The Effect Of Parents’ Insurance
Enrollment On Health Care
Utilization: Evidence From Ghana
"Leaders in Development" Seminar Series
February 27, 2013
Gissele Gajate-Garrido
Clement Ahiadeke
IFPRI
ISSER-University of Ghana
Research question
What is the impact of parental
participation in the National Health
Insurance Scheme (NHIS) on health care
utilization in Ghana?
Contributions to the literature



This paper is the first attempt to look at the impact of
parental participation in the NHIS on both curative and
preventive health care utilization in Ghana using
information from the whole country.
It uses exogenous variation in insurance participation to
correct for the potential endogeneity
The variation comes from the variability in membership
rules of the District Mutual Health Insurance Schemes
(DMHIS) located in every district in Ghana
Summary of findings

Insurance membership increases the probability of
parents:
Seeking higher quality, but not greater quantity of prenatal
services
 Becoming more active users of child curative care (covered
by the NHIS)
 Becoming more active users of child preventive care (not
covered by the NHIS)


These findings are consistent with the idea that when
curative services become available for free, parents
become more proactive towards other kinds of health
care utilization practices (sibling hypothesis, access
hypothesis, information hypothesis)
Summary of findings




IV estimates are larger than OLS ones indicating
underlying heterogeneity in returns to NHIS participation
“Compliers” have much higher returns to participating in
the NHIS than the average participant
We verify the exogeneity of the instrument using pre
scheme data.
We also collect data on barriers to health care access in
districts, as well as challenges to participate in a DMHIS
so we can control for district characteristics that determine
both membership rules and health care seeking behaviors
Outline

Motivation

The NHIS and the district membership rules

Data

Identification Strategy

Specification

Results and Discussion

Instrument Strength and Validity

Conclusion
Motivation





Access to and utilization of health services continues to be a
main concern in poor countries such as Ghana.
Delaying medical treatment or choosing self-treatment can
generate serious health consequences (Hadley 2002).
Young children are particularly susceptible to negative
health shocks that can generate nutritional deficits and
cognitive disabilities.
Early experiences are particularly important for future skill
development and establishment of mental potential
(Knudsen, Heckman, Cameron and Shonkoff 2006 and
Heckman 2007).
If implemented correctly health insurance systems could
provide an effective solution to this challenge.
The NHIS


The National Health Insurance Act (Act 560) which
created Ghana’s National Health Insurance Scheme
(NHIS) was passed in 2003. Yet, the scheme became
operational only in March 2004 (Hsiao & Shaw 2007).
Before this scheme was implemented Ghana had a
“cash and carry system” of health delivery. Under this
system, patients – even those who had been brought into
the hospital on emergencies – were required to pay
money at every point of service delivery.
The NHIS



1
The most popular insurance plan is the District-Wide
Mutual Health Insurance Scheme1 which operates in
every district in Ghana. This is the insurance scheme
analyzed in this paper.
It covers over 95% of disease conditions that afflict
Ghanaians and around 40% of our sample is enrolled
in it.
People pay an annual fee according to their income,
yet the government subsidizes pensioners, the old
(70+), the core poor or indigent and the children and
dependents below 18 of parents that participate in the
system.
97% of people with an insurance plan are registered in a DMHIS (DHS 2008)
District Mutual Health Insurance Schemes





The law establishes that the District Mutual Health
Insurance Schemes (DMHIS) are autonomous entities.
Each scheme in each district is completely
independent of any other in the country with
independent boards of directors.
Schemes do not pool risks together in any way.
Hence membership rules vary across different DMHIS.
This source of exogenous variation is used for
identification of parental participation in the NHIS
Membership rules


The existence of exceptions that allowed children to
benefit from the NHIS without their parents being
registered  this could increase child participation
in the scheme .
The verification methods employed by the DMHIS to
ensure parents were registered in the NHIS in order
for their children to benefit  Easier methods, more
participation.
Membership rules


The amount of money paid to renew insurance cards
both for adults and children vary across districts 
Districts with higher renewal fees would probably
discourage participation of marginal participants.
The waiting periods for both adults and children
when registering  Longer wait, less participation.
Data


The 2008 Ghana Demographic and Health Survey
(GDHS) is the main source of data for this study. This
survey is carried out by the Ghana Statistical
Service and the Ghana Health Service.
This is a national survey designed to collect
information on housing and household
characteristics, education, maternal health and child
health, nutrition, family planning and gender.
Data



In addition we carried out a census of all the District
Mutual Health Insurance Schemes (DMHIS) in
existence in 2008. The census contains information
on the membership rules in each of the 145 DMHIS
licensed by 2008.
This census also contains district level information on
health services delivery in the community during
2008.
Despite the availability of a health insurance
program, families might not decide to join if they
perceive the services they could access through this
system as deficient.
Outcomes analyzed

Curative child health care utilization
 If
mothers chose to deliver at a health facility
 If they treat their children with antimalarial medication
when they are sick with either a fever or cough
 If they seek advice from a health provider when their
children experience a fever or cough
Outcomes analyzed

Quantity and quality of maternal health services
Number of prenatal visits sought
 If pregnant women were told about pregnancy complications
during prenatal visits
 If they had urine and blood samples taken during pregnancy


Preventive child health care utilization.
Vaccination rates (for yellow fever, BCG, DPT/HepB/
Influenza 2)
 If the child slept under a mosquito bed net the previous night.

Descriptive Statistics by insurance participation
Child Gender (male)
Child Age (in years)
Child is a twin
Mother's educational attainment
Not registered in the
DMHIS
N
Mean
1681
0.49
1681
1.96
1808
0.04
1808
1.29
Registered in the
DMHIS
N
Mean
1107
0.53
1107
1.85
1178
0.05
1175
1.90
Diff
-0.04
0.11
-0.01
-0.60
P-value
0.04**
0.05**
0.39
0.00***
Mother currently pregnant
1681
0.08
1107
0.08
0.00
0.86
Children born in the last 5 years
Number of living children
Mother’s Age (in years)
1808
1808
1808
1.67
1.75
29.84
1178
1178
1178
1.57
1.61
30.44
0.11
0.15
-0.60
0.00***
0.03**
0.02**
Mother currently working
1808
0.86
1178
0.88
-0.02
0.09*
Household head is male
Wealth Index
Type of place of residence (0 rural, 3 urban)
Population in millions (region)
1808
1808
1808
1808
0.73
-45869
0.79
2.20
1178
1178
1178
1178
0.75
-1957
1.03
2.10
-0.02
-43912
-0.23
0.10
# of facilities per 100,000 people (region)
1808
7.86
1178
7.92
-0.06
0.61
Distance to doctor (km)
1808
4.48
1178
3.30
1.18
0.03**
Distance to private hospital (km)
1808
21.89
1178
17.92
3.97
0.02**
Dummy for poor regions
1808
0.33
1178
0.35
-0.02
0.26
Source: DHS 2008
0.20
0.00***
0.00***
0.00***
Descriptive Statistics by insurance participation
Not registered in Registered in the
the DMHIS
DMHIS
N
Mean
N
Mean
Curative child health care
Delivery at health facility
Anti-malarial medication taken for fever/cough
Seek medical treatment - child has fever/cough
Quantity and quality of maternal health care
Number of prenatal visits
Told about pregnancy complications
Urine sample taken during pregnancy
Blood sample taken during pregnancy
Preventive child health care
BCG rate
DPT/HepB/ Influenza 2 rate
Yellow fever vaccine rate
Child slept under mosquito net
Diff
P-value
1357
494
494
0.45
0.27
0.33
909
326
326
0.74
0.37
0.61
-0.29
-0.10
-0.28
0.00***
0.00***
0.00***
992
962
965
965
5.21
0.65
0.85
0.85
707
705
706
706
6.49
0.74
0.92
0.95
-1.28
-0.08
-0.07
-0.10
0.00***
0.00***
0.00***
0.00***
1672
1670
1665
1808
0.90
0.83
0.71
0.45
1101
1098
1103
1178
0.95
0.87
0.74
0.51
-0.05
-0.04
-0.02
-0.06
0.00***
0.00***
0.24
0.00***
Source: DHS 2003 and 2008, P-values are reported from t-test on the equality of means for each variable.
* p<.10; ** p<.05; *** p<.01
Identification Strategy



One of the main difficulties in identifying the effect of
joining a health insurance scheme on child health care
utilization is selection bias or omitted variable bias.
Omitted variable bias is an issue because people who
participate less in formal health insurance systems may
do so because, for example, of their wealth or
education level which in turn could affect child health
care use. To address this concern we include a wide
range of covariates as controls.
Nonetheless, unobserved characteristics that cannot
directly be controlled for remain a concern.
Identification Strategy


For example the household level of risk aversion could
induce parents to choose to enroll in a health insurance
plan and also affect their ability to care for their child.
Moreover, a household could decide to spend more or
less in health inputs (such as health insurance) according
to the health capital of their children. Parents with
healthier children might participate more (favorable
selection) in health insurance schemes. Conversely,
parents with frailer children might decide to enroll more
frequently (adverse selection).
Identification Strategy


To address this endogeneity problem we use an
instrumental variable approach. The exogenous
variation in the decision to join the National Health
Insurance system comes from the variations in the
membership rules in each DMHIS.
We would expect parental participation to be
higher in districts where membership rules are less
stringent or more straightforward.
Testing the IV’s exogeneity

Before the creation of these offices, government officials in the
different DMHIS could have chosen to establish their membership
rules to make them less stringent in districts that had:




Lower child health care utilization rates or worse maternal health care
Higher poverty, malaria or fertility rates
To test this hypothesis we use the 2003 DHS data (the best health
indicators available in Ghana at the time the NHIS was created
and before the district rules were actually established) and run
regressions for each instrument chosen on each of the outcomes
examined in the paper, as well as on district level poverty, malaria
and fertility rates.
The results show the instruments chosen are indeed exogenous.
Testing the IV’s exogeneity


The results could also be biased by the existence of district
characteristics that determine both membership rules and health
care seeking behaviors.
Higher barriers to access health care in a district as well as greater
challenges to participate in a DMHIS could also imply lower health
care utilization rates. E.g.





Low quality/quantity health care
Poor road network
High rate divorce/single parenthood
Lack of education/preference for traditional medicine
To discard such a hypothesis we test that the results obtained are not
affected after introducing both barriers affecting children's access
to health care in 2008, as well as challenges to participation in the
DMHIS. The results confirm that the IV coefficients are not biased.
Specification

First stage
I ij   0  1 Z1 j   2 Z 2 j   3 Z 3 j   4 X ij   5 X ij   6 X ij   7 X j   ij
c




H
R
(1)
Iij is a dummy indicating whether mother i is registered in the DMHIS in district j,
Z1j to Z3j = membership rules in district j
Xt c, XtM, XtH, XtR = child, mother, household and community characteristics
Second stage
Cij   0  1 I ij   2 X ij   3 X ij
c



M
M
  4 X ij   5 X j   ij
H
R
(2)
Cij = child or maternal health care utilization choice
ij = unobserved attributes related to both the mother and the child such as
resilience, maternal risk aversion, initiative and determination.
Robust standard errors are used.
Results
OLS and IV Estimation - Impact of NHIS participation on child
health care practices
Child health care
utilization
NHIS Participation
Other Controls
N
R2
Delivery at health
facility
OLS
IV
(1)
(2)
0.192
0.322
(0.020)*** (0.134)**
YES
YES
2133
2133
0.303
0.288
Anti-malarial
medication taken
OLS
IV
(3)
(4)
0.044
0.334
(0.035) (0.161)**
YES
YES
820
820
0.063
Seek medical treatment
OLS
IV
(5)
(6)
0.225
0.423
(0.037)*** (0.162)***
YES
YES
820
820
0.125
0.092
Source: DHS 2008 and DMHIS 2008 Census.
Note: The estimations include the following controls: child’s gender and age, mother’s age and educational attainment, the number of children born in the
last 5 years, the number of living children older than 5, whether the mother is currently pregnant and whether she is currently working, sex of the
household head, the household’s wealth level, the place of residence, the population level in each region (in millions), and a dummy if the household
belongs to one of the three poorest regions in Ghana, the distance from the district to the nearest doctor and the nearest private hospital (in km), the
number of facilities per 100,000 people in each region. The first 2 estimations (delivery at a health facility) also include as a control if the child was a
twin, but do not include a control for whether the mother is currently pregnant. Instruments chosen for specification 2: the existence of an exception for
children born out of wedlock; the renewal fees for non-exempt adults and the renewal fees for children. Instruments chosen for specification 4 and 6: the
existence of an exception for children born out of wedlock; the existence of non-standard verification method and the renewal fees for non-exempt
adults. Robust standard errors in parentheses, * significant at 10%; ** significant at 5%; *** significant at 1%.
OLS and IV Estimation-Impact of NHIS participation on
maternal health care practices
Maternal
health care
Number of
prenatal care
visits
OLS
IV
(1)
(2)
NHIS
0.698
0.059
Participation
(0.529)
(0.878)
Other Controls
N
R2
YES
1698
0.183
YES
1698
0.175
Told about
pregnancy
complications
OLS
IV
(3)
(4)
0.042
0.416
Urine sample taken
during pregnancy
Blood sample taken
during pregnancy
OLS
(5)
IV
(6)
OLS
(7)
IV
(8)
0.053
0.333
0.085
0.346
(0.023)* (0.136)** (0.015)*** (0.090)*** (0.019)*** (0.092)***
YES
1666
0.078
YES
1666
YES
1670
0.213
YES
1670
0.052
YES
1670
0.165
YES
1670
0.014
Source: DHS 2008 and DMHIS 2008 Census.
Note: The estimations include the following controls: if the child was a twin, mother’s age and educational attainment, the number of children born in
the last 5 years, the number of living children older than 5, whether the mother is currently working, sex of the household head, the household’s
wealth level, the place of residence, the population level in each region (in millions), and a dummy if the household belongs to one of the three
poorest regions in Ghana, the distance from the district to the nearest doctor and the nearest private hospital (in km) as well as the number of
facilities per 100,000 people in each region. Instruments chosen for all specifications except 3 and 4: the existence of an exception for children
born out of wedlock; the renewal fees for non-exempt adults and the renewal fees for children. Instruments chosen for specifications 1 and 2: the
existence of a non-standard verification method; the renewal fees for non-exempt adults and the renewal fees for children.
Robust standard errors in parentheses, * significant at 10%; ** significant at 5%; *** significant at 1%.
OLS and IV Estimation-Impact of NHIS participation on
preventive child health care practices
Vaccination rates
Preventive child
health care
practices
NHIS
Participation
Controls
BCG
DPT/HepB/ Influenza 2
Yellow fever vaccine
Child slept under a
mosquito bed net
OLS
IV
OLS
IV
OLS
IV
OLS
IV
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
0.027
0.153
0.024
0.214
0.015
0.218
0.063
0.371
(0.008) ** (0.074) **
(0.014) * (0.102) ** (0.015) * (0.110) **
(0.020) ** (0.121) **
YES
YES
YES
YES
YES
YES
YES
YES
N
2770
2770
2765
2765
2765
2765
2785
2785
R2
0.052
0.006
0.060
0.001
0.280
0.236
0.077
Source: DHS 2008 and DMHIS 2008 Census.
Note: The estimations include the following controls: child’s gender and age, mother’s age and educational attainment, the number of children born in the
last 5 years, the number of living children older than 5, whether the mother is currently pregnant and whether she is currently working, sex of the household
head, the household’s wealth level, the place of residence, the number of facilities per 100,000 people in each region, the population level in each region
(in millions), and a dummy if the household belongs to one of the three poorest regions in Ghana. Instruments chosen for all specifications except 7 and 8:
the existence of an exception for children born out of wedlock; the renewal fees for non-exempt adults and the renewal fees for children. Instruments
chosen for specifications 7 and 8: the existence of a non-standard verification method; the renewal fees for non-exempt adults and the renewal fees for
children.
Robust standard errors in parentheses, * significant at 10%; ** significant at 5%; *** significant at 1%.
Discussion




The OLS coefficients are biased downwards: parents with frailer
children could decide to enroll more frequently in the NHIS to
provide their children with more curative care.
Adverse selection means underlying heterogeneity in returns to
NHIS participation.
IV estimates are substantially larger than OLS estimates because
IV calculates local average treatment effects (LATE) in contrast to
average treatment effects (ATE).
Membership rules variances do not affect participation decisions
of all parents. Parents that to begin with have positive health
behaviors are probably not discouraged to join the NHIS by
slight variations in membership rules.
Discussion



On the other hand, a subgroup of parents did decide to
participate in the NHIS merely because the membership
rules in their district were less stringent, but would not
have participated otherwise. These are probably parents
that are not so keen on health practices.
The IV estimation measures the impact of participating in
the NHIS on the health care behaviors of these
“compliers”.
If “compliers” have much higher returns to participating in
the NHIS, then our IV estimates will be larger than the
average marginal return to NHIS membership.
Compliers are more likely to …


Have children born out of wedlock  they have a
higher cost of becoming members of the NHIS (single
income household). Hence, reductions to membership
costs and easier access would motivate these parents
much more than the average parent to enroll and take
advantage of the health services provided by the
system.
Have children who are frailer at birth  a frailer baby
would benefit much more from both preventive and
curative care if his/her parents have access to health
insurance, which means they would be much more active
users of these services.
Compliers are more likely to …


Live in a district with lower levels of education and a
higher reliance on traditional medicine  they would
probably be less eager consumers of professional
health services unless prompted by easier access to
insurance services.
Live in districts with poor road networks  with health
insurance parents can afford more expensive health
facilities closer to home. Hence parents who live in
areas with particularly bad transportation networks
might value access to health insurance much more than
the average parent and use health services more
frequently.
Instrument Strength and Validity

The instruments chosen are relevant for all specifications:
They are significant at a 1% level and display a nonimmaterial effect on the probability of being covered by the
NHIS.
 The signs of the coefficients are in line with expectations.


They are strong according to Stock et al. (2002)


The F statistic ensured that the maximal bias of the IV
estimator relative to OLS is no larger than 5%.
The results of over identification tests are consistent with
instrument exogeneity.
First stage estimation - Impact of membership rules on the
probability of being covered by the NHIS
Dependent variable:
Exception for children
born out of wedlock
Renewal fees adults
Covered by NHIS
Sample defined by the availability of the second stage variable
Seek medical
treatment /
medicine
Delivery at
health
facility
0.101
(0.036)***
-0.008
(0.002)***
0.110
(0.023)***
-0.005
(0.001)***
-0.029
(0.008)***
Pregnancy
complications
Urine/
Blood
sample
BCG
DPT/H/I 2/
Yellow
fever
0.119
(0.027)***
-0.006
(0.001)***
-0.030
(0.009)***
0.096
(0.020)***
-0.005
(0.001)***
-0.028
(0.007)***
0.094
(0.020)***
-0.005
(0.001)***
-0.028
(0.007)***
1670
2770
2765
-0.003
(0.001)***
-0.027
(0.007)***
-0.168
(0.025)***
2785
Mosquito
net
-0.175
(0.045)***
820
2133
-0.003
(0.001)***
-0.030
(0.009)***
-0.178
(0.032)***
1666
Partial R2 of excluded
instrument
0.052
0.023
0.031
0.026
0.020
0.020
0.029
F-test for weak
identification
16.21
16.90
20.82
15.13
19.93
19.72
32.60
Hansen J statistic 2 Pvalue
0.105
0.288
0.134
0.312
0.328
0.148
0.566
Renewal fees for
children
Non-standard
verification method
N
Source: DHS 2008 and DMHIS 2008 Census
Conclusion





The main objective of this research paper is to look at the
impact of parental participation in the NHIS on health care
utilization in Ghana.
This is the first attempt to look into this particular outcome.
This study overcomes the difficulties in identifying the effect
of joining a health insurance scheme on child health care
utilization by using an instrumental variables approach.
The instrumental variable estimations show positive and
statistically significant coefficients which are much bigger
than the ones obtained using OLS regressions.
The results are consistent with coefficients biased downwards
in the initial estimations.
Conclusion



This paper calculates local average treatment effects
(LATE) in contrast to average treatment effects (ATE).
Since “compliers” have much higher returns to
participating in the NHIS, IV estimates are larger than
the average marginal return to NHIS membership.
This is not a shortcoming of this paper, because, for
policy evaluation purposes, it might be more relevant to
calculate the average return to NHIS participation for
the group who will be impacted by changes in
membership rules than to calculate the average
marginal return for the whole population.