
education, some college education, two year college degree, four year college degree, graduate degree,
age, age squared, youngest child age dummies, year dummies, and state dummies.

insignificant, though the full-time work participation results are significant below the 10% level.
According to the point estimates, moving all children in a family to public coverage reduces a
mother’s likelihood of full-time work 38.1 to 43.7 percentage points. This reflects a decline of
full-time work participation of at least 58.6% relative to the sample mean.
61

Columns 5-8 of Table 2.5 reports results for the hours and weeks dependent variables.
2SLS estimates for hours are never significant and have large standard errors. Estimates for
weeks worked are significant and negative. The coefficients imply that when all children in a
family are induced into public coverage, mothers work approximately 24 to 28 fewer weeks each
year. Relative to the sample mean of just over 45 weeks, this reflects a 54% to 62% reduction in
work weeks. The significant effect of SCHIP on weeks worked may reflect job exits among
working mothers, though the large size of the coefficients and standard errors may be due to the
weak first stage association between public coverage and the eligibility threshold (Bound, Jaeger,
and Baker, 1995).

VI. Conclusion
The State Children’s Health Insurance Program is one of the largest recent reforms to the
public health insurance system in the United States, providing insurance to children living in
families with incomes above Medicaid eligibility thresholds. Because eligibility is means-tested,
some worry that SCHIP distorts the labor supply decisions of parents. In this paper, I examine
the effect of the SCHIP expansion on public insurance coverage for children and the labor supply
behavior of single mothers. First stage estimates imply that a significant fraction, nearly 40%, of
public insurance take-up associated with SCHIP is the result of private insurance crowd-out.
Labor supply estimates from equation (1) indicate little evidence of an extensive or intensive
margin response, although the program is associated with a negative effect on annual weeks
worked. These estimates should be interpreted with caution, however, as the first stage
relationship between public coverage and the threshold variable is weak. In addition, the
instrumental variables strategy used to estimate equation (1) relies on the assumption that the

62

SCHIP threshold variable affects labor supply only through the public insurance take-up
decision. If the program also affects maternal work behavior through another channel, my
instrumental variable estimates are difficult to interpret. Reduced form estimates from equation
(3) also indicate little evidence of a labor supply response among single mothers. Only the
negative coefficient on annual weeks worked is significant.
Overall, estimates from equations (1) and (3) show almost no evidence of a relationship
between SCHIP and maternal labor supply. This conclusion may reflect changes in employer
sponsored health insurance behavior among eligible families and their employers. Based on
Figure 3, a large fraction of families made eligible for public coverage because of SCHIP were
privately insured. This suggests that the potential for crowd-out with the program is higher than
with previous Medicaid expansions. Vistnes et al. (2010) attribute decreasing rates of employer
coverage between 2000 and 2008 to declines in both coverage offers from employers, which
includes offers of dependent coverage, and take-up among employees. Importantly, the authors
find that these trends correspond with large increases in employee premium costs.
There is also substantial heterogeneity in the coverage and design of employer insurance
plans. So-called “mini-med” plans, for instance, offer limited benefits for covered employees
(Levitt and Claxton, 2011). SCHIP insurance is less heterogeneous. States are required to cover
the same services as Medicaid or offer a set of services actuarially equivalent to several
benchmark plans, such as insurance offered to federal or state government employees (CBO,
2007). The CBO argues that states often elect to provide vision and dental services under
SCHIP, services not often covered by private insurance (CBO, 2007). Limited benefit plans and
increasing premium costs may have encouraged mothers to forgo employer coverage for their
children in favor of public insurance, without changing work behavior.

63

These results point to future research topics. The availability of public child health
insurance programs like SCHIP may increase the incidence of job turnover for eligible families.
This can happen if, before program implementation, parents were constrained to stay in their
current job to retain insurance coverage for their children. The availability of SCHIP may allow
parents to transition to a more preferred job without loss of insurance benefits.29 Additionally,
methods from this analysis can be adapted to a broader population of households. Results from
the health insurance literature (Lo Sasso and Buchmueller, 2004; Gruber and Simon, 2008) show
that SCHIP is an important determinant of children’s health insurance coverage for all household
types. These authors estimate high rates of private insurance crowd-out, suggesting that families
substitute employer insurance with public coverage. What these SCHIP induced insurance
coverage effects imply for household labor supply are unknown.

29

Bansak and Raphael (2008) investigate parental job lock in the context of SCHIP using 1996 – 2001 SIPP data.
They find married fathers of SCHIP eligible children, whose wives did not have employer sponsored insurance,
were 5-6% more likely to separate from their job after implementation of the program than married fathers whose
wives did have their own insurance coverage.

64

APPENDICES

65

APPENDIX A

Analysis with Additional Medicaid and SCHIP Policy Variables

66

New policy variables:
1. Parent thresholds: This variable captures the income eligibility cutoff for parental public
insurance. It is measured in terms of the income cutoff as a proportion of the FPL.
2. Waiting periods: This variable captures the length of time a child must be uninsured
before receiving SCHIP coverage, and is measured in months.
3. Premium payments: This is a dummy variable that indicates if the SCHIP program in
state s required families to pay a premium payment during year t.
4. Cost sharing payments: This is a dummy variable that indicates if the SCHIP program in
state s required families to pay co-payment/co-insurance payments for any medical
services during year t.
This analysis uses the same equations as the main analysis. In addition to the child threshold
variable I add policy variables for parental public insurance thresholds, waiting periods, premium
payments, and cost sharing payments. Since these variables vary at the state by year level, my
estimates do not include state by year interaction terms. Consequently, some coefficients will
vary from corresponding estimates in main analysis.
LSiast = αs + δt + μa + γ PublicCoverageiast + Xiβ + εiast

(1)

PublicCoverageiast = αs + δt + μa + γ1EligibilityThresholdast + γ2ParentThresholdst +
γ3WaitPeriodst + γ4Premiumst + γ5CostSharest + Xiβ + εiast
(2)
LSiast = αs + δt + μa + γ1EligibilityThresholdast + γ2ParentThresholdst + γ3WaitPeriodst +
γ4Premiumst + γ5CostSharest + Xiβ + εiast
(3)

Table 2.6: Means of additional policy variables
Variable
Mean
Parental threshold (FPL)
0.781
Waiting period (dummy)
0.463
Waiting period (months)
2.005
Premium (dummy)
0.597
Cost sharing payments (dummy)
0.376
Observations
67,016

67

SD
0.512
0.499
2.558
0.490
0.484

Variable
SCHIP Threshold
Parental Threshold
Waiting Period (months)
Waiting Period (dummy)
Premium
Cost sharing

1
0.027*
(0.013)

Table 2.7: First stage (maternal sample), equation (2)
Share of children with public coverage
2
3
4
5
6
7
0.028*
0.032*
0.035*
0.031*
0.027*
0.035*
(0.013)
(0.014)
(0.014)
(0.012)
(0.012)
(0.013)
0.004
(0.011)
-0.003
-0.003
(0.002)
(0.002)
-0.021
(0.015)
-0.020
-0.017
(0.016)
(0.016)
0.003
(0.017)

F-Statistic
4.796
2.920
R-Squared
0.1732
0.1732
Observations
Cluster robust standard errors in parentheses
** p<0.01, * p<0.05

2.950
0.1733

2.980
0.1733

68

3.830
2.490
0.1733
0.1732
67,016

2.620
0.1733

8
0.036*
(0.014)

-0.017
(0.016)
-0.013
(0.016)

2.380
0.1733

9
0.032*
(0.013)
0.003
(0.011)
-0.003
(0.002)

10
0.033*
(0.014)
0.002
(0.011)

-0.020
(0.016)
0.011
(0.017)

-0.019
(0.016)
-0.016
(0.017)
0.014
(0.017)

1.890
0.1734

1.880
0.1734

Variable
SCHIP Threshold
Parental Threshold

1
-0.025
(0.013)

2
-0.024
(0.012)
0.016
(0.012)

Waiting Period (months)
Waiting Period (dummy)

Table 2.8: Labor supply, equation (3)
Labor Force Participation
3
4
5
6
7
-0.023
-0.024*
-0.027*
-0.021
-0.025*
(0.012)
(0.012)
(0.012)
(0.011)
(0.011)

-0.001
(0.002)

8
-0.025*
(0.011)

-0.001
(0.002)
-0.002
(0.013)

Premium

0.012
(0.011)

Cost sharing

0.014
(0.010)
-0.013
(0.012)

Observations
Cluster robust standard errors in parentheses
** p<0.01, * p<0.05

67,016

69

-0.007
(0.014)
0.015
(0.011)

9
-0.021*
(0.010)
0.015
(0.011)
-0.001
(0.001)

0.019
(0.010)
-0.017
(0.012)

10
-0.022*
(0.010)
0.015
(0.011)

-0.002
(0.013)
0.019
(0.010)
-0.017
(0.012)

Variable
SCHIP Threshold
Parental Threshold

1
-0.014
(0.011)

2
-0.014
(0.010)
0.014
(0.010)

Waiting Period (months)
Waiting Period (dummy)

Table 2.9: Labor supply, equation (3)
Full time work
3
4
5
6
-0.016
-0.022
-0.017
-0.014
(0.011)
(0.012)
(0.011)
(0.010)

7
-0.018
(0.011)

0.001
(0.002)

0.001
(0.002)
0.021
(0.014)

Premium

0.012
(0.010)

Cost sharing

0.011
(0.010)
-0.002
(0.010)

Observations
Cluster robust standard errors in parentheses
** p<0.01, * p<0.05

67,016

70

8
-0.022
(0.012)

0.020
(0.016)
0.004
(0.011)

9
-0.017
(0.010)
0.015
(0.009)
0.001
(0.002)

0.013
(0.011)
-0.005
(0.011)

10
-0.020
(0.011)
0.017
(0.009)

0.024
(0.015)
0.005
(0.011)
-0.009
(0.011)

Variable
SCHIP Threshold
Parental Threshold

1
0.276
(0.182)

2
0.287
(0.178)
0.340*
(0.155)

Waiting Period (months)
Waiting Period (dummy)

Table 2.10: Labor supply, equation (3)
Usual weekly work hours
3
4
5
6
7
0.189
0.082
0.226
0.236
0.159
(0.222)
(0.234)
(0.205)
(0.209)
(0.233)

0.054
(0.038)

8
0.082
(0.240)

0.049
(0.036)
0.583*
(0.224)

Premium

0.258
(0.284)

Cost sharing

0.198
(0.267)
0.138
(0.214)

Observations
Cluster robust standard errors in parentheses
** p<0.01, * p<0.05

54,513

71

0.583**
(0.214)
-0.001
(0.251)

9
0.142
(0.243)
0.390*
(0.147)
0.056
(0.037)

0.169
(0.261)
0.083
(0.182)

10
0.084
(0.248)
0.422*
(0.164)

0.637**
(0.211)
-0.013
(0.250)
-0.013
(0.178)

Variable
SCHIP Threshold
Parental Threshold

1
-0.438
(0.223)

2
-0.452*
(0.222)
-0.413
(0.376)

Waiting Period (months)
Waiting Period (dummy)

Table 2.11: Labor supply, equation (3)
Weeks worked
3
4
5
6
-0.557*
-0.606*
-0.567*
-0.453
(0.249)
(0.269)
(0.247)
(0.280)

7
-0.647*
(0.265)

0.074
(0.044)

0.058
(0.044)
0.507
(0.272)

Premium

0.662*
(0.317)

Cost sharing

0.591
(0.308)
0.053
(0.336)

Observations
Cluster robust standard errors in parentheses
** p<0.01, * p<0.05

54,513

72

8
-0.653*
(0.275)

0.351
(0.299)
0.506
(0.339)

9
-0.605*
(0.292)
-0.380
(0.388)
0.053
(0.045)

0.661
(0.348)
-0.206
(0.338)

10
-0.607*
(0.294)
-0.378
(0.394)

0.343
(0.331)
0.589
(0.370)
-0.252
(0.343)

Variable

1

2

Share public coverage

-0.896*
(0.414)

-0.824*
(0.390)

Share public coverage

Share public coverage

Share public coverage

-0.521
(0.323)

20.038
(23.113)

-31.834
(32.573)

-0.464
(0.314)

22.718
(18.875)

-32.463
(24.891)

SCHIP Threshold
Y
Y
Parental Threshold
X
Y
Waiting Period (months)
X
x
Waiting Period (dummy)
X
x
Premium
X
x
Cost sharing
X
x
Observations
Cluster robust standard errors in parentheses
** p<0.01, * p<0.05

Table 2.12: Labor supply, equation (1)
3
4
5
6
7
Labor force participation
-.602
-.579*
-.829*
-.922*
-.630*
(0.321)
(0.294)
(0.352)
(0.419)
(0.292)

-0.502
(0.278)

-3.513
(7.960)

-0.680*
(0.289)

-11.182
(8.836)

-20.345
(12.676)

-19.942*
(8.645)

Y
x
Y
x
x
x

Y
x
x
Y
x
x

8

9

10

-.619*
(0.282)

-.663*
(0.284)

-.654*
(0.269)

-0.524*
(0.255)

-0.654*
(0.270)

-0.506*
(0.248)

-0.643*
(0.258)

Usual weekly work hours
2.470
18.034
-4.888
(12.064) (21.084)
(8.080)

-10.873
(8.843)

-1.656
(7.567)

-7.569
(7.739)

-23.086*
(11.679)

-21.279*
(8.788)

-24.417*
(12.058)

-22.594*
(9.305)

Y
x
Y
x
Y
x

Y
x
x
Y
Y
x

Y
Y
Y
x
Y
Y

Y
Y
x
Y
Y
Y

Full time work
-0.545*
-0.522
(0.274)
(0.323)

Weeks worked
-32.788
-31.385
(18.480) (30.527)
Y
x
x
x
Y
x

Y
x
x
x
x
Y
67,016

73

APPENDIX B

Additional Tables

74

Table 2.13: Health insurance coverage (maternal sample), equation (2), 1996-2002 sample years only
Variable
Public
Private
Any
Share of children
SCHIP threshold
0.069**
-0.027
0.042*
(0.024)
(0.031)
(0.016)
[2.875]
[0.871]
[2.625]
All children
SCHIP threshold

0.062*
(0.024)
[2.583]

-0.030
(0.032)
[0.938]

0.035
(0.017)
[2.059]

Observations
31,082
Cluster robust standard errors in parentheses, absolute value of t-statistic in brackets
** p<0.01, * p<0.05

Table 2.14: Group and non-group private coverage (maternal sample), 1996-2002 sample years only
Variable
Group Private
Non-Group Private
Share of children
SCHIP threshold
-0.070**
0.043**
(0.024)
(0.015)
[2.917]
[2.867]

SCHIP threshold

All children
-0.077**
(0.026)
[2.962]

Observations
31,082
Cluster robust standard errors in parentheses, absolute value of t-statistic in brackets
** p<0.01, * p<0.05

75

0.036*
(0.015)
[2.400]

Table 2.15: Labor supply, equation (3), 1996-2002 sample years only
Policy variable
LFP
FT
Hours~
SCHIP threshold
-0.040*
-0.053**
-0.178
(0.018)
(0.018)
(0.371)
[2.222]
[2.944]
[0.480]

Weeks~
-1.461
(0.770)
[1.897]

Observations
31,082
25,434
Cluster robust standard errors in parentheses, absolute value of t-statistic in brackets
** p<0.01, * p<0.05
~Estimates conditional on positive values

Table 2.16: Labor supply 2SLS only, equation (1), 1996-2002 sample years only
LFP
FT
Hours~
Weeks~
Share of children
Public
-0.583**
-0.770**
-3.030
-24.950**
coverage
(0.212)
(0.218)
(6.048)
(9.011)
Variable

Public
coverage

-0.656**
(0.233)

Observations
Cluster robust standard errors in parentheses
** p<0.01, * p<0.05
~Estimates conditional on positive values

All children
-0.866**
(0.266)
31,082

76

-3.379
(6.734)

-27.820**
(10.250)
25,434

Table 2.17: Health insurance coverage (maternal sample), equation (2), no state-year interactions
Variable
Public
Private
Any
Share of children
SCHIP threshold
0.028*
-0.008
0.019**
(0.013)
(0.014)
(0.007)
[2.154]
[0.571]
[2.714]
Mean value

0.356

0.504

0.860

All children
SCHIP threshold

0.026*
(0.012)
[2.167]

-0.011
(0.014)
[0.786]

0.016*
(0.007)
[2.286]

Mean value

0.340

0.488

0.845

Observations
67,016
Cluster robust standard errors in parentheses, absolute value of t-statistic in brackets
** p<0.01, * p<0.05

Table 2.18: Group and non-group private coverage (maternal sample), no state-year interactions
Variable
Group Private
Non-Group Private
Share of children
SCHIP threshold
-0.032**
0.024*
(0.011)
(0.009)
[2.909]
[2.667]
Mean value

SCHIP threshold

Mean value

0.417

0.086

All children
-0.035**
(0.011)
[3.181]

0.018
(0.009)
[2.000]

0.397

0.075

Observations
67,016
Cluster robust standard errors in parentheses, absolute value of t-statistic in brackets
** p<0.01, * p<0.05

77

Policy variable
SCHIP threshold

Mean value

Table 2.19: Labor supply, equation (3), no state-year interactions
LFP
FT
Hours~
-0.025
-0.014
0.276
(0.013)
(0.011)
(0.182)
[1.923]
[1.273]
[1.516]
0.806

0.650

Weeks~
-0.438
(0.223)
[1.964]

38.050

45.356

Observations
67,016
54,513
Cluster robust standard errors in parentheses, absolute value of t-statistic in brackets
** p<0.01, * p<0.05
~Estimates conditional on positive values

Table 2.20: Labor supply 2SLS only, equation (1), no state-year interactions
Variable
LFP
FT
Hours~
Weeks~
Share of children
Public
-0.896*
-0.521
20.040
-31.830
coverage
(0.414)
(0.323)
(23.110)
(32.570)

Public
coverage

-0.953*
(0.429)

Observations
Cluster robust standard errors in parentheses
** p<0.01, * p<0.05
~Estimates conditional on positive values

All children
-0.555
(0.332)
67,016

78

23.80
(30.020)

-37.810
(43.020)
54,513

Table 2.21: Health insurance coverage (maternal sample), equation (2), results with overlap category
Variable
Public
Private
Overlap
Any
Share of children
SCHIP threshold
0.080*
-0.044
0.015*
0.051**
(0.031)
(0.033)
(0.006)
(0.015)
[2.580]
[1.333]
[2.500]
[3.400]

SCHIP threshold

All children
-0.050
(0.034)
[1.471]

0.070*
(0.030)
[2.333]

0.022**
(0.008)
[2.750]

0.040*
(0.016)
[2.500]

Observations
67,016
Cluster robust standard errors in parentheses, absolute value of t-statistic in brackets
** p<0.01, * p<0.05

Table 2.22: Health insurance coverage (maternal sample), equation (2), no weights
Variable
Public
Private
Any
Share of children
SCHIP threshold
0.083*
-0.017
0.065**
(0.033)
(0.034)
(0.012)
[2.515]
[0.500]
[5.417]
All children
SCHIP threshold

0.073*
(0.032)
[2.281]

-0.026
(0.036)
[0.722]

Observations
73,855
Cluster robust standard errors in parentheses, absolute value of t-statistic in brackets
** p<0.01, * p<0.05

79

0.054**
(0.013)
[4.154]

Table 2.23: Group and non-group private coverage (maternal sample), no weights
Variable
Group Private
Non-Group Private
Share of children
SCHIP threshold
-0.070*
0.053**
(0.032)
(0.011)
[2.188]
[4.818]

SCHIP threshold

All children
-0.087*
(0.036)
[2.417]

0.036**
(0.013)
[2.769]

Observations
73,855
Cluster robust standard errors in parentheses, absolute value of t-statistic in brackets
** p<0.01, * p<0.05

Policy variable
SCHIP threshold

Table 2.24: Labor supply, equation (3), no weights
LFP
FT
Hours~
-0.017
-0.001
0.629
(0.020)
(0.028)
(0.582)
[0.085]
[0.036]
[1.081]

Weeks~
-0.753
(0.850)
[0.886]

Observations
73,855
60,032
Cluster robust standard errors in parentheses, absolute value of t-statistic in brackets
** p<0.01, * p<0.05
~Estimates conditional on positive values

80

Public
coverage

Table 2.25: Labor supply 2SLS only, equation (1), no weights
LFP
FT
Hours~
Share of children
-0.200
-0.018
10.380
(0.204)
(0.325)
(13.410)

Public
coverage

All children
-0.020
(0.367)

Variable

-0.227
(0.227)

Observations
Cluster robust standard errors in parentheses
** p<0.01, * p<0.05
~Estimates conditional on positive values

81

12.250
(16.320)

Weeks~
-12.410
(9.405)

-14.650
(11.040)

WORKS CITED

82

WORKS CITED

Bansak, C., & Raphael, S. (2008, July). The State Children's Health Insurance Program and Job
Mobility: Identifying Job Lock among Working Parents in Near-Poor Households.
Industrial and Labor Relations Review, 61(4), 564-579.
Bernard, D., & Banthin, J. (2009, March). Family Level Expenditures on Health Care and
Insurance Premiums among the Nonelderly Population, 2006. Retrieved July 24, 2014,
from http://meps.ahrq.gov/mepsweb/data_files/publications/rf29/rf29.pdf
Bound, J., Jaeger, D. A., & Baker, R. M. (1995, June). Problems with Instrumental Variables
Estimation When the Correlation between the Instruments and the Endogeneous
Explanatory Variable is Weak. Journal of the American Statistical Association, 90(430),
443-450.
Broaddus, M., Blaney, S., Dude, A., Guyer, J., Ku, L., & Peterson, J. (2001, December).
Expanding Family Coverage: States’ Medicaid Eligibility Policies for Working Families
in the Year 2000. In The Center on Budget and Policy Priorities. Retrieved January 28,
2013
Congressional Budget Office. (2007). The State Children’s Health Insurance Program.
Washington, DC: Congress of the United States.
Cutler, D. M., & Gruber, J. (1996, May). Does Public Insurance Crowd Out Private Insurance.
The Quarterly Journal of Economics, 111(2), 391-430.
Dave, D. M., Decker, S. L., Kaestner, R., & Simon, K. I. (2013, June). The Effect of Medicaid
Expansions in the Late 1980s and Early 1990s on the Labor Supply of Pregnant Women.
In The National Bureau of Economic Research. Retrieved July 1, 2013, from
http://www.nber.org/papers/w19161#navDiv=1
Gruber, J., & Simon, K. (2008). Crowd-Out 10 Years Later: Have Recent Public Insurance
Expansions Crowded Out Private Health Insurance? Journal of Health Economics, 27,
201-217.
Ham, J. C., & Shore-Sheppard, L. D. (2005a, April). Did Expanding Medicaid Affect Welfare
Participation? Industrial and Labor Relations Review, 58(3), 452-470.

83

Ham, J. C., & Shore-Sheppard, L. (2005b). The Effect of Medicaid Expansions for Low-Income
Children on Medicaid Participation and Private Insurance Coverage: Evidence from the
SIPP. Journal of Public Economics, 89, 57-83.
Hudson, J. L., Selden, T. M., & Banthin, J. S. (2005). The Impact of SCHIP on Insurance
Coverage of Children. Inquiry, 42(3), 232-254.
Ku, L., Ullman, F., & Almeida, R. (1999). What Counts? Determining Medicaid and CHIP
Eligibility for Children. In The Urban Institute. Retrieved January 28, 2013, from
http://www.urban.org/uploadedpdf/discussion99-05.PDF
Levitt, L., & Claxton, G. (2011, July 5). What is a Mini-Med Plan? In Kaiser Family Foundation.
Retrieved July 24, 2014, from http://kff.org/health-reform/perspective/what-is-a-minimed-plan/
Lo Sasso, A. T., & Buchmueller, T. C. (2004). The Effect of the State Children’s Health
Insurance Program on Health Insurance Coverage. Journal of Health Economics, 23,
1059-1082.
Meyer, B. D., & Rosenbaum, D. T. (2001, August). Welfare, the Earned Income Tax Credit, and
the Labor Supply of Single Mothers. The Quarterly Journal of Economics, 116(3), 10631114.
Miriam King, Steven Ruggles, J. Trent Alexander, Sarah Flood, Katie Genadek, Matthew B.
Schroeder, Brandon Trampe, and Rebecca Vick. Integrated Public Use Microdata Series,
Current Population Survey: Version 3.0. [Machine-readable database]. Minneapolis:
University of Minnesota, 2010.
Moffitt, R., & Wolfe, B. (1992, November). The Effect of the Medicaid Program on Welfare
Participation and Labor Supply. The Review of Economics and Statistics, 74(4), 615-626.
Montgomery, E., & Navin, J. C. (2000, July). Cross-State Variation in Medicaid Programs and
Female Labor Supply. Economic Inquiry, 38(3), 402-418.
Rosenbach, M., Ellwood, M., Czajka, J., Irvin, C., Coupe, W., & Quinn, B. (2001, January).
Implementation of the State Children’s Health Insurance Program: Momentum Is
Increasing After a Modest Start . In Mathematica Policy Research. Retrieved January 28,
2013, from http://www.mathematica-mpr.com/pdfs/schip1.pdf
Ross, D. C., & Cox, L. (2002, June). Enrolling Children and Families in Health Coverage: The
Promise of Doing More. In The Kaiser Commission on Medicaid and the Uninsured.
84

Retrieved January 28, 2013, from http://kff.org/medicaid/report/enrolling-children-andfamilies-in-health-coverage/
Ross, D. C., & Cox, L. (2003, July). Preserving Recent Progress on Health Coverage for
Children and Families: New Tensions Emerge. In The Kaiser Commission on Medicaid
and the Uninsured. Retrieved January 28, 2013, from
http://kff.org/medicaid/report/preserving-recent-progress-for-health-coverage-of/
Ross, D. C., & Cox, L. (2004, October). Beneath the Surface: Barriers Threaten to Slow Progress
on Expanding Health Coverage of Children and Families. In The Kaiser Commission on
Medicaid and the Uninsured. Retrieved January 28, 2013, from
http://kff.org/medicaid/report/beneath-the-surface-barriers-threaten-to-slow/
Ross, D. C., & Cox, L. (2005, October). In a Time of Growing Need: State Choices Influence
Health Coverage Access for Children and Families. In The Kaiser Commission on
Medicaid and the Uninsured. Retrieved January 28, 2013, from
http://kff.org/medicaid/poll-finding/in-a-time-of-growing-need-state/
Ross, D. C., & Cox, L. (2007, January). Resuming the Path to Health Coverage for Children and
Parents. In The Kaiser Commission on Medicaid and the Uninsured. Retrieved January
28, 2013, from http://kff.org/medicaid/poll-finding/resuming-the-path-to-healthcoverage-for/
Ross, D. C., Horn, A., & Marks, C. (2008a, January). Health Coverage for Children and Families
in Medicaid and SCHIP: State Efforts Face New Hurdles. In The Kaiser Commission on
Medicaid and the Uninsured. Retrieved January 28, 2013, from
http://kff.org/medicaid/report/health-coverage-for-children-and-families-in/
Ross, D. C., Horn, A., Rudowitz, R., & Marks, C. (2008b, May 14). Determining Income
Eligibility in Children’s Health Coverage Programs: How States Use Disregards in
Children’s Medicaid and SCHIP. In The Kaiser Commission on Medicaid and the
Uninsured . Retrieved January 28, 2013, from
http://kaiserfamilyfoundation.files.wordpress.com/2013/01/7776.pdf
Selden, T. M., Hudson, J. L., & Banthin, J. S. (2004). Tracking Changes in Eligibility and
Coverage among Children, 1996-2002. Health Affairs, 23(5), 39-50.
Strumpf, E. (2011). Medicaid’s Effect on Single Women’s Labor Supply: Evidence from the
Introduction of Medicaid. Journal of Health Economics, 30, 531-548.
Vistnes, J., Zawacki, A., Simon, K., & Taylor, A. (2010, September). Declines in Employer
Sponsored Coverage Between 2000 and 2008: Offers, Take-Up, Premium Contributions,
and Dependent Options. In Center for Economic Studies, U.S. Census Bureau Working
Papers. Retrieved October 1, 2013, from http://ideas.repec.org/p/cen/wpaper/10-23.html.
85

Winkler, A. E. (1991). The Incentive Effects of Medicaid on Women's Labor Supply. The
Journal of Human Resources, 26(2), 308-337.
Yelowitz, A. S. (1995, November). The Medicaid Notch, Labor Supply, and Welfare
Participation: Evidence from Eligibility Expansions. The Quarterly Journal of
Economics, 110(4), 909-939.

86

CHAPTER 3
HEALTH INSURANCE COVERAGE DURING THE GREAT RECESSION
I. Introduction
The Great Recession is an extensive period of reduced economic activity in the United
States, beginning in December 2007 and continuing until June 2009 (The National Bureau of
Economic Research, 2010). The recession led to a large increase in the unemployment rate,
peaking at 10% in October 2009 (United States Department of Labor, 2014). To date,
unemployment is still well above pre-recession levels. The economic downturn was also
associated with a large decrease in health insurance coverage among working age adults.
Approximately 20.3% of 19 to 64 year old adults were without insurance in 2009, an increase of
2.6 percentage points from 2007.30
The recession worsened an existing insurance coverage decline among adults that began
almost a decade earlier. Researchers attribute this decline to a large decrease in private coverage
(Fronstin, 2013; Holahan and Chen, 2011; Vistnes, Zawacki, Simon, and Taylor, 2010). Existing
research of the Great Recession, however, does not establish the extent to which the fall in
coverage between 2007 and 2009 is the result of the recession itself or other factors, such as the
existing secular decline in private coverage. In this paper, I investigate the degree to which
business cycle variation accounts for the insurance coverage decline among adults during the
Great Recession. I accomplish this by using state level employment information in a regression
model of insurance coverage to predict insurance trends from the Great Recession. A deviation
between predicted and actual insurance trends is evidence that insurance coverage either changed
in a way that is not fully implied by the economic downturn, or that its relationship with the

30

Based on March Current Population Survey data collected for this analysis. Adult insurance coverage trends are
shown in Table 1.2.

87

economy is different than in previous downturns.31 In a separate regression model, I directly test
whether the Great Recession had a differential effect on insurance coverage relative to past
recessions.
Results show evidence that the proportion of adults without health insurance during the
Great Recession is greater than the level predicted by regression models using state level
employment information. This finding is driven by a deviation between actual and predicted
insurance coverage trends in the years leading up to the Great Recession, suggesting that factors
other than the economic decline were important determinants of adult insurance coverage during
the recession. These results, however, are sensitive to the choice of state level employment and
time trend variables.
Empirical methods and data are discussed in section II. Results of the empirical analysis
are presented in section III. Section IV concludes.

II. Methods
II.A. Data
This analysis uses data from the IPUMS CPS, a version of the March Current Population
Survey (CPS) managed by the University of Minnesota. The extract drawn from this data set
spans survey years 1988 to 2013, and includes information on working age (19-64 years) adults.
Elderly adults are not included in this extract, since individuals in this group are likely to be
Medicare eligible. Insurance coverage variables are defined to be mutually exclusive; summing

31

In their paper, Cawley et al. (2011) seek to estimate the relationship between insurance coverage and the economy
during the Great Recession. They do not establish whether the proportion of adults without insurance coverage by
the end of the recession is due entirely to the economic downturn or some other factor.

88

across both private and public coverage equals the overall insurance category.32 Because the
March CPS collects coverage information from the year preceding the survey, these variables
measure individual insurance coverage data from 1987 to 2012. Demographic variables such as
age are lowered by one year to reflect the age concurrent with insurance information. Finally, I
use state employment information from both the Bureau of Labor Statistics Local Area
Unemployment Statistics (BLS LAUS) program and the IPUMS March CPS. BLS LAUS data
reflects employment information from the civilian non-institutionalized population at least 16
years in age (Bureau of Labor Statistics, 2014). I include March CPS employment information
in order to create a set of aggregate state employment variables based on the 19 to 64 year old
working age population.33
Table 3.1 shows means of key variables used in this analysis. At the top of this table are
state employment variables from the BLS LAUS and the working age (19 to 64 year old) adult
extract from the March CPS. For the BLS LAUS data, the mean state unemployment rate is
5.9% and state employment to population (EP) ratio is 62.7%. The mean state unemployment
rate among 19 to 64 year olds is approximately 5.8%, slightly lower than the BLS LAUS
measure. The average state EP ratio for 19 to 64 year olds also differs from the BLS LAUS,
indicating an employment rate of 74.3%. Table 3.1 also indicates health insurance coverage
means. Among the adults in the sample, 73.6%, have private insurance coverage, 17.3% have no
insurance coverage, and only 9.1% of have public insurance. A minority of the sample, 35.7%,
have a two year college degree or above. Finally, 62.4% of working age adults indicate being
currently married, and 52.2% are parents.

32

Some individuals report both public and private sources of coverage during a year. For this analysis, these
“overlap” individuals are coded as having private insurance coverage.
33
Because labor force questions in the March CPS are based on current, rather than previous year, employment
status, results that use age 19-64 state employment variables correspond to years 1988-2012.

89

Table 3.1: Means of key variables (1987-2012)
Variable
Unemployment rate (BLS LAUS)
EP ratio (BLS LAUS)
Unemployment rate (19-64 year olds)^
EP ratio (19-64 year olds)^
Private HI
Public HI
Uninsured
Family size
Female
Age
White
Black
Parent
Less than HS education
HS education
Some college education
2 year college degree
4 year college degree
Graduate degree
Married
Full-time employment
Self-employed
Private sector work
Public sector work

Mean
5.940
62.716
5.765
74.303
0.736
0.091
0.173
3.066
0.520
39.769
0.823
0.106
0.522
0.186
0.276
0.182
0.091
0.179
0.087
0.624
0.672
0.082
0.593
0.130

Observations

2,684,632

^ 1988-2012 data

Table 3.2 presents insurance coverage trends for working age adults during the Great
Recession. The proportion of uninsured adults grew 2.6 percentage points between 2007 and
2009, a 14.7% increase relative to the 2007 mean. The loss in overall insurance coverage is
driven by a 4.2 percentage point decline in private coverage over this time period. Private
coverage losses are largest among adults with low educational attainment or who are not
currently married. Gains in public coverage offset only a fraction of the decline in private
coverage, with an overall increase of 1.6 percentage points. The largest gains occur among
90

adults who are young and have low educational attainment. Consequently, declines in private
coverage for adults translate to large increases in uninsured status. These results are similar to
insurance coverage trends reported elsewhere in the health insurance literature (Fronstin, 2013;
Holahan and Chen, 2011).
Table 3.2: Insurance coverage trends during the Great Recession, 19-64 adults
Group
2007 % of sample
2007 mean
2007 - 2009 change
Uninsured
Overall
1.000
0.177
0.026
Education
Less than HS
0.115
0.406
0.027
HS/some college
0.496
0.197
0.034
2-4 year degree
0.292
0.094
0.018
Graduate degree
0.097
0.050
0.007
Age
19-26
0.163
0.293
0.020
27-54
0.677
0.164
0.030
55-64
0.160
0.112
0.015
Marital status
Married
0.614
0.122
0.017
Not married
0.386
0.264
0.034
Private coverage
Overall
1.000
0.727
-0.042
Education
Less than HS
0.115
0.374
-0.048
HS/some college
0.496
0.694
-0.055
2-4 year degree
0.292
0.857
-0.029
Graduate degree
0.097
0.922
-0.013
Age
19-26
0.163
0.598
-0.042
27-54
0.677
0.755
-0.045
55-64
0.160
0.741
-0.028
Marital status
Married
0.614
0.810
-0.030
Not married
0.386
0.595
-0.050
Public coverage
Overall
1.000
0.096
0.016
Education
Less than HS
0.115
0.220
0.021
HS/some college
0.496
0.109
0.021

91

Table 3.2 (cont’d)
2-4 year degree
Graduate degree

0.292
0.097

0.048
0.029

0.010
0.005

0.163
0.677
0.160

0.109
0.081
0.146

0.022
0.015
0.013

0.614
0.386

0.068
0.141

0.014
0.016

Age
19-26
27-54
55-64
Marital status
Married
Not married

II.B. Empirical methodology
The insurance coverage model is specified as follows (where i denotes an individual
observation, s state, and t year):
Yist = αs + δ1t + δ2t2 + Xistβ + Employmentstγ + εist

(1)

Yist is an indicator variable for one of three insurance coverage categories: uninsured status,
public coverage, and private coverage. Employmentst is a vector of employment variables meant
to capture the effect of business cycle variation on insurance coverage. These variables reflect
information on either the annual unemployment rate or EP ratio in state s during year t.34 This
approach follows methods from other researchers who estimate the effect of business cycle
variation on insurance coverage (Cawley et al., 2011; Glied and Jack, 2003). These authors use
only contemporaneous state unemployment rates. Equation (1) allows additional flexibility in
modeling the effect of business cycle variation on insurance coverage, including a quadratic
employment variable and employment lags.
Equation (1) includes right hand side variables in addition to the employment
information. Xist is a vector of demographic controls including information on parent status

34

The unemployment rate does not account for discouraged workers who leave the labor force, perhaps as a result of
long term unemployment. In case the unemployment rate understates the true effect of the recession on
employment, I include state employment to population ratios in my analysis.

92

(whether the individual has a child living in the household), gender, age, family size, race,
education, and marital status. Also included are state fixed effect indicators, αs, and linear and
quadratic time trends, t and t2. In case the model is sensitive to the specification of the time
trend, I present results from models with only a linear time trend or with a full set of state
specific linear time trends. Finally, standard errors are clustered at the state level.
I also directly test whether the Great Recession has a differential effect on insurance
coverage, relative to other years. I use the following equation:
Yist = αs + δ1t + δ2t2 + Xistβ + γ1Employmentst + λ1GreatRecessiont +
λ2Employmentst*GreatRecessiont + εist

(2)

Here, γ1 captures the effect of the contemporaneous employment information in years outside of
the Great Recession period of 2007 to 2009. GreatRecessiont is an indicator variable equal to
one for observations belonging to years 2007, 2008, or 2009. λ2 is the main coefficient of
interest, indicating whether the effect of the employment variable during the recent recession is
consistent with other time periods. If the relationship between the economy and insurance
coverage during the Great Recession is similar to past downturns, then λ2 should not be
statistically significant. The remaining variables in equation (2) are defined as in equation (1).

III. Results
In order to determine whether observed insurance trends during the Great Recession are
consistent with past recessions, I estimate equation (1) using data in the years before 2007.35 I

35

Employment variable coefficients from equation (1) are shown in Table 3.8.

93

Figure 3.1: Proportion uninsured, unemployment rate (BLS LAUS)

Figure 3.2: Proportion uninsured, EP ratio (BLS LAUS)

94

use the resulting estimates to create out of sample health insurance predictions from 2007 to
2012. Figures 3.1 and 3.2, which use BLS LAUS information on state unemployment rates and
EP ratios, respectively, display trends in the actual and predicted uninsured rates among working
age adults. Predicted trends in Figure 3.1 are sensitive to the specification of the employment
variable.36 The specification including both dynamic and quadratic terms, denoted “All”,
predicts a substantial increase in the proportion of uninsured adults during the Great Recession.
This increase is large relative to the actual trend and remaining predicted trends, although the
“Square term” trend also over-predicts the loss in insurance coverage. Results using state EP
ratios, indicated in Figure 3.2, are more consistent across specifications. Predictions based on EP
ratio information appear to accurately predict the magnitude of the increase in uninsured status
during the Great Recession.
Both Figures 3.1 and 3.2 show a predicted proportion of uninsured adults that is less than
the actual amount during the recession. This suggests that the regression model, which generally
captures the change in insurance coverage during the Great Recession, under-predicts the level of
adults without coverage. This result is due to a divergence between actual and predicted trends
in the period before 2007. Throughout the decade, the actual uninsured rate among working age
adults is increasing. Meanwhile, the predicted uninsured rate reverses its upward trend in the
middle of the decade, in approximately 2004 or 2005. One possible reason for this result is the
secular decline in private insurance coverage. Figures 3.3 and 3.4 indicate that private coverage
is decreasing throughout the decade, despite the model predicting a large increase in private
coverage in the years immediately prior to the Great Recession. This is unlike the previous

36

Figures 1 to 4 contain 3 different predicted trend lines, each representing a different specification of the
employment variable in equation (1). “Square term” uses both a contemporaneous and quadratic employment
variable. “1yr Int Dynamic” includes a contemporaneous term, 1 year lag, and an interaction of both variables.
Finally, “All” includes employment variables from the “1yr Int Dynamic” specification as well as a quadratic.

95

Figure 3.3: Private coverage, unemployment rate (BLS LAUS)

Figure 3.4: Private coverage, EP ratio (BLS LAUS)

96

Figure 3.5: 1988-2012 Proportion uninsured, unemployment rate (19-64 year old employment variable)

Figure 3.6: 1988-2012 Proportion uninsured, EP ratio (19-64 year old employment variable)

97

period of economic recovery during the middle and late 1990s, where private coverage shows a
strong upward trend. Instead, private coverage during the middle 2000s declines despite the
improvement in the economy.
Figures 3.5 and 3.6, which display results from equation (1) using 19 to 64 year old
employment data, show only small differences with the BLS LAUS estimates. Predicted trends
in these figures are based on the “All” specification of the employment variable. In Figure 3.5,
the predicted increase in uninsured status during the Great Recession is smaller when using 1964 year old state unemployment data than with the BLS LAUS information. This increase is
now more in line with the actual uninsured trend. Figure 3.6 displays a similar result, with a
slightly smaller predicted increase in the uninsured rate when using the 19 to 64 year old state EP
ratio.
Figures 3.7 and 3.8 show results from equation (1) with varying specifications of the time
trend variable.37 Results from this exercise indicate that predictions from the main analysis are
sensitive to the type of time trend included in the model. Results from the main analysis use
both a linear and quadratic time trend. The two additional time trend specifications in Figures
3.7 and 3.8 are a linear and state specific linear time trend. In both figures, models using the
linear or state specific trends produce similar predictions of the uninsured rate. However, these
results show a predicted proportion of uninsured adults that is greater than the actual level during
the Great Recession. This is the opposite of the finding from the main analysis. Unlike results
from the main analysis, predicted rates of uninsurance in the years just before 2007 do not
decline. Instead, the predicted rate from equation (1) using either of the alternative trends
appears to increase throughout the decade. The sensitivity of the predictions to different time

37

As with Figures 3.5 and 3.6, predicted trends in these figures are based on the “All” specification of the
employment variable.

98

Figure 3.7: Proportion uninsured trend comparison, unemployment rate (BLS LAUS)

Figure 3.8: Proportion uninsured trend comparison, EP ratio (BLS LAUS)

99

trend specifications implies that the model fails to control for omitted variables which also affect
insurance coverage.
Finally, Table 3.3 displays estimates from equation (2) using BLS LAUS employment
data. This equation is intended to evaluate whether the relationship between the economy and
insurance coverage is different during the Great Recession, relative to past recessions. As with
the predicted insurance trends, results depend on the choice of employment variable. EP ratio
results, presented in even numbered columns, show no evidence that the relationship between
employment and insurance is different during the Great Recession. Results based on state
unemployment information are shown in odd numbered columns. Uninsured status estimates,
shown in column 1, indicate that the effect of the unemployment variable between 2007 and
2009 is smaller than in other years. This appears to match trends from Figure 3.1, where some of
the specifications indicate that the actual increase in uninsured status during the Great Recession
is smaller than predicted. Private and public coverage estimates show similar results. The
coefficient on the interaction term in column 3, for instance, suggests that negative relationship
between private coverage and unemployment is smaller in magnitude during the Great
Recession. Results from Table 3.4, which uses 19 to 64 year old employment data, are similar.

IV. Conclusion
The Great Recession is associated with large declines in employment and insurance
coverage. Between 2007 and 2009, working age adults experienced a 14.7% increase in
uninsured status relative to the mean number of uninsured in 2007. In this paper, I evaluate the
degree to which the increase in the number of uninsured working age adults is explained by the
economic downturn. Using a regression model to estimate the relationship between employment

100

Table 3.3: Great Recession and insurance coverage (BLS LAUS employment variables)
Variable
Uninsured
Private coverage
Public coverage
1
2
3
4
5
6
07-09 indicator
0.0094*
0.0161
-0.0269***
-0.0243
0.0176***
0.0082
(0.0048)
(0.0293)
(0.0069)
(0.0318)
(0.0035)
(0.0201)
Unemployment rate 0.0077***
-0.0135***
0.0058***
(0.0007)
(0.0009)
(0.0005)
EP ratio
-0.0060***
0.0010***
-0.0040***
(0.0008)
(0.0009)
(0.0005)
Interaction
-0.0016**
-0.0002
0.0040***
0.0003
-0.0024***
-0.0001
(0.0006)
(0.0005)
(0.0008)
(0.0005)
(0.0005)
(0.0003)
R-squared
0.108
0.108
Observations
Cluster robust standard errors in parentheses.
*** p<0.01, ** p<0.05, * p<0.1

0.166
0.166
2,684,632

0.068

0.068

Table 3.4: Great Recession and insurance coverage (1988-2012; 19-64 year old employment variables)
Variable
Uninsured
Private coverage
Public coverage
1
2
3
4
5
6
07-09 indicator
0.0108**
0.0143
-0.0264***
-0.0159
0.0156**
0.0016
(0.0046)
(0.0257)
(0.0064)
(0.0327)
(0.0031)
(0.0237)
Unemployment rate 0.0070***
-0.0117***
0.0047***
(0.0006)
(0.0008)
(0.0004)
EP ratio
-0.0050***
0.0086***
-0.0036***
(0.0006)
(0.0004)
Interaction
-0.0013**
-0.0002
0.0033***
0.0001
-0.0019***
0.0000
(0.0006)
(0.0004)
(0.0008)
(0.0004)
(0.0005)
(0.0003)
R-squared
0.110
0.109
Observations
Cluster robust standard errors in parentheses.
*** p<0.01, ** p<0.05, * p<0.1

0.168
0.168
2,593,396

0.069

0.069

and insurance coverage in previous recessions, I create out of sample predictions of insurance
coverage between 2007 and 2012. I find evidence that the proportion of adults without insurance
coverage during the recession is higher than implied by predictions from the regression model.
This result is sensitive to the specification of the time trend variable, however. Estimates from
101

regression models which include linear or state specific linear time trends suggest that level of
uninsured adults during the Great Recession is lower than the predicted.
Overall, evidence from the main analysis indicates that at least some of the decline in
insurance coverage during the Great Recession is due to factors other than the economic
downturn. A possible explanation for this finding lies with the secular decline in private
insurance coverage, which predates the Great Recession. In addition, I find little evidence for an
alternative explanation of the main results discussed above. Predicted insurance trends from
equation (1), as well as estimates from equation (2), suggest that the relationship between adult
insurance coverage and the economy during the Great Recession is consistent with past
downturns. This conclusion depends on the choice of employment variable. In both equations,
state unemployment rate data results in an over-prediction of the increase in uninsured status.
Unlike the EP ratio results, however, estimates based on unemployment rate data are sensitive to
the specification of the employment variable.
Among adults, there are several factors contributing to the decline in private and overall
insurance coverage levels. Vistnes et al. (2010) find that decreasing rates of employer coverage
in the 2000s is the result of a decline in both coverage offers from employers and take-up among
employees. The authors also find that these trends correspond with rising employee premiums.
In addition, there were few policy changes during this time period to help preserve insurance
coverage among working age adults. The State Children’s Health Insurance Program (CHIP),
implemented in all states by 2000, acted as an important health insurance safety net for children
throughout the 2000s and in the Great Recession. For adults, however, CHIP is associated with
only a limited expansion of parental public insurance eligibility in a small number of states.
Subsequent health insurance reforms, such as the Affordable Care Act (ACA), should help

102

preserve health insurance coverage among working age adults. The ACA expands adult
eligibility for Medicaid and provides subsidized private insurance coverage. In the years to
come, this should reduce the impact of declining employer insurance coverage and lessen the
sensitivity of adult health insurance coverage to variations in the business cycle.

103

APPENDIX

104

Table 3.5: Great Recession, proportion of adults unemployed at least 1 week in year
Group
2007 % of sample
2007 mean
2007 - 2009 change
Overall
1.000
0.060
0.035
Education
Less than HS
0.115
0.088
0.043
HS/some college
0.496
0.069
0.039
2-4 year degree
0.292
0.046
0.030
Graduate degree
0.097
0.030
0.019
Age
19-26
0.163
0.105
0.039
27-54
0.677
0.056
0.037
55-64
0.160
0.033
0.020
Marital status
Married
0.614
0.044
0.033
Not married
0.386
0.086
0.035

Table 3.6: Trends in main employment variables over past decade
Variable
2002
2007
Unemployment rate (BLS LAUS)
5.567
4.477
Unemployment rate (19-64 year olds)
5.437
4.122
EP ratio (BLS LAUS)
63.589
63.915
EP ratio (19-64 year olds)
75.617
75.930

105

2012
7.728
7.807
59.633
71.319

Group
Overall
Education
Less than HS
HS/some college
2-4 year degree
Graduate degree
Age
19-26
27-54
55-64
Marital status
Married
Not married

Table 3.7: Great Recession, parental insurance coverage trends
2007 % of sample
2007 mean
2007 - 2009 change
Uninsured
1.000
0.150
0.016
0.118
0.467
0.310
0.106

0.402
0.166
0.069
0.036

0.017
0.025
0.010
0.002

0.081
0.838
0.081

0.274
0.140
0.130

0.004
0.017
0.022

0.808
0.192

Overall
Education
Less than HS
HS/some college
2-4 year degree
Graduate degree
Age
19-26
27-54
55-64
Marital status
Married
Not married

1.000

Overall
Education
Less than HS
HS/some college
2-4 year degree
Graduate degree
Age
19-26
27-54
55-64
Marital status
Married

1.000

0.126
0.250
Private coverage
0.761

0.013
0.022
-0.035

0.118
0.467
0.310
0.106

0.401
0.729
0.886
0.942

-0.053
-0.050
-0.018
-0.008

0.081
0.838
0.081

0.524
0.786
0.739

-0.063
-0.034
-0.029

0.808
0.192

0.813
0.545
Public coverage
0.089

-0.030
-0.044
0.019

0.118
0.467
0.310
0.106

0.197
0.105
0.045
0.022

0.036
0.026
0.008
0.006

0.081
0.838
0.081

0.202
0.074
0.131

0.059
0.016
0.007

0.808

0.061

0.017

106

Table 3.7 (cont’d)
Not married

0.192

0.205

0.021

Table 3.8: 19-64 Adults results from equation (1) used for prediction, uninsured (1987-2006)
Variable
Uninsured
1
2
3
4
5
Employment variable: unemployment rate (BLS LAUS)
Current year
0.0073*** 0.0112*** 0.0062*** 0.0083***
0.0044
(0.0006)
(0.0036)
(0.0008)
(0.0018)
(0.0035)
Current year, squared
-0.0003
0.0004*
(0.0002)
(0.0002)
1 year lag
0.0011
0.0061*** 0.0094***
(0.0009)
(0.0017)
(0.0013)
2 year lag
0.0012
(0.0008)
Interaction (current and 1 year lag)
-0.0005* -0.0009***
(0.0003)
(0.0002)
R-squared

Current year
Current year, squared
1 year lag
2 year lag
Interaction (current and 1 year lag)

0.108

0.108

0.108

0.108

0.108

Employment variable: EP ratio (BLS LAUS)
-0.0061***
-0.0114
-0.0054***
-0.0071
-0.0364***
(0.0007)
(0.0101)
(0.0008)
(0.0052)
(0.0106)
0.0000
0.0005***
(0.0001)
(0.0002)
-0.0020**
-0.0005
0.0282***
(0.0008)
(0.0053)
(0.0103)
0.0019**
(0.0008)
0.0000
-0.0005***
(0.0001)
(0.0002)

R-squared
0.108
Observations
Cluster robust standard errors in parentheses.
*** p<0.01, ** p<0.05, * p<0.1

0.108

107

0.108
2,684,632

0.108

0.108

Table 3.9: Great Recession and insurance coverage (full set of results from Table 3.3)
Variable
Uninsured
Private coverage
Public coverage
1
2
3
4
5
6
07-09 indicator
0.0094*
0.0161
-0.0269***
-0.0243
0.0176***
0.0082
(0.0048)
(0.0293)
(0.0069)
(0.0318)
(0.0035)
(0.0201)
Unemployment rate 0.0077***
-0.0135***
0.0058***
(0.0007)
(0.0009)
(0.0005)
EP ratio
-0.0060***
0.0010***
-0.0040***
(0.0008)
(0.0009)
(0.0005)
Interaction
-0.0016**
-0.0002
0.0040***
0.0003
-0.0024***
-0.0001
(0.0006)
(0.0005)
(0.0008)
(0.0005)
(0.0005)
(0.0003)
Parent
-0.0455*** -0.0456***
-0.0004
-0.0004
0.0460*** 0.0460***
(0.0039)
(0.0039)
(0.0030)
(0.0030)
(0.0035)
(0.0035)
Family size
0.0129*** 0.0129*** -0.0136*** -0.0136***
0.0007
0.0007
(0.0017)
(0.0017)
(0.0020)
(0.0020)
(0.0006)
(0.0006)
Female
-0.0264*** -0.0265*** 0.0028**
0.0029**
0.0236*** 0.0236***
(0.0015)
(0.0015)
(0.0012)
(0.0012)
(0.0016)
(0.0016)
Age
-0.0005
-0.0005
0.0102*** 0.0102*** -0.0097*** -0.0097***
(0.0004)
(0.0004)
(0.0007)
(0.0007)
(0.0005)
(0.0005)
Age squared
-0.0000*** -0.0000*** -0.0001*** -0.0001*** 0.0001*** 0.0001***
(0.0000)
(0.0000)
(0.0000)
(0.0000)
(0.0000)
(0.0000)
White
-0.0447*** -0.0447*** 0.0663*** 0.0663*** -0.0216*** -0.0216***
(0.0126)
(0.0126)
(0.0176)
(0.0177)
(0.0073)
(0.0074)
Black
-0.0333*** -0.0334***
-0.0178
-0.0174
0.0510*** 0.0509***
(0.0095)
(0.0095)
(0.0163)
(0.0164)
(0.0088)
(0.0088)
HS education
-0.1140*** -0.1130*** 0.1880*** 0.1850*** -0.0738*** -0.0728***
(0.0112)
(0.0113)
(0.0089)
(0.0090)
(0.0057)
(0.0056)
Some college
-0.1840*** -0.1830*** 0.2830*** 0.2810*** -0.0986*** -0.0978***
(0.0148)
(0.0148)
(0.0132)
(0.0132)
(0.0068)
(0.0067)
2 year degree
-0.1930*** -0.1920*** 0.3010*** 0.2990*** -0.1080*** -0.1070***
(0.0144)
(0.0144)
(0.0125)
(0.0126)
(0.0076)
(0.0075)
4 year degree
-0.2250*** -0.2250*** 0.3600*** 0.3580*** -0.1340*** -0.1340***
(0.0163)
(0.0164)
(0.0143)
(0.0144)
(0.0076)
(0.0076)
Graduate degree
-0.2390*** -0.2380*** 0.3780*** 0.3770*** -0.1400*** -0.1390***
(0.0167)
(0.0168)
(0.0151)
(0.0152)
(0.0080)
(0.0080)
Married
-0.1020*** -0.1020*** 0.1810*** 0.1820*** -0.0797*** -0.0798***
(0.0034)
(0.0034)
(0.0038)
(0.0038)
(0.0051)
(0.0051)
Time trend
0.0118*** 0.0119*** -0.0157*** -0.0156*** 0.0039*** 0.0037***
(0.0007)
(0.0008)
(0.0008)
(0.0009)
(0.0005)
(0.0006)
Time trend squared -0.0003*** -0.0003*** 0.0004*** 0.0004*** -0.0001*** -0.0001***
(0.0000)
(0.0000)
(0.0000)
(0.0000)
(0.0000)
(0.0000)
R-squared
0.108
0.108
0.166
0.166
0.068
0.068
Observations
2,684,632

108

Table 3.9 (cont’d)
Cluster robust standard errors in parentheses.
*** p<0.01, ** p<0.05, * p<0.1

109

Figure 3.9: Public coverage, unemployment rate (BLS LAUS)

Figure 3.10: Public coverage, EP ratio (BLS LAUS)

110

WORKS CITED

111

WORKS CITED

Bureau of Labor Statistics. (2014, April 18). Civilian Noninstitutional Population and Associated
Rate and Ratio Measures for Model-Based Areas. In Local Area Unemployment
Statistics. Retrieved July 28, 2014, from http://www.bls.gov/lau/rdscnp16.htm
Cawley, J., Moriya, A. S., & Simon, K. I. (2011, November). The Impact of the Macroeconomy
on Health Insurance Coverage: Evidence from the Great Recession. In NBER Working
Paper Series. Retrieved April 29, 2014, from http://www.nber.org/papers/w17600
Congressional Budget Office. (2007). The State Children’s Health Insurance Program.
Washington, DC: Congress of the United States.
DeNavas-Walt, C., Proctor, B. D., & Smith, J. (2007, August). Income, Poverty, and Health
Insurance Coverage: 2006. In United States Census Bureau. Retrieved April 29, 2014,
from http://www.census.gov/hhes/www/hlthins/data/incpovhlth/2006/index.html
DeNavas-Walt, C., Proctor, B. D., & Smith, J. C. (2008, August). Income, Poverty, and Health
Insurance Coverage: 2007. In United States Census Bureau. Retrieved April 29, 2014,
from http://www.census.gov/hhes/www/hlthins/data/incpovhlth/2007/index.html
DeNavas-Walt, C., Proctor, B. D., & Smith, J. C. (2009, September). Income, Poverty, and
Health Insurance Coverage: 2008. In United States Census Bureau. Retrieved April 29,
2014, from http://www.census.gov/hhes/www/hlthins/data/incpovhlth/2008/index.html
DeNavas-Walt, C., Proctor, B. D., & Smith, J. C. (2010, September). Income, Poverty, and
Health Insurance Coverage: 2009. In United States Census Bureau. Retrieved April 29,
2014, from http://www.census.gov/hhes/www/hlthins/data/incpovhlth/2009/index.html
DeNavas-Walt, C., Proctor, B. D., & Smith, J. C. (2011, September). Income, Poverty, and
Health Insurance Coverage: 2010. In United States Census Bureau. Retrieved April 29,
2014, from http://www.census.gov/hhes/www/hlthins/data/incpovhlth/2010/index.html
DeNavas-Walt, C., Proctor, B. D., & Smith, J. C. (2012, September). Income, Poverty, and
Health Insurance Coverage: 2011. In United States Census Bureau. Retrieved April 29,
2014, from http://www.census.gov/hhes/www/hlthins/data/incpovhlth/2011/index.html
DeNavas-Walt, C., Proctor, B. D., & Smith, J. C. (2013, September). Income, Poverty, and
Health Insurance Coverage: 2012. In United States Census Bureau. Retrieved April 29,
2014, from http://www.census.gov/hhes/www/hlthins/data/incpovhlth/2012/index.html
Elsby, M. W.L., Hobijn, B., Sahin, A., Katz, L. F., & Shimer, R. (2010). The Labor Market in
the Great Recession. Brookings Papers on Economic Activity, 1-69.

112

Fronstin, P. (2013, September). Sources of Health Insurance and Characteristics of the
Uninsured: Analysis of the March 2013 Current Population Survey. In Employee Benefit
Research Institute. Retrieved April 29, 2014, from
http://www.ebri.org/pdf/briefspdf/EBRI_IB_09-13.No390.Sources.pdf
Glied, S., & Jack, K. (2003, October). Macroeconomic Conditions, Health Care Costs, and the
Distribution of Health Insurance Coverage. In NBER Working Paper Series. Retrieved
June 4, 2014, from http://www.nber.org/papers/w10029
Holahan, J., & Chen, V. (2011, December 1). Changes in Health Insurance Coverage in the Great
Recession, 2007-2010. In Kaiser Family Foundation. Retrieved April 29, 2014, from
http://kff.org/medicaid/issue-brief/changes-in-health-insurance-coverage-in-the
The National Bureau of Economic Research. (2010, September 20). US Business Cycle
Expansions and Contractions. Retrieved April 29, 2014, from
http://www.nber.org/cycles/cyclesmain.html
United States Department of Labor, Bureau of Labor Statistics. (2014). Labor Force Statistics
from the Current Population Survey. In Economic Releases. Retrieved April 29, 2014,
from http://data.bls.gov/cgi-bin/surveymost
Vistnes, J., Zawacki, A., Simon, K., & Taylor, A. (2010, September). Declines in Employer
Sponsored Coverage Between 2000 and 2008: Offers, Take-Up, Premium Contributions,
and Dependent Options. In Center for Economic Studies, U.S. Census Bureau Working
Papers. Retrieved April 29, 2014, from http://ideas.repec.org/p/cen/wpaper/10-23.html

113