Does the State Children’s Health Insurance Program Reduce School Absenteeism?
Ph.D. Student in Public Administration
Graduate Research Associate
Center for Policy Research
The Maxwell School of Citizenship and Public Affairs
426 Eggers Hall
Syracuse, NY 13244-1020
Senior Policy Officer
New Visions for Public Schools
Hospital for Special Surgery
State of New Jersey
Manager, Political Programs
American Institute of Architects
The authors would like to thank Ross Rubenstein, Amy Ellen Schwartz, John Yinger, Tim Smeeding, and attendees
of the American Education Finance Association Annual Meeting for their helpful comments and suggestions, and
Tejas Shukla of the Centers for Medicare & Medicaid Services for providing the Medicaid data, which have all
contributed to improving this study.
Enacted in 1997, the State Children’s Health Insurance Program (SCHIP) represented the
largest expansion of U.S. public health care coverage since the passage of Medicare and
Medicaid 32 years earlier. Even though the reauthorization of the program is currently up in the
air, there is a considerable lack of thorough and well-designed evaluations of the program. In this
study, we use school attendance as measure of the program’s impact. Utilizing state-level data
and the use of fixed-effects regression techniques, we conclude that SCHIP has had a positive
and significant effect on state average daily attendance rates, as measured by both SCHIP
participation and eligibility rates. The results support the renewal and expansion of the program.
Enacted in 1997, the State Children’s Health Insurance Program represented the largest
expansion of public health insurance since the creation of Medicare and Medicaid in 1965. The
program was intended to provide coverage for children whose families were too wealthy to
qualify for Medicaid but not wealthy enough to purchase private insurance. In fiscal year 2006,
SCHIP served 7.4 million people at some point, including 6.6 million children (Pear, 2007a;
In this paper, we examine the relationship between SCHIP and student attendance.
School attendance is crucial to achieving America’s educational goals because it ensures
continuity in learning and consistency of curricula. Research has demonstrated a positive
correlation between student attendance and achievement on standardized examinations (Lamdin,
1996). There is also a long literature demonstrating a link between early student absenteeism and
later drop out (Epstein & Sheldon, 2002).
The panel nature of our dataset allows us to control for elements in the states that do not
change over time but may be correlated with SCHIP programs and school attendance. We isolate
the effect of SCHIP on school attendance by investigating the association between within-state
variation in SCHIP programs and school attendance. SCHIP is shown to be effective in reducing
school absenteeism, which may indicate that the benefits of SCHIP go beyond improving access
to care for children. Moreover, we conclude that different state approaches toward implementing
and operating SCHIP programs vary in their efficacy. In particular, the effect of SCHIP on
school attendance appears to be greatest when eligibility is most inclusive.
Section two provides further background on the State Children’s Health Insurance
Program. Section three reviews the literature on student absenteeism. Section four presents the
empirical methodology and data used in this analysis. Our results are then summarized and
presented in section five. Section six concludes and offers suggestions for further research and
implications for public policy.
Before discussing whether the State Children’s Health Insurance Program has been
successful in reducing school absences for children, it is useful to understand the climate in
which the program arose. In 1997, the year SCHIP was enacted by Congress and signed into law
by President Clinton as part of the Balanced Budget Act, nearly 11 million children – defined as
those age 18 and under– were without any health insurance coverage, up 1.2 million from only
two years earlier. SCHIP was intended to expand health care coverage to children in these
families who were not poor enough to qualify for Medicaid but not wealthy enough to pay for
To remedy this problem of pediatric uninsurance, the State Children’s Health Insurance
Program was enacted with the purpose of providing at least basic public health insurance
coverage to children and adolescents through age 18 in families living above the poverty line. By
fiscal year 2005, over 6.1 million children were enrolled in some form of SCHIP at some point
according to the U.S. Centers for Medicare and Medicaid Services (CMS) (see Figure 1).
Within broad federal guidelines, each state determines eligibility requirements such as
income limits, the benefits package available, payment levels for coverage, and operating
procedures (Centers for Medicare & Medicaid Services, 2006). The states were also given three
ways in which to implement their version of the program: Expand their existing Medicaid
programs, create a separate SCHIP program, or a combination of the two approaches.
The generosity of the program varies widely from state to state, especially in terms of
cost-sharing mechanisms, such as co-payments and premiums, as well as eligibility
requirements. Because of these differences, we believe it is important to examine the variations
in generosity to determine not only if SCHIP itself has impacted school attendance rates but also
whether the structure of the program has a significant effect on its success.
The State Children’s Health Insurance Program has generally been considered a success
by most researchers and policymakers as well. Particularly compelling have been the effects of
SCHIP on measures of access to care. According to data from the Centers for Disease Control
and Prevention, while 28% of children with no health insurance do not have access to a usual
place of care, only three percent of children enrolled in Medicaid and SCHIP have the same
problem, only one percent more than children with private insurance. Only 2% of children
insured by SCHIP and Medicaid are reported to have unmet medical needs versus 12% for
uninsured children and one percent for children enrolled in private insurance (Lyons, 2007).
A study by Kempe, Beaty, Crane, Stokstad and Barrow (2005) of Colorado’s SCHIP
program, also demonstrated improved utilization figures. This study found significant increases
in the percentage of families who received any routine care or specialty care in the year after
enrollment in SCHIP when compared to the year before. However, the authors did not find
significant differences in the mean number of visits for any type of care, emergency room care,
or hospitalizations. These results are consistent with the findings of the congressionally
mandated evaluation of the program. The report found that 91 percent of SCHIP enrollees had a
usual source of medical care. Compared to their experiences before enrolling in SCHIP, SCHIP
enrollees received more preventative care, had fewer unmet needs, and had better access to, as
well as communication with, health care providers (Wooldridge, Kenney, & Trenholm, 2005).
There is also evidence to suggest that SCHIP has had the effect of reducing racial
disparities on measures of access. Using multiple regression techniques, Shone, Dick, Klein,
Zwanziger and Szilagyi (2005) find that New York State’s SCHIP program has had the effect of
reducing disparities between white, black, and Hispanic children on various measures of medical
care access including access to a usual source of care and preventive care. The share of children
within each racial subgroup with unmet health care needs also fell and equalized across the
various subgroups after enrolling in SCHIP.
Despite the gains that have been made since the program’s inception in 1997, the long
term outlook of SCHIP is uncertain. Originally set to expire on September 30, 2007, the program
continues to operate at current spending levels with few policy changes through a continuing
resolution. A full reauthorization of the program is up in the air while President Bush and
Congress resolve significant differences regarding the mission of the program. Both Houses of
Congress have passed a bill that would expand SCHIP coverage to nearly four million additional
children at a cost of an additional $35 billion over five years, $30 billion more than the President
has proposed (Pear 2007b, 2007c).
This paper contributes to the debate on the future of the State Children’s Health Insurance
Program by examining the existence of externalities from SCHIP. We look at the impact of
SCHIP on school attendance. We believe attendance is an important educational goal and an
example of an externality that SCHIP might generate to be effective. In the next section, we
review the academic literature on student absenteeism before relating it to SCHIP.
According to data from the 2005 National Assessment of Educational Progress (NAEP),
19 percent of American fourth-graders and 20 percent of American eighth-graders report missing
three or more days of school in the previous month. Though a significant number, these children
are far from the norm. Fifty-two percent of fourth-graders and 45 percent of eighth-graders
report having perfect attendance in the previous month. These patterns have remained fairly
stable since 1994 and differ along racial lines with American Indian and African-American
students having the highest rates of absenteeism (see Figure 2) (U.S. Department of Education,
National Center for Education Statistics, 2006). Nevertheless much work remains to be done as
internationally the United States falls in the middle of the pack of nations when it comes to
school attendance, with only 18 percent of students attending schools with a high level of good
school and class attendance (Mullis, Martin, Gonzales, & Chrostowski, 2004). Similarly,
American children perform close to the international average on assessments in science and
mathematics, and one cannot help but to suggest a relationship between the two variables.
Within the United States, the academic literature has increasingly documented a link
between student absenteeism and adverse outcomes inside and outside of school. Using school-
level data from Baltimore public elementary schools and education production function
techniques, Lamdin found a positive and statistically significant effect of attendance on
achievement, though he acknowledges endogeneity may also be driving the results. A similar
methodology was used by Caldas (1993), who found attendance to be a positive and significant
factor in secondary school achievement, particularly in central city schools. In this study of
Louisiana public schools, attendance by itself accounted for 5.5% of the variance in secondary
school achievement. Lamdin and Caldas’ findings are supported by research from Minneapolis,
Minnesota. Data from Minneapolis indicate that 63% of children who attended school 95-100%
of the time passed the Minnesota Basic Standards Test in reading, versus only 33% of children
who attended school less than 85% of the time. The same trends are also evident for the
mathematics examination (Minneapolis Public Schools, 2007). Similarly, researchers in
Rochester, New York found that on average, students who scored in the 85th to 100th percentile
attended school 93 percent of the time. Students who attended school only 85 percent of the time
scored below the 54th percentile on the statewide English regents exam (Skip class, lose ground,
The mechanisms that motivate this correlation between attendance and achievement are
unclear. There is little doubt that absenteeism has negative implications for the continuity of
learning, which may be necessary to perform well in school and on examinations. In addition,
good attendance may help children to learn good work and study habits, habits that may prove
useful in preparing for exams (Mohonasen Central School District, 2003). However, it is more
than plausible that attendance is merely capturing the effect of another variable, like motivation,
that is the true causal determinant of achievement.
The academic literature has also documented a relationship between student absenteeism
and high school dropout rates. Both cross-sectional studies including Kaplan, Peck and Kaplan
(1995); Rumberger (1987); and Rumberger, Ghatak, Poulos, Ritter, and Dornbusch (1990) and
longitudinal studies including Alexander, Entwisle, and Horsey (1997); Barrington and
Hendricks (1989); Ensminger (1992); and Rumberger (1995) have found that children who
eventually end up dropping out of school are absent more often than other students as early as
first grade (Epstein & Sheldon, 2002). If the link is true, reducing student absenteeism may also
have the benefit of reducing school dropout later in a child’s life.
In general, strategies for addressing the problem of student absenteeism fall into two
categories: the behavior modification approach and the needs-based approach. The behavior
modification focuses on policies that are directly within a school’s control like detentions, grade
reductions and parent-teacher conferences. The needs-based approach takes a more holistic view
on student absenteeism and tries to address the underlying issues behind absenteeism (Muir,
2005). Increasingly, this more holistic view has meant taking a look at the interaction of health
status and educational outcomes.
Coleman, et al. (1966) concluded that the strongest predictors of academic performance
were not school based factors as per pupil expenditures, but individual student characteristics
like household income and parental socio-economic status (Center for Advanced Social Science
Research, 2001). Despite this finding, education reform has continued to emphasize
improvement within schools and classrooms and not by more external factors as health status.
This fact has been particularly true in the case of attendance policy, despite a well-established
relationship between health and school attendance.
A study by the Johns Hopkins Medical Center found that 56 percent of children suffering
from nighttime asthma attacks were awakened one or more nights and 30 percent reported
missing one or more days of school (Johns Hopkins Medical Institutions, 1999). Researchers
from the University of Pennsylvania and Temple University found that elementary school
children who were overweight or obese missed more days of school than their peers (Loviglio,
2007). Treating school attendance as endogenous within a simultaneous three-stage least squares
model, Wolfe (1985) finds that various health characteristics, including having problems with
strenuous activity and number of doctors visits have significant effects on the number of days
missed. She also finds that each day of school missed in this model was negatively and
significantly associated with declines in achievement. Her findings are reinforced by Kahn and
Nursten (1996), who find that students with chronic health problems are more likely to miss
school, and by the Institute of Medicine (2003).
While the link between health and absenteeism may seem clear, the relationship between
health insurance and absenteeism is more opaque. A lack of confirmatory research linking health
insurance and health status is partially responsible for this difficulty. In a review of the literature
on health insurance and health status, Levy and Meltzer (2004) find evidence strongly suggesting
that policies to expand health insurance coverage can also promote health, but that the effect size
of health insurance on health greatly depends on the population that is being studied. It is also
entirely possible that the link between health insurance and attendance is unrelated to health.
Especially in the case of public health insurance programs, expansions in coverage may allow
parents to take jobs that do not offer health insurance for dependents, which can in turn mean
work hours more conducive to transporting children to school.
Lykens and Jargowsky (2002) find a positive link between private insurance and
attendance for non-Hispanic white children with private insurance relative to non-Hispanic white
children without private insurance, but their estimates of Medicaid insurance are not statistically
significant. Kahn and Nursten find a similarly ambiguous result: 21 percent of children with
private insurance missed one or more days of school, compared with 34 percent of children
covered by Medicaid and 30 percent of children without insurance. On the other hand, an
evaluation of Iowa’s child health insurance program found that the program did significantly
decrease the number of days that a child was unable to perform his or her normal activities
because of illness. However, when measuring days of school missed, they find that the sign was
the expected direction but the coefficient was not significant (Damiano & Tyler, 2005).
While the link between public health insurance programs and school absenteeism has not
been well established, other studies suggest a connection between the absence of any insurance
and absence from school. A Government Accountability Office (GAO) report concludes that
uninsured children are more frequently hospitalized for conditions that, had the children had a
primary care physician, could have been treated on an outpatient basis and thereby enabled the
children to miss fewer days of school (Gutowski, Ratner, Avruch, & Lamb, 1997).
In addition, a report by Florida’s Healthy Kids’ initiative claims that while the average
child misses four days per year due to illness, children who are uninsured are 25 percent more
likely to miss school than children who have insurance (Florida Healthy Kids Corporation,
1997). So while there is some support for the direct link between health insurance and
absenteeism, it rests primarily on the strength of privately insured children; the evaluations of
children with public insurance are neither conclusive nor comprehensive regarding the effect that
public insurance may have on attendance rates.
As summarized above, research has consistently demonstrated that attendance matters.
Attendance is associated with higher achievement. Absenteeism in primary school is associated
with dropout later in high school. Given these strong correlations, public health insurance
programs like SCHIP may be insuring against not only against future health problems but also
absenteeism and a host of other adverse associated academic outcomes as well.
This study aims to build on previous research in several important ways. First, it offers an
evaluation not only of SCHIP in general, but of the structure of state SCHIP programs, which
can and do vary to a considerable degree (Dubay, Hill, & Kenney, 2002). Such a comparison is
largely lacking in the current body of literature, as SCHIP evaluations tend to focus on a small
sample of states (Kenney & Chang, 2004). Second, the measurement of variation in the data over
time should help reduce the probability that any potential difference in outcomes is due to
unobserved characteristics of a particular state or a particular year instead of the SCHIP program
Using SCHIP participation as the determinant of attendance requires controlling for a
number of variables that may also affect attendance rates. Some of them – for example,
educational institutions – do not need to be controlled for directly, because they are likely time-
invariate and therefore will be controlled for through panel data techniques. Other variables, such
as socioeconomic status and race, are incorporated in the model and thus necessitated data
collection and analysis. Overall, we controlled for several variables in our study that we believe
– and the existing literature on attendance and children’s health insurance suggest – are
substantial factors in explaining absence.
The structure of our project and its intended goal – to measure the impact of SCHIP on
school attendance rates – necessitated the retrieval of three types of data: 1) a measurement of
school attendance rates, 2) measurements of SCHIP, and 3) other control variables that are
theorized to have an impact on attendance rates and may be correlated with measures of SCHIP.
We dropped the states of Hawaii and Alaska in our study due to their unusual funding
and districting characteristics. Hawaii, is the only state that acts as one unified school district,
while Alaska is the only state where residents receive tax revenues from the U.S. government
instead of paying taxes to the government. We also did not include Washington, D.C. in our
analysis, due to the distinct government structure of this federal district.
States primarily implemented their distinct SCHIP programs in 1998 or 1999. To
ascertain the implementation’s effect, we aimed to gather data on each variable in each state for
several years before the program’s initiation (i.e. dating back to 1992) to observe attendance
rates and its determinants prior to program implementation, and for four to five years after
implementation took place. In total, we gathered data on all of our variables for eleven years:
1992 through 2003. Descriptive statistics for each of these variables can be found in Table 1.
Dependent Variable. We used a measurement of attendance known as ―average daily
attendance,‖ or total number of students who attend school on a given day, averaged over the
entire school year. We then computed an average daily attendance rate for each state for each
year, expressed as a percentage of ―average‖ enrollment. These data are from National Center for
Education Statistics (NCES).
Average enrollment was calculated by averaging initial enrollment during the school year
the data were collected and the year following. This adjustment to enrollment was necessary to
correct for potential problems resulting from within-year enrollment growth. All percentage
measures in this study are expressed either as a percentage of average enrollment or average
The attendance rate varies from a low of 85.3 percent in Kentucky for 2003 to a high of
99.9 percent for several years in New Mexico and Virginia. Because of the adjustment to the
denominator of this variable as well as unaccounted for movement into the state, an observation
can theoretically exceed 100 percent attendance. These observations were reset to 99.9. Overall,
we believe that state and year fixed effects will control for such attendance rate inflation.
The mean attendance rate is 92.5 percent with a standard deviation of 2.7 percent. While
attendance rates change slowly from year to year within states, the sample affords enough
variation over time and across states to allow for the measurement of various effects on
attendance rates in our fixed-effects model.
Independent Variables of Interest. The crucial data on SCHIP were obtained from two
sources. We retrieved data on participation in SCHIP programs from the U.S. Centers for
Medicare and Medicaid Services (CMS) web site, and we procured data on state expenditures on
SCHIP programs and eligibility from the Kaiser Family Foundation’s Commission on Medicaid
and the Uninsured.
We used three variables to measure the overall effect of SCHIP on attendance rates: 1)
the participation rate, or the number of children ever enrolled in the program expressed as a
percentage of ―average‖ total child population (19 years of age and under) in each state for each
year. Population data were obtained from the U.S. Census Bureau’s Current Population Surveys;
2) the log of total state expenditures per participant on SCHIP for each year, which represents the
sum of state spending and federal funding; and 3) eligibility requirements, expressed as the
percentage of the poverty level at which states accept participants in SCHIP.
SCHIP participation rates vary from zero before the programs were implemented – and
less than one percent in several states in the year after program implementation – to 10.3 percent
of the state’s child population in New York in the year 2001. Even after the federal law
authorizing SCHIP was passed, variation was large primarily because SCHIP programs had few
enrollees in their first, nascent years of implementation. The mean participation rate is 1.3
percent with a standard deviation of 2.0 percent.
SCHIP expenditures per participant — which as for all expenditure amounts were
adjusted for inflation using the GDP Deflator and are expressed in 2000 constant dollars and also
logged — range from zero to 15.9, in Minnesota in 2002, with a mean of 2.9. After the program
was implemented, several states had expenditures below $1,000 per participant in the first year
of program implementation. By 1999, every state except for Oklahoma, Tennessee, Washington,
and Wyoming had implemented SCHIP programs – either as separate initiatives or as part of
their Medicaid programs. Those four states launched SCHIP in 2000. Expenditure data were
obtained from the State Health Expenditure Report published by the Milbank Memorial Fund,
the National Association of State Budget Officers, and the Reforming States Group.
Most states set eligibility requirements at 200 percent of the poverty level. The most
generous state was New Jersey, accepting participants whose families earn up to 350 percent of
the poverty level between 1999 and 2003. There was significant variation in eligibility
requirements both between states and within states over time. In New York State for example,
eligibility ranged from 185 percent of the poverty line in 1997 to 250 percent in 2003.
Other Independent Variables. Measurements of SCHIP are insufficient by themselves to test
the effects of the program on attendance rates. To form a more complete picture of attendance,
we gathered data on a number of variables that, we believed to have substantive effects on school
attendance. Most of these variables pertain to state education expenditures, state education
requirements, and the demographic composition of students in each state. These variables were
obtained from the U.S. Department of Education’s National Center for Education Statistics’
Common Core of Data and include:
The racial composition of the student bodies in each state as a percentage of average enrollment.
The number of students in kindergarten through fifth grade as a percentage of average enrollment.
The number of students who are eligible for free lunch programs as a percentage of average enrollment.1
The number of students who are considered special-education students as a percentage of average
Log of constant dollar local revenues for education from property taxes per pupil.
Log of constant dollar expenditures for school food services per pupil.
Log of constant dollar expenditures on school transportation services per pupil.
Many states did not provide free-lunch data to the U.S. Department of Education. We therefore
sought the data from state health, nutrition, and education departments themselves. Other
education-related variables obtained through direct contact with state education departments
The proportion of total school districts that engage in year-round schooling, which was used to create a
binary variable for states with either greater than or less than 10 percent of districts with year-round
Whether or not average daily attendance is an element in the state funding formula for districts and public
schools. If so, states might provide further incentive for schools to enforce policies that encourage kids to
attend school as much as possible.2
The compulsory ages at which students are required to start school and through which they are required to
remain in school.
For some of the education-related and demographic variables, we were missing observations in
certain years for particular states. For these missing observations, we interpolated the data, using
bivariate linear regressions of the existing data on the state in question and the variable in
question to estimate the missing observations. These missing data points account for less than
one percent of all observations.
Data were also obtained to measure a state’s level of economic activity. These variables
include the log of real gross state product per capita, the log of real per capita personal income,
unemployment rate, and the log of population in a state 19 years of age and under. These are
variables that are time variant at the state level and may be correlated with both SCHIP and
attendance. Controlling for the growth in child population was another way we correct for any
interstate movement during the year that would have the effect of biasing our results.
Finally, we control for the share of children who are Medicaid beneficiaries in a state.
Because states must also screen applicants for Medicaid eligibility before enrolling children in
SCHIP, some states have seen marked increase in enrollment in Medicaid because of SCHIP. In
Alabama, for example, approximately 121,000 additional children have enrolled in Medicaid
since the SCHIP program began (Allen, 2007). If not adequately controlled for, Medicaid
participation may be an important source of omitted variable bias. Medicaid participation ranges
from 5.4 percent to 50.1 percent with mean of 22.4 percent and standard deviation of 7.6 percent.
We should note that Medicaid participation rate is likely to be overestimated—and hence is a
conservative measure—as the measure was calculated as Medicaid beneficiaries under 21 in a
state as a percentage of population 19 years of age and under.
As with almost any social research study, our challenge was to overcome hurdles in
attributing a particular effect to a particular cause. In this case, confounding variables correlated
with SCHIP and attendance rate may, if absent, bias our estimates of the causal effect of the
program. This problem is reflected in the equation below:
ADAsy SCHIPsy STATE sy STUDENT sy ECONOMY sy s sy
This equation models the average daily attendance rate in state s for year y as a function of state
SCHIP program configuration (SCHIP), state student characteristics (STUDENT), state
education characteristics (STATE), state economic development variables (ECONOMY) and an
error term. The error term itself is composed of ( sy ) a transitory component, which changes
randomly between and within states, and a fixed component within a state ( s ) that does not
change over time.
Ordinary Least Squares (OLS) estimation of this equation would identify δ or the
magnitude of the effect of SCHIP, using cross-sectional variation in SCHIP. Given the limited
number of controls, OLS estimates of δ are likely to conflate the causal effect of SCHIP with
unobserved factors that are also associated with attendance. If, for example, states with
especially high expenditures for SCHIP also tend to have especially high expenditures for
education, and education expenditures indeed have a positive effect on attendance rates, and if
those education expenditures are not adequately controlled for, then the estimate of SCHIP
impact would be biased positively by the omitted education expenditure variable.
This study uses the panel nature of the dataset to address this issue of unobserved
heterogeneity; that is differences between states that are not adequately controlled for by
observed characteristics. In addition to an OLS cross-sectional estimation of SCHIP, some of our
regression specifications also control for state and year fixed effects, a method that we believe
allows more accurate estimates of a causal relationship between SCHIP and attendance rates.
Interactions between participation rate and eligibility were not significant and not included in the
results of this paper for reasons of parsimony but are available upon request.
Consider first the inclusion of state fixed effects: this removes between-state differences
in SCHIP measurements, such as participation and expenditures, and uses only within-state
variation to identify the coefficients. Fully accounted for are variables that do not change over
time, such as education and political institutions, which may bias estimates of SCHIP. However
state fixed effects do not control for year-to-year changes in attendance rates related to SCHIP.
Hence, the inclusion of year fixed effects takes advantage of the structure of the data and
accounts for many of the systematic differences across years. We should note however, that we
cannot rule out the possibility that variables that do change over time that do not affect all
observations within a single year are influencing our results.
Because variation in SCHIP may have different consequences for small versus large
states, heteroskedasticity is a concern. To address this concern, all regressions are weighted by
average enrollment to mitigate the effects of heteroskedasticity, after a White Test was unable to
reject the possibility of heteroskedasticity. Additionally, we calculated Cook’s distance statistics
to measure the influence of individual observations. Because no single case had a Cook’s
distance value greater than 1, we chose to use all the information provided in the dataset to
generate our analyses.
We examined correlations to establish basic relationships between our variables of interest.
These correlations are presented in Table 2.
The correlations indicate strong, positive and significant associations between each of the
SCHIP related variables, and suggest that each is capturing some aspect of the SCHIP program.
Attendance rates are positively and significantly associated with SCHIP expenditures and
eligibility, leading one to suspect SCHIP may have beneficial effects on attendance. Also
positively and significantly associated with average daily attendance rates are Medicaid
participation rate, the share of Asian, Hispanic, and special education students, the log of per
capita personal income, the log of real food expenditures per pupil, and minimum ending age.
Negatively and significantly correlated with attendance are elementary school share,
unemployment rate, the log of state population under 19 years of age, pupil-teacher ratio,
compulsory starting age, and whether average daily attendance is in a state’s finance formula. Of
course, all these correlations lack the ability to control for omitted variables as our regressions
attempt to do.
This section presents the results of our weighted least squares regression estimates.
Overall, the findings are striking in terms of their consistency. In each of the specifications,
estimates of SCHIP participation rate and eligibility are positive and significant at the .01 level,
providing strong support for an important role for SCHIP in reducing student absenteeism. The
lack of significance for any of the SCHIP expenditure variables also imply that expanding
participation in the program may not equal to expanding the cost of the program.
Column I of Table 3 presents the results of a cross-sectional regression controlling only
for SCHIP participation rate. In this regression, on average, a one percent increase in
participation rate is associated with a 0.170 percent increase in attendance rate. The adjusted R-
squared is very low however, with the regression model explaining only 2.0% of the variation in
average daily attendance rate. The results indicate that there are a large number of variables that
are missing from this specification.
Column II adds Medicaid participation rate, a set of state demographic variables, a set of
state economic characteristics, a group of state education variables, expenditures and SCHIP
participation rate. This specification also controls for state and year fixed effects, which appear
to be an extremely important contributor to a state’s average daily attendance rate, as indicated
by an adjusted R-squared value of 0.885. Eligibility was omitted from this specification to more
easily interpret the effects of participation in SCHIP on attendance.
After controlling for these additional variables, the estimate of participation rate actually
increases in magnitude, with each additional percentage point increase in participation rate
associated with an increase in average daily attendance rate of 0.390 percent. The estimate is
approximately the same in specifications II through V, each of which analyzes the effects of
within-state variation in SCHIP and other variables over time on attendance rates. This consistent
finding suggests that the effects of participation in SCHIP on attendance are stronger within
states than across states.
Surprising is the finding that regression estimates do not appear to be affected to any
considerable extent by the inclusion or omission of either the eligibility or participation
variables. This finding may mean participation rate and rte of eligibility are capturing different
aspects of the SCHIP program. For example, increasing participation, may mean better
marketing to uninsured children and their families, rather than expanding eligibility.
To more clearly interpret the impact of eligibility expansions (the policy lever that states
are best able to control) have on attendance rates, column III presents regression results that do
not include participation rate. The findings suggest that each 100 percent increase over the
federal poverty line results in an increase in attendance of approximately one percentage point.
The estimate is virtually identical for columns IV and V.
The results presented in column IV remove the state economic development variables
from the regression analysis as a check on the robustness of the SCHIP regression results. Our
overall SCHIP results are bolstered by the fact that there appears to be very little difference in
the SCHIP regression results from column IV to column V. In other words, the state and year
fixed effects appear to be adequately controlling for omitted variables.
Column V presents the results from the full model, which includes all of the SCHIP
related variables, Medicaid participation rate, the full set of state demographic, economic, and
education variables, as well as state and year fixed effects. As in the other specifications, the log
of real expenditures per participant is not significant, but both eligibility and participation and
eligibility are significant and positive.
In the final specification, black, Hispanic, white, and the log of real gross state product,
are all positively and significantly (at the .05 level or better) related to attendance. The estimates
of the log of real personal income and the log of state population, 0-19, however, are in
opposition to the estimate of the log of real gross state product. These variables are negatively
and significantly associated with attendance rate. Overall, the effect of economic development on
attendance is not clear. Also significant at a .01 level is the negative effect of elementary school
share on state attendance rates, which is consistent with previous research that has documented a
negative correlation between age and absenteeism.
In this study, we find that SCHIP participation rates and eligibility requirements are
positively and significantly associated with average daily attendance rates. Overall, it does seem
that enrolling more participants in SCHIP, which might be accomplished through raising the
eligibility requirements, leads to lower student absenteeism and improved attendance. Hence, we
conclude the State Children’s Health Insurance Program has had an impact, at least when it
comes to school attendance. The improvement in school attendance may indicate an
improvement in the health of children or at the very least the existence of a positive externality
resulting from SCHIP. Whatever the mechanism, the results of this study support renewal and a
significant expansion in the program.
Nevertheless, even in states with large, well-funded programs, the effect of SCHIP on the
attendance rate is relatively small. This small effect is likely the result of the immeasurably large
number of variables that could explain attendance, in combination with the small degree of
variation in attendance rates within states: the average state in this study had a range of
attendance rates of less than three percentage points over 11 years. While the effect seems small
in terms of the attendance rate, even small changes in the attendance rate can translate into a
substantial number of students. For example, specification II of our regression model predicts
that increasing the SCHIP participation rate by just one percentage point would lead to an
additional 24,875 students attending class each day in California, 11,032 students attending class
each day in New York, and 16,906 students in Texas, holding all else constant.
These results suggest that the State Children’s Health Insurance Program is more than an
insurance program; it is an attendance program. In addition, because increases in participation
and eligibility appear to have a larger effect on improving attendance than do expenditures,
enrolling more children in SCHIP may not be as costly as one may expect. Children on the whole
are healthier than adults and therefore may require a less comprehensive level of benefits from
their health insurance carriers. Indeed, basic coverage for children may go a long way in
improving the quality of their lives.
Suggestions for Further Research
While the coefficients on participation and eligibility confirm our hypotheses, we were
not able to fully measure the effect of the different types of SCHIP programs – those operating as
separate programs, those that are a part of Medicaid, and those that are a combination of the two
options. While we initially intended to include this aspect of SCHIP in our model, the sample
size of 48 states over 11 years is too small to divide into three groups, especially because states
have not frequently changed from one type of program to another. Examining how the different
types of programs affect the quality of SCHIP services is therefore a potential area for further
Although the relationship between a state’s level of generosity, as measured by more
inclusive eligibility requirements and the effect on the attendance rate seems clear, this study is
limited in its ability to assess the extent of each state’s generosity. As measured by the
percentage of the federal poverty level that the state uses as its cutoff for eligibility, states that
have a higher cutoff, such as New Jersey, Delaware, and Michigan, have reaped greater benefits
— at least in terms of attendance rates — from SCHIP than those with lower cutoffs. However,
eligibility requirements are more complex than these figures. Some states, for instance, allow
additional children to participate only on a cost-sharing basis, using methods such as deductibles
and co-payments; others informally waive eligibility requirements; and still other states enroll
parents as well as children. To better assess the relationship between state eligibility
requirements and SCHIP effectiveness, we propose the creation of an index of each state’s
generosity as another area for future study.
Finally, this study does not address the issue of SCHIP ―crowding out‖ private health
insurance, where some children that are enrolled in SCHIP would otherwise be enrolled in
private insurance. While some previous studies have suggested that this might be occurring, the
amount of crowding out has been determined to be relatively small and the large majority of
participants in SCHIP would most likely be otherwise uninsured (Lo Sasso & Buchmueller,
2004). Our results indicate these children would also be missing school more frequently if the
program did not exist.
Alexander, K. L., Entwisle, D. R., & Horsey, C. S. (1997). From first grade forward: Early
foundations of high school dropout. Sociology of Education, 70 (2), 87-107.
Barrington, B. L., & Hendricks, B. (1989). Differentiating characteristics of high school
graduates, dropouts, and nongraduates. The Journal of Educational Research, 82 (6),
Broaddus, M., & Park, E. (2007). Freezing SCHIP funding in coming years would reverse recent
gains in children’s health coverage. Washington, DC: Center on Budget and Policy
Caldas, S. J. (1993, March/April). Reexamination of input and process factor effects on public
school achievement. Journal of Educational Research, 206-214.
Center for Advanced Social Science Research. (2001, June 5). Conferences: After the bell:
education solutions outside the school. Retrieved August 25, 2007, from
Centers for Medicare & Medicaid Services. (2006, July 12). Overview. Retrieved July 21, 2007,
Coleman, J., Campbell, E., Hobson, C., McPartland, J., Mood, A., Weinfeld, F., et al. (1966).
Equality of educational opportunity. Washington, DC: Department of Health, Education
Cooper, H., Valentine, J. C., Charlton, K., & Melson, A. (2003). The effects of modified school
calendars on student achievement and on school and community attitudes. Review of
Educational Research, 73, 1-52.
Damiano, P., & Tyler, M. (2005). hawk-i: Impact on access and health status. Public Policy
Center. Iowa City, IA: University of Iowa.
Dubay, L., Hill, I., & Kenney, G. (2002). Five things everyone should know about SCHIP.
Washington, DC: The Urban Institute.
Ensminger, M. E. & Slusarcick, A. L. (1992). Paths to high school graduation or dropout: a
longitudinal study of a first-grade cohort. Sociology of Education, 65 (2), 95-113.
Epstein, J. L., & Sheldon, S. B. (2002). Present and accounted for: Improving student attendance
through family and community involvement. The Journal of Educational Research, 95
Florida Healthy Kids Corporation. (1997). Health kids annual report, 1997. Tallahassee, FL.
Gutowski, M., Ratner, J., Avruch, S., & Lamb, S. (1997). Coverage leads to increased health
care access for children. Washington, DC: United States General Accounting Office.
Institute of Medicine. (2003). Hidden costs, value lost: Uninsurance in America. Washington,
DC: National Academies Press.
Johns Hopkins Medical Institutions. (1999, April 29). Nighttime asthma squeezes school
attendance. Retrieved August 25, 2007, from ScienceDaily:
Kahn, J. H., & Nursten, J. (1996). Unwillingly to school. New York: American Psychiatric
Kaplan, D. S., Peck, B. M., & Kaplan, H. B. (1995). A structual model of dropout behavior: A
logitudinal analysis. Applied Behavioral Science Review, 3 (2), 177-193.
Kempe, A., Beaty, B. L., Crane, L. A., Stokstad, J., & Barrow, J. (2005). Changes in access,
utilization, and quality of care after enrollment into a state child health insurance plan.
Pediatrics, 115 (2), 364-371.
Kenney, G., & Chang, D. (2004). The State Children’s Health Insurance Program: Successes,
shortcomings, and challenges. Health Affairs, 23 (5).
Lamdin, D. J. (1996). Evidence of student attendance as an independent variable in education
production functions. Journal of Educational Research, 89 (3), 155-162.
Levy, H., & Meltzer, D. (2004). What do we really know about whether health insurance affects
health? In C. G. McLaughlin, Health Policy and the Uninsured (pp. 179-204).
Washington, DC: Urban Institute Press.
Lo Sasso, A. T., & Buchmueller, T. C. (2004). The effect of the State Children’s Health
Insurance Program on health insurance coverage. Journal of Health Economics, 23 (5),
Loviglio, J. (2007, August 11). Obesity is a predictor of school absenteeism, research team
reports. The Arizona Republic.
Lykens, K., & Jargowsky, P. (2002). Medicaid matters: Children’s health and Medicaid
expansions. Journal of Policy Analysis and Management, 21 (2), 219-238.
Lyons, B. (2007). The State Children’s Health Insurance Program: Lessons and outlook. The
Kaiser Commission on Medicaid and the Uninsured. Billings, MT: The Henry J. Kaiser
Minneapolis Public Schools. (2007). Attendance matters! Minneapolis.
Mohonasen Central School District. (2003). SchoolAttendance. (E. McNulty, Editor, & Capital
Region BOCES Communications Service) Retrieved August 24, 2007, from
Muir, M. (2005). Research Brief: Strategies for dealing with tardiness. Omaha: The Principals’
Mullis, I. V., Martin, M. O., Gonzales, E. J., & Chrostowski, S. J. (2004). TIMSS 2003
international mathematics report: Findings from IEA’s trends in international
mathematics and science study at the fourth and eighth Grades. Lynch School of
Education, TIMSS & PIRLS International Study Center. Chestnut Hill, MA: Boston
National Center for Policy Analysis. (2001). NCPA - Education - Taking a second look at
student attendance. Retrieved August 24, 2007, from
Pear, R. (2007a, July 9). A battle over expansion of children’s insurance. New York Times.
Pear, R. (2007b, September 26). House passes children’s insurance measure. New York Times.
Pear, R. (2007c, September 27). Houses passes a stopgap bill to pay for programs. New York
Rumberger, R. W. (1995). Dropping out of middle school. American Educational Research
Journal, 32 (3), 583-625.
Rumberger, R. W. (1987). High school dropouts: A review of issues and evidence. Review of
Educational Research, 57 (2), 101-121.
Rumberger, R. W., Ghatak, R., Poulos, G., Ritter, P., & Dornbusch, S. (1990). Family influences
on dropout behavior in one California high school. Sociology of Education, 63 (4), 283-
Ryan, J. (2007). The basics: State Children’s Health Insurance Program (SCHIP). The George
Washington University. Washington, DC: National Health Policy Forum.
Shone, L. P., Dick, A. W., Brach, C., Kimminau, K. S., LaClair, B. J., Shenkman, E. A., et al.
(2003). The role of race and ethnicity in the State Children’s Health Insurance Program
(SCHIP) in four states: Are there baseline disparities, and what do they mean for
SCHIP?. Pedatrics, 112 (6), 521-532.
Shone, L. P., Dick, A. W., Klein, J. D., Zwanziger, J., & Szilagyi, P. G. (2005). Reduction in
racial and ethnic disparities After enrollment in the State Children’s Health Insurance
Program. Pediatrics, 115 (6), 697-705.
Skip class, lose ground. (2000, June 27). USA Today, p. 14A.
The Kaiser Commission on Medicaid and the Uninsured. (2007). Impacts of Medicaid and
SCHIP on low-income children’s health. Washington, DC: The Kaiser Family
U.S. Department of Education, National Center for Education Statistics. (2006). The condition of
education 2006. Washington, DC: U.S. Government Printing Office.
Wolfe, B. L. (1985). The influence of health on school outcomes: A multivariate approach.
Medical Care, 23 (10), 1127-1138.
Wooldridge, J., Kenney, G., & Trenholm, C. (2005). Congressionally mandated evaluation of
the State Children’s Health Insurance Program. Princeton, NJ: Mathematica Policy
1. Arizona provided data for free and reduced lunch eligibility combined.
2. 11 states used average daily attendance of each district as an element in their state
education financing formula but only one state, Tennessee, changed whether or not it was
included during the duration of our data collection.
Variable Mean Deviation Minimum Maximum
Average Daily Attendance Rate 92.484 2.714 85.295 100.000
SCHIP Related Variables
Participation Rate 1.349 1.964 0.000 10.329
Log of Real Expenditures Per Participant 2.859 3.417 0.000 15.879
Eligibility (% of Federal Poverty Level) 114.623 103.924 0.000 350.000
Medicaid Participation Rate 22.358 7.577 5.366 50.056
State Demographic Variables
Asian Share 2.316 1.995 0.260 10.976
Black Share 14.405 13.251 0.453 51.311
Hispanic Share 8.652 11.052 0.176 52.206
Native American Share 1.947 3.460 0.052 18.440
White Share 72.154 16.164 31.835 98.105
Free Lunch Eligible Share 28.054 10.062 0.000 57.253
Special Education Share 12.515 2.704 1.254 21.175
Kindergarten through Fifth Grade Share 46.633 2.009 40.726 50.765
State Economic Variables
Log of Real Gross State Product 10.297 0.188 9.853 10.908
Log of Real Per Capita Personal Income 10.148 0.163 9.732 10.644
Unemployment Rate 5.049 1.399 2.300 11.300
Log of State Population, Age 0-19 13.832 0.984 11.821 16.173
State Education Variables
Pupil-Teacher Ratio 16.185 2.314 11.300 24.700
Log of Local Revenues for Education Per Pupil 7.902 0.470 6.279 8.920
Log of Real Food Expenditures Per Pupil 5.500 0.258 4.827 6.180
Log of Real Transportation Expenditures Per Pupil 5.491 0.393 4.303 6.581
Compulsory Starting Age 6.377 0.777 5.000 8.000
Minimum Ending Age 16.568 0.806 16.000 18.000
Year-Round Schools > 10% of All Schools? 0.045 0.208 0.000 1.000
ADA in State Finance Formula? 0.189 0.392 0.000 1.000
Note: There are 576 observations.
Average Log of Real (% of
Daily Expenditures Federal
Attendance Participation Per Poverty
Rate Rate Participant Level)
Average Daily Attendance Rate 1.000 0.032 0.113 * 0.233 *
SCHIP Related Variables
Participation Rate 0.032 1.000 0.766 * 0.622 *
Log of Real Expenditures Per Participant 0.113 * 0.766 * 1.000 0.737 *
Eligibility (% of Federal Poverty Level) 0.233 * 0.622 * 0.737 * 1.000
Medicaid Participation Rate 0.111 * 0.365 * 0.322 * 0.317 *
State Demographic Variables
Asian Share 0.086 * 0.149 * 0.101 * 0.200 *
Black Share -0.042 0.161 * 0.037 0.014
Hispanic Share 0.114 * 0.147 * -0.127 * 0.137 *
Native American Share 0.057 -0.030 0.015 -0.031
White Share -0.050 -0.243 * -0.127 * -0.112 *
Free Lunch Eligible Share -0.047 0.151 * 0.036 -0.005
Special Education Share 0.179 * 0.311 * 0.348 * 0.360 *
Kindergarten through Fifth Grade Share -0.171 * -0.203 * -0.310 * -0.300 *
State Economic Variables
Log of Real Gross State Product 0.078 0.306 * 0.383 * 0.487 *
Log of Real Per Capita Personal Income 0.083 * 0.385 * 0.448 * 0.562 *
Unemployment Rate -0.189 * -0.108 * -0.233 * -0.393 *
Log of State Population, Age 0-19 -0.148 * 0.179 * 0.061 0.055
State Education Variables
Pupil-Teacher Ratio -0.222 * -0.202 * -0.218 * -0.266 *
Log of Local Revenues for Education Per Pupil -0.039 0.368 * 0.354 * 0.394 *
Log of Real Food Expenditures Per Pupil 0.157 * 0.577 * 0.630 * 0.612 *
Log of Real Transportation Expenditures Per Pupil 0.050 0.409 * 0.464 * 0.520 *
Compulsory Starting Age -0.184 * -0.104 * -0.070 -0.068
Minimum Ending Age 0.127 * 0.021 0.060 0.067
Year-Round Schools > 10% of All Schools? 0.053 0.065 0.023 0.027
ADA in State Finance Formula? -0.163 * 0.161 * 0.028 0.002
Notes: 1) There are 576 observations. 2) *Significant at 5% level.
WLS Regression Estimates, Average Daily Attendance Rate as Dependent Variables
I II III IV V
SCHIP Related Variables
Participation Rate 0.170 *** 0.390 *** 0.377 *** 0.387 ***
(0.045) (0.043) (0.044)
Log of Real Expenditures Per Participant -0.013 0.001 -0.024 -0.033
(0.040) (0.043) (0.040) (0.040)
Eligibility (% of Federal Poverty Level) 0.011 *** 0.010 *** 0.010 ***
(0.002) (0.002) (0.002)
Medicaid Participation Rate 0.041 ** -0.010 0.024 0.028
(0.017) (0.018) (0.017) (0.017)
State Demographic Variables
Asian Share 0.554 ** 0.334 -0.243 0.133
(0.245) (0.275) (0.235) (0.258)
Black Share 0.362 ** 0.144 0.275 ** 0.357 ***
(0.132) (0.137) (0.127) (0.130)
Hispanic Share 0.347 *** 0.342 *** 0.124 ** 0.339 ***
(0.102) (0.108) (0.053) (0.100)
White Share 0.123 ** 0.178 *** 0.124 ** 0.120 **
(0.054) (0.056) (0.053) (0.053)
Free Lunch Eligible Share -0.004 0.032 *** 0.006 0.005
(0.012) (0.013) (0.012) (0.012)
Special Education Share -0.014 -0.073 ** 0.004 -0.023
(0.030) (0.031) (0.028) (0.029)
Kindergarten through Fifth Grade Share -0.362 *** -0.320 *** -0.319 *** -0.311 ***
(0.065) (0.069) (0.065) (0.065)
State Economic Variables
Log of Real Gross State Product 8.229 *** 6.381 *** 8.002 ***
(2.347) (2.462) (2.302)
Log of Real Per Capita Personal Income -8.369 ** -12.077 *** -9.438 **
(3.609) (3.792) (3.547)
Unemployment Rate -0.192 * -0.305 *** -0.061
(0.102) (0.108) (0.104)
Log of State Population, Age 0-19 -11.346 *** -9.836 *** -10.796 ***
(2.916) (3.069) (2.862)
State Education Variables
Pupil-Teacher Ratio -0.195 *** -0.169 * -0.185 * -0.131
(0.098) (0.105) (0.096) (0.098)
Log of Local Revenues for Education Per Pupil 0.369 1.031 ** -0.008 0.388
(0.450) (0.467) (0.434) (0.442)
Log of Real Food Expenditures Per Pupil -2.579 ** -2.598 ** -1.171 -2.017 *
(1.163) (1.228) (1.121) (1.147)
Log of Real Transportation Expenditures Per Pupil -0.033 -0.255 0.129 0.158
(0.727) (0.765) (0.713) (0.714)
Compulsory Starting Age -0.017 -0.127 0.145 -0.065
(0.236) (0.249) (0.230) (0.232)
Minimum Ending Age 0.103 0.051 0.068 0.023
(0.140) (0.148) (0.137) (0.138)
Year-Round Schools > 10% of All Schools? -0.988 -0.657 -1.318 -0.903
(0.991) (1.043) (0.987) (0.972)
ADA in State Finance Formula? -0.801 -1.421 -0.848 -1.472
(1.190) (1.262) (1.186) (1.176)
Constant 92.115 *** 267.062 *** 300.482 *** 96.595 *** 265.962 ***
(0.152) (53.104) (55.724) (11.454) (52.078)
Adjusted R-Squared 0.020 0.885 0.873 0.885 0.890
State Fixed Effects No Yes Yes Yes Yes
Year Fixed Effects No Yes Yes Yes Yes
Notes: 1) There are 576 observations. 2) Standard errors in parentheses. 3) Regressions are weighted by total student enrollment. 4) ***
p<0.01, ** p<0.05, * p<0.10.