TUI MHA Health Care Finance Patient Protection and Affordable Care Act Discussion

Journal of Health Economics 53 (2017) 72–86Contents lists available at ScienceDirect
Journal of Health Economics
journal homepage:
Premium subsidies, the mandate, and Medicaid expansion: Coverage
effects of the Affordable Care Act
Molly Frean a , Jonathan Gruber b , Benjamin D. Sommers c,∗
Harvard T.H. Chan School of Public Health, 677 Huntington Ave, Kresge 4th Floor, Boston, MA 02115, United States
Massachusetts Institute of Technology and National Bureau of Economic Research, Department of Economics, E52-434, 77 Massachusetts Avenue,
Cambridge, MA 02139, United States
Harvard T.H. Chan School of Public Health, 677 Huntington Ave, Kresge Room 406, Boston, MA 02115, United States
a r t i c l e
i n f o
Article history:
Received 15 September 2016
Received in revised form 17 February 2017
Accepted 24 February 2017
Available online 6 March 2017
JEL classification:
Health insurance
Tax credits
Individual mandate
a b s t r a c t
Using premium subsidies for private coverage, an individual mandate, and Medicaid expansion, the
Affordable Care Act (ACA) has increased insurance coverage. We provide the first comprehensive
assessment of these provisions’ effects, using the 2012–2015 American Community Survey and a tripledifference estimation strategy that exploits variation by income, geography, and time. Overall, our model
explains 60% of the coverage gains in 2014–2015. We find that coverage was moderately responsive to
price subsidies, with larger gains in state-based insurance exchanges than the federal exchange. The
individual mandate’s exemptions and penalties had little impact on coverage rates. The law increased
Medicaid among individuals gaining eligibility under the ACA and among previously-eligible populations
(“woodwork effect”) even in non-expansion states, with no resulting reductions in private insurance.
Overall, exchange premium subsidies produced 40% of the coverage gains explained by our ACA policy
measures, and Medicaid the other 60%, of which 1/2 occurred among previously-eligible individuals.
© 2017 Elsevier B.V. All rights reserved.
One of the most significant policy issues facing the United States
over the past forty years has been the high number of those without health insurance. The percentage of uninsured Americans rose
steadily from the 1980s through 2010, through both recessions
and economic growth (DeNavas-Walt et al., 2013). A major policy focus during this era was intervening in insurance markets to
expand coverage and offset this trend. This mostly happened using
public insurance via Medicaid and the Children’s Health Insurance Program, with little private sector intervention (Gruber and
Levitt, 2000). This pattern of incremental public coverage expansion changed dramatically with the passage of the Affordable Care
Act (ACA) in 2010.
The ACA enacted enormous expansions of both public and private insurance. The former was to take place through a nationwide
expansion of Medicaid to all those with incomes below 138% of the
Federal Poverty Level (FPL); however, the Supreme Court ruled in
2012 that states could refuse this expansion. The private insurance
expansion takes place through sizeable income-based tax credits
∗ Corresponding author.
E-mail addresses: (M. Frean),
(J. Gruber), (B.D. Sommers).
0167-6296/© 2017 Elsevier B.V. All rights reserved.
for those with incomes from 100–400% of FPL who are not eligible
for Medicaid, to subsidize premiums for private insurance purchased on newly established insurance exchanges. Underlying the
expansion are new insurance regulations that end discrimination
on the basis of pre-existing conditions, coupled with an individual
mandate that requires most Americans to obtain insurance (with
several exemptions, most notably related to affordability). These
principal pieces of the ACA took effect in January 2014.1
National data from multiple sources strongly support the notion
that the ACA has reduced the uninsurance rate substantially beginning in 2014, reaching an historic low by 2015 (Cohen and Martinez,
2014; Smith and Medalia, 2015; Sommers et al., 2015a). This drop
has generally been attributed to the ACA, but most analyses of the
ACA to date have been largely descriptive (Cohen and Martinez,
2014; Long et al., 2014) or limited to a particular aspect of the ACA
such as the Medicaid expansion (Black and Cohen, 2015; Kaestner
The earliest coverage expansion enacted under the ACA was the dependent coverage provision, which mandated that private insurers allow parents to cover their
children on their insurance until age 26. This provision took effect September 2010.
We do not examine this policy here, since it had essentially reached steady-state by
2012 and has already been examined thoroughly elsewhere (Antwi et al., 2013).
M. Frean et al. / Journal of Health Economics 53 (2017) 72–86
et al., 2015; Wherry and Miller, 2016). No studies have disentangled the different coverage effects of the ACA’s various provisions.
Even as the 2016 election results cast uncertainty over the ACA’s
future, these issues remain critically important to understanding
the potential impact of a partial or complete rollback of the law, as
well as the potential consequences of future state or federal efforts
to expand coverage.
In this paper, we provide the first comprehensive model that
identifies the causal impact of the ACA’s numerous provisions on
insurance coverage. In doing so, we also offer an empirical template
for future research on this wide-ranging law. We use data from the
American Community Survey (ACS) for the two years before and
two years after full ACA implementation. We estimate rich models that examine both the public coverage expansions and private
coverage subsidies that are put in place by the ACA, as well as the
individual mandate. Public insurance expansions are identified by
state decisions about whether to take up the Medicaid expansions
and by differential impacts of Medicaid expansions across income
groups and family types. Private insurance subsidies are identified by the variation in effective subsidy rates by income group
and area of the country. Mandate effects are identified by variation in the incidence of the penalty (related to the law’s several
mandate exemptions) and the magnitude of the penalty (tied to
income and family structure). Our models allow us to control for
fixed differences and trends by income group and geographic area.
Overall, we find that our policy parameterization can explain
roughly 60% of the increase in insurance coverage from 2012–2013
to 2014–2015. The remaining 40% of coverage gains in 2016 does
not appear to be mediated by the economic recovery, based on prior
research by Blumberg, Garrett, and Holohan. A reasonable interpretation is therefore that as much as 40% of the ACA’s coverage
gains could be attributable to the combined effects of increased
insurance purchase rates related to the new marketplace, increased
value of coverage based on the law’s essential benefits, community
rating, and a generalized effect of the mandate not tied to its specific
exemptions and penalty amounts. Any one of these policies in isolation could, in theory, be contributing up to 40% of the overall ACA
effect, though more likely each is contributing a portion. Unfortunately, given the simultaneous implementation of these policy
provisions, it is impossible to disentangle each of these components.
Within these ACA coverage changes, we have several key findings. The impact of tax credits to private insurance was fairly
modest but grew over time, with each 10% increase in subsidy
reducing the uninsured rate by roughly 0.5 percentage points
in 2014 and 0.9 percentage points in 2015. Premium tax credits produced much larger effects in states operating state-based
insurance exchanges, as opposed to using the federal exchange
(, suggesting potential benefits to local implementation of the law. All told, exchange insurance subsidies accounted
for approximately 40% of the reduction in the uninsured rate
attributable to our ACA policy parameters. In contrast, the mandate
penalty had a negligible impact on coverage.
Meanwhile, Medicaid accounted for the other 60% coverage
change attributable to our ACA policy measures, via three distinct
pathways. Medicaid expansion increased coverage among newlyeligible individuals by roughly 14 percentage points in 2015, which
accounted for nearly 20% of the observed ACA effect on the uninsured rate. Another 10% came from the ACA’s early expansions of
Medicaid that occurred in 6 states between 2011–2013. Nearly
30% of the ACA policy impact on coverage in 2014–2015 came
via the less discussed “woodwork effect” of increased insurance
enrollment among those who were previously eligible for Medicaid
before the ACA but not enrolled. This phenomenon was evident
in all states, whether or not they had expanded Medicaid, and
occurred for both adults and children. Finally, we find no evidence
that the expansion of Medicaid led to offsetting reductions in private insurance.
Our paper proceeds as follows. Section 1 describes the ACA’s
main coverage provisions. Section 2 reviews the existing literature
on how these policies may impact health insurance coverage. Section 3 describes our data and policy variables. Section 4 presents
our empirical strategy. Section 5 presents our results. Section 6
discusses policy implications and concludes.
1. Background on the Affordable Care Act
The ACA represents the largest transformation of the U.S. health
care system since the introduction of Medicare and Medicaid in
the mid-1960s. While the legislation also addressed issues such
as health care costs and quality of care, we focus on the coverage
provisions of the ACA. There are three key provisions that form the
law’s “three legged stool”:
The first is a federal overhaul of private insurance market
regulation. Among other changes, the ACA guarantees the issue
of insurance regardless of pre-existing conditions, bans medical
underwriting, and eliminates annual or lifetime benefit limits.
These provisions apply to the entire non-group insurance market,
as well as to non-self-insured employers.
The second is the individual mandate. Under the ACA, legal residents of the U.S. are mandated to obtain insurance, subject to a
number of exemptions, and those who do not are subject to a tax
penalty. This penalty was modest in 2014, equal to the larger of
$95 or 1% of income; it has grown more sizeable since, rising to
the larger of $695 or 2.5% of income in 2016. Exemptions exist for
those with incomes below the threshold for filing federal income
taxes, low-income residents in states that have not expanded Medicaid under the ACA, and those who cannot find insurance on the
exchange for less than 8% of income.
The third is comprised of policies to make health insurance more
affordable. This includes a massive expansion of public insurance
through a universal extension of Medicaid eligibility to all those
below 138% of the federal poverty level.2 Medicaid was previously
categorically restricted: some groups (such as children and pregnant women) were typically eligible above this income level, others
(such as disabled adults and low-income parents) were only eligible at much lower income levels, and the remaining low-income
adults (so-called “childless adults”) were not eligible at all in most
states. This expansion had differential impacts by state, income,
and family type. An additional element of variation in Medicaid eligibility was the result of a Supreme Court decision in 2012, which
made the ACA’s Medicaid expansions voluntary. As a result, only 24
states plus Washington D.C. expanded by January 2014; since then,
another 7 states have expanded (Kaiser, 2015).
The other source of financial support for insurance was through
the introduction of new tax credits for private insurance purchased
through the exchanges. Individuals are eligible for tax credits if they
are ineligible for Medicaid and have incomes between 100% and
400% of FPL. These credits cap the share of income that individuals
must pay for coverage (at the “silver” level described below) at
between 2% and 9.5% of income on a sliding scale basis. In addition,
the ACA provides cost-sharing subsidies to enrollees with incomes
below 250% of FPL.
The ACA included other provisions that are harder to quantify, but which might have significant effects. The first is the
The statutory cutoff for Medicaid eligibility under the ACA is 133% of FPL, but
requires that states disregard a portion of applicants’ income equal to an additional
5% of FPL, producing an effective eligibility threshold of 138% of FPL. Also, note that
Medicaid coverage is not available to individuals without either U.S. citizenship or
legal permanent residency status for at least 5 years.
M. Frean et al. / Journal of Health Economics 53 (2017) 72–86
introduction of private insurance exchanges, which brought organized shopping to a fractured non-group insurance market. On
these exchanges, individuals can compare options at four different “metal” levels based on the plans’ actuarial value (the share of
expected medical costs covered): 60% for bronze, 70% for silver, 80%
for gold, and 90% for platinum. States had the option of establishing
their own exchanges or using the federal exchange; 13 states plus
Washington D.C. operated state-based exchanges in 2014–2015,
though two states (Kentucky and Hawaii) have since reverted to
the federal exchange.
The ACA includes an “employer mandate” as well. This is a
charge levied on firms based on the share of their employees that
are not offered affordable coverage who end up receiving exchange
tax credits. However, this provision was delayed until 2015, and
given the lack of information on employer offers of coverage in the
ACS, we did not model that policy directly.
2. Literature review
Our paper focuses on three main policy levers: public expansions, private insurance subsidies, and an individual mandate. In
this section we review the literature on the effects of these policies
on health insurance coverage and what is known to date about the
ACA’s effects.
Previous research on public insurance expansions focuses on the
sizeable expansions of Medicaid and the Children’s Health Insurance program (CHIP) over the late 1980s-early 1990s, and again
in the late 1990s-early 2000s (Gruber and Simon, 2008; Sommers
et al., 2012a). The literature finds that take-up of these Medicaid
and CHIP expansions was moderate, with roughly 25–35% of those
who became newly-eligible for public insurance coverage choosing
to enroll. One reason is that many of those made eligible for public
insurance already had private insurance coverage. Complex application processes and informational barriers also contribute to low
participation (Sommers et al., 2012b).
Some individuals, however, may have dropped their private
coverage for free or heavily subsidized public insurance, a phenomenon known as “crowd out” (Cutler and Gruber, 1996).
Estimates of the share enrolling in public insurance who would
otherwise have private insurance vary. Some studies have found
rates ranging from 20–60% (Gruber and Simon, 2008; Lo Sasso and
Buchmueller, 2004), while others have found little to no crowd-out
(Hamersma and Kim, 2013; Thorpe and Florence, 1998). In general, crowd-out has been found to be greater among expansions to
higher-income groups (Kronick and Gilmer, 2002).
There has been much less work on the impact of private insurance subsidies. One well-cited study (Marquis and Long, 1995)
used geographic variation in the price of individual insurance to
assess the correlation with insurance coverage, estimating an elasticity of demand of −0.4. This is problematic, however, since other
factors correlated with insurance demand may drive this price variation. There has been more work on tax policy and the demand
for employer-sponsored insurance; see Gruber (2005) for a review.
Massachusetts’ 2006 health reform law, which featured premium
subsidies and a state exchange, led to large reductions in the
uninsured rate (Long et al., 2009). However, the state law’s other
features (including individual and employer mandates) complicate
the interpretation of these findings, and previous research has not
disentangled the effects of subsidies vs. these other provisions.
There is also less understanding of how the individual mandate
impacts coverage and interacts with the ACA’s other provisions.
Again, the best evidence comes from Massachusetts, which introduced an individual mandate as part of its 2006 health reform. In
addition to a general decline in the uninsured rate, prior research
shows several spillover effects of the mandate. First, individu-
als who were already eligible for the state’s Medicaid program
but not yet enrolled significantly increased their take-up (Sonier
et al., 2013). Second, despite generous non-employer insurance
subsidies and a weak employer mandate, there was no erosion of
employer-sponsored coverage – and some evidence that such coverage increased (Kolstad and Kowalski, 2012). This may reflect a
response to the individual mandate, in which workers accept lower
wages in return for employer coverage (Hackmann et al., 2015).
In terms of the ACA itself, a growing body of research has
begun to document changes in coverage under the law. Several
states opted to expand Medicaid under the ACA prior to 2014,
and studies indicate small marginal changes in coverage with variable crowd-out – little among those with health problems, but
significant among younger adults (Sommers et al., 2014). For the
2014 expansion, federal survey data (Cohen and Martinez, 2014;
Smith and Medalia, 2015) and private data sources (Shartzer et al.,
2015; Sommers et al., 2015a) all confirm a large drop in the uninsured rate, particularly among lower-income adults. A time-series
analysis estimated nearly equal coverage gains in 2014 due to
exchange insurance and Medicaid (Carman et al., 2015), though this
study simply presented descriptive trends. Finally, several analyses describe moderate coverage gains in 2014 due to the Medicaid
expansion (Courtemanche et al., 2016; Kaestner et al., 2015). To our
knowledge, no research has yet developed an identification strategy to assess the ACA’s coverage provisions simultaneously and
disentangle their effects.3
3. Data and policy measurement
3.1. Data
Our primary source of data for this analysis is the 2012–2015
American Community Survey (ACS). The ACS, conducted by the
United States Census Bureau, is the largest household survey in
the country, with approximately 3 million individuals surveyed in
the public-use file each year. Within-state geographical information is available in the ACS based on approximately 2350 “public
use microdata areas” (PUMAs). PUMAs are mutually exclusive areas
within states that are populated with at least 100,000 individuals;
PUMA boundaries were redrawn after 2011 using the Decennial
Census, which precludes us from using data prior to 2012. The ACS is
one of the primary sources used by the federal government to evaluate health insurance coverage (Finegold and Gunja, 2014; Smith
and Medalia, 2015).
Our study sample includes all non-elderly (age under 65) individuals residing in the U.S., other than in Massachusetts. We
exclude the elderly from our analysis because the ACA’s coverage expansions did not apply to individuals 65 and over. We
excluded Massachusetts because the state’s 2006 health reform law
already included many ACA-like features; our results are essentially
unchanged by this exclusion.
Our dependent variables of interest in the ACS are four measures
of insurance coverage: no health insurance (uninsured), Medicaid,
employer-sponsored insurance (ESI, including military and union
coverage), and non-group private insurance. Together, these four
categories are inclusive of 98% of non-elderly individuals in the
survey, with the remainder insured by the VA or Medicare. Regarding Medicaid, the ACS’s question asks about “Medicaid, Medical
Assistance, or any kind of government-assistance plan for those
Another strand of research examines effects of public insurance on labor supply, with conflicting findings (Baicker et al., 2014; Garthwaite et al., 2014). Early
evidence on the ACA suggests that labor market effects have been minimal (Garrett
and Kaestner, 2015; Gooptu et al., 2016; Moriya et al., 2016), and we do not focus
on this issue here.
M. Frean et al. / Journal of Health Economics 53 (2017) 72–86
with low incomes or a disability.” Thus, some respondents may
answer “yes” to this question based on their receipt of government subsidized exchange coverage, while others may report this
as non-group coverage (i.e., “Insurance purchased directly from an
insurance company”).4
3.2. Policy measures
The ACA marks an enormous policy change toward insurance
coverage. Fortunately for research purposes, many of the changes
embodied in the law vary substantially across individuals in a way
that can be parameterized. Other factors are more uniform and difficult to separate from non-ACA conditions that may more generally
impact insurance coverage.
Medicaid Eligibility. Our first policy measure is eligibility for
Medicaid, which we combine with eligibility for the related Children’s Health Insurance Program (CHIP). We decompose eligibility
into three parts: eligibility prior to the ACA’s Medicaid expansion (i.e., using 2010 income thresholds and criteria), eligibility
under the so-called “early expansions” that occurred under the
ACA between 2011–2013 in 6 states5 ; and new eligibility as a
result of the 2014 Medicaid expansion. The first group is known
as the “woodwork” (or “welcome mat”) population that may
newly take up Medicaid coverage due to increased awareness
of coverage options under the ACA, the law’s attempt to reduce
administrative-related barriers to applying that have previously
reduced enrollment (Aizer, 2007), and the individual mandate
(Sommers and Epstein, 2011). In addition, the existence of Medicaid
expansion may increase participation in this group as individuals
know they are less likely to lose eligibility for small changes in
Our approach distinguishes between these categories of Medicaid eligibility because all three may have plausibly experienced
coverage changes as a result of the ACA, but likely with heterogeneous take-up rates. Most analyses of Medicaid expansions prior
to the ACA have ignored any potential woodwork effect, which in
many cases was likely negligible given the lack of other systemic
policy changes (such as the ACA’s mandate and application streamlining). However, some analyses have identified similar spillover
effects in previous expansions (Aizer and Grogger, 2003; Sonier
et al., 2013). All measures of Medicaid eligibility are constructed
using state rules based on age, income, disability, and parental status obtained from the Centers for Medicare and Medicaid Services
(CMS) and the Kaiser Family Foundation.6
Fig. 1 depicts the percent of the sample eligible for Medicaid/CHIP, based on state expansion status for children (Panel A)
and adults (Panel B). All children under the poverty line are eligible, regardless of state expansion decision. In the 200–300% FPL
The ACS, while generally quite reliable at assessing health insurance coverage
and used by the Census in its annual reports on insurance of the U.S. population,
does produce overestimates of non-group coverage compared to other data sources
(Mach and O’Hara, 2011). However, our study design effectively subtracts out any
time-invariant over-reporting bias for this form of coverage in the survey.
The early expansion states are CA, CT, DC, MN, NJ, and WA. See Sommers et al.
(2013) for expansion details and timing.
Medicaid and CHIP eligibility for children, parents, and childless adults was
obtained for each state, as of 2013, from a pre-ACA survey of all 50 states conducted by the Kaiser Family Foundation (Heberlein et al., 2013), supplemented by
information on the six states adopting the ACA’s early-expansion option to expand
prior to 2014 (Meng et al., 2012; Sommers et al., 2013). Information on disabilityrelated eligibility is also from Kaiser (Kaiser, 2010); adult disability was identified
in the ACS using their disability recode variable. 2014 eligibility was updated with
information from CMS (2014). The ACS does not report pregnancy, so we do not
attempt to model that pathway of eligibility here. We apply the ACA’s statutory 5%
income disregard to all MAGI-eligible groups (groups who income is totaled using
the notion of Modified Gross Adjusted Income).
Fig. 1. Eligibility for Medicaid/CHIP by Income and State Medicaid Expansion Status. Notes: Top panel represents child eligibility and bottom panel represents adult
eligibility. Dashed vertical line indicates 138% of the Federal Poverty Level (FPL).
range, coverage is typically via CHIP and eligibility trails off – more
steeply in non-expansion states (which have traditionally been less
generous with coverage). For adults, expansion states offer eligibility to everyone with incomes up to 138% of FPL, while a minority of
adults in non-expansion states meet both income and categorical
criteria for eligibility. Even prior to the ACA expansion, eligibility
standards for adults were more generous in expansion states than
non-expansion states.
Exchange Premium Subsidies. Our second policy measure is the
subsidy rate for insurance purchased through the ACA’s exchanges.
Since exchange premiums are defined based on the family unit, our
analysis models the premiums and subsidies using the notion of the
health insurance unit (HIU) – defined as an adult, his/her spouse,
and their dependent children in the household, excluding unrelated
roommates or other adult relatives (such as grandparents). This
corresponds to the family unit upon which premium subsidies and
Medicaid eligibility is based, and we use the term “family” and HIU
interchangeably below.
To construct the subsidy measure, we first calculate an unsubsidized premium for each HIU based on the ACA rating area they
resided in. We directly matched premiums to individuals in cases
where a single rating area mapped directly to a PUMA and used
population-weighted premium averages in cases where multiple rating areas spanned a single PUMA. The HIU unsubsidized
premium is the sum of the individual premiums for each of its members, with no more than three covered children included in the sum
based on federal regulations. Individual premiums are based on the
M. Frean et al. / Journal of Health Economics 53 (2017) 72–86
second-lowest-cost silver plan in the rating area, obtained from
the Robert Wood Johnson Foundation. We use this plan for two
reasons: (1) the silver tier is the most commonly purchased tier,
selected by 65% of consumers in the first open enrollment period
(ASPE, 2014); and (2) the second-lowest cost silver plan is the one to
which the ACA’s premium tax credits are pegged. All unsubsidized
premiums are age-adjusted using state-specific age-rated premium
curves obtained from CMS.
Then, we calculate the net subsidized premium for each family. Families with incomes outside of 100–400% of FPL and those
eligible for Medicaid or CHIP are ineligible for subsidies. For the
remainder of the sample, net premiums are calculated based on the
ACA’s subsidy schedule, which determines premium payments on
a sliding scale percentage of income.7 In addition to net premiums,
we also calculate each HIU’s Percent Subsidy, equal to 1 – (Net Premium/Unsubsidized Premium). While both are measures of premium
subsidy generosity, they have important differences. The net premium only captures cost, while the percent subsidy incorporates
both the cost of coverage and its effective value to the consumer;
holding net premiums constant, the percent subsidy is higher for
older adults and those living in areas with more expensive health
insurance. A priori, we hypothesize consumers respond to both the
cost and value of coverage. Thus, percent subsidy is our preferred
parameter, but we test both and allow the data to indicate which
is a better predictor of behavior.8
Fig. 2 shows the percent subsidy as a function of FPL and Medicaid expansion status. In non-expansion states, premium subsidies
are available starting at 100% of FPL; in expansion states, where
such families are eligible for Medicaid, subsidies begin at 138% of
FPL.9 Subsidy rates peak at about two-thirds for non-expansion
states for those between 100–138% of FPL, and slightly over half
for those just above 138% of FPL in expansion states. The subsidy
rate then declines steadily but not quite linearly until going to zero
above 400% of FPL. Overall, the variation in premium subsidies is
a result of two factors – unsubsidized premiums and the ACA’s
income-based subsidy rules. While the former may vary across
markets based on potentially endogenous factors including health
care costs, PUMA fixed effects should address this concern, and it
is only in combination with the ACA’s tax credit schedule – which
is plausibly exogenous – that we obtain variation used to identify
the policy impact of the subsidies.10
Premium tax credits are pegged to the following thresholds: 2% of income
for individuals with incomes up to 133% of FPL; 3–4% of income for individuals with incomes between 133–150% of FPL; 4.0–6.3% of income for individuals
with incomes between 150–200% of FPL; 6.3–8.05% of income for individuals
with incomes between 200–250% of FPL; 8.05–9.5% of income for individuals with
incomes between 250–300% of FPL; and 9.5% of income for individuals with incomes
between 350–400% of FPL.
Yet a third measure of premium subsidy is also possible, taking into account the
ACA’s cost-sharing reductions (CSR) for individuals with incomes from 100–250%
FPL who are eligible for exchange subsidies. In an alternative analysis, we created a
premium subsidy measure that takes this into account by inflating the value of coverage (the unsubsidized premium) based on the CSR’s legislative increase in actuarial
value. Silver plans have an actuarial value of 70%, but the CSR increases this to 94%
for those with incomes 100–150% FPL, 87% for incomes 150–200% FPL, and 73% for
incomes 200–250% FPL. The resulting measure therefore reflects a higher percent
subsidy for those who would receive CSRs, and is equal to our original premiumbased measure for those ineligible to receive CSRs. Overall, the CSR’s increase the
mean subsidy rate from 16% to 17%, and the results of the model using this variable
are quite similar to our primary specification.
Legal permanent residents are not eligible for Medicaid until after a five-year
waiting period, but premium tax credits are available under the ACA for those with
incomes under 138%. The ACS does not enable us to distinguish between legal and
undocumented immigrants, though it does self-reported citizenship. We test the
robustness of our results by excluding non-citizens from our sample, and the results
are quite similar.
Another important component of the ACA is community rating of premiums. The
main effect of community rating is difficult to capture as it is essentially a time series
Fig. 2. Exchange percent subsidy in 2015 by Income and State Medicaid Expansion
Status. Notes: Dashed vertical lines indicate 138% and 400% of the Federal Poverty
Level (FPL).
Mandate. Our third policy measure is the tax penalty associated
with the individual mandate. Fundamentally, the existence of the
mandate is a time series change that cannot be separately identified
in our model. To the extent that the mandate creates a generalized
“taste for compliance” (Saltzman et al., 2015), our model is unable
to capture that effect. However, in principle, the mandate does not
impact those who are exempted, and due to non-linearities in the
mandate penalty, families may be exposed to different levels of tax
penalties for forgoing health insurance. We therefore construct a
measure representing each family’s tax penalty in dollars due to the
mandate. The penalty is equal to $0 for families exempt due to any
of the following (with the percentage of the sample affected by each
exemption in 2014 listed in parentheses): (1) family income below
the federal tax-filing threshold11 (20.7%); (2) family income below
138% of FPL in a state that elected not to expand Medicaid (5.5%);
(3) Native Americans (0.6%); or (4) no affordable coverage available,
defined as the lowest-cost option having a premium greater than
8% of family income (10.2%).12 For the roughly 64% of our sample
subject to the mandate, the family-level mandate penalty is calculated per ACA criteria: the greater of $95 per uninsured adult (half
that per child) or 1% of taxable income in 2014, and $325 per adult
or 2% of taxable income in 2015.13
Fig. 3 shows the average mandate penalty per family in 2015,
by income and Medicaid expansion status, while Appendix Fig. 1
effect. One could compute the “effective” subsidy to include the implicit subsidy of
community rating, but since we do not observe health status, we cannot include
implicit subsidization of the sick in such a calculation. But we do include the implicit
subsidies from compressed age rating, since our existing subsidy variable is largest
within a given income band for the most expensive (i.e., oldest) group.
In 2014, the tax-filing thresholds were $10,150 for single non-elderly individuals; $20,300 for married couples filing jointly; and $13,050 for ‘heads of household’
(i.e., multi-individual HIUs without a married couple).
The last exemption was based on the lowest-cost bronze-level plan in each rating area. provides county-level bronze premium data for states on
the federal exchange, which are not available in the data source we use for our silverlevel premiums. Thus, for the 16 states using state-based exchanges, we imputed
the lowest-cost bronze premiums for each rating area using a regression model to
predict the ratio of second-lowest-cost silver plan to lowest-cost bronze plan as a
function of the following variables: number of silver plans, ratio of maximum to minimum silver premium, ratio of maximum to second-lowest silver premium, ratio of
median to second-lowest silver premium, ratio of second-lowest to minimum silver
premium, and PUMA-level demographic measures from the ACS for age, sex, race,
citizenship, education, disability, parental status, marital status, and household size.
The mandate penalty is additionally capped at the national average premium
for bronze-level health plans offered by the health insurance exchanges, and those
M. Frean et al. / Journal of Health Economics 53 (2017) 72–86
+ˇ9 PercentSubsidy2014ij ∗ Yr2014t
+ˇ10 MandatePenalty2014ij ∗ Yr2014t
+ˇ11 McaidEligiblePreACAij ∗ Yr2014t
+ˇ12 McaidEarlyExpansionEligibleij ∗ Yr2014t
+ˇ13 McaidNewlyEligible2014ij ∗ Yr2014t
+ˇ14 PercentSubsidy2015ij ∗ Yr2015t
+ˇ15 MandatePenalty2015ij ∗ Yr2015t
+ˇ16 McaidEligiblePreACAij ∗ Yr2015t
+ˇ17 McaidEarlyExpansionEligibleij ∗ Yr2015t
+ˇ18 McaidNewlyEligible2015ij ∗ Yr2015t
+˝ Areaj ∗ HIU Typei + ∂ Yeart ∗ HIU Typei
Fig. 3. Individual mandate penalty in 2015 by Income And State Medicaid Expansion
Status. Notes: Dashed vertical lines indicate 138% and 400% of the Federal Poverty
Level (FPL).
depicts the percentage of families subject to any mandate penalty.
No one below 138% of FPL in non-expansion states is subject to the
mandate, while in expansion states, the mandate takes effect at
the tax-filing threshold. Between 138% and 400% of FPL, most families are subject to the mandate, with the penalty increasing with
income. Near and above the 400% FPL subsidy cutoff, substantial
portions of families are exempt based on the affordability criterion. At higher incomes, most families are subject to the mandate
and the average penalty approaches $1500 per family.
4. Empirical strategy
Our overall empirical strategy consists of a longitudinal design
that uses geographical and income-based variation in the ACA policy levers to identify changes in coverage over time, adjusting for
time, geography, and income. We use the 2012–2013 period to control for geographic and income group differences that might be
correlated with our outcomes of interest. Essentially, this allows us
to do a difference-in-difference-in-difference (DDD) model across
PUMAs, income groups, and time. Our model also separately identifies the policy effects in 2014 vs. 2015 since the policies themselves
evolved over time.
We have 8 policy parameters – two versions each (2014 and
2015) of the mandate penalty, new Medicaid eligibility based on
state expansion decisions, and premium subsidy rate; and then single measures of pre-ACA Medicaid eligibility and early expansion
eligibility (since neither policy changed between 2014 and 2015).
We model the direct effects of these policies in all four years of
the study (which includes 2 years of the pre-ACA baseline) and the
DDD estimates by interacting each term with post-ACA year fixed
%Uninsuredijt = ˇ0 + ˇ1 PercentSubsidy2014ij
+ Incomei ∗ HIU Typei +  AreaUnemploymentRatejt
+ˇx Xijt + εijt
Subscript i indexes the family (HIU), which is the unit of observation; j indexes the geographical area; and t indexes time (year). The
dependent variable is the percent of each HIU without insurance at
the time of the survey; for single adults, this is a binary variable,
for families with multiple members this is a continuous fraction
ranging from 0 to 1. ˇ1 through ˇ8 capture the baseline (pre-ACA)
direct effects of the PUMA-income policy variables. The coefficients
of interest are ˇ9 through ˇ13 , which measure the impact of the ACA
policy variables in 2014, and ˇ14 through ˇ18 , which measure the
policy impacts in 2015.14
˝ is a vector of area fixed effects (either PUMA or state, depending on the model), ∂ is a vector of year fixed effects, and  is a
vector of fixed effects for different income groups; all three fixed
effects were interacted with HIU type (single adults, adult couples,
and families with children), since each group has its own coverage
trends and policy responses. Xijt is a vector of the demographics
based on the adult(s) in the family: race/ethnicity, marital status,
citizenship, age, educational attainment, and number of children.
Finally, the model adjusts for annual county-level unemployment
rates from the Bureau of Labor Statistics.
Even with the DDD model, Eq. (1) raises several identification
concerns. Primary among these is state- or PUMA-level differences
in the income distribution that may be related to both premiums
and Medicaid expansion, as well as omitted factors correlated with
both family income and tastes for insurance. Another flaw is that the
mapping of survey-reported income onto ACA-related eligibility is
imprecise, creating measurement error biased toward the null.
We address many of these concerns through the use of a “simulated” measure of eligibility (Currie and Gruber, 1996a,b; Cutler and
Gruber, 1996). For this measure, we first group all families into 12
income bands.15 For each income band, we randomly select from
the national sample up to 200 families of each of three types –
+ˇ2 PercentSubsidy2015ij + ˇ3 MandatePenalty2014ij
+ˇ4 MandatePenalty2015ij + ˇ5 McaidEligiblePreACAij
+ˇ6 McaidEarlyExpansionEligibleij + ˇ7 McaidNewlyEligible2014ij
+ˇ8 McaidNewlyEligible2015ij
with only short periods without insurance (less than 3 months per year) are also
exempt from the fine.
A simpler alternative is to run two separate models: one for 2012–2014, and
the other for 2012–2015 that omits 2014 as a washout period. The overall results of
these separate models are essentially identical to the single model specified here.
The income bands were: 0–50% FPL, 50–100% FPL, 100–138% FPL, 138–200%
FPL, 200–250% FPL, 250–300% FPL, 300–350% FPL, 350–400% FPL, 400–500% FPL,
500–600% FPL, 600–800% FPL, and greater than 800% FPL. Tweaking these to stagger them across the ACA’s key income thresholds (e.g., using 100–150% FPL and
350–450% FPL instead) produces similar results in our regression models. In all models, we recoded negative incomes as $0, and incomes above the 99th percentile were
top-coded as the 99th percentile.
M. Frean et al. / Journal of Health Economics 53 (2017) 72–86
single adults, adult couples,16 and families with children – such
that the total number of individuals sampled per group is approximately 200. We then assign this same sample to each PUMA in
our dataset and estimate the value of our policy variables for that
family type-PUMA-income group cell.
The resulting measure computes, for example, the average subsidy for a representative set of single adults from 150–200% of FPL,
in each PUMA in the nation. Critically, this approach allows us to
capture the variation in subsidies by income group and PUMA,
but to also rigorously control for any direct influences of income
and PUMA by putting in a full set of 12 income category dummies
and PUMA dummies. That is, the only variation that identifies this
model is interactions of PUMA, income, and year, and not direct
effects of any of these factors.17
This approach mitigates several potential sources of bias. First,
it eliminates any endogeneity based on state-level differences in
income distributions. Some states have poorer populations or worsening economic conditions over time, which may affect premiums
as well as the share of the population eligible for Medicaid; using a
standardized population in all states removes any bias from this
source. Second, this approach reduces concerns about the measurement of survey income, since we are longer reliant on precise
family-level estimates of premium subsidies and other measures;
instead, we use an aggregate measure of the ACA’s policy features
for families within a given income band. Third, it reduces the potential for individual-level endogeneity of income in response to ACA
policies. We assign families to fairly broad income bands using their
actual income, but then use the simulated measure for the whole
income band to assess policy impact. In other words, the model
assumes that income group is not endogenous, even if a family’s
specific income within that group may be. Fortunately, other work
on the ACA suggests that employment responses to the law have
been minimal on both the intensive and extensive margins (Gooptu
et al., 2016; Kaestner et al., 2015; Moriya et al., 2016).
These simulated policy measures can serve as instruments for
each family’s actual premium subsidy, mandate penalty, and Medicaid eligibility as described in Eq. (1). The first-stage regression for
such a 2SLS estimate is close to one for each policy measure (see
Appendix Table 2), so that IV and reduced form estimation yield
almost identical answers. Thus, for most analyses we focus on a
reduced form model identical to Eq. (1), except that for each of the
policy variables and interaction terms (ˇ1 through ˇ18 ) we use simulated policy measures as the independent variables of interest. For
brevity, we do not re-state this equation, but this and the full 2SLS
equation are described in Appendix A.
Our primary model focuses on the percentage of each HIU that
is uninsured. In all models, we use ACS survey weights aggregated
at the HIU-level and robust standard errors clustered at the level of
the PUMA.
5. Results
5.1. Summary statistics and coverage trends
Table 1 presents summary statistics for our sample in 2014 and
2015. Nearly one-quarter of the population was Medicaid eligible
before the ACA, 2% gained eligibility under the ACA’s early expan-
Table 1
Summary statistics of simulated policy variables in 2014 and 2015.
Medicaid eligibility
Percent previously eligiblea
Percent eligible under ACA early expansion
Percent newly eligible in 2014
Individual mandate
Family mandate penalty
Subject to mandate penalty
Exchange premiums
Unsubsidized family premium
Net subsidized family premium
Percent subsidy
23.0% (31.9%)
2.0% (11.1%)
4.5% (18.2%)
22.7% (31.7%)
1.9% (10.9%)
5.5% (19.7%)
$458 ($632)
63.7% (41.0%)
$956 ($1210)
64.5% (40.5%)
$8023 ($3282)
$6631 ($3488)
16.2% (24.4%)
$8114 ($3298)
$6715 ($3519)
16.1% (24.3%)
Notes: Table presents weighted means, with standard deviations in parentheses, for
the population 0 to 64 years old. All measures are assessed at the level of the Health
Insurance Unit and use ACS survey weights, excluding the state of Massachusetts.
Based on state eligibility criteria as of 2013.
Table 2
Time series change in insurance outcomes by family type (2012–2015).
Employer sponsored insurance
Non-group private
Single adults
Employer sponsored insurance
Non-group private
Adult couples
Employer sponsored insurance
Non-group private
Families with children
Employer sponsored insurance
Non-group private
Notes: Table presents weighted means for the population 0–64 years old. All measures are assessed at the level of the Health Insurance Unit and use ACS survey
weights, excluding the state of Massachusetts.
sions, while approximately 5% became eligible in 2014. Overall,
nearly two-thirds of the sample was subject to the mandate, and
the size of the average mandate penalty more than doubled from
2014 to 2015 ($458 to $956). The mean unsubsidized premium
was slightly more than $8000 in both years. The subsidy rate was
approximately 16% in both 2014 and 2015.
Table 2 shows the time series for insurance outcomes. There
was a net decrease in the uninsured rate of roughly 3.4 percentage
points in 2014 and 6.0 percentage points in 2015, both compared
to the 2012–2013 period. By 2015, there had been a 3.3 percentage point increase in Medicaid, 0.9 percentage point increase in ESI,
and 1.9 percentage point increase in non-group private coverage.
Overall coverage gains were largest for single adults (10.8 percentage points by 2015), with smaller changes for couples (4.6) and
families with children (4.2).
5.2. IV and reduced form model results
This group contains families with 2 adults and no children 18 or younger.
Approximately 99% of the HIUs in this group are married couples. The others are
typically single parents with adult dependents (e.g., a 20 year-old student).
Unlike the original Currie-Gruber approach, our model simulates eligibility at
the PUMA rather than state-level. However, collapsing our simulated eligibility measure to a state-level estimate for each income band produces similar findings as our
main model.
Table 3 shows the ACA-related coefficients for the 2SLS and
reduced form approaches; coefficients for demographic covariates
are in Appendix Table 1. Due to computational constraints when
attempting to run the 2SLS model with the full set of PUMA fixed
effects, we control for state rather than PUMA in the IV model. The
estimates are virtually identical between the IV and reduced form
M. Frean et al. / Journal of Health Economics 53 (2017) 72–86
Table 3
Reduced form and IV estimates for ACA effects on percent uninsured.
2014 policy interactions
Family percent subsidy* 2014
Family mandate penalty* 2014 (in $100s)
Previously Medicaid-eligible* 2014
Early expansion Medicaid-eligible* 2014
Newly Medicaid-eligible* 2014
2015 policy interactions
Family percent subsidy* 2015
Family mandate penalty* 2015 (in $100s)
Previously Medicaid-eligible* 2015
Early expansion Medicaid-eligible* 2015
Newly Medicaid-eligible* 2015
Direct effects
Family percent subsidy2014
Family percent subsidy2015
Family mandate penalty2014 (in $100s)
Family mandate penalty2015 (in $100s)
Previously Medicaid-eligible
Early expansion Medicaid-eligible
Newly Medicaid-eligible2014
Newly Medicaid-eligible2015
(1) 2-stage
least squares
(2) Reduced
Notes: Standard errors in parentheses are clustered at the PUMA level. Dependent
variable was the percentage of each Health Insurance Unit without any health insurance. All variables are expressed at the level of the Health Insurance Unit (HIU)
and use ACS survey weights, excluding the state of Massachusetts, for the population aged 0–64 years old. Models control for HIU type (single adult, couple, family
with children); number of men and women in the family; number of children; educational attainment, age, and race/ethnicity of adults in the family; area-specific
annual unemployment rates; and year, income group, and state fixed effects each
interacted with HIU type. N = 5,458,170.
Significant at the 10% level.
Significant at the 5% level.
Significant at the 1% level.
models for each policy measure. For simplicity, we focus on the
reduced form estimates for the remainder of the paper, which also
enables us to consider a more robust set of PUMA-level fixed effects
and various interaction terms as discussed below.
We estimate a significant negative effect of the subsidy rate on
the risk of being uninsured. The subsidy rate estimate shows that
for each 1.0 percentage-point of subsidy, the uninsured rate fell
by 0.051 percentage points in 2014. This effect was nearly twice as
large in 2015, with a coefficient of 0.089. Put another way, each 10%
increase in average subsidy produced a decrease in the uninsured
rate of 0.89 percentage points in 2015, equal to roughly 2.4 million
Americans (given 273 million non-elderly Americans).
The coefficient on the mandate penalty is quite small in magnitude and presumably wrong-signed (i.e., higher mandate leads
to more uninsured). The magnitude of the coefficient implies that
each $100 in mandate in 2014 (when the average penalty was
roughly $460) increases the uninsured rate by 0.04 percentagepoints, which is negligible. The coefficient in 2015 was similarly
small – 0.03. This could be because individuals are not aware of the
precise exemption parameters, or because they do not respond to
the affordability exemption. It does not necessarily imply that the
mandate had no effect, though it does suggest that individuals did
not respond to their income-specific mandate. This still leaves open
the possibility of a more general impact of a “taste for compliance”
that some have hypothesized (Saltzman et al., 2015).
The coefficients on all three Medicaid eligibility variables are
highly significant. The results indicate a marginal reduction in the
uninsured rate of 8.9 percentage points in 2014 and 13.7 percentage points in 2015 among individuals made newly for Medicaid.
Take-up rates were even higher (10.7 and 19.7 points in 2014 and
2015, respectively) for those who became eligible under the ACA’s
early Medicaid expansions. Meanwhile, we also detect smaller but
significant insurance changes among those who were previously
eligible for Medicaid. Our coefficient suggests that the ACA expansion led to 2.6 and 4.6 percentage-point increases in coverage in
2014 and 2015, respectively, for those who were already eligible
for Medicaid prior to the ACA – the so-called “woodwork effect.”
Overall, our Medicaid findings build on prior analyses of
the Medicaid expansion, which used standard difference-indifferences methods and several data sources to estimate
2014 coverage gains ranging from 3 to 6 percentage points
(Courtemanche et al., 2016; Kaestner et al., 2015; Sommers et al.,
2015a). Importantly, those analyses did not attempt to model other
aspects of the law simultaneously and did not disentangle the various types of Medicaid eligibility – they present estimates for overall
take-up among the broad group of low-income adults. One paper
(Simon et al., 2016) that took an approach closer to our “newlyeligible” estimate separately identified childless adults, who were
the main group to become newly eligible; they found larger take-up
rates on the order of 15 percentage points, closer to our newlyeligible coefficient.
The next set of coefficients show the direct impact of our policy
measures when not interacted with 2014 or 2015 – i.e., the impact
in 2012–2013. Pre-existing and early expansion Medicaid eligibility
are negatively associated with uninsurance as one would expect.
It is somewhat surprising that there are significant coefficients
on several other policy measures – though these point estimates
are generally small. This suggests the possibility of omitted factors across PUMA-income cells that are correlated with both our
policy measures and coverage. We can address this concern by further enriching the model to incorporate interactions of PUMA and
income category, so that the identification purely comes from differences in effects within each PUMA-income category. We do so
in Table 5, with minimal impact on the results.
Fig. 4 plots the ACA policy coefficients from the reduced form
model for each year of our study as a visual test of our DDD
approach. For clarity, we have separated the 2014 and 2015 results.
We find generally flat trend lines for 2012–2013 before large
changes in 2014 and 2015 for premium subsidies, new Medicaid eligibility, and previous Medicaid eligibility. For the early expansions,
we see a slight downward trend in the uninsured coefficient from
2012–2013, consistent with the implementation of those expansions during that period, before a much larger drop occurred in 2014
and 2015. Meanwhile, the trend for the mandate penalty is essentially flat throughout the study period. This offers strong evidence
that our model is capturing a discontinuous change in outcomes
related to these policy measures in 2014 and 2015, rather than
spurious variation in outcomes that predated the ACA’s implementation.
5.3. Decomposing coverage changes by ACA policy provision
In Table 4, we apply our estimates to model the populationlevel changes in insurance coverage in 2014 that are accounted
M. Frean et al. / Journal of Health Economics 53 (2017) 72–86
Fig. 4. Policy coefficients by year. Note: Estimates are the reduced-form coefficients for each policy measure’s direct effect (see Appendix for reduced-form regression
equation), with the regression separately estimated for each year of the sample. Panel A shows results for the 2014 policy measures, and Panel B shows results for the 2015
policy measures.
Table 4
Projected time series impact of ACA policy variables on percent uninsured.
2014 effects
Family percent subsidy × 2014
Family mandate penalty × 2014 (in $100s)
Previously Medicaid-eligible × 2014
Early expansion Medicaid-eligible × 2014
Newly Medicaid-eligible × 2014
2015 effects
Family percent subsidy × 2015
Family mandate penalty × 2015 (in $100s)
Previously Medicaid-eligible × 2015
Early expansion Medicaid-eligible × 2015
Newly Medicaid-eligible × 2015
Reduced form
Population mean
(simulated measure)
Implied percentage
point change
Share of total
ACA-related change
Notes: Dependent variable was the percentage of each Health Insurance Unit without any health insurance. All variables are expressed at the level of the Health Insurance
Unit (HIU) and use ACS survey weights, excluding the state of Massachusetts, for the population aged 0–64 years old. Models control for HIU type (single adult, couple, family
with children); number of men and women in the family; number of children; educational attainment, age, and race/ethnicity of adults in the family; area-specific annual
unemployment rates; and year and state fixed effects both interacted with HIU type.
M. Frean et al. / Journal of Health Economics 53 (2017) 72–86
for by these aspects of the ACA. Over the period from 2012–2013
to 2014, the rate of uninsurance as measured by the ACS fell by
3.4 percentage points. We find that the average 16% subsidy to
exchange coverage in the full sample implies a reduction in uninsurance of 0.83 percentage points. The estimates for the mandate
are small (as well as inconsistent and non-significant in robustness
checks we present below), so we ignore this in our calculations. We
estimate that the 2014 Medicaid expansion to 4.5% of our sample
reduced uninsurance by 0.40 percentage points; the early expansions affected 2% of the sample and reduced uninsurance by 0.21
percentage points; and the “woodwork effect” – impacting 23%
of our sample – led to a decline in uninsurance of 0.60 percentage points. Taken together, the policy variables in our model sum
to nearly a 2.1 percentage-point reduction in the uninsured rate.
Of this total, 41% is attributable to premium subsidies, 20% to the
expansion of Medicaid eligibility in 2014, 10% the early expansion,
and 29% to the woodwork effect. The relative magnitudes of the
changes for each policy were quite similar in 2015. The prominence
of the woodwork effect is due to the fact that nearly 5 times as many
people (disproportionately children) were already eligible for Medicaid before the ACA than those made newly eligible in 2014. Our
estimates are slightly more oriented toward Medicaid gains than
previous estimates by Carman et al. (2015), but broadly consistent
with their results, despite different analytical approaches and data
sources. However, Carman and colleagues only assessed Medicaid
and premium subsidies, without examining different types of Medicaid coverage gains, heterogeneity across state Marketplace type,
or effects of the mandate penalty.
Overall, our parameterization of the ACA explains approximately 60% of the 3.4 percentage-point decrease observed in 2014
and a nearly equal fraction of the 6.0 percentage-point change
in 2015. Several other analyses have attributed nearly all of the
national change in coverage in 2014 to the ACA, even after adjustment for the improving economy (Blumberg et al., 2016; Sommers
et al., 2015a). In our model, the area unemployment rate is a significant predictor of coverage (with each percentage point drop
in unemployment reducing the uninsured rate by 0.2 percentage
points), but our ACA policy coefficients are nearly identical with
or without this adjustment. Thus, the remaining decline in uninsurance in 2014–2015 may be due to other unmeasured aspects of
the ACA, such as the social effect of the individual mandate, guaranteed issue requirements, simplification of purchasing coverage
due to the creation of the exchanges, and any measurement error
in our policy variables.
5.4. Robustness checks
Table 5 considers the robustness of our estimates; for simplicity, we list only the 2015 policy effects (the 2014 coefficients
follow a similar pattern, but with smaller magnitudes). Column
1 shows the same reduced form model used in Table 3, but
replaces state fixed effects with PUMA fixed effects. The results are
nearly identical to the baseline model. Columns 2–4 include various second-level interactions to test whether omitted variables
may be driving the results. Column 2 allows for an interaction
between PUMA and income categories. This allows us to drop
the direct effects of our simulated policy variables (set at the
PUMA-income level), and leaves only the policy interactions with
Year2014 and Year2015. The results again are nearly identical. Column 3 allows for PUMA-income interactions and income-year
fixed effects, to address possible time-varying differences in insurance trends across income groups unrelated to the ACA. This
model reduced the point estimates somewhat though with the
same basic pattern, except the mandate penalty becomes negative and non-significant. Column 4 tests PUMA-year interactions
and income-year interactions, with generally similar estimates.
We also consider replacing the Mandate Penalty variable with Any
Mandate (i.e., percent of families that are not exempt from the mandate) or the mandate penalty as a percentage of income. These
results (Columns 5 and 6) again demonstrate small and inconsistent
impacts of the mandate.
Finally, we consider an alternative premium subsidy measure.
Column (7) replaces Percent Subsidy with Net Premium (in $1000s),
which shows a significant positive effect of the premium on uninsurance – but the implied effect is much smaller than that captured
by the percent subsidy in our main model. Applying this coefficient
to the type of analysis shown in Table 4, we estimate a change in
uninsured in 2015 due to subsidies of just 0.73 percentage points,
as opposed to the 1.53 percentage points from the percent subsidy. Column (8) shows that when both measures are included
together, the coefficient for Percent Subsidy is unchanged from our
primary model, while the coefficient for Net Premium is essentially
zero. These results indicate that the percent subsidy – by capturing
information on both price and the potential benefits of coverage –
better reflects exchange consumer decision-making than the outof-pocket premium alone. This suggests, for example, that older
individuals or those in areas with more costly insurance are more
likely to take up exchange coverage than younger individuals or
those in cheaper rating areas, conditional on facing a similar net
premium after subsidies.
5.5. Results by type of insurance
Next, we decompose our findings on uninsurance into changes
among three types of coverage: Medicaid/government assistance
plan, employer-sponsored insurance, and non-group private insurance (Table 6). As discussed earlier, the ACS survey wording makes
it reasonable for respondents receiving subsidized exchange coverage to report either “Medicaid/government assistance plan for
those with low incomes” or “insurance purchased directly from an
insurance company.” The positive coefficient on percent subsidy
in the “Medicaid/government assistance plan” regressions indeed
suggests that some individuals report their publicly-subsidized
exchange coverage in this way. This is consistent with evidence
from the Census Bureau that some respondents in the ACS describe
their private coverage using this option (Pascale et al., 2016).
Still, as expected, we find the largest effect of the premium subsidies on non-group insurance. We estimate that each 10% rise
in subsidy increased the share of the population with non-group
insurance by 0.48 percentage points by 2015. At the mean subsidy
(16.2%) and baseline non-group coverage rate (8.8%), this implies
an elasticity of demand for non-group coverage of −0.09. If we treat
the subsidy coefficients on Medicaid as part of the exchange effect,
the elasticity is −0.17. While lower than the elasticity used in typical microsimulation modeling of the ACA (Gruber, 2011), the 2015
estimate is nearly twice as large as the 2014 estimate, suggesting
that increased awareness of the law and resolving technical challenges in the exchanges likely improved consumer responsiveness
over time. Our data source does not allow us to detect whether an
individual has an “affordable” offer of employer coverage (defined
by the ACA to be ≤9.5% of income), which precludes a person from
receiving a premium subsidy. However, previous research suggests
that only 1.1% of those who are uninsured or with individual coverage and potentially income-eligible for premium subsidies have an
employer offer for coverage (Dorn and Buettgens, 2013). Accounting for this omission and multiplying our elasticity by 1/.989 is
within the rounding error on our overall estimate.
The coefficient on the mandate penalty remains small, wrongsigned, and only statistically significant for non-group coverage.
We estimate highly significant impacts of all three Medicaid
variables on Medicaid coverage. These coefficients reflect marginal
take-up rates among those eligible for the program. Strikingly,
M. Frean et al. / Journal of Health Economics 53 (2017) 72–86
Table 5
Robustness to alternative specifications for ACA effects on percent uninsured.
Premium subsidies
Family percent subsidy* 2015
Family net subsidized premium ($1000s)* 2015
Individual mandate
Family mandate penalty ($100s)* 2015
Subject to mandate* 2015
Early expansion Medicaid-eligible* 2015
Newly Medicaid-eligible* 2015
Fixed effects (all interacted with HIU-type)
Year fixed effects
Income fixed effects
PUMA fixed effects
PUMA-income fixed effects
Income-year fixed effects
PUMA-year fixed effects
Family mandate penalty (percent of income)* 2015
Medicaid eligibility
Previously Medicaid-eligible* 2015



Notes: PUMA, public use microdata area. Standard errors in parentheses are clustered at the PUMA level. Dependent variable was the percentage of each Health Insurance Unit
without any health insurance. All variables are expressed at the level of the Health Insurance Unit (HIU) and use ACS survey weights, excluding the state of Massachusetts, for
the population aged 0–64 years old. Models control for HIU type (single adult, couple, family with children); number of men and women in the family; number of children;
educational attainment, age, and race/ethnicity of adults in the family; and area-specific annual unemployment rates. All fixed effects are interacted with HIU type. All
coefficients refer to 2015 policy estimates.
Significant at the 10% level.
** Significant at the 5% level.
Significant at the 1% level.
Table 6
ACA policy effects in 2015 by type of coverage.
Family percent subsidy* 2015
Family mandate penalty* 2015 ($100s)
Previously Medicaid-eligible* 2015
Early expansion Medicaid-eligible* 2015
Newly Medicaid-eligible* 2015
(1) Uninsured
(2) Medicaid or “government
assistance plan” for
low-income families
(3) Employer
(4) Non-group
Notes: Regressions in table include fixed effects from Model 1 described in Table 5. Standard errors in parentheses are clustered at the PUMA level. Dependent variable was
the percentage of each Health Insurance Unit without any health insurance. All variables are expressed at the level of the Health Insurance Unit (HIU) and use ACS survey
weights, excluding the state of Massachusetts, for the population aged 0 to 64 years old. Models control for HIU type (single adult, couple, family with children); number of
men and women in the family; number of children; educational attainment, age, and race/ethnicity of adults in the family; area-specific annual unemployment rates; and
year, income group, and PUMA fixed effects each interacted with HIU type. All coefficients refer to 2015 policy estimates.
Significant at the 10% level.
Significant at the 5% level.
Significant at the 1% level.
these Medicaid effects are very close to the effects for overall
insurance coverage; that is, we estimate virtually no crowd-out of
private coverage by the Medicaid expansion or woodwork effect.
This is illustrated further in the next two columns of Table 6. We
observe no negative impact of the Medicaid eligibility variables on
either ESI or non-group insurance (and in fact detect a significant
but small positive effect of pre-ACA eligibility on both, on the order
of 0.5–0.8 percentage points). This is a notable finding, as most previous literature suggested at least some crowd-out was likely under
the ACA.
At least one analysis of the ACA has detected moderate crowdout – roughly 25% for parents, less for childless adults (Kaestner
et al., 2015). However, this appears to be an artifact of modeling
Medicaid eligibility alone. In the 100–138% income range, individuals in non-expansion states are able to receive premium subsidies
as a fallback to Medicaid. This increases private coverage in those
states. In a simple differences-in-differences model with a simple
binary Medicaid expansion vs. non-expansion independent variable, this larger increase in private coverage in non-expansion
states appears as a form of “crowd-out” (i.e., a negative DD coef-
M. Frean et al. / Journal of Health Economics 53 (2017) 72–86
ficient on private insurance – see Appendix Table 3). However, this
is not crowd-out in the traditional sense: Medicaid is not leading
some to drop private insurance coverage. Rather, in the absence of
Medicaid expansion, some adults obtain premium subsidies. In our
model, which explicitly accounts for both Medicaid and premium
subsidies, we find no crowd-out at all.18
One previous coverage expansion without much crowd-out
occurred in Massachusetts, suggesting that the individual mandate may play an important role here (Hackmann et al., 2015). In
addition, Clemens (2015) showed that community rating in private
insurance – as required by the ACA – may also reduce crowd-out
from Medicaid.19
5.6. Heterogeneity in coverage changes
We examine patterns of ACA effects across different demographic groups and states (Table 7). To do so, we repeat our reduced
form analysis for several stratified samples, based on family type
(single adults, adult couples without children, and families with
children) and state policies. We compared states that had established their own exchanges in 2014 (n = 14) to those using the
federal exchange.20 We also compared states based on their ACA
Medicaid policies, classified into three groups – non-expansion
states (n = 21); early (2011–2013) expansion states (n = 6, including Washington DC); and states that expanded eligibility in 2014
or 2015 (n = 24).
Coverage gains associated with premium subsidies were significantly larger for adult couples (ˇ = −0.108) than single adults
(ˇ = −0.085) or for families with children (ˇ = −0.070). The 2015
effects of Medicaid eligibility were largest in for adult couples, with
take-up rates of over 18% for both previously-eligible and newlyeligible adults, compared to 5.1% and 11.7% for single adults. Among
families with children, the woodwork effect was smaller – 4.1% –
which likely reflects the fact that Medicaid/CHIP take-up rates for
children were already quite high (Kenney et al., 2011). However,
since children make up such a large portion of Medicaid eligibility, this is a non-trivial population effect. In children-only models,
we estimate that they account for 28–45% of the overall population
woodwork effect in 2015.21
In our analysis by state policy, exchange subsidies were significantly more effective at reducing the uninsured rate in states with
state-based exchanges than in states using the federal exchange.
The simple D-in-D model also shows a reduction in ESI associated with Medicaid
expansion. This effect also disappears in our full model, in which we do not simply
analyze Medicaid expansion as a binary variable but take into account how many
people actually gained eligibility.
Table 6 also includes some counterintuitive results that are statistically significant but of such small magnitude as to be economically negligible. For instance,
the coefficient on “Previously Medicaid Eligible * 2015” for ESI is significantly positive. However, the mean level of this independent variable is .227, which means
that the coefficient in question relates to a predicted increase in ESI of 0.0018 – less
than 2/10ths of a percentage point. The statistical significance here most likely just
reflects the very large sample size of the ACS.
The 14 states with state-based exchanges were CA, CO, CT, DC, HI, ID, KY, MA,
MD, MN, NY, RI, VT, and WA. For 2016, Hawaii has reverted to the federal exchange,
and Kentucky will do so for 2017.
This requires adapting our HIU-level model to run an individual-level analysis
for children only. If we use the same simulated instruments as our main model
for premium subsidies and the mandate, but substitute child-level Medicaid/CHIP
eligibility for the three-part Medicaid eligibility modeled in Eq. (1), we estimate
that the uninsured rate among children eligible for Medicaid or CHIP fell by 1.0
percentage point in 2014 and 1.8 percentage points in 2015, with no significant
private insurance crowd-out. At the population level, this accounts for 28% of the
overall woodwork effect in Table 4 (given that 57% of children were already eligible),
or equivalently, 800,000 additionally insured children in 2015. An alternative model
that includes only Medicaid/CHIP eligibility provides an upper bound of 1.4 million
additionally insured children, which would represent roughly 45% of the woodwork
Conditional on the subsidy amount, gains in coverage were essentially twice as high in the state exchanges (ˇ = −0.129) as in the
federal exchange (ˇ = −0.076). While technical difficulties plagued
the launch of the federal website, several state exchanges were
similarly affected, and these difficulties would not explain the
differences we observe well into 2015. More likely is that states
that implemented their own exchanges were more consistent
supporters of coverage expansion, with greater outreach efforts
and stronger application assistance programs (Shin et al., 2014;
Sommers et al., 2015b). However, our analysis here is merely suggestive and without a clear causal interpretation.
The pattern by Medicaid expansion decision also showed significantly larger effects of exchange subsidies in states more supportive
of the ACA (expansion states, particularly the early expansion
states). This finding suggests that the early eligibility expansions
from 2011–2013 laid the groundwork for increased Medicaid participation later on. Notably, we find large and similarly-sized
woodwork effects in all groups of states, regardless of Medicaid
expansion status.
6. Policy implications and conclusions
In what we believe is the most comprehensive analysis to date
of coverage changes under the ACA related to the law’s primary
policy measures, we identify several notable findings. First, of the
ACA’s reduction in the uninsured rate in 2014 and 2015 that is
attributable to our policy measures, roughly 40% was due to the
creation of premium subsidies for exchange coverage. The other
60% was due to increased Medicaid coverage – much of it the result
of enrolling individuals eligible for Medicaid before 2014, including many children. While some policymakers and researchers had
anticipated this potential “woodwork effect,” the fact that it is such
a large policy lever is somewhat surprising, and simple differencesin-differences models of the Medicaid expansion obscure this
important policy heterogeneity across eligibility groups.
In part, our large woodwork estimate may reflect some measurement error in Medicaid eligibility, and if some share of our
sample appeared eligible based on 2013 data but in fact was not
eligible until 2014 or 2015, this could bias our findings toward a
larger woodwork effect.22 However, federal administrative data on
Medicaid enrollment confirm that a substantial woodwork effect
is evident, and this effect exists whether or not a state expanded
Medicaid under the ACA. Even in non-expansion states, Medicaid
enrollment by January 2015 had increased by 8% over pre-ACA levels. In expansion states, of course, it had increased even more – by
26% – but our results suggest that a sizable portion of the gains
in these states was in fact from the woodwork effect (CMS, 2014).
Moreover, even among the childless adult group that comprises
the bulk of the newly-eligible population, as of late 2014 roughly
1/3 of this enrollment group was eligible under pre-ACA criteria
(CMS, 2016).23 These findings are also consistent with enrollment
spillovers detected in pre-ACA Medicaid expansions (Aizer and
Grogger, 2003; Dubay and Kenney, 2003; Sonier et al., 2013), as
well as one recent analysis of children’s coverage under the ACA
(Kenney et al., 2016).
Of course, the converse is also possible – our approach may define some
individuals as ineligible in 2013 even though they were eligible. But these two mismeasurement effects are likely to be asymmetric, since the marginal take-up rate
in 2014 among newly-eligible individuals should (and does) exceed the marginal
take-up rate among previously eligible individuals. Essentially, mismeasurement of
pre-ACA Medicaid eligibility should bias the woodwork coefficient upwards and the
newly-eligible coefficient downwards.
Massachusetts, due to its 2006 health reform law, and New York and Arizona,
due to their large 2002–2003 expansions of Medicaid under Section 1115 waivers,
were the largest contributors to this group in the CMS statistics.
M. Frean et al. / Journal of Health Economics 53 (2017) 72–86
Table 7
Uninsured results by family type, exchange type, and Medicaid expansion status.
Family type
Single adults
Number of
Family percent
subsidy* 2015
Family mandate
penalty* 2015
Early expansion
Adult couples
Families with children
Exchange type
Medicaid expansion status
No expansion
Early expansion in 2011–2013 1,130,446
Expansion in 2014 or 2015
Notes: Regressions in table include fixed effects from Model 1 described in Table 5. Standard errors in parentheses are clustered at the PUMA level. Dependent variable was
the percentage of each Health Insurance Unit without any health insurance. All variables are expressed at the level of the Health Insurance Unit and use ACS survey weights,
excluding the state of Massachusetts, for the population aged 0–64 years old. Models control for family type (single adult, couple, family with children); number of men
and women in the family; number of children; educational attainment, age, and race/ethnicity of adults in the family; and area-specific annual unemployment rates. All
coefficients refer to 2015 policy estimates.
Significant at the 10% level.
Significant at the 5% level.
Significant at the 1% level.
Another key finding is the lack of private insurance crowd-out.
We find no evidence of significant crowd-out of employersponsored coverage by the new premium subsidies, and no
evidence of crowd-out of either employer coverage or non-group
private coverage by the Medicaid expansion. These results have
implications for the ACA’s efficiency and effects on social welfare,
as expanding coverage without crowding out alternative sources
of health insurance reduces the law’s total cost and potential deadweight loss (Gruber, 2008).
In terms of premium subsidies, our findings offer some useful
insights for policy and future research. We find that modeling the
net premiums is a fairly weak approach to predicting enrollment
behavior, with coverage gains much more responsive to the percent subsidy received. It also suggests that much of the recent
attention to absolute premium rate increases may be less relevant than the subsidy rate received by most exchange customers.
By necessity, our model only examined a single representative
premium in each market – the second lowest cost silver plan. Further research is needed into how consumers enrolling in exchange
plans choose among their various options, in terms of the relative tradeoffs between overall subsidy rates, net premiums, and
other plan features such as cost-sharing requirements and provider
We find small and inconsistent effects of the individual mandate
in 2014 and 2015. In some models, the coefficient is wrong-signed,
while in others it is in the expected direction, with varying levels
of statistical significance. Overall, it is the least robust of our estimates, and in all models, the coefficients indicate minimal policy
impact. In part, this may indicate a lack of consumer awareness
about the intricacies of the tax penalty rules and exemptions. It
may also reflect the low levels of the mandate penalty in the law’s
early years, though we saw no increase in the mandate’s effect in
2015 even with steeper penalties. Finally, the mandate likely exerts
a generalized effect that encourages people to obtain coverage in
a way that is independent of its precise details and whether one is
even subject to it. In Massachusetts, for instance, researchers have
shown an increase in Medicaid participation among adults after the
implementation of the mandate, even though most had incomes too
low to make them subject to it (Sonier et al., 2013).
One of our paper’s main contributions is its comprehensive
framework for rigorous causal evaluation of the ACA’s effects. Given
the intense interest in many outcomes from this wide-ranging
law, including health care utilization, labor market outcomes, and
health effects, our approach will likely be useful for many subsequent analyses of the law.
However, the 2016 election results have cast significant doubt
over the law’s future. There has been discussion of a full repeal,
as well as more targeted changes such as repealing the individual mandate, shifting Medicaid to block grants, and scaling back
or eliminating premium tax credits. Our findings offer insight
into how each of these pieces of the law are interacting to produce coverage gains and how their repeal would likely do the
opposite. Overall, we find that the bulk of the coverage gains in
2014–2015 are directly attributable to expanded eligibility for subsidized insurance via exchanges and Medicaid. While we also detect
large gains in coverage for previously-Medicaid eligible populations (including many children), the underlying mechanism for
these gains in 2014–2015 is presumably a combination of the
ACA’s other features, such as a streamlined application process, the
elimination of Medicaid asset tests, the mandate, and expanded
eligibility for parents that likely improved Medicaid take-up rates
for children as well. It is therefore possible that much of these
latter coverage gains would also unravel after a repeal of the
ACA. Undoubtedly, how patterns of coverage evolve over time in
this policy environment – and how they affect other domains of
health care and the economy – will remain worthy of continued
This project was supported by grant number K02HS021291
from the Agency for Healthcare Research and Quality (AHRQ).
M. Frean et al. / Journal of Health Economics 53 (2017) 72–86
We are grateful to Kosali Simon, Jenny Kenney, participants in
the economics research seminar series at the University of Virginia, and the 6th Biennial Conference of the American Society of
Health Economists for helpful comments. Dr. Sommers has received
research grants from AHRQ, the Commonwealth Fund, and the
National Institute for Health Care Management. He served as a paid
part-time advisor for the U.S. Department of Health and Human
Services from Sept. 2012 through June 2016, and he is an unpaid
member of the National Advisory Board for the non-profit Institute
for Medicaid Innovation. The authors have no other disclosures to
Appendix A. Supplementary data
Supplementary data associated with this article can be found, in
the online version, at
Aizer, A., 2007. Public health insurance, program take-up, and child health. Rev.
Econ. Stat. 89 (3), 400–415.
Aizer, A., Grogger, J., 2003. Parental Medicaid Expansions and Child Medicaid
Coverage NBER Working Paper 21836. National Bureau of Economic Research,
Cambridge, MA.
Antwi, Y.A., Moriya, A.S., Simon, K., 2013. Effects of federal policy to insure young
adults: evidence from the 2010 Affordable Care Act’s dependent coverage
mandate. Am. Econ. J.: Econ. Policy 5 (4), 1–28.
ASPE, 2014. Health Insurance Marketplace: Summary Enrollment Report for the
Initial Annual Open Enrollment Period, From
reports/2014/MarketPlaceEnrollment/Apr2014/ib 2014Apr enrollment.pdf.
Baicker, K., Finkelstein, A., Song, J., Taubman, S., 2014. The impact of Medicaid on
labor market activity and program participation: evidence from the Oregon
Health Insurance Experiment. Am. Econ. Rev. 104 (5), 322–328.
Black, L.I., Cohen, R.A., 2015. Insurance Status by State Medicaid Expansion Status:
Early Release of Estimates From the National Health Interview Survey,
2013–September 2014. National Center for Health Statistics.
Blumberg, L.J., Garrett, B., Holahan, J., 2016. Estimating the counterfactual: how
many uninsured adults would there be today without the ACA? Inquiry (ePub
ahead of print).
Carman, K.G., Eibner, C., Paddock, S.M., 2015. Trends in health insurance
enrollment, 2013–15. Health Affairs 34 (6) (online before print).
Clemens, J., 2015. Regulatory redistribution in the market for health insurance.
Am. Econ. J.: Econ. Policy 7 (2), 109–134.
CMS, 2014. State Medicaid and CHIP Income Eligibility Standards. Centers for
Medicaid & CHIP Services, Baltimore, MD.
CMS, 2016. October–December 2014 Medicaid MBES Enrollment Report.
Department of Health & Human Services.
Cohen, R.A., Martinez, M.E., 2014. Health Insurance Coverage: Early Release of
Estimates from the National Health Interview Survey, January–March 2014.
National Center for Health Statistics.
Courtemanche, C., Marton, J., Ukert, B., Yelowitz, A., Zapata, D., 2016. Impacts of the
Affordable Care Act on Health Insurance Coverage in Medicaid Expansion and
Non-Expansion States NBER Working Paper 22182. National Bureau of
Economic Research, Cambridge, MA.
Currie, J., Gruber, J., 1996a. Health insurance eligibility, utilization of medical care
and child health. Q. J. Econ. 111 (2), 431–466.
Currie, J., Gruber, J., 1996b. Saving babies: the efficacy and cost of recent expansions
of medicaid eligibility for pregnant women. J. Polit. Econ. 104 (6), 1263–1296.
Cutler, D., Gruber, J., 1996. Does public insurance crowd out private insurance? Q. J.
Econ. 11 (1), 391–460.
DeNavas-Walt, C., Proctor, B., Smith, J., 2013. Income, Poverty, and Health
Insurance Coverage in the United States: 2012. U.S. Census Bureau,
Washington, DC.
Dorn, S., Buettgens, M., 2013. Verifying eligibility for affordable care act subsidies:
access to employer-sponsored insurance. Health Affairs Blog.
Dubay, L., Kenney, G., 2003. Expanding public health insurance to parents: effects
on children’s coverage under Medicaid. Health Serv. Res. 38 (5), 1283–1301.
Finegold, K., Gunja, M.Z., 2014. Survey Data on Health Insurance Coverage for 2013
and 2014. ASPE, Washington, DC.
Garrett, B., Kaestner, R., 2015. Recent Evidence on the ACA and Employment: Has
the ACA Been a Job Killer? Urban Institute, Washington, DC.
Garthwaite, C., Gross, T., Notowidigdo, M.J., 2014. Public health insurance, labor
supply, and employment lock. Q. J. Econ. 129 (2), 653–696.
Gooptu, A., Moriya, A.S., Simon, K.I., Sommers, B.D., 2016. Medicaid expansion did
not result in significant employment changes or job reductions in 2014. Health
Affairs 35 (1), 111–118,
Gruber, J., 2005. Tax policy for health insurance. In: Poterba, J. (Ed.), Tax Policy and
the Economy. MIT Press, Cambridge, MA, pp. 39–63.
Gruber, J., 2008. Covering the uninsured in the United States. J. Econ. Lit. 46 (3),
Gruber, J., 2011. The impacts of the affordable care act: how reasonable are the
projections. Natl. Tax J. 64 (3), 893–908.
Gruber, J., Levitt, L., 2000. Tax subsidies for health insurance: costs and benefits.
Health Affairs 19 (1), 72–85.
Gruber, J., Simon, K., 2008. Crowd-out 10 years later: have recent public insurance
expansions crowded out private health insurance? J. Health Econ. 27 (2),
Hackmann, M.B., Kolstad, J.T., Kowalski, A.E., 2015. Adverse selection and an
individual mandate: when theory meets practice. Am. Econ. Rev. 105 (3),
Hamersma, S., Kim, M., 2013. Participation and crowd out: assessing the effects of
parental Medicaid expansions. J. Health Econ. 32, 160–171.
Heberlein, M., Brooks, T., Aiker, J., Artiga, S., Stephens, J., 2013. Getting into Gear for
2014: Findings from a 50-State Survey of Eligibility, Enrollment, Renewal, and
Cost-Sharing Policies in Medicaid and CHIP, 2012–2013. Kaiser Family
Foundation, Washington, D.C.
Kaestner, R., Garrett, B., Gangopadhyaya, A., Fleming, C., 2015. Effects of ACA
Medicaid Expansions on Health Insurance Coverage and Labor Supply NBER
Working Paper 21836. National Bureau of Economic Research, Cambridge, MA.
Kaiser, 2010. Medicaid Financial Eligibility: Primary Pathways for the Elderly and
People with Disabilities. Kaiser Family Foundation, Washington, D.C.
Kaiser, 2015. Status of State Action on the Medicaid Expansion Decision, From
Kenney, G.M., Haley, J., Pan, C., Lynch, V., Buettgens, M., 2016. Children’s Coverage
Climb Continues: Uninsurance and Medicaid/CHIP Eligibility and Participation
Under the ACA. Urban Institute/Robert Wood Johnson Foundation.
Kenney, G.M., Lynch, V., Haley, J., Huntress, M., Resnick, D., Coyer, C., 2011. Gains
for Children: Increased Participation in Medicaid and CHIP in 2009. Urban
Institute/Robert Wood Johnson Foundation.
Kolstad, J.T., Kowalski, A.E., 2012. The impact of health care reform on hospital and
preventive care: evidence from Massachusetts. J. Public Econ. 96 (11–12),
Kronick, R., Gilmer, T., 2002. Insuring low-income adults: does public coverage
crowd out private? Health Affairs 21 (1), 225–239.
Lo Sasso, A.T., Buchmueller, T.C., 2004. The effect of the State Children’s Health
Insurance Program on health insurance coverage. J. Health Econ. 23 (5),
Long, S.K., Kenney, G.M., Zuckerman, S., Wissoker, D., Goin, D., Karpman, M.,
Anderson, N., 2014. QuickTake: Number of Uninsured Adults Falls by 5.4
Million since 2013. Urban Institute, Washington, DC.
Long, S.K., Stockley, K., Yemane, A., 2009. Another look at the impacts of health
reform in Massachusetts: evidence using new data and a stronger model. Am.
Econ. Rev.: Papers Proc. 99 (2), 508–511.
Mach, A., O’Hara, B., 2011. Do People Really Have Multiple Health Insurance Plans?
Estimates of Nongroup Health Insurance in the American Community Survey
SEHSD Working Paper Number 2011–28. U.S. Census Bureau, Washington, D.C.
Marquis, M.S., Long, S.H., 1995. Worker demand for health insurance in the
non-group market. J. Health Econ. 14 (1), 47–63.
Meng, Y.Y., Cabezas, L., Roby, D.H., Pourat, N., Kominski, G.F., 2012. Successful
Strategies for Increasing Enrollment in California’s Low Income Health
Program (LIHP). UCLA Center for Health Policy Research, Los Angeles, CA.
Moriya, A.S., Selden, T.M., Simon, K.I., 2016. Little change seen in part-time
employment as a result of the Affordable Care Act. Health Affairs 35 (1),
Pascale, J., Call, K.T., Fertig, A., Oellerich, D., 2016. Validating Self-Reported Health
Insurance Coverage: Preliminary Results on CPS and ACS. United States Census
Bureau, Washington, D.C.
Saltzman, E.A., Eibner, C., Enthoven, A.C., 2015. Improving the Affordable Care Act:
an assessment of policy options for providing subsidies. Health Affairs 34 (12),
Shartzer, A., Long, S.K., Karpman, M., Kenney, G.M., Zuckerman, S., 2015.
QuickTake: Insurance Coverage Gains Cross Economic, Social, and Geographic
Boundaries. Urban Institute, Washington, DC.
Shin, P., Sharac, J., Zur, J., Alvarez, C., Rosenbaum, S., 2014. Assessing the Potential
Impact of State Policies on Community Health Centers’ Outreach and
Enrollment Activities. George Washington University, Washington, DC.
Simon, K., Soni, A., Cawley, J., 2016. The Impact of Health Insurance on Preventive
Care and Health Behaviors: Evidence from the 2014 ACA Medicaid Expansions
NBER Working Paper 22265. National Bureau of Economic Research, Cambridge,
Smith, J.C., Medalia, C., 2015. Health Insurance Coverage in the United States: 2014.
U.S. Census Bureau, Washington, DC.
Sommers, B.D., Arntson, E., Kenney, G.M., Epstein, A.M., 2013. Lessons from early
medicaid expansions under health reform: interviews with medicaid official.
Medicare Medicaid Res. Rev. 3 (4), E1–E23.
Sommers, B.D., Epstein, A.M., 2011. Why states are so miffed about Medicaid –
economics, politics, and the “woodwork effect”. N. Engl. J. Med. 365 (2),
Sommers, B.D., Gunja, M.Z., Finegold, K., Musco, T., 2015a. Changes in self-reported
insurance coverage, access to care, and health under the Affordable Care Act.
JAMA 314 (4), 366–374,
M. Frean et al. / Journal of Health Economics 53 (2017) 72–86
Sommers, B.D., Kenney, G.M., Epstein, A.M., 2014. New evidence on the affordable
care act: coverage impacts of early medicaid expansions. Health Affairs 33 (1),
Sommers, B.D., Kronick, R., Finegold, K., Po, R., Schwartz, K., Glied, S., 2012a.
Understanding Participation Rates in Medicaid: Implications for the Affordable
Care Act. U.S. Department of Health and Human Services (ASPE), Washington,
Sommers, B.D., Maylone, B., Nguyen, K.H., Blendon, R.J., Epstein, A.M., 2015b. The
impact of state policies on ACA applications and enrollment among
low-income adults in Arkansas, Kentucky, and Texas. Health Affairs 34 (6),
Sommers, B.D., Tomasi, M.R., Swartz, K., Epstein, A.M., 2012b. Reasons for the wide
variation in medicaid participation rates among States hold lessons for
coverage expansion in 2014. Health Affairs 31 (5), 909–919,
Sonier, J., Boudreaux, M.H., Blewett, L.A., 2013. Medicaid ‘welcome-mat’ effect of
affordable care act implementation could be substantial. Health Affairs 32 (7),
Thorpe, K.E., Florence, C.S., 1998. Health insurance among children: the role of
expanded Medicaid coverage. Inquiry 35 (4), 369–379.
Wherry, L.R., Miller, S., 2016. Early coverage, access, utilization, and health effects
associated with the Affordable Care Act Medicaid expansions: a
quasi-experimental study. Ann. Intern. Med.,

How to place an order?

Take a few steps to place an order on our site:

  • Fill out the form and state the deadline.
  • Calculate the price of your order and pay for it with your credit card.
  • When the order is placed, we select a suitable writer to complete it based on your requirements.
  • Stay in contact with the writer and discuss vital details of research.
  • Download a preview of the research paper. Satisfied with the outcome? Press “Approve.”

Feel secure when using our service

It's important for every customer to feel safe. Thus, at HomeworkGiants, we take care of your security.

Financial security You can safely pay for your order using secure payment systems.
Personal security Any personal information about our customers is private. No other person can get access to it.
Academic security To deliver no-plagiarism samples, we use a specially-designed software to check every finished paper.
Web security This website is protected from illegal breaks. We constantly update our privacy management.

Get assistance with placing your order. Clarify any questions about our services. Contact our support team. They are available 24\7.

Still thinking about where to hire experienced authors and how to boost your grades? Place your order on our website and get help with any paper you need. We’ll meet your expectations.

Order now Get a quote