where yij is a dummy variable indicating a particular type of health insurance outcome
for person i in state j (such as enrollment in Medicaid); Xij a vector of socioeconomic
characteristics defined below; tij a dummy variable set to unity if the observation refers
to the post-PRWORA period (i.e., calendar years 1998–2000); Iij a vector of two dummy
variables indicating if the person is a naturalized citizen or a non-citizen (the left-out variable
indicates if the person is native-born); and Gj is the dummy variable indicating the state’s
generosity towards immigrants, set to unity if the state did not go beyond-the-minimum level
of assistance offered to pre-enactment or post-enactment immigrants during the 5-year bar.
Specifically, Gj is set to unity if the state did not offer any of the programs listed in the first
two columns of Table 2. Finally, the standard errors are clustered by state-immigration cells
to adjust for possible serial correlation in insurance outcomes at the state level for each of
the three immigration status groups.
For simplicity, the regression specification in (1) uses a three-way classification of the
immigration status of the population (i.e., natives, naturalized citizens, and non-citizens). I
account for the immigrant’s refugee status as well as year of entry into the United States
by including these characteristics as regressors in the vector X. The other socioeconomic
characteristics in this vector include: the person’s age, gender, race, and educational attainment,
the number of persons in the household, and the number of children, elderly persons,
and disabled persons in the household.15 The regression also includes the state’s unemployment
rate at time t, as well as the unemployment rate interacted with the dummy variables in the immigration vector I. These interactions control for the possibility that immigrant
outcomes are more sensitive to the business cycle than those of natives (as well as net out
any potential correlation between the generosity variable, G, and the state unemployment
rate).16 Because the generosity dummy variable is set to one for states that did not replace
the lost federal benefits, the coefficient vector θ in Eq. (1) measures the impact of the federal
cutbacks on the relative trend in immigrant health coverage. In particular, it measures the
extent to which the pre- and post-PRWORA change in coverage differs between states that
were less generous and states that were more generous.
Table 4 reports the triple-difference coefficient vector θ estimated from a number of
alternative specifications of the model. The specification reported in the first column of
the table includes only the variables in the vector X, while the specification reported in the second column adds a vector of state fixed effects, and these fixed effects are interacted
with both the time dummy variable (ti), as well as with the immigrant status vector
(I). The state-time interactions capture not only state-specific differences in the level of
health insurance, but also state-specific changes in health insurance coverage rates (induced
perhaps by varying economic and political conditions). Similarly, the state-immigration
status interactions net out the possibility that there may be state differences in health
insurance coverage (and in the trends) across the various immigration status groups. Finally,
the last two columns of Table 4 replicate the regression analysis in the sample of
children.
The top panel of the table estimates the impact of the state policies on the relative
change in Medicaid enrollment. In the full-interaction specification, the triple-difference
coefficient for non-citizens is −0.049 (with a standard error of −0.025) in the sample of all
persons, and −0.105 (0.048) in the children’s sample. The state policies, therefore, had a
significant impact on Medicaid participation in the non-citizen population. In other words,
non-citizens residing in states that did not offer state-funded assistance programs to their
immigrant populations experienced a significant decline in their Medicaid participation
rates, and the decline was particularly steep for non-citizen children. In contrast, these
programs did not affect the relative Medicaid participation rate of citizens or of the children
of citizens.
The middle panel of the table estimates the regression using a different dependent
variable, namely an indicator of whether the person has any type of health insurance
coverage. To the extent that the Medicaid cutbacks generate a larger pool of uninsured
non-citizens, one would expect the relevant coefficient in the vector θ to be negative
and significant. However, this coefficient is positive. In particular, it takes on a value
of 0.024 (0.021) in the sample of all persons, and 0.022 (0.031) in the sample of children.
In other words, there is no evidence that the welfare cutbacks significantly reduced
the aggregate health insurance coverage rate in the targeted group of non-citizens. In
contrast, the health insurance coverage rate actually increased in the states that were
the least generous and did not attempt to attenuate the presumed adverse impacts of
PRWORA.
Finally, the bottom panel helps to resolve the puzzle of declining Medicaid participation
and stable (or increasing) health insurance coverage by showing how the state-funded assistance
programs influenced the probability that immigrants were covered by employersponsored
insurance. The coefficient for non-citizens in this regression is 0.101 (0.026)
in the sample of all persons, and 0.147 (0.049) in the sample of children.17 In other
words, immigrants who lived in states that did not provide generous assistance programs
to their immigrant populations after 1996 became substantially more likely to be covered
by employer-sponsored insurance. This increase in ESI helped to greatly attenuate the potential
adverse impact of the welfare cutbacks on the number of non-citizens who lack health insurance. In contrast, the probability that citizens are covered by ESI does not
strongly depend on the provision of state-funded assistance (the coefficient is negative, but
insignificant)