Content uploaded by Sandra G Sosa-Rubí
Author content
All content in this area was uploaded by Sandra G Sosa-Rubí on Aug 03, 2015
Content may be subject to copyright.
HEDG Working Paper 08/12
THE IMPACT OF UNIVERSAL HEALTH INSURANCE ON
CATASTROPHIC AND
OUT-OF-POCKET HEALTH EXPENDITURES IN MEXICO:
A MODEL WITH AN ENDOGENEOUS TREATMENT
VARIABLE
Omar Galárraga
Sandra G. Sosa-Rubí
Aarón Salinas
Sergio Sesma
June 2008
ISSN 1751-1976
http://www.york.ac.uk/res/herc/research/hedg/wp.htm
THE IMPACT OF UNIVERSAL HEALTH INSURANCE ON
CATASTROPHIC AND
OUT-OF-POCKET HEALTH EXPENDITURES IN MEXICO:
A MODEL WITH AN ENDOGENEOUS TREATMENT VARIABLE
Omar Galárraga1,2, Sandra G. Sosa-Rubí1, Aarón Salinas1, Sergio Sesma3
For Submission to:
17th European Workshop on Econometrics and Health Economics
1 June 2008
From the:
1 National Institute of Public Health (INSP), Cuernavaca, Mexico
2 Institute of Business and Economic Research (IBER), University of California, Berkeley, CA.
3 Carso Institute for Health, Mexico City, Mexico.
Correspondence to: Sandra G. Sosa-Rubí, Health Economics Unit, Center for Evaluation and Survey
Research, National Institute of Public Health (INSP), Av. Universidad 655, Cuernavaca, C.P. 62508,
Mexico. E-mail: srubi@insp.mx
2
SUMMARY
Introduction: The main goal of Seguro Popular is to improve the financial protection
of the uninsured population against excessive health expenditures. Seguro Popular (SP)
covers a variety of preventive and curative procedures, as well as medicines, and
hospital care for the poorest segment of the Mexican population.
Data: This paper estimates the impact of Seguro Popular on catastrophic health
expenditures, as well as out-of-pocket health expenditures, from three different sources:
National Household Survey of Income and Expenditures (ENIGH 2006); National
Health and Nutrition Survey (ENSANUT 2006); and SP Impact Evaluation Survey.
Methods: We first estimate naive probit models, and then compare them against
bivariate probit models which use instrumental variables that take advantage of the
specific SP implementation mechanisms to address the endogeneity of insurance
selection choices.
Results: No effect on catastrophic health expenditures is observed in the ENIGH
sample. However, we find a statistically significant effect on the reduction of
household’s expenditures on medicines and outpatient care. On the other hand, Seguro
Popular reduces the probability of catastrophic health expenditures using the other two
datasets: SP Impact Evaluation Survey, and ENSANUT. We also observe a reduction
of the probability of expenditures on medicines and outpatient care among the SP
insured families.
AEA-JEL classification: 012, 038, 054, I18, I38
Key terms: catastrophic health expenditures; health insurance; instrumental variables;
non-linear methods; Mexico
3
1. INTRODUCTION
This paper has two main goals. First, we provide estimates of the treatment effect of
universal health insurance targeted to the poorest families on catastrophic health
expenditures in Mexico using three different data sources. Second, we compare the
results and survey methodologies, and try to explain some of the differences
encountered in the estimation of the treatment effect. We draw on previous analyses of
the issues around catastrophic health expenditures in Mexico (Parker and Wong 1997;
Perez-Rico, Sesma-Vazquez et al. 2005; Sesma-Vazquez, Perez-Rico et al. 2005;
Gakidou, Lozano et al. 2006; Knaul, Arreola-Ornelas et al. 2006), but we add a new
dimension in the literature by using non-linear models with selection correction for the
potentially endogenous treatment variable. To the best of our knowledge, this paper
provides the first non-experimental evidence of a causal effect of Seguro Popular on
catastrophic health expenditures. Previous analyses using non-experimental data
demonstrated strong associations but have not used methods to show causal effects.
The results presented here are relevant in the Mexican context, but they go beyond that.
The issue of the impact of universal health insurance on financial protection is of wide
relevance across Latin America and other regions.
This paper proceeds as follows. In Section 2 we present a brief background of the
Seguro Popular program in Mexico. Section 3 shows the methods used including data
sources and study population. Section 4 presents the results, followed by a discussion.
The paper ends with some conclusions for policy and future research.
4
2. BACKGROUND
The Popular Health Insurance (or “Seguro Popular”) was implemented in Mexico as a
comprehensive health reform effort to provide financial protection in health for the
poorest segment of the population (Frenk, Gonzalez-Pier et al. 2006). Until 2001 health
insurance coverage in Mexico was directed only to employees working in the formal
sector of the economy. Coverage for formal sector workers included the Mexican
Social Insurance System (Instituto Mexicano del Seguro Social, or IMSS), the
Government Workers’ Social Security and Services Institute (Instituto de Seguridad y
Servicios Sociales de los Trabajadores del Estado, or ISSSTE), as well as insurance
programs for employees of such state-run enterprises as PEMEX (petroleum) and
SEDENA (national defence). Participants in the informal sector of the economy had to
attend government-sponsored facilities through the Ministry Health (Secretaría de
Salud, or SSA) or pay out-of-pocket for medical care at private health services. These
private facilities varied considerably in price, quality, and availability. Whilst a modern
network of private health services for the middle and upper classes served those
individuals who had insurance coverage or could pay out-of-pocket (OOP) for their
health care, there was also a lower-priced private health providers of variable quality.
By 2002 there was evidence of excessive health-related spending for the poorest rural
families in Mexico, particularly for the care of older adults (over 60 years of age) and
the care of pregnancy (Torres and Knaul 2003; Knaul, Arreola-Ornelas et al. 2005;
Sesma-Vazquez, Perez-Rico et al. 2005). Most of the catastrophic expenditures among
the poor were attributed to outpatient care and medications. This situation is common
among the poorest segments of the population in most developing countries where “a
relatively small payment can mean financial catastrophe to a poor person or household,
forcing them to reduce other basic expenses such as food, shelter, or their children’s
5
education” (Xu, Evans et al. 2007), or even suffer financial catastrophe (Van Damme,
Van Leemput et al. 2004).
The explicit goal of the Seguro Popular program was to financially protect the poorest
families (within the poorest two income deciles) that did not have any health insurance
coverage. Although the campaigns to be enrolled were targeted to the poorest sections
of the population in rural an urban areas, the decision of enrolling to SP was a family’s
voluntary choice (Frenk, Gonzalez-Pier et al. 2006). The program objectives were to
assure:
(i) the protection of poor families against catastrophic health
expenditures and its impoverishing effect; and
(ii) the universal access to adequate secondary and tertiary medical care.
Additionally, on the supply-side, all SP-sponsored health facilities from the public
health providers had to offer a minimum level of health-services quality in order to
belong to the SP-sponsored health facilities network.
The process of health unit accreditation to Seguro Popular was rolled out gradually
during 2001–2005. Five states (Aguascalientes, Campeche, Colima, Jalisco and
Tabasco) were incorporated into the program in 2001 as part of a pilot study. An
additional 15 states were integrated in the program in 2002; four more states were
incorporated in 2003; and the remaining states were incorporated in during 2004 and
2005. By the end of 2005, all 32 of Mexico’s states had been incorporated, and
approximately 4 million families (comprising about 12 million individuals) had signed
up for the voluntary program (SSA 2006a).
6
Overall, through the first quarter of 2007, approximately 5.2 million (44 %) of the
estimated 11.9 million eligible households nationwide had enrolled in the program.
Although the indicators of coverage have widely shown the proven capacity of the SP
programme to enrol a large group of uninsured households, there has been limited
evidence about the medium-term improvements in the financial protection of the
poorest households.
Analyses about the trends and evolution of catastrophic and impoverishing health
spending have shown a decreased incidence of catastrophic spending among the poorest
households, but this trend was not clearly found in the case of the out-of-pocket
expenditures (Knaul, Arreola-Ornelas et al. 2005).
There have been only few studies that have empirically estimated the effect of the health
insurance coverage on the incidence of catastrophic health spending in developing
countries with experimental data (Wagstaff and Yu 2007), and fewer with observational
data (Jowett, Contoyannis et al. 2003).
3. METHODS
3.1 Data Sources and Study Population
We analysed the impact of Seguro Popular on the incidence of catastrophic health
expenditures and out-of-pocket health expenditures among poor households using three
different data sources of household expenditure and insurance enrolment. We used data
from:
7
• Encuesta Nacional de Ingreso y Gasto de los Hogares (ENIGH 2006) [National
Household Survey of Income and Expenditures]; and
• Encuesta Nacional de Salud y Nutrición (ENSANUT 2005-2006) [National
Health and Nutrition Survey].
• Encuesta de Impacto del Seguro Popular [SP Impact Evaluation Survey];
For all surveys, we had strict selection criteria such that we identified households where
all members of the household were enrolled into Seguro Popular (“insured” group); that
is, our “treatment” is that everyone in the household be officially enrolled into the SP
program. Then, we created a comparison group of households with no insurance
coverage at all (“uninsured” group); that is, our controls are households where no
person had any type of health insurance.
The ENIGH 2006 is a cross-sectional dataset with nationally representative data, with a
sample of 20,875 dwellings (INEGI 2006). Its main purpose was to obtain information
about income and expenses of the households. It collected information about
occupational characteristics, socio-demographic characteristics of the members of the
households, infrastructure characteristics of the dwellings and household assets. It also
has information about the access to public programmes by households (including
transfers and subsidies). From the original sample with health expenditure data and
health insurance data, we selected 1,736 SP-insured households and 12,936 uninsured
households.
The ENSANUT 2006 is also a cross-sectional dataset with nationally representative
data, which was collected for 48,304 dwellings (INSP 2006). This dataset contains
8
information about individual’s health, use of health services, socio-economic
characteristics of households, access to health programmes, and biological health
indicators. From the original sample, we took those 45,699 households with health
expenditure data and health insurance data. After this, we end up with the analytical
samples: 4,440 SP-insured households and 16,376 uninsured households.
The SP Impact Evaluation Survey is a panel dataset composed with 36,000 dwellings
for which there was baseline information (August 2005) in 32,506 dwellings, and first-
wave of data collected in mid-2006 with information in 29,836 dwellings (King and et
al 2006; SSA 2006b; King, Gakidou et al. 2007). The data were collected in seven states
in Mexico (Sonora, San Luis Potosi, Jalisco, Estado de Mexico, Guerrero, Morelos and
Oaxaca). Its main purpose was to measure the impact of Seguro Popular among eligible
households (poor households without any coverage of health insurance). The criteria to
select the location of the treatment and control clusters were: i) to include those zones
were the rate of penetration of the programme was very low; ii) to include those places
where the incorporation of the SP programme was being postponed. Note that this data
was experimental in design with a baseline and follow-up rounds of data collection.
However, we are using only the follow-up data, as if it was a cross-section, so that we
can maximize the comparability with the other two data sources. (Details on the
experimental design have been presented elsewhere (King, Gakidou et al. 2007)). From
the 29,836 households with the relevant data (in the first-wave of data follow-up), we
selected the following analytical samples: 7,952 SP-insured households and 21,884
uninsured households.
9
3.2 Variables and Analysis
Based on the literature, the basic econometric specification we used to analyze the
impact of health insurance on catastrophic health expenditures (CHE) was of the form:
Y = X + T + + e1 (1)
T = X + + Z + e2 (2)
where:
Y = catastrophic health expenditure
X = covariate vector
T = household enrolment into Seguro Popular
Z= instrumental variables
We defined expenditure as being catastrophic with a dummy variable equal to the unity
if a household's financial contributions to the health system exceed 30% of income
remaining after subsistence needs ($2 USD per capita) have been met; and zero
otherwise. This definition is the most widely used in the literature (Murray, Knaul et al.
2000; Xu, Klavus et al. 2003), but there are other alternatives (Xu, Evans et al. 2003)4.
In the naïve probit models, we assumed that the error terms e1 and e2 were not correlated,
and thus we could directly estimate equation (1), independently of equation (2).
However, considering a potentially endogenous treatment variable, we also used a model
with a bivariate normal distribution for the error terms (Maddala 1994; Wooldridge 2002;
Greene 2003), with the variance normalized to the unity and the correlation coefficient
denoted as , in the following manner:
4 Other criteria that have been used to define catastrophic expenditures is when the household’s financial
contributions to the health system exceed 40% of income and subsistence needs of $1 USD per capita).
10
1
1
,
0
0
~
2
1
ρ
ρ
BVN
e
e (3)
The correlation between e1 and e2 captures the correlation between the likelihood of
having catastrophic health expenditures, and at the same time having enrolled in Seguro
Popular. We hypothesized that >0 because those households which may be more
prone to have high health care expenditures (relative to the level of household income),
would also have more incentives to sign up for Seguro Popular.
To select an appropriate covariate vector we searched the literature (Makinen, Waters et
al. 2000; Kawabata, Xu et al. 2002; Xu, Evans et al. 2003) and found that the main
determinants of CHE seem to relate to poverty, aging, chronic illnesses, low levels of
insurance coverage, urban/rural differences, socio-economic status, types of illnesses,
demographic composition of the household, and characteristics of the household head
(age, sex, education). Health spending would be affected by the family’s
wealth/physical assets, its income or financial assets as well as their insurance coverage.
We have used an asset index as a proxy for household wealth (McKenzie 2004).
Additionally, we have used the deprivation index at the municipality/locality level to
control for general levels of wellbeing at the local level (CONAPO 2005).
Thus, the main explanatory or “treatment” variable (T) to be analysed was enrolment
into Seguro Popular (SP), a public insurance scheme for the poor and otherwise
uncovered population in Mexico. Enrolment into Seguro Popular would be determined
by equation (2) above, which has the same covariate vector X, and a set of instrumental
variables (Z). These set of “instruments” would strongly affect the probability of a
household being part of the SP program, but they would not be correlated with the
11
outcome of interest (catastrophic expenditures in health) through channels other than the
enrolment into SP. The instrumental variables approach has been used in several
studies of CHE (Pradhan and Prescott 2002; Jowett, Contoyannis et al. 2003; Jowett,
Deolalikar et al. 2004; Sepehri, Sarma et al. 2006)
The instrumental variables we used take advantage of the fact that Seguro Popular was
implemented gradually across the different 32 Mexican states. First, we used the year
of incorporation as a proxy for the length of time that a particular state has had Seguro
Popular. For example, if a state was incorporated by 2003, a dummy variable for 2001
would be zero, a dummy for 2002 would also be zero, but the dummy variables for
2003 and 2004 would be equal to the unity. By 2005 all states were incorporated, so
that 2005 serves as the reference year (except in the SP Impact Survey where the
reference year was 2002). The marginal effect of the incorporation dummy measures
the effect incorporation a year earlier on the household SP enrolment probability. This
instrument has been implemented successfully in a similar context (Sosa-Rubi,
Galarraga et al. 2007).
We recognize that the official dates of incorporation (SSA 2006b) may not be
necessarily exact indicators of program availability, because some health centres may
still have been going through accreditation by the end of the year. Thus, we view these
instruments as indicators of likelihood that SP was available to each household in the
survey. This use of year-of-incorporation dummies as instruments follows the spirit of
Duflo’s use of distance from schools as instruments a study of impact of school
construction on educational attainment and wages in Indonesia (Duflo 2001).
12
Second, similar to the first set of instrumental variables, we used the level of penetration
of the program at the locality level to help us determine the probability of enrolment.
The logic was that households living in localities with higher SP penetration or
coverage had higher probabilities of enrolling into the program. We constructed the
variable with a ratio of SP enrolled households over eligible (uninsured households) at
the locality level using the latest round of Census data (INEGI 2005). Households
located in areas where the level of diffusion of Seguro Popular was higher tended to
have higher probability of being enrolled into the SP program. We see this instrumental
variable as an aggregate continuous proxy for program participation at the household
level. We are implicitly assuming that the level of program diffusion or penetration has
a direct impact on the behavioural choices of households; but assuming that there is no
underlying aggregate effect over expenditures, other than through the channel of
insurance choice (Angrist, Imbens et al. 1996; Heckman 1997; Angrist and Krueger
2001).
The regional and temporal variations in incorporation and coverage rates helped us to
identify Seguro Popular household enrolment, independently of the outcomes of
interest: catastrophic health expenditures. Thus, the year-of-incorporation dummies as
well as SP coverage rates (as a continuous variable: 0-100%) were excluded from the
equation (1).
We have encountered no evidence of “policy endogeneity”. That is, there is no reason
to believe that states made harder efforts to enrol earlier because of evidence that
families living there where particularly prone to excessive health expenditures. Neither
do we have evidence that excessive health expenditures at the aggregate level have been
13
correlated with higher levels of program diffusion, penetration or coverage. The rates of
incorporation and program coverage at the state and locality levels were driven
primarily by political considerations (party in power); administrative issues (rate at
which clinics and hospitals where accredited to belong to the SP program) and budget
availability.
In addition to estimating the impact of SP on CHE, we also used the same econometric
framework to estimate the effect of SP on out-of-pocket expenditures. We defined out-
of-pocket expenditures as any positive expenditure related to outpatient care, inpatient
care, and medicines. Thus, we created three dichotomous variables equal to the unity if
there was positive spending, and equal to zero otherwise.
To control for important predictors and enrolment and CHE, in the main results for
ENSANUT we controlled for covariates regarding to indigenous language and
indigenous self-identification, as well as the presence of a chronic illness by someone in
the household. Unfortunately those variables were available neither in the ENIGH nor
in the SP Impact Evaluation Survey. Thus, sensitivity analyses exercises tested all the
models using the same set of covariates for the three surveys.
For all model specifications, we compared the results from “naïve” estimates where the
choice of health insurance use was assumed to be exogenous, to the results we obtained
using instrumental variables. Analyses were conducted using STATA™ (StataCorp
2005), including the probit and biprobit procedures.
14
4. RESULTS
Table 1 shows the description of the dependent and explanatory variables for the three
datasets. The main dependent variable was catastrophic health expenditures which were
more comparable in the ENIGH and ENSANUT surveys, than in the SP Impact Survey,
which showed higher CHE. Similarly, out-of-pocket expenditures were also more
prevalent in the SP Impact Survey dataset.
The characteristics of the household-head we analysed included: age, female-headed,
formal education, indigenous self-identity, speak indigenous language. For these
variables we did not find important differences between SP insured vs. uninsured
households, particularly with the ENSANUT and the SP Impact Survey; differences
between the two groups were more noticeable for the ENIGH survey. For both
uninsured and insured groups the mean of the age of the household-head in ENSANUT,
ENIGH and SP impact survey fluctuated between 44 to 50 years. However, the
percentage of household that were female-headed was considerably higher among the
uninsured in the ENIGH with approximately 29% with respect to 21-22% found in the
ENSANUT and SP Impact Survey. In the three surveys the percentage of female-
headed households among the insured was between 19 and 22%. The number of years
of education for the household head was around 5 to 6 years, although in the ENIGH
this variable was considerable lower among the sample of insured households:
approximately only 3.7 years of schooling.
We also included characteristics of the household such as: the household asset index as
a proxy of family income, household size, and benefits from other social programmes,
particularly the Oportunidades programme. At this level, we considered those variables
15
that denoted the composition of the family: children who were one year old or younger,
and children who were 7 years old or younger, as well as adults 65 years-old or older.
For the specific case of ENSANUT we included additional variables that informed us
about the presence of some chronic health conditions among at least one of the members
of the family (diabetes, hypertension and gastritis). The mean for the household asset
index was generally lower for the SP insured population, indicating lower levels of
family wealth. The mean household size was around 3 to 4 members; although this
indicator was slightly higher among SP insured households, particularly in the SP
Impact Survey dataset (with almost 5 members). Differences in the composition of the
household were notable in the percentage with children 7 years-old or younger. While in
the ENIGH this percentage was around 37% in the insured sample, it was considerable
higher (about 53%) in the insured sample in the SP Impact Survey. The percentage of
families with adults 65 years-old or older varied in the three surveys between 15 to 35%
in both insured and uninsured sub-samples.
Comparing insured and uninsured, we did not find statistically significant differences
between households reporting at least one member of with a chronic condition in the
ENSANUT survey. However, we found differences in the percentage of families who
reported to be beneficiaries of the Oportunidades programme among the different
surveys. Specifically, we found a considerable lower percentage of families from the
ENSANUT survey who reported to be benefited from this programme than the rest of
the surveys. Generally, though, across the surveys, SP insured households seemed to be
consistently more enrolled in Oportunidades.
16
At locality level we incorporated variables that described the rural or urban condition of
the municipalities and the deprivation index (CONAPO 2005). While there was 56 to
53% of SP insured households, and 30 to 34% of the uninsured living in rural localities
reported in the ENIGH and ENSANUT surveys, the SP Impact Survey showed 96% of
the insured, and about 89% of uninsured living in rural areas. Most of the households
from the SP Impact Survey were poor, and lived in rural areas with a high deprivation
index. Insured and uninsured households were much poorer in the SP Impact Survey
than in the other two surveys.
Table 2 reports the effect of SP on CHE results for the ENIGH survey. The first two
columns represent the coefficients and marginal effects for the naïve probit model.
Note that the enrolment into Seguro Popular was not a statistically significant
determinant of catastrophic health expenditures. Having an infant less one year of age
or younger increased the probability of CHE by five percentage points; similarly, the
presence of an older adult (65 years of age or older) also increased that probability by
2.2 percentage points.
The last three columns in Table 2 present the bivariate probit model which corrects for
endogeneity in the selection into SP using instrumental variables. First, note that in the
SP equation, the instrumental variables were significant and of the expected sign. Since
there are no formal tests for instrumental variables in the non-linear context, we
conducted informal checks with linear specifications. In the linear models (with all
three data sources), the IVs were highly relevant. The F test statistics of excluded
instruments were very high; all much higher than the “rule of thumb” of 10 to detect
weak instruments (Bound, Jaeger et al. 1995). (Additional details for the linear models
and formal tests are presented elsewhere: Salud Pública de México, forthcoming).
17
Even though the IVs seemed to be working well (longer time of program exposure at
the state level, and higher coverage at the locality level increased the household
probability of having SP insurance), after we addressed the selection issue, we found no
effect of SP on CHE using the ENIGH sample: the effect was not statistically different
from zero.
Table 3 presents the results from the National Health and Nutrition Survey (ENSANUT
2006). As before, we contrasted naïve versus selection-correction non-linear models.
In the naïve probit model, where we estimated the CHE equation independently, we can
see that SP enrolees had a 2 percentage-point lower probability of incurring CHE. On
the other hand, having small children or older adults in the household made CHE more
likely. Similarly, the presence of someone with chronic health conditions (particularly,
diabetes and hypertension) also increased the likelihood of CHE.
In the bivariate probit model, once we addressed selection bias, the effect of SP on
CHE, was even more protective. Households insured with SP had a probability of CHE
that was 3.6 percentage points lower than uninsured households.
The instruments in the ENSANUT data were also significant and worked well. The
seemingly unusual negative sign of “Incorporation by 2003” can be explained as the
effect with respect to the following year. Furthermore, the aggregated effect (adding the
yearly effects) was positive as expected. Households in states that incorporated earlier
to the SP network had a higher probability of enrolling into SP.
18
Note also in the SP equation of Table 3 that that the estimates for the coefficient of
correlation () were positive and significant, meaning that unobservable factors
associated with the outcome (excessive health expenditures) were also associated with
the treatment variable (SP enrolment). That is, household with SP insurance had
unobserved characteristics that made them more likely to incur into CHE.
Table 4 presents the naïve and selection correction models using the SP Impact
Evaluation Survey. Note that the naïve estimate of the protective effect of SP on CHE
was of about 1.8 percentage points, while the effect grew to a 5.3 percentage-point
difference once we corrected for endogeneity.
Table 4 also shows that both instruments were statistically significant, and of the
hypothesized sign. As the SP coverage increased at the locality level, the probability
households to participate in SP also increased. The year of incorporation dummy also
worked as expected. If the state incorporated into SP by 2001, the household
probability of enrolment was larger than if the state incorporated by 2002 (the reference
year in this sample, given that all seven states included in the SP Evaluation Survey
incorporated to SP by 2002). The coefficient of correlation was also positive and
significant as hypothesized.
Table 5 shows the results for the effect of SP on out-of-pocket (OOP) expenditures,
including outpatient care, inpatient care, and medicines. Note that the effect of SP on
outpatient care expenditures was protective for all the samples, but was much stronger
for the SP Impact Survey population. On the other hand, we found a protective effect
against inpatient care expenditures only in the SP Impact Survey population; the effect
19
was not different from zero in the other two datasets. Finally, for out-of-pocket
expenditures related to medicines, we found again that SP reduced the probability of
spending out-of-pocket for medicines for all samples, but the effect was stronger in the
SP Impact Survey.
5. DISCUSSION
The overall larger protective effect of Seguro Popular on catastrophic health
expenditures (CHE) and out-of-pocket expenditures in the SP Impact Evaluation Survey
can be explained as that survey was targeted to a population with lower socio-economic
status. As seen in Table 1, the SP Impact Survey population was poorer, more rural,
and in more marginalized communities. On the other hand, the ENSANUT and ENIGH
surveys are nationally representative samples of the entire Mexican population
(including middle- and high-income households). It is expected then that the effect of
SP on the general population be somewhat diminished.
The ENIGH survey has the advantage of being more accurate in collecting information
about the total household’s expenditure in comparison to the other two surveys.
Although ENSANUT and the SP Impact Evaluation Survey had similar contents,
mainly related to health and use of health services; the first one is representative at
nationally level, whilst the second one was gathered in a few states in Mexico where the
SP programme had a low rate of SP penetration at the time of the development of the
survey. Note that the differences between ENSANUT and ENIGH were not due to the
additional control variables in ENSANUT. In Appendix Tables A1 and A2 we show
that the protective effect of SP on CHE is still present when we do not control for
20
indigenous language and self-identity and the chronic health condition variables in
ENSANUT.
By using the same non-experimental (IV) approach and the same covariates with all
three datasets, we can ascribe the differences in results to the differences in data
collection methods only. Nevertheless, it is still puzzling that no effect is observed
using the ENIGH, but that a significant protective effect of SP against CHE is present in
both ENSANUT and the SP Impact Survey.
On the other hand, note that our results of SP program participation on CHE are similar
to those found by King and collaborators (King and et al 2006; King, Gakidou et al.
2007), when analyzing experimental data with the SP Impact Evaluation Survey. They
found that SP reduces the probability of incurring catastrophic health expenditures by
1.5% (intention to treat analysis) and by 3% using the complier causal average effect
(CACE). These differences with our results are explained from the treatment effect
methodology used, although in both cases the effects were significant.
21
6. CONCLUSION AND LIMITATIONS
The results in this paper show that the universal insurance system for the poor in
Mexico, Seguro Popular, has a protective effect on excessive health expenditures, as
well as out-of-pocket health-related expenditures, particularly for the poorest segments
of the population.
Given these results, the SP program should continue to concentrate in enrolling the
poorest households, in the most marginalized localities, because it is in those
populations where we observe a higher prevalence of excessive health expenditures, but
also where the SP program seems to be more effective for financial protection.
The fact that we have defined our analytical sample for insured households as only
those where all members of households are enrolled in SP, is likely to be under-
estimating the number of families enrolled in SP, and the potential effect of the
programme. In that sense, given the strict criteria to define the analytical sample, we
view the results provided as conservative estimates.
One of the limitations of the definition of catastrophic health expenditures is the fact
that it does not consider all those households that postpone their health care for the lack
of financial resources. We do not address the issue of selection into CHE. Health
expenditure is, by definition, conditional on utilization. Thus, there is another possible
econometric specification to take into account of the endogeneity generated by health
seeking behavior. Such correction requires specific data on non-users (which was not
available to us); just a few papers attempt to correct such inherent endogeneity (Pradhan
and Prescott 2002; Hatt 2006).
22
The present analysis does not include other alternative indicators that can also describe
the effect of the incidence of health expenditures on the household’s financial status
such as the impoverishing expenditures; which are the expenditures do not provoke
family’s bankruptcy but do move the household’s economic status below the poverty
line.
Lastly, at the household level, the cross-sectional surveys do not provide information
about the length of treatment (i.e., how long the households have been enrolled in the
SP programme), which could have an important effect on excessive health expenditures.
ACKNOWLEDGEMENTS
We thank Martha Maria Tellez-Rojo, Gustavo Nigenda, and Miguel Angel González-
Block for inviting us to participate in the 2007 Seguro Popular Program Evaluation
Team, sponsored by the Mexican Ministry of Health. The results presented in this
paper constitute additional work conducted by the authors. The views do not represent
the official position of any of the institutions mentioned. The authors are solely
responsible for the contents.
23
REFERENCES
Angrist, J. D., G. W. Imbens, et al. (1996). "Identification of Causal Effects Using
Instrumental Variables." Journal of the American Statistical Association
91(434): 444-455.
Angrist, J. D. and A. B. Krueger (2001). "Instrumental Variables and the Search for
Identification: From Supply and Demand to Natural Experiments." Journal of
Economic Perspectives 15(4): 69-85.
Bound, J., D. A. Jaeger, et al. (1995). "Problems with Instrumental Variables Estimation
When the Correlation between the Instruments and the Endogenous Explanatory
Variable Is Weak." Journal of the American Statistical Association 90(430):
443-50.
CONAPO. (2005). "Índice de marginación a nivel localidad, 2005 ", from
http://www.conapo.gob.mx/publicaciones/indice2005xloc.htm.
Duflo, E. (2001). "Schooling and Labor Market Consequences of School Construction
in Indonesia: Evidence from an Unusual Policy Experiment." American
Economic Review 91(4): 795-813.
Frenk, J., E. Gonzalez-Pier, et al. (2006). "Comprehensive reform to improve health
system performance in Mexico." Lancet 368(9546): 1524-34.
Gakidou, E., R. Lozano, et al. (2006). "Assessing the effect of the 2001-06 Mexican
health reform: an interim report card." Lancet 368(9550): 1920-35.
Greene, W. H. (2003). Econometric analysis. Upper Saddle River, N.J., Prentice Hall.
Hatt, L. E. (2006). Measuring Risk Factors for Catastrophic Health Expenditures in
Peru, and their Effects on Families over Time. Ph.D. Dissertation. Department
of International Health Baltimore, MD, Johns Hopkins University. Ph.D.: 281.
Heckman, J. J. (1997). "Instrumental Variables: A Study of Implicit Behavioral
Assumptions Used in Making Program Evaluations." Journal of Human
Resources 32(3): 441-462.
INEGI. (2005). "Conteo de población y vivienda 2005 ", from
http://www.inegi.gob.mx/inegi/default.aspx?s=est&c=10215.
INEGI. (2006). "Encuesta Nacional de Ingresos y Gastos de los Hogares (ENIGH)."
from
http://www.inegi.gob.mx/est/contenidos/espanol/proyectos/encuestas/hogares/en
igh/enigh_2006/Propuesta/default.asp.
INSP. (2006). "Encuesta Nacional de Salud y Nutrición (ENSANUT)." from
http://www.insp.mx/ensanut/.
Jowett, M., P. Contoyannis, et al. (2003). "The impact of public voluntary health
insurance on private health expenditures in Vietnam." Soc Sci Med 56(2): 333-
42.
Jowett, M., A. Deolalikar, et al. (2004). "Health insurance and treatment seeking
behaviour: evidence from a low-income country." Health Econ 13(9): 845-57.
Kawabata, K., K. Xu, et al. (2002). "Preventing impoverishment through protection
against catastrophic health expenditure." Bull World Health Organ 80(8): 612.
King, G. and et al (2006). Evaluation of the System for Social Protection in Health.
Boston, MA, Harvard University, Mexican Ministry of Health, and National
Institute of Public Health (INSP).
King, G., E. Gakidou, et al. (2007). "A "politically robust" experimental design for
public policy evaluation, with application to the Mexican universal health
insurance program." J Policy Anal Manage 26(3): 479-506.
24
Knaul, F. M., H. Arreola-Ornelas, et al. (2006). "Evidence is good for your health
system: policy reform to remedy catastrophic and impoverishing health spending
in Mexico." Lancet 368(9549): 1828-41.
Knaul, F. M., H. Arreola-Ornelas, et al. (2005). "[Financial protection in health:
Mexico, 1992 to 2004]." Salud Publica Mex 47(6): 430-9.
Maddala, G. S. (1994). Econometric methods and applications. Aldershot, Hants,
England ; Brookfield, Vt., USA, E. Elgar.
Makinen, M., H. Waters, et al. (2000). "Inequalities in health care use and expenditures:
empirical data from eight developing countries and countries in transition." Bull
World Health Organ 78(1): 55-65.
McKenzie, D. J. (2004). "Measuring inequality with asset indicators." Journal of
Population Economics 18(2): 229-260.
Murray, C., F. Knaul, et al. (2000). Defining and measuring fairness of financial
contribution. Global Programme on Evidence Working Paper No. 24. Geneva,
World Health Organization.
Parker, S. W. and R. Wong (1997). "Household income and health care expenditures in
Mexico." Health Policy 40(3): 237-55.
Perez-Rico, R., S. Sesma-Vazquez, et al. (2005). "[Catastrophic health expenditures in
Mexico: comparative study by social exclusion level]." Salud Publica Mex 47
Suppl 1: S47-53.
Pradhan, M. and N. Prescott (2002). "Social Risk Management Options for Medical
Care in Indonesia." Health Economics 11(5): 431-446.
Sepehri, A., S. Sarma, et al. (2006). "Does non-profit health insurance reduce financial
burden? Evidence from the Vietnam Living Standards Survey Panel." Health
Econ 15(6): 603-16.
Sesma-Vazquez, S., R. Perez-Rico, et al. (2005). "[Catastrophic health expenditures in
Mexico: magnitude, distribution and determinants]." Salud Publica Mex 47
Suppl 1: S37-46.
Sosa-Rubi, S. G., O. Galarraga, et al. (2007). Heterogeneous Impact of the 'Seguro
Popular' Program on the Utilization of Obstetrical Services in Mexico, 2001-
2006: A Multinomial Probit Model with a Discrete Endogenous Variable,
National Bureau of Economic Research, Inc, NBER Working Papers: 13498.
SSA (2006a). Sistema de Protección Social en Salud. Vols. 1-3. México, D.F., México
Secretaría de Salud (SSA), Gobierno de México.
SSA. (2006b). "Seguro Popular: Encuesta de Evaluación de Impacto." from
http://www.salud.gob.mx/unidades/evaluacion/seguropopular/seguropopular.ht
m.
StataCorp (2005). Stata Statistical Software: Release 9. College Station TX, StataCorp
LP.
Torres, A. C. and F. M. Knaul (2003). Determinantes del gasto de bolsillo en salud e
implicaciones para el aseguramiento universal en México: 1992–2000.
Caleidoscopio de la salud: De la investigación a las políticas y de las políticas a
la acción. . F. M. Knaul and G. Nigenda. México. D.F., México Fundación
Mexicana para la Salud: 209–225.
Van Damme, W., L. Van Leemput, et al. (2004). "Out-of-pocket health expenditure and
debt in poor households: evidence from Cambodia." Trop Med Int Health 9(2):
273-80.
Wagstaff, A. and S. Yu (2007). "Do health sector reforms have their intended impacts?
The World Bank's Health VIII project in Gansu province, China." J Health Econ
26(3): 505-35.
25
Wooldridge, J. M. (2002). Econometric analysis of cross section and panel data.
Cambridge, Mass., MIT Press.
Xu, K., D. B. Evans, et al. (2007). "Protecting households from catastrophic health
spending." Health Aff (Millwood) 26(4): 972-83.
Xu, K., D. B. Evans, et al. (2003). "Household catastrophic health expenditure: a
multicountry analysis." Lancet 362(9378): 111-7.
Xu, K., J. Klavus, et al. (2003). Household health system contributions and capacity to
pay: Definitional, empirical and technical challenges. Health system
performance assessment: Debates, methods and empiricism. C. Murray and D.
Evans. Geneva, World Health Organization: 533-563.
26
!" #$#% #$#% #$#% #$#% #$#% #$#%
& #$#% #$#% #$#% #$#% #$#% #$#%
' #$#% #$#% #$#% #$#% #$#% #$#%
(!" #$#% #$#% #$#% #$#% #$#% #$#%
)* #$#% #$#% #$#% #$#% #$#% #$#%
+,-!! #$#% #$#% #$#% #$#% #$#% #$#%
+,! $."% #$#% #$#% #$#% #$#% #$#% #$#%
'!*""/-!. #$#% #$#%
0!*"** #$#% #$#%
)""'! -#$#% -#$#% -#$#% -#$#% -#$#% #$#%
!1 #$#% #$#% #$#% #$#% #$#% # $#%
!1 #$#% #$#% #$#% #$#% #$#% #$#%
)!"2 #$#% #$#% #$#% #$#% #$#% #$#%
,3!4" #$#% #$#%
,3." #$#% #$#%
,3*"" #$#% #$#%
"!"5$6/""% #$#% #$#% #$#% #$#% #$#% #$#%
!&!!" #$#% #$#% #$#% #$#% #$#% #$#%
7 #$#% #$#% #$#% #$#% #$#% #$#%
89! -#$#% -#$#% -#$#% -#$#% -#$#% -#$#%
:.9* #$#% #$#% #$#% #$#% #$#% #$#%
'!4. #$#% #$#% #$#% #$#% #$#% #$#%
'!4. #$#% #$#% #$#% #$#%
'!4. #$#% #$#% #$#% #$#%
'!4. #$#% #$#% #$#% #$#%
";)<" "4"!=)=>? $'=%='@$'=@'%',9.$)4%
""!$'= @'%!9!$)%#
=";!!""#
'"!" ,,4"9!9."*,3"
"!"
/ ,,4"9../ "#
!"# $ % !
27
A
//"
(*//"
//"
//"
(*
//"
!*$!% # # # #
B#C B#C B#C B#C
)* # # -# # #
B#CDD B#CDD B#CDD B#CDD B#CDD
+,-!!$!% # # -# # #
B#C B#C B#CDD B#C B#C
+,!$."% -# -# -# -# -#
B#CDD B#CDD B#CDD B#CDD B#CDD
)""'! -# -# -# -# -#
B#CD B#CD B#CE B#CD B#CD
!1$!% # # -# # #
B#CDD B#CDD B#CDD B#CDD B#CDD
!1$!% -# -# # -# -#
B#C B#C B#CDD B#C B#C
)!"2$!% # # -# # #
B#CDD B#CDD B#C B#CDD B#CDD
+,."5 -# -# -# -# -#
B#CDD B#CDD B#CDD B#CDD B#CDD
!&!!"$!% -# -# # -# -#
B#CDD B#CDD B#CDD B#CD B#CD
7$!% # # # # #
B#CDD B#CDD B#CDD B#CDD B#CDD
89! # # -# # #
B#C B#C B#CDD B#C B#C
:.9* #
B#CDD
'!4.$!% #
B#C
'!4.$!% #
B#CDD
'!4.$!% #
B#C
'!4.$!% -#
B#CD
#
B#C
" -# -# -#
B#CDD B#CDD B#CDD
&4"9"
=";,*//"",!,/99
&
#
$!%,*"/!"*/!,,.94/,
4""!!"40"
E"*/F GD"*/FGDD"*/F
AA
' !"#'(()*+++# ,
=94( ! H94(!
28
A
//"
(*//"
//"
//"
(*//"
!*$!% -# -# -# -#
B#CDD B#CDD B#CDD B#CDD
)* # # -# # #
B#CDD B#CDD B#CDD B#CDD B#CDD
+,-!!$!% -# -# -# -# -#
B#C B#C B#CD B#C B#C
+,!$,4/."% -# -# -# -# -#
B#C B#C B#CDD B#CE B#CE
'!*""/-!.$!% # # -# # #
B#C B#C B#CDD B#C B#C
0!*"**$!% -# -# -# -# -#
B#CDD B#CDD B#C B#CDD B#CDD
)""'! # # -# # #
B#C B#C B#CD B#C B#C
!1$!% # # -# # #
B#CDD B#CDD B#CDD B#CDD B#CDD
!1$!% # # # # #
B#C B#C B#CDD B#C B#C
)!"2$!% # # -# # #
B#CDD B#CD B#C B#CDD B#CD
,3!4"$!% # # # # #
B#CD B#CD B#CD B#CD B#CD
,3."$!% # # # # #
B#CE B#CE B#CDD B#CE B#CE
,3*""$!% # # # # #
B#CE B#C B#CE B#CE B#C
+,."5 -# -# -# -# -#
B#CDD B#CDD B#CDD B#CDD B#CDD
!& !!"$!% -# -# # -# -#
B#C B#C B#CDD B#C B#C
7$!% # # -# # #
B#CD B#CD B#CDD B#CDD B#CDD
89! # # -# # #
B#CDD B#CDD B#CDD B#CDD B#CDD
:.9* #
B#CDD
'!4.$!% #
B#CDD
'!4.$!% #
B#C
'!4.$!% -#
B#CDD
'!4.$!% #
B#C
///$% #
B#CDD
" -# -# -#
B#CDD B#CDD B#CDD
&4"9"
=";,*//"",!,/99
&
#
$!%,*"/!"*/!,,.94/,
4""!!"40"
E"*/F GD"*/FGDD"*/F
- $ %'(()*+++# ,
=94(! H94(!
A A
29
A
//"
(*
//"
//"
//"
(*
//"
!*$!% -# -# -# -#
B#CDD B#CDD B#CDD B#CDD
)* # # -# # #
B#CDD B#CDD B#CE B#CDD B#CDD
+,-!!$!% # # # # #
B#C B#C B#CE B#C B#C
+,!$,4/."% -# -# -# -# -#
B#CDD B#CDD B#C B#CDD B#CDD
)""'! -# -# -# -# -#
B#CDD B#CDD B#CDD B#CDD B#CDD
!1$!% # # # # #
B#CDD B#CDD B#C B#CDD B#CDD
!1$!% -# # # # #
B#C B#C B#CDD B#C B#C
)!"2$!% # # -# # #
B#CDD B#CDD B#C B#CDD B#CDD
+,."5 -# -# # -# -#
B#CDD B#CDD B#CDD B#CDD B#CDD
!&!!"$!% -# -# # # #
B#CD B#CD B#CDD B#C B#C
7$!% # # -# # #
B#CDD B#CDD B#CDD B#CDD B#CDD
89! # # # # #
B#CDD B#CDD B#CDD B#CDD B#CDD
:.9* #
B#CDD
'!4.$!% #
B#CD
///$% #
B#CDD
" -# -# -#
B#CDD B#CDD B#CDD
&4"9"
=";,*//"",!,/99
&
#
$!%,*"/!"*/!,,.94/,
4""!!"40"
E"*/FGD"*/FGDD"*/F
.!*+++#,
=94( !
A
H94(!
A
30
///&&
!"
//"
(*
//"
//"
(*
//"
//"
(*
//"
& -# -# -# -# -# -#
B#CDD B#CDD B#CDD B#CDD B#CDD B#CDD
' -# -# # # -# -#
B#C B#C B#C B#C B#CDD B#CDD
(!" -# -# -# -# -# -#
B#CE B#CE B#CDD B#CDD B#CDD B#CDD
=";
H94,!"////&& !"*/9"?4#
(*//"",!,/9"
74""!!"40"
E"*/FGD"*/FGDD"*/F
/++++012,
!"# $ % !
31
A
//"
(*
//"
//"
//"
(*
//"
!*$!% -# -# -# -#
B#CDD B#CDD B#CDD B#CDD
)* # # -# # #
B#CDD B#CDD B#CDD B#CDD B#CDD
+,-!!$!% -# -# -# -# -#
B#C B#C B#CD B#C B#C
+,!$,4/."% -# -# -# -# -#
B#CE B#CE B#CDD B#CE #
)""'! # # -# # #
B#C B#C B#C B#C B#C
!1$!% # # -# # #
B#CDD B#CDD B#CDD B#CDD B#CDD
!1$!% # # # # #
B#C B#C B#CDD B#C B#C
)!"2$!% # # -# # #
B#CDD B#CD B#C B#CDD B#CD
+,."5 -# -# -# -# -#
B#CDD B#CDD B#CDD B#CDD B#CDD
!&!!"$!% -# -# # -# -#
B#C B#C B#CDD B#C B#C
7$!% # # -# # #
B#CDD B#CD B#CDD B#CDD B#CDD
89! # # -# # #
B#CDD B#CDD B#CDD B#CD B#CD
:.9* #
B#CDD
'!4.$!% #
B#CDD
'!4.$!% #
B#C
'!4.$!% -#
B#CDD
'!4.$!% #
B#C
///$% #
B#CDD
" -# -# -#
B#CDD B#CDD B#CDD
&4"9"
=";"9..""!""!*"$**!"/-!.%"/!"#
(*//"",!,/99
&
#
$!%,*"/!"*/!,,.94/,
4""!!"40"
E"*/FGD"*/FGDD"*/F
$,$*$*
+++# , $ %
=94( ! H94(!
A A
32
///&& !"
//"
(*
//"
& -# -#
B#CDD B#CDD
' # #
B#C B#C
(!" -# -#
B#CDD B#CDD
=";
"9..""!""!*"$**!"/-!.%"/!"#
H94,!"////&& !"*/9"?4#
(*//"",!,/9"
74""!!"40"
E"*/F GD"*/F GDD"*/F
$,$'$'*+++
+012, $ %
$ %
