Health insurance and health the effect of medicaid eligibility expansions on parent health outcomes

AbstractThis paper uses exogenous variation in Medicaid eligibility laws to evaluatethe effect of Medicaid eligibility expansions on the self-reported health of low-income parents in exp

WILLIAMS COLLEGE LIBRARIESYour unpublished thesis, submitted for a degree at Williams College and administered by theWilliams College Libraries, will be made available for research use You may, through this form,provide instructions regarding copyright, access, dissemination and reproduction of your thesis._ The faculty advisor to the student writing the thesis wishes to claim joint authorship

in this work

Ineach section, please check theONEstatement that reflects your wishes

I.PUBLICATION AND QUOTATION: LITERARY PROPERTY RIGHTS

A student author automatically owns the copyright to his/her work, whether or not a copyright symbol and date are placed on the piece The duration of U.S copyright on a manuscript and Williams theses are considered manuscripts is the life of the author plus 70 years.

_ I/we do not choose to retain literary property rights to the thesis, and I wish to assignthem immediately to Williams College

Selecting this option will assign copyright to the College This in no \vay precludes a student author from later publishing his/her work; the student would, however, need to contact the

Archives for a permission form The Archives would be free in this case to also grant perml',slcm

to another researcher to publish small sections from the thesis Rarely would there be any reason for the Archives to gram permission to another party to publish the thesis in its entirely; if such a situation arose, the Archives would be in touch with the author to let them know that such a request had been made.

_I/we wish to retain literary property rights to the thesis for a period of three years, atwhich time the literary property rights shall be assigned to Williams College

Selecting this option gi ves the author a few years to make exclusi ve use of the thesis in up-coming projects: articles, later research, etc.

, ,/I/we wish to retain literary property rights to the thesis for a period of 1-7 years, oruntil my death, whichever is the later, at which time the literary property rights shall beassigned to Williams College

Selecting this option allows the author great flexibility in extending or shortening the time of his/her automatic copyright period Some students are interested in using their thesis in graduate school work In this case it would make sense for them to enter a number such as ' l O i n the blank, and line out the words 'or until my death, whichever is the later,' In any event it is easier for the Archives to administer copyright on a manuscript if the period ends with the individual's death our staff won't have to search for estate executors in this case but this is entirely up to each student.

II.ACCESS

The Williams College Libraries are investigating the posting of theses online, as well as their retention in hardcopy.

~illiamsCollege is grantedpermi~sionto maintain and provide access to my thesis

in hardcopy and via the Web both on and off campus

Selecting this option allows researchers around the world to access the digital version of your work.

Williams College is granted permission to maintain and provide access to my thesis

in hardcopy and via the Web for on-campus use only

Selecting this a]]ows access to the version of your work from the on-campus

network only.

_ The thesis is to be maintained and made available in hardcopy form only

Selecting this option allows access to your work only from the hardcopy you submit Such access

pertains to the of your work including any media that it comprises or includes.

III.COPYING AND DISSEMINATION

Because theses are listed on FRA.~CIS, the Libraries receive numerous requests every year for copies of

works If/when a hardcopy thesis is duplicated for a researcher, a copy of the release form always

accompanies the copy Any digital version of your thesis will include the release form.

~Piesof the thesis may be provided to any researcher

Selecting this allows any researcher to request a copy from the Williams College Libraries.

or to make one from an electronic version.

_ Copying of the thesis is restricted for _ years, at which time copies may be provided to

any researcher

This option allows the author to set a time limit on restrictions During this period an

electronic version of thesis vvili be protected against duplication.

Copying of the thesis or portions thereof, except as needed to maintain an adequate

number of research copies available in the Williams College Libraries, is expressly

prohibited The electronic version of the thesis will be protected against duplication

Selecting this option allows no reproductions to be made for researchers The electronic version

of the thesis will be protected against duplication This does not dis-allow researchers from

reading/viewing the work in either hardcopy or digital form.

Signed (student author) _

Signed (faculty advisor)_

Signatures Removed

Signature Removed

Thesis title +-\E';Ll\-\ Ij\,S~ Li~ t j (J;' N-J D \-\5:: J.\lru: T~t;:" EF~T ()i=

Date S! F? loti

Accepted for the Libraries

Date accepted _ _ -'S " r'-_· / ::.t.; -,Gb'\,.l '-(ci _I _

HEALTH INSURANCE AND HEALTH: THE EFFECT OF MEDICAIDELIGIBILITY EXPANSIONS ON PARENT HEALTH OUTCOMES

byJessica Beck

Sara LaLumia, Advisor

A thesis submitted in partial fulfillment

of the requirements for theDegree of Bachelor of Arts with Honors

in Economics

WILLIAMS COLLEGEWilliamstown, MA

May 19,2008

AbstractThis paper uses exogenous variation in Medicaid eligibility laws to evaluatethe effect of Medicaid eligibility expansions on the self-reported health of low-

income parents in expansion states between 1994 and 2006 I find that Medicaideligibility expansions are positively correlated with parental take-up rates althoughthey have no effect on parental health outcomes I measure this relationship for fourdifferent parent samples; mothers, parents, poor health parents, and parents withincome below 130% of the federal poverty line The insignificant relationship

between eligibility expansions and health holds across all specifications except forthe sickest parents For this population I find that the eligibility expansions reducethe probability of being in poor health by 0.7 to 0.9 percentage points, or by 35-45%

TABLE OF CONTENTS

Acknowledgements

Introduction

Background and Literature Review

A Medicaid and Medicaid Eligibility Expansions

B. Literature Review

B.l Eligibility and Insurance Coverage

B.2 Eligibility and Health

B.3 Measures of Health: Self-Reported Health Status

Methodology and Data

99

11 11

1213151521

252929

3539

48

5264

67 68

71

Despite improvements in public health insurance coverage, in 2006 47

million people below age 65, or about 15.8% of the non-elderly population, wereuninsured The majority of this population (79%) is adults below 300% of the federalpoverty line (Holahan 2007) Public policy and health care reform activists havelong been concerned with improving health insurance coverage At the foundation ofthese issues is the assumption that offering coverage to the uninsured will improvetheir health by increasing access to quality medical care Health insurance eligibilityexpansions, the policy mechanism for improving coverage, are often motivated bythe desire to increase medical care utilization and health among the poor However,economists have not conclusively determined if there is a relationship between

insurance coverage and health Eligibility expansions can lead to improvements incoverage, but this does not necessarily increase utilization In addition, even if

eligibility expands, this does not ensure that the uninsured actually take up insurancecoverage

There is some reason to believe that extending insurance to the uninsured canimprove health Relative to individuals with private insurance or Medicaid, the

uninsured population reports less access to primary care and preventive services.According to a survey by the Kaiser Association, in 2004 44% of the uninsured didnot have a usual source of care In contrast, only 18% of individuals with insurancereport lacking a routine care site Additionally, the uninsured are also more likely todelay or forgo care, allowing conditions to get worse; 51%report not having seen adoctor in the past year Individuals with insurance coverage, through Medicaid or

private providers, report higher levels of access and use regardless of provider Incontrast to the uninsured, 18% of those with Medicaid and 17% of individuals withprivate insurance report not having a routine care source and 27% of individuals inboth plans report not having seen a doctor in the past year (Holohan 2007) Thesimilarity in these numbers and the sharp contrast between the behaviors of the

insured and uninsured suggest that expanding coverage is a viable way to improveaccess to care, and subsequently, health I test the hypothesis that Medicaid eligibilityexpansions will improve the health of the newly eligible

Economists have previously studied the effect of Medicaid insurance

expansion on insurance take-up and coverage The majority of the work in this field

is concerned with the effect of expansions on children's health insurance coverageand only a small number of papers focus on the coverage of non-elderly poor adults

In studies that do focus on adult take-up, Medicaid expansions are believed to haveincreased coverage among the previously ineligible (Grogger and Aizer (2003)).Additional work that has analyzed the relationship between coverage, access to care,and utilization of care suggest that insurance coverage is positively correlated witheach

Few studies, however, have actually assessed the relationship between

insurance expansion and health and those that do use very small and specific

samples Despite some conflicting results, most of these studies find small significanteffects of insurance on health Using changes in California's Medicaid (Medi-Cal)law, Lurie (1986) finds a small improvement in health for patients at LA clinics.More notably, Currie and Gruber ((1996a) (l996b) (1997)) also find small positive

effects when measuring the effect of Medicaid expansions on the health of pregnantwomen, children, and infants Despite these positive findings, it is hard to generalizethis research because infants, pregnant women, and patients at LA clinics are

arguably quite different from the general population As Levy and Meltzer (2001)suggest, insurance expansions affect "vulnerable populations" who have the most togain from increased resources The effect of insurance on the health of the generalpopulation is less clear

In this paper, I add to the small amount of literature that assesses the effect of

insurance coverage on health I address the limitations of previous works by using ageneral sample and a very general measure of health To do this, I utilize changes instate Medicaid eligibility requirements that occurred as a result of federal law

changes in 1996 During this time, the Personal Responsibility and Work

Opportunity Reconciliation Act (PRWORA) separated Medicaid eligibility

requirements from welfare eligibility rules Between 1996 and 2001 many statesraised their income eligibility cutoffs and expanded coverage to parents of previouslyeligible children These changes varied in scale and occurred at different times indifferent states, creating variation in eligibility laws both across and within states

I use data from the CPS March Supplement to assess the effect of the 1996Medicaid eligibility expansions on health status I first measure the effect of theeligibility expansions on insurance coverage and then, unlike previous work, onhealth My model is consistent with that used by Grogger and Aizer (2003) although

I expand it to measure the health of additional samples The March Supplement is anideal data source because it over samples low-income individuals (those that qualify

for Medicaid) and provides in depth information on income, health insurance, andstate of residence It also includes some information on health conditions Unlikeother authors, I supplement my CPS health data with information from the IntegratedHealth Interview Series (IHIS) in order to establish that my measure of health isvalid and provide additional insight on the health behaviors and conditions of thosethat use Medicaid This allows me to comprehensively assess the effect of insuranceexpansions on the health status of the sample population.

BACKGROUND AND LITERATURE REVIEW

A Medicaid and Medicaid Eligibility Expansions

Founded in 1965, Medicaid is a federal and state funded program that provides

health insurance to low income non-elderly adults and children Medicaid eligibility

was historically linked to welfare participation and individuals or families who

qualified for AFDC aid were automatically enrolled in Medicaid This population

mostly included unemployed and very income single mothers Other

low-income working parents and families were above Medicaid low-income eligibility cutoffs

and did not qualify for the program Over time, Medicaid became more inclusive and

the scope and scale of coverage greatly increased

In the mid 1980s and 1990s, Medicaid coverage expanded to cover children and

pregnant women in both welfare and non-welfare families during "poverty related

expansions." By 1990, all states were required to cover pregnant women and children

up to eighteen years of age with income equal to or below 133% of the federal

poverty line Vv'hile these expansions expanded coverage, they still targeted the same

historic population: low income pregnant women and their children

In 1996, the passage of the Personal Responsibility and Work Opportunity

Reconciliation Act (PRWORA) fundamentally changed the eligibility requirements

for Medicaid Better known for replacing AFDC with TAI\'F, the act also separated

Medicaid eligibility requirements from the requirements of other welfare programs

In order to ensure that individuals with Medicaid did not lose coverage, states were

required, at a minimum, to use the initial AFDC income-eligibility requirements for

Medicaid after the passage of PRWORA In addition, states were allowed to set their

own income cutoffs and redefine their income eligibility cutoffs as long as they wereabove AFDC requirements Funding for expansions came from a combination ofstate and federal sources Thus, as a result of PRWORA, states were able to offerhealth insurance coverage to previously ineligible low income working adults.

This paper focuses on the 24 states that expanded eligibility between 1996 and

2004 Eligibility expansions occurred through three mechanisms: a redefinition ofincome-eligibility requirements, a redefinition of earnings disregards, and an

application for research and demonstration waivers from the federal government.1Information on the timing of state expansions and changes in income-eligibility istaken from Grogger and Aizer (2003) \Vhile it is hard todetermine what exactly ledcertain states to expand coverage, it is likely that they were responding to nationalconcerns regarding the number of uninsured adults and the rising cost of health care.For the purposes of this study, I have not found any evidence that state expansionswere motivated by fears that the population in the expansion states were

experiencing declining health relative to non-expansion states

Finally, it is also important to note that some members of the previously ineligible population could qualify for Medicaid under the "medically needy"

income-program prior to 1996 The medically needy income-program covers certain elderly, childrenand very sick low-income adults that are otherwise not covered by Medicaid Beforethe passage ofPRWORA, the program was also used by states to expand coverage to

1Income-eligibility is defined as a percent of annual family income relative to thefederal poverty line Earnings disregards refer to certain amounts of money that afamily is allowed to deduct from their annual income prior to computing income-eligibility Research and demonstration waivers allow states to disregard certainpreviously established limitations on care

individuals and families above the AFDC income cutoffs (Holohan 2007) As aresult, it is possible that low-income adults in certain states actually had Medicaidcoverage despite otherwise appearing income ineligible.

B Background Literature

B.l Eligibility and Insurance Coverage

Economists frequently use Medicaid expansions to assess the relationship

between insurance eligibility and insurance coverage The majority of this work isconcerned with the effect of the expansions on insurance coverage and take-up oflow-income children Cutler and Gruber (1996) find a 24% take-up rate amongnewly eligible children between 1987 and 1992 but attribute 30-40% of the increasedenrollment to the crowding out of private insurance After Medicaid expansions inthe 1980s and 90s, Dubay and Kenney (1996) find a 51 % take-up rate among newlyeligible children They also find that 15-22% of this increase is attributableto

crowding out of private insurance In contrast, later researchers including Dubayand Kenney (1996), Yaziei and Kaestner (2000), and Shore-Sheppard and Card(2004) find lower rates of child take-up and crowd out

A smaller number of works assess the relationship between eligibility expansionsand adult insurance coverage Cutler and Gruber (1996) find no statistically

significant increase in coverage among newly eligible women age 15-44 between1987-1992 Grogger and Aizer (2003) analyze the same relationship after the 1996expansions and find a statistically significant 2.7 percentage point increase in

coverage among newly eligible mothers and a small crowd out effect

Unlike the majority of the works in this field, I do not look at the effect of

eligibility expansions on child coverage or health Instead, I use Grogger and Aizer'sdifference-in-difference model to assess the relationship between eligibility

expansions and insurance coverage for parents prior to analyzing changes in health

In addition to replicating Grogger and Aizer's experiment, I also expand my sample

to include fathers

B.2 Eligibility and Health

A small number of studies analyze the relationship between insurance coverageand health using Medicaid expansions These typically assess changes within onestate or focus on very specific segments of the population Lurie (1986) uses

variation in California's 1982 Medicaid laws to assess the relationship betweeninsurance loss and health for the medically needy by measuring diastolic blood

pressure and self-reported health status She finds that individuals who lost insurancecoverage experienced a statistically significant increase in blood pressure after 6months and reported lower health status Currie and Gruber (l996a) analyzes theeffect of the 1987-1992 Medicaid expansions on infant mortality rates, and pre-natalpractices of pregnant women Using aggregate data from Vital Statistics and a

simulated eligibility variable from the CPS, they find that eligibility expansionstargeting pregnant women slightly reduced the incidence of low birth weight anddecreased infant mortality by 8.5% for every 30-percentage point increase in

eligibility Similarly, Currie and Gruber (1997) demonstrate that increasing eligibilityfor teen mothers and high school dropouts increased the use of obstetric procedures

and decreased neonatal mortality Eligibility expansions are strongly correlated withpre-birth treatments and for a 10 percentage point increase in eligibility neonatalmortality falls by 2.3 percentage points In a final study using the late 1980s

Medicaid expansions, Currie and Gruber (1996b) also find that increasing child (1-15years) eligibility decreases child mortality They estimate that a 15.1 percentagepoint increase in child eligibility reduces child death resulting from disease by 8%

\\11ile these works provide a wealth of information on the relationship betweeneligibility expansions and health, they are limited in two key ways First, each workfocuses on very specific populations: pregnant women, high school drop-outs, infantsand children While these are important target groups for health care policy, they nolonger represent the majority of the uninsured population Second, each study usesspecific measures of health: mortality, birth weight, and diastolic blood pressure.These are strong indicators of health, but they do not take into account the variety ofother factors that are part of health such as chronic diseases, disabilities, or

psychological well-being This paper addresses these limitations by focusing on asegment of the population that still reports high levels of uninsurance, parents below150% of the poverty line, and using a general measure of overall health, self-reportedhealth status

B.3 Measures of Health: Self-Reported Health Status

A few papers also assess the relationship between self-reported health andother health variables These papers frequently analyze the relationship between self-reported health status, work disability variables, and labor force participation Dwyer

and Mitchell (1999) find that self-reported health is strongly correlated with existinghealth conditions while disability variables are more closely linked to variables thatmeasure the physical ability to perform work.2 They ultimately report that generalself-reported health measures are slightly more variable than objective health

measures but both are viable and adequate proxies for health Disability variables arealso adequate indicators of health, although they measure more physical components

of health Baker, Stabile and Deri (2004) assess the relationship between generalself-reported health variables, objective health self-reports and actual health

conditions They conclude that self-reported objective health variables are subject tomeasurement error in similar ways as disability variables and general health

measures.3

I use a self-reported health indicator as the key dependent variable in mystudy and perform similar tests to ensure that it is a valid measure of health Giventhe relative accuracy of both objective self-report variables and disability variables Iregress these indicators on the self-reported health variable in the CPS I use

additional data from the National Health Interview Survey (NHIS) to supplement theCPS data This dataset is well suited to analyzing the health of working age parents.Case, Lubotsky and Paxson (2002) use a combination of objective self-reportedhealth indicators and a general self-reported health variable from the NHIS as

2 Generally, disability variables report the answer to "have you missed work in thepast year due to a disability or sickness?" Self-reported health status variables ask anindividual to rank his health on a 1 to 5 scale where 1 is typically excellent and 5 is

fOor

Self-reported objective health variables report answers to "have you had X disease

in the past year."Itis also possible that questions may extend the time horizon

beyond a year and ask if an individual has ever missed work or ever had a disease inhis entire lifetime

proxies for child and parent health I use the same variables when I regress reported objective health indicators on general health status.

self-METHODOLOGY AND DATA

A. Methodology

Health insurance lowers the price of health care for the individual and/or

family when consumption of medical care is necessary (Weissman and Epstien

1994) Conceptually, health insurance coverage allows an individual to access and

utilize care, which in tum is believed to improve health (Levy and Meltzer 2001) I

empirically assess the relationship between health insurance and health status by

testing the hypothesis that expanding eligibility for Medicaid leads to health

improvements for low-income parents Itis important to note that this paper only

briefly addresses the relationship between health insurance and medical care

utilization because data on utilization is not included in the CPS Thus, while it is

possible that health insurance affects health status, it is not possible to specifically

identify how it does so in this experiment

A major problem that occurs when assessing the relationship between

insurance and health is that health insurance may be endogenous to health status.Itis

possible that poor health status motivates an individual to purchase health insurance,

or, equally likely, it is possible that people with health insurance are healthier

because they have insurance In order to establish a causal relationship, I identify a

natural experiment that provides variation in insurance which is exogenous to health

status I use changes in Medicaid eligibility laws as a source of variation because law

changes should be unconelated with health status.4

4Itis possible that certain states chose to expand eligibility as a result of failinghealth among their populations, however I do not find evidence of this phenomenon.Instead, it seems likely that eligibility expansions were motivated by concerns about

Endogeneity is partially driven by adverse selection For unobservable

reasons, some individuals may choose to purchase insurance while others may not Inthe typical example, the consumer of health care has more information about hishealth status than the insurance provider Individuals with poorer health may

purchase insurance while individuals with better health may not and the insurancesupplier cannot identify one from the other In other words, there is the potential thatunobservable and unquantifiable characteristics motivate some people to buy

insurance and some people to remain uninsured Conceptually, it is possible that thetraits which drive adverse selection also impact health Ifthese traits are

indistinguishable from insurance coverage it will be impossible to determine theeffect insurance coverage has on health Thus, in order to establish a causal

relationship between health insurance and health status it is necessary to control forthe possibility of adverse selection To address this issue, I use eligibility laws as asource of variation instead of actual insurance coverage status I compare individualswho were always income-eligible for Medicaid to those that become income-eligibleafter the 1996 expansions A major limitation of this method, however, is that I amnot measuring the effect of actual coverage on health

The 1996 Medicaid eligibility expansions occurred at different times in

different states Additionally, the scope of the expansions varied greatly by state

health insurance inequity and the high rate of uninsurance in the United States Prior

to expansion, the mean health in expansion and non-expansion states is very similar;

in 1995-1996 the mean health was equal to 2.31 and 2.35, respectively Additionally,the correlation between health and expansion is only -.01 This suggests that whileexpansion and non-expansion states may differ in key ways, these differences are notcorrelated with the dependent variable I wish to study

Consistent with Grogger and Aizer (2003), my sample only includes states thatexpanded eligibility between 1996 and 2004 Table 1 lists the expansion states alongwith the expansion years and eligibility requirements In addition to the Medicaidexpansions, welfare reform, and an economic boom also occurred during the 1990s.

In order to control for different attitudes towards welfare, different economic

environments in each state, and other potential policy endogeneity issues I use

Grogger and Aizer' s method of comparing populations within expansion statesinstead of across all states 5

States determine Medicaid eligibility on a monthly basis but normally

calculate eligibility income as a percentage of annual family income relative to thefederal poverty line To separate the always eligible from the newly eligible I createtwo mutually exclusive groups within each state, control and treatment Members ofthe control group are always eligible for Medicaid and have income below the pre-expansion income cutoff With a few exceptions, members of the treatment group areonly eligible for Medicaid after the expansions have occurred I also define a secondcontrol group, control 2, which includes individuals that are always income ineligiblefor Medicaid These individuals are above both the original income-eligibility cutoffand the expanded income-eligibility cutoff Medicaid eligibility expansions primarilytargeted parents although a minority of states later changed income-eligibility rules

5Medicaid eligibility rules in non-expansion states are based on and extremely

similar to the AFDC income-eligibility requirements prior to 1996 These may bedifferent in expansion and non-expansion states Prior to the passage of PR\VORAexpansion states provided higher mean AFDC benefits than non-expansion states In

1994, the mean AFDC benefit for a family of three was $444.67 in expansion statesand $353.85 in non-expansion states

for childless adults I use the more generous of these two eligibility expansions and

restrict my sample to only low-income parents.6

Prior to measuring the relationship between eligibility and health, it is

important to assess what affect the expansions had on coverage rates I replicate part

of Grogger and Aizer (2003) to test the relationship between eligibility expansions

and Medicaid coverage.7Based on Grogger and Aizer's model, I estimate the

following regression:

Medicaid Coverage=~l +~2Treatment +~3Interact+~3Years +~4Age+~sRace+

~6# Child+~7Education+~8Married+~9Sex+~9Unemp+f.l (1)

where MEDICAID COVERAGE is a dummy variable equal to one if the individual

has coverage and TREATMENT is a dummy variable equal to one if income is

between the pre expansion and post-expansion cutoffs INTERACT is a variable that

captures the effect of being in the treatment group after eligibility expansion The

coefficient on this variable represents a difference-in-difference estimate that

compares the change in coverage between the control and treatment groups

MARRIED is a dummy variable equal to one if an individual is married, AGE is

6In 2004, there were 11.1 million uninsured parents and 25.5 uninsured childlessadults in the United States with income below 300% of the FPL (Holahan et al

2007) Despite representing a larger uninsured population, eligibility expansions didnot target childless adults and I remove them from the sample

7Grogger and Aizer (2003) restrict their sample to low-income mothers In order tocompare my results to their findings I initially restrict my sample to low-incomemothers I then run a separate regression for both mothers and fathers

measured by year, and SEX is equal to one if the individual is male RACE is

composed of WHITE and BLACK dummy variables equal to one if the individualfalls into either category and OTHER is the omitted race category.8 EDUCATION isequal to years of education and the YEAR variable is a vector of individual yeardummy variables equal to one for each year 1995-2006 (1994 is the omitted year).9

To account for the effect of business cycles and economic conditions on Medicaidcoverage I also include the annual unemployment rate in each state, signified by thevariable UN'EMP

I am also interested in the effect of eligibility expansion on private insurancecoverage and I estimate equation (1) with private insurance as the dependent

variable

After assessing the relationship between eligibility and insurance coverage Iestimate the relationship between eligibility and health status given the followingequation:

Health=~l +~2Treatment +~3Interact+~3Married+~4Age+~5Race+

~6#Child+~7Education+~8Years +~9Sex+~9Unemp+fl (2)

where HEALTH is a dummy variable equal to 1 if parents self-report a health of 1 or

2 Itis based on a categorical variable HEALTH STATUS, which has a value of 1-5where 1 corresponds to excellent health and 5 corresponds to poor health The other

8 This group includes Hispanics, Asians, Native Americans, Pacific Islanders and anyindividual who responded to the CPS survey question in race by marking "other."

9 I also use a specification by age group: 19-29,30-40, and 41-50 and include an squared term

age-variables are defined in the same way as those in equation(1). Given the hypothesisthat becoming insured will improve health, one would expect the health of the

treatment group to improve after expansion relative to the health of the control

group Because HEALTH is a dummy variable equal to 1 for good health, I predictthat the coefficient on INTERACT will be positive if the health of the treatmentgroup improved after the expansions

I also include other variables that may affect health I control for

demographic characteristics such as race, sex, marital status, number of children, andeducation Because Medicaid has different eligibility requirements for mothers,fathers, and childless adults, I restrict the sample to only include mothers and fathersaged 19-50 To control for year effects and the business cycle, I also include a full set

of year dummies and the annual unemployment rate in each state for 1994-2006

In addition to estimating equations (1) and (2) I am also interested in the role

of time on coverage and health status I hypothesize that it may take more than oneyear for eligibility expansions to affect coverage and health To investigate thispossibility, I replace the year dummies with three dummies equal to 1 for

observations 1,2, and 3 or more years after the eligibility expansions I replace myprevious interaction term with the interaction of the treatment variable and the newdummies

To explore the hypothesis put forth by other authors that changes in insurancestatus will have the greatest impact on medically vulnerable populations, I alsoestimate this equation using the dependent variable SICK which is a dummy variable

equal to 1 for parents that report a health status of 5 In this specification, I wouldexpect the coefficient on INTERACT to be negative after expansion.

I first use OLS to estimate how eligibility expansions affect coverage andhealth Then, because both MEDICAID COVERAGE and HEALTH are limiteddependent variables I run probit regressions in order to address the possibility that

my OLS coefficients are biased I also include weights in each regression to ensurethat my sample is representative of the general population

Finally, because my unit of observation is the individual but the variation intreatment is at the state level, residuals may be correlated at the state level, biasing

my standard errors (Moulton 1990) To address this possibility I report clusteredstandard errors

B Data

My primary dataset is the CPS March Supplement The CPS March

Supplement is an annual survey that provides detailed information on income,demographics, and health insurance status.Italso contains individual, family, andhousehold data about health status and state of residence, making it a good datasource to study state-by-state Medicaid expansions \Vhile I use data from 1995-

2007, the CPS income and insurance statistics lag by one year so the sample yearsare actually 1994-2006 This time period is ideal for two reasons First, Medicaideligibility expansions began in the early 1990s but the early expansions suffered

10Case, Lubostky and Paxson (2002) use self-reported health status to target lowerhealth populations

from under-funding and flawed implementation (Grogger and Aizer 2003).11 Incontrast, the 1996-2001 period is characterized by a large number of substantial andfully funded expansions made possible by the passage of PRWORA Additionally,the CPS March Supplement was redesigned in 1994 To avoid discrepancies in

question interpretation and to ensure variable comparability across all years in mysample I only use data from the redesigned survey

The CPS collects information on a variety of individual and family incometypes throughout the year Specifically, information is provided on weekly wages,annual earnings, and additional non-wage income The CPS also contains a

constructed variable that describes annual family earnings as a percentage of thefederal poverty line This is helpful because Medicaid eligibility is based on familymonthly earnings but is frequently expressed in annual terms relative to the federalpoverty line I utilize the CPS variables for annual family earnings relative to thepoverty line to generate income measures by family and use them to assign parents toeither the treatment or control group All income-eligibility measures in this studyare expressed as percentages of the federal poverty line

Despite the detailed income information in the CPS, using CPS data doespresent some problems J'v1edicaid enrollment and eligibility is based on monthlyfamily income while the income data available in the CPS is reported annually

\\illile I use variables that relate annual family income to the federal poverty line,

11 Five states, Washington, Minnesota, Oregon, Tennessee, and Hawaii, chose toexpand Medicaid coverage to low income adults prior to 1996 (Grogger and Aizer2003) These expansions, however, were severely under funded and resulted incoverage cutbacks Itis hard to determine exactly who among the technically

income-eligible received care and coverage

there is still the possibility that families may have varying coverage during the course

of the survey year despite reporting Medicaid coverage at the time of the CPS

survey

The CPS also contains a wealth of information on insurance providers andcoverage status A variety of questions ask if family members, individuals, andchildren are covered by private insurance, government insurance, Medicaid, andMedicare.12 Government insurance is defined as military or veterans insurance andIndian health service insurance For the purposes of this study, I remove individualsthat have non-Medicaid government insurance from the sample.13

Additionally, the CPS includes a question asking individuals to rate theirhealth status over the past year on a scale of "poor" to "excellent." In this question 1

is equal to "excellent," 2 equals "very good," 3 equals "good," 4 equals "fair," and 5equals "poor." The CPS also asks if an individual missed work as a result of illness

or disability I use this variable to check the relationship between the CPS healthstatus variable and medical conditions.14

12The CPS defines private insurance as employer provided or privately purchased

13 I do this mostly as a precaution If low-income parents are covered by military orIndian health service insurance they are distinctly different from the civilian non-elderly population that this study targets Additionally, the only individuals that mayqualify for Medicare that are within my sample age range are defined as "disabled"and are also different from the target population of this paper In total, only 2.7% of

my sample reports Medicare coverage and 3.71 report military or Indian healthservice insurance; dropping these individuals does not substantially change mysample size

14Dwyer and Mitchell (1999) analyze the difference between using work disabilityvariables and health conditions to check the validity of self-reported health status.They find that disability variables are not a good indicator of self reported healthratings but health conditions are I address this later in the paper by using data fromthe IBIS to supplement the CPS data

Unfortunately, the CPS does not include other measures of medical

conditions beyond work disability variables that allow fie to check the validity of theself-reported health variable Self-reported health status is a helpful measure ofhealth because it is extremely inclusive and reflects an individual's overall health bytaking into account a variety of factors such as medical conditions, smoking habits,mental health etc However, it is also a very subjective variable and individuals maydefine "health" differently Because health status is the dependent variable in thisexperiment, I run a variety of tests to ensure that it represents actual concrete, non-subjective, health conditions such as hypertension, diabetes, and heart conditions

To do this, in addition to using data from the CPS, I use data from the

Integrated Health Interview Series (IHIS) Data in the IHIS is based on informationfrom the National Health Interview Survey (NHIS) which provides annual

information on health conditions, insurance coverage, and medical care access andutilization behaviors of the US civilian non-institutionalized population It alsocontains income and demographic information but does not report state of residence.Like the CPS, it asks participants to self report their health on a 1 to 5 scale where 1corresponds to "excellent" and 5 to "poor" health I use data from 1995-2007 andrestrict my sample to parents aged 19-50 with annual family income of $30,000 orlower and I drop individuals who are covered by another type of government

insurance This roughly mirrors the CPS sample, although it includes the entire USpopulation, not only those that live in expansion states.IS

IS The NHIS survey does not ask for an individual's state of residence so I cannotremove individuals in non-expansion states from this data set

The IHIS provides detailed information on whether an individual experiencedcertain health conditions within either the past year or ever in an individual's

lifetime These range from kidney disease and cancer to arthritis, hypertension,diabetes, heart conditions and stroke I regress self-reported health status on everhaving one of these health conditions All regressions are preformed using OLS andresults are reported in Appendix 1.16I find a negative relationship between chronichealth conditions, such as cancer, arthritis, and diabetes, and self-reported health.The coefficients on some short-term conditions are insignificant This suggests thatwhile there may be some inconsistencies with the self-reported health measure in theshort term it does capture the relationship between chronic health conditions andlong-term health I use it as the dependent variable because, for the purposes of thisstudy, I believe that long-term health is the more important public policy concern

To eliminate problems that may arise from different individual interpretations

of "excellent" "good" "average" "fair" and "poor" I also create two health statusdummy variables The first, "Healthy" includes individuals who self report theirhealth as 1 or 2 The second, "Unhealthy" includes those who report health

equivalent to 3, 4 or 5 I use these as dependent variables and regress them on thehealth conditions listed above In this specification, while coefficients on each healthcondition are slightly different than those in the initial specification, the relationshipbetween the conditions and health status is the same: chronic diseases are

significantly correlated with poor health OLS and probit estimation yield similarmarginal effects and significance Results are reported in Appendix 1

16I perform an ordered probit regression and find similar marginal effects and

significance

C Sample

I restrict my sample to individuals with characteristics typical of Medicaidrecipients I include low-income parents below 150% of the poverty line, aged 19-50that live in states that expanded Medicaid coverage between 1996-2001.17For afamily of four with two children this corresponds to an annual family income of

$23,866 in 1996 Additional sample characteristics are described in Table 2 Thesample includes 76518 observations, 15032 in the treatment group, 43965 in thealways-eligible control group, and 17521 are the alternative never-eligible controlgroup The income cutoff between treatment and control groups varies across statesbut the demographic characteristics are very similar Most women are white and haveabout 2 children and 11 years of education By design, the groups do differ

substantially by income Families in the treatment group report an average familyincome of $20,148 while families in the control group have an average annual familyincome of $7,968

Reported health status varies more by insurance type than by income group

As expected, more parents in the treatment group have private insurance than in thecontrol group and those in the control group have high rates of Medicaid coverage

17 Despite the fact that some state eligibility expansions were either far below orquite above 150% of the federal poverty line, I choose this cutoff for two reasons.First, I am interested in targeting members of the treatment group that were

previously just above the old AFDC income eligibility cutoff for Medicaid I

hypothesize that these individuals p.re the most similar to the parents in the controlgroup Second, I want to minimize! the number of individuals who are always

income-ineligible for Medicaid in certain states (their income is too high) but mayhave income below my sample income cutoff These individuals are part of myalternative control group

Those in the treatment group are slightly more likely to be uninsured than those inthe control group Interestingly, parents with Medicaid in both the treatment andcontrol groups report worse health (mean health of 2.7) than those with private

insurance and the uninsured (mean health of 2.1 and 2.3 respectively) This maysuggest that adverse selection influences the take-up of Medicaid coverage by the18

poor

Initial calculations suggest that expanding eligibility did increase Medicaidcoverage in the treatment group (Table 3) Prior to expansion, about 17% of mothersand 14% of parents in the treatment group reported Medicaid coverage; after

expansion those numbers rose to 24%and 21.6%respectively.19 The number ofindividuals with private insurance fell after expansion while the percent of the groupthat is uninsured remained essentially the same This suggests that the eligibilityexpansions may have allowed individuals with private insurance to switch to

Medicaid while those who were uninsured did not necessarily take up coverage This

is consistent with previous findings that take-up rates among previously uninsured

18Poor individuals who are uninsured but need medical care can be directly emolled

in Medicaid at the time of service in a hospital or doctor's office Because this is alarge source of Medicaid take-up, the population reporting Medicaid coverage will

be the individuals who are most in need of care and that report worse health

19 Itis possible for members of the treatment group to have Medicaid pre-expansionbecause low-income adults may qualify for other types of Medicaid programs such

as "medically needy" which covers very sick individuals and has different incomeeligibility requirements 9.2% of the treatment population with Medicaid pre-

expansion reports a health status of4(relative to 5.5%of the pre-expansion

treatment population overall) and 5.2% report health equal to 5(relative to 1.9%ofthe pre-expansion treatment population) Additionally, Medicaid covers programsthat provide medical support to pregnant women and new mothers 11.4% of thetreatment population with Medicaid before expansion is mothers with children lessthan one year of age The remainder may be pregnant mothers who have not givenbirth, although I cannot identify these individuals given my data set

children and adults were relatively low (Card and Shore-Sheppard (2004)), Groggerand Aizer (2003)) As expected, insurance coverage in the control group did notchange drastically, although there is a small decline in Medicaid coverage and anincrease in uninsurance (Table 3) The decline in insurance coverage among thecontrol group is probably a result of welfare reform Ellwood and Ku (2002) find thatreductions in welfare caseloads resulting from welfare reform also led to a decline inMedicaid coverage among the previously insured Similarly, Grogger and Aizer(2003) find that Medicaid coverage fell among the always eligible by 5 percentagepoints after the passage ofPRWORA 20

The treatment group's insurance coverage also varies greatly by health status.Table 4 shows insurance coverage before and after expansion by health status Mostnotably, after the eligibility expansions, individuals with better health report a

decline in private insurance and an increase in Medicaid coverage and uninsuredstatus In contrast, individuals with poor health report an increase in Medicaid andprivate insurance coverage and a decrease in uninsured status While it is impossible

to determine what drives this trend by using cross-section data from the CPS, onehypothesis may be that healthy individuals and families switch from high cost privateinsurance to lower cost Medicaid when they become income-eligible after expansion

20Prior to the passage of PRWORA, families and individuals who qualified forAFDC were automatically enrolled in Medicaid After Medicaid and AFDC

requirements were separated, it is possible that previously covered families exitedboth welfare and Medicaid Studies that analyze the relationship between PRWORAand welfare caseloads find that caseloads fell after the passage of the act, although it

is hard to distinguish between the effect of the new welfare programs and a strongeconomy Overall, welfare reform was believed to lower both Medicaid coverage andcaseloads for the previously eligible population (Moffitt 2001)

In contrast, individuals who were in need of care, but could not previously afford it,may take up Medicaid after they become income-eligible.

RESULTS AND DISCUSSION

A Medicaid Coverage

I find that the Medicaid eligibility expansions had a positive and significanteffect on Medicaid coverage Table 6 reports the OLS results from equation (1) forMedicaid coverage Column (1) states the results of equation(1)when replicatingGrogger and Aizer's experiment and the sample includes only mothers The

coefficients on the treatment variable demonstrate that, as expected, mothers in thetreatment group are less likely (21.1 percentage points) to have Medicaid coveragethan mothers in the omitted control group.21 After expansion, Medicaid coverageincreased 8.3 percentage points among previously ineligible low-income mothers, asmeasured by the coefficient on the INTERACT variable This number is relativelyconsistent with Grogger and Aizer's initial difference-in-difference estimate (7.7percentage points).22 This result answers the preliminary question of what effectexpanding eligibility had on Medicaid coverage rates and lays the foundation foranalyzing the relationship between eligibility expansion and health status

In addition to replicating the work of Grogger and Aizer, I perform additionalregressions using alternative sample specifications In column(2) I report results for

a sample that includes mothers and fathers below 150% of the federal poverty line I

21 I always compare the treatment group to the primary control group of income-eligible parents I run a few regressions using only the alternative never-eligible group but ultimately drop these parents from the sample and do not includethem in any of my results

always-22 Grogger and Aizer develop this model further and control for the effect of welfarereform on coverage They find that it caused coverage in the control group to fall by

5 percentage points As a result, they find a direct effect of Medicaid expansions onmothers' coverage of only 2.7 percentage points The sum of these two numbers isroughly equal to the effect that I find with my model

find that Medicaid coverage increased by 7.5 percentage points for this group afterexpansion and that fathers are 5.6 percentage points less likely to have Medicaidcoverage than mothers Column (3) states results for mothers and fathers below150% of the federal poverty line who report a health status of 3,4, or 5.23Take-uprates were stronger amongst this group and expansions increased coverage by8.9

percentage points.24Finally, in column (4) I present results for a sample that includesmothers and fathers below 130% of the poverty line After expansion, Medicaidcoverage increased by 8.6 percentage points for these parents These groups are ofparticular interest because individuals with poor health may be especially likely totake-up Medicaid when it becomes available Higher take-up rates for this group arealso to be expected as they were previously only slightly above income-eligibilitycutoffs and are generally less likely to have private insurance coverage.25

The signs of the coefficients on demographic and time variables are consistentwith those in previous research across all specifications Individuals with higherlevels of education are less likely to have Medicaid than those with lower levels ofeducation Using the sample of parents, I find that for every year of education anindividual is 1.9 percentage points less likely to have Medicaid Younger parents are

23The composition of this group may be changing over time as result of the

eligibility expansions which could potentially make these results harder to interpret

24I also try a specification where I restrict the sample to parents that only report ahealth status of 4 or 5 For these individuals the coefficient on INTERACT is 4.7percentage points although it is insignificant (t-statistic = 1.24) I attribute this tothedecrease in sample size (N=6823) However, it is also possible that the sickest

individuals already had Medicaid coverage because they qualified for the "medicallyneedy" Medicaid programs.Ifthis is the case, then one would not expect eligibilityexpansions to have a large effect on coverage among the sickest members of thesample

25 I also control for state-fixed effects in column (5) The results are essentially thesame

also more likely to have Medicaid For every additional year of life a parent is about

1 percentage point less likely to report Medicaid coverage?6 \Vhites are less likely tohave Medicaid than the omitted "other" race group although the coefficients vary insignificance Black mothers are more likely to have Medicaid than the "other"

group.27 Number of children is positively correlated with Medicaid coverage; forevery additional child a parent is 3.3 percentage points more likely to have Medicaid.State unemployment rates are positively correlated with Medicaid coverage althoughthe coefficient on Ul\TEMP is not statistically different than zero

To explore the timing of take-up after the eligibility expansions I report resultsfor equation (l) in Table 7 where I replace linear year variables with dummies for 1,

2, and 3 or more years after expansion This is motivated by the possibility thatknowledge of the expansions may diffuse slowly Using the sample of mothers, I findthat coverage increased 9.9 percentage points the first year after expansion, 7.8percentage points in the second year after expansion and 9.9 percentage points three

or more years after expansion I test each interaction term and find that they are notstatistically different from each other This suggests that coverage increased at aconstant rate after the eligibility expansion for newly eligible mothers below 150%

of the poverty line I also assess this relationship for the other sample specifications

26 I try a few different specifications for age Parents between 30 and 40 years of ageare 12percentage points less likely to have Medicaid than parents between 19 and29

years of age Parents aged 41-50 are 16.15 percentage points less likely to haveMedicaid than parents 19-29 years

27 wilen I restrict the sample to the "poor health" specification the coefficients on therace variables become insignificant Technically, one would not expect the

coefficients on demographic controls to change very much Itis possible that at thislevel of demand for medical care race will not affect coverage However, it is morelikely that this is caused by the specification I use which decreases variation andsample size

using the initial Medicaid coverage equation For all parents I find that coverage

increased between 6.7 percentage points and 9.1 percentage points each year after

expansion (column (2)) and for mothers and fathers reporting poor health I find an

increase between 8.5 percentage points and 13.3 percentage points each year (column

(3)) For newly eligible parents below 130% of the poverty line I find an increase

between 8.6 percentage points and 9.4 percentage points in each year after expansion

(column (4)) In no case is there evidence of increased growth in take-up over time

My initial difference-in-difference calculations presented in Table 3 and the

findings of Grogger and Aizer (2003) suggest that private insurance coverage rates

may also have been affected by the Medicaid eligibility expansions Due to the

possibility that increases in Medicaid coverage could be a result of private insurance

crowd out I run equation (1) using a dummy for private insurance coverage as a

dependent variable and report these results in Table 8 Column (1) presents the

results when I use only mothers sample and column (2) presents results when

mothers and fathers are included in the sample I find that the eligibility expansions

caused private insurance rates to fall by 4.7 percentage points for both mothers and

all parents The coefficient on INTERACT falls slightly to 3.4 and 3.7 percentage

points and is not statistically different than zero when I restrict the sample to the

sickest individuals and lower the sample income cutoff to 130% of the poverty line I

report these results in column (3) and (4) respectively

When I replace the single interaction term with the three post expansion

interaction terms, I find that private insurance loss occUlTed for mothers in the

second and third years after expansion (Table (9)) Column (2) shows that private

insurance loss occuned for both parents in all three periods, but got stronger in

periods 2 and 3 While it is hard to say what drives this trend, it is possible that someindividuals with private insurance who became eligible for Medicaid decided to droptheir private coverage after learning about their eligibility This story is supported bythe gradually increasing magnitudes of the coefficients on the interaction terms.Individuals with private insurance could take their time in applying for and taking upMedicaid coverage while uninsured individuals may have acted more quickly or out

of more pressing need Given this story, I hypothesize that there may be a smallamount of private insurance crowd out among parents in later years as a result of theMedicaid eligibility expansions This is consistent with findings in Grogger andAizer (2003)

The coefficients and signs of the demographic control variables are also

consistent with those in previous literature Using the sample that includes mothersand fathers, I find that for every additional year of education an individual is 1.5percentage points more likely to have private insurance Fathers are 5.6 percentagepoints less likely to have private insurance relative to mothers and for every

additional child a parent is 09 percentage points less likely to have private coverage.The state unemployment rate is surprisingly insignificant Age is positively

conelated with private insurance and both white and black individuals are morelikely to have private insurance than the omitted "other" race category although thecoefficients onVlHITE are not statistically different from zero \\nen compared tothe results of the Medicaid coverage regressions, individuals with private insuranceare more likely to be educated, older, and have fewer children

The dependent variables for Medicaid and private insurance are both dummy

variables and I check the accuracy of the OLS coefficients by performing probit

regressions I report results from the probit regressions in Appendix 2 Table 2A,

column(1)shows results from mothers only and column (2) presents results for all

parents for Medicaid coverage I report the marginal effect of each variable on

Medicaid coverage Vlhen compared to the OLS marginal effects, the probit effects

are slightly larger but do not change in sign or significance In the probit regression

for the mothers only sample, members of the treatment group are 23.0 percentage

points less likely than the control grouptoreport Medicaid coverage and 11.7

percentage points more likely to have coverage after the eligibility expansion The

corresponding values for the OLS marginal effects are 21.0 percentage points and 9.9

percentage points Similarly, in the prabit regression for all parents, the treatment

group is 11.2 percentage points more likely to have coverage after expansion and the

corresponding OLS value is 7.5 percentage points I report prabit results for the poor

health parent sample and the sample with income below 130% of the federal poverty

line in columns (3) and (4) respectively Across all variables I find the same signs

and significance which suggests that the OLS coefficients are relatively accurate

Table 2B in Appendix 2 presents the prabit results for private insurance As in

the Medicaid regressions, the OLS and the probit marginal effects are not very

different which allows me to continue using the OLS specification

Overall, I find that eligibility expansions increased Medicaid coverage among the

treatment group This finding holds acrass all sample specifications and regression

types, although the magnitude of the effect varies Given this starting point, I