Content uploaded by Enyew Assefa
Author content
All content in this area was uploaded by Enyew Assefa on Jun 19, 2020
Content may be subject to copyright.
Content uploaded by Enyew Assefa
Author content
All content in this area was uploaded by Enyew Assefa on Jun 19, 2020
Content may be subject to copyright.
Full Terms & Conditions of access and use can be found at
http://www.tandfonline.com/action/journalInformation?journalCode=wwah20
Download by: [Enyew Assefa] Date: 27 October 2017, At: 00:06
Women & Health
ISSN: 0363-0242 (Print) 1541-0331 (Online) Journal homepage: http://www.tandfonline.com/loi/wwah20
Factors related to the use of antenatal care
services in Ethiopia: Application of the zero-
inflated negative binomial model
Enyew Assefa & Mekonnen Tadesse
To cite this article: Enyew Assefa & Mekonnen Tadesse (2017) Factors related to the use of
antenatal care services in Ethiopia: Application of the zero-inflated negative binomial model,
Women & Health, 57:7, 804-821, DOI: 10.1080/03630242.2016.1222325
To link to this article: http://dx.doi.org/10.1080/03630242.2016.1222325
Accepted author version posted online: 11
Aug 2016.
Published online: 11 Aug 2016.
Submit your article to this journal
Article views: 74
View related articles
View Crossmark data
Factors related to the use of antenatal care services in
Ethiopia: Application of the zero-inflated negative
binomial model
Enyew Assefa, MSc
a
and Mekonnen Tadesse, MSc
b
a
Department of Statistics, Dire Dawa University, Dire Dawa, Ethiopia;
b
Department of Statistics, Addis
Ababa University, Addis Ababa, Ethiopia
ABSTRACT
The major causes for poor health in developing countries are
inadequate access and under-use of modern health care ser-
vices. The objective of this study was to identify and examine
factors related to the use of antenatal care services using the
2011 Ethiopia Demographic and Health Survey data. The num-
ber of antenatal care visits during the last pregnancy by
mothers aged 15 to 49 years (n=7,737) was analyzed. More
than 55% of the mothers did not use antenatal care (ANC)
services, while more than 22% of the women used antenatal
care services less than four times. More than half of the women
(52%) who had access to health services had at least four
antenatal care visits. The zero-inflated negative binomial
model was found to be more appropriate for analyzing the
data. Place of residence, age of mothers, woman’s educational
level, employment status, mass media exposure, religion, and
access to health services were significantly associated with the
use of antenatal care services. Accordingly, there should be
progress toward a health-education program that enables
more women to utilize ANC services, with the program target-
ing women in rural areas, uneducated women, and mothers
with higher birth orders through appropriate media.
ARTICLE HISTORY
Received 21 September 2014
Revised 12 June 2016
Accepted 14 June 2016
KEYWORDS
Antenatal care; count model;
zero-inflated negative
binomial
Introduction
Maternal and child health care begins with the immediate health problems of
mothers and children and extends to health throughout life and to the health
of the community (WHO 1976). The WHO estimates that about 536,000
women of reproductive age die each year from pregnancy-related complica-
tions. Nearly all of these deaths (99%) occur in the developing world (WHO
2005).
In Ethiopia, the levels of maternal and infant mortality and morbidity are
among the highest in the world. Maternal mortality rate in the country
continues to be at an unacceptably high level. An estimated 2.9 million
CONTACT Mekonnen Tadesse mekonnentadesse@yahoo.com Department of Statistics, Addis Ababa
University, P.O. Box 1176, Addis Ababa, Ethiopia.
Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/WWAH.
WOMEN & HEALTH
2017, VOL. 57, NO. 7, 804–821
https://doi.org/10.1080/03630242.2016.1222325
© 2017 Taylor & Francis
Downloaded by [Enyew Assefa] at 00:06 27 October 2017
women give birth every year; of these approximately over 25,000 women and
girls die each year and more than 500,000 suffer from serious injuries and
permanent damage to their health, such as obstetric fistulas (FMOH 2005).
The health risks associated with pregnancy and childbirth are far greater in
developing countries than in industrialized ones (UNICEF 2008). The pur-
pose of antenatal care is to screen for signs of illness or other complications
that may occur during pregnancy. For instance, blood-pressure measure-
ments and urine analysis done during antenatal care visits can screen preg-
nant women for hypertensive disorders of pregnancy (including pre-
eclampsia and eclampsia) and to seek medical attention when the condition
appears (WHO 2006).
ANC motivates a pregnant woman facing any pregnancy complication to
seek advice for her problems (Pell et al. 2013). Two indicators often used to
assess care during pregnancy and delivery are antenatal care (ANC) and
skilled birth attendance during delivery (Mangeni et al. 2013). A study in
rural Uttar Pradesh showed that the likelihood of women with high ANC use
delivering in an institution was three times higher than for women with no
ANC use (Fausdar and Abhishek 2006).
According to the WHO’s recommendation, a minimum of four antenatal
visits (at least 20 minutes duration for each) is needed to accomplish the
essential level of ANC (Overbosch et al. 2002). The proportion of pregnant
women in low- and middle-income countries who had at least one antena-
tal care visit increased from less than 55% in the early 1990s to almost 75%
in a decade (WHO 2008). Although this is an improvement, the recom-
mended norm of four antenatal visits is still not accessible to many preg-
nant women worldwide: for example, 55% of those in sub-Saharan Africa
(WHO 2008).
One explanation for poor health outcomes among women and children is
the non-use of modern health care services by a sizable proportion of
women in Ethiopia. According to the 2007 report of the Ministry of
Health of Ethiopia, about 52% of Ethiopian women received ANC, less
than 17% received professionally-assisted delivery care. One of the key
factors that can contribute to reducing maternal and perinatal mortality is
access to antenatal care services. These services permit timely detection of
complications that can arise during pregnancies or childbirth, as well as
ensure that women have access to educational programs, vaccinations,
diagnostic tests, and treatment for infectious diseases (Di Mario 2005;
Gross et al. 2011).
Poor women in remote areas are the least likely to receive adequate health
care (WHO 2014). This is especially true for regions with low numbers of
skilled health workers, such as sub-Saharan Africa and South Asia. While
levels of antenatal care have increased in many parts of the world during the
past decade, only 46% of women in low-income countries benefit from
WOMEN & HEALTH 805
Downloaded by [Enyew Assefa] at 00:06 27 October 2017
skilled care during childbirth (WHO 2014). A study by Regassa (2011)
revealed that women with a high level of education and exposure to mass
media and low parity had higher usage of antenatal and postnatal care.
However, unlike previous findings that revealed low use of antenatal care
in the rural area, this study showed high antenatal care use in the rural
population of Ethiopia.
Dina and Mohamed (2012) used zero-inflated negative binomial regres-
sion and identified that women’s age, women’s education, terminated preg-
nancy, reading newspaper, watching TV, relationship between spouses, place
of residence, region, parental education, and wealth index as significant
factors related to the number of ANC visits. Their study also indicated that
women’s background characteristics had a significant relation to the prob-
ability that women would have zero visits. A study in Ibadan, Nigeria by
Dairo and Owoyokun (2010) revealed significant differences by residence,
religion, and age in relation to use of antenatal care. Findings by Awusi,
Anyanwu, and Okeleke (2009) and Assfaw (2010) in Morocco-Tunisia,
Nigeria, and Ethiopia, respectively, revealed that parity was negatively related
to antenatal care usage. A study conducted by the Ethiopian Society of
Population Studies (2008) revealed that household autonomy in decision-
making was not related to women’s health-seeking behavior. Adolescent
mothers were more likely to drop out of school due to pregnancy, less likely
to earn a salary, and more likely to attend ANC fewer times compared to
adult mothers (Atuyambe et al. 2008). On the other hand, older adolescents
who have had uneventful previous pregnancies and deliveries might see no
reason to attend ANC. In 19 out of 26 developing countries, women who
were 19 years or younger were reportedly less likely than older women to
seek ANC from health professionals (Reynolds, Wong, and Tucker 2006).
Despite the fact that use of antenatal care is essential for further improve-
ment of maternal and child health, little is known about the current magni-
tude of use and factors influencing the use of ANC services in Ethiopia.
Therefore, the factors related to the use of these services need to be clearly
understood. The aim of this study was, therefore, to address this gap using
data from the 2011 Ethiopia Demographic and Health Survey (EDHS;
Central Statistical Agency and ICF International 2012) and also demonstrate
the use of zero-inflated negative binomial regression model to investigate use
of associated with use of antenatal care services.
This study examined whether use of antennal care was related to factors
assumed to have positive or negative associations with the use of antenatal
care services. It was not possible to use the entire spectrum of such factors,
because sufficient data were not available. Insights on the major factors
associated with antenatal care use will help in targeting interventions on
factors that could potentially reduce maternal death.
806 E. ASSEFA AND M. TADESSE
Downloaded by [Enyew Assefa] at 00:06 27 October 2017
Methods
Study participants
This study was based on data from the 2011 EDHS. The 2011 EDHS sample was
selected using a stratified, two-stage cluster design; enumeration areas (EAs)
were the sampling units for the first stage. Administratively, regions in Ethiopia
are divided into zones, and zones are divided into administrative units called
weredas; each wereda is further subdivided into the lowest administrative unit,
called kebele, and each kebele was subdivided into enumeration areas (EAs). An
enumeration area is an area consisting of an average of 150–200 households or
housing units. The sample included 624 randomly selected EAs, 187 in urban
areas and 437 in rural areas. Households comprised the second stage of
sampling. A complete listing of households was carried out in each of the 624
selected EAs from September 2010 through January 2011. A random sample of
17,817 households was selected for the 2011 EDHS, of which 17,018 were
covered during data collection. Of these, 16,702 were successfully interviewed,
yielding a household response rate of 98%. All women aged 15–49 years and all
men aged 15–59 years were eligible for interview. In the interviewed house-
holds, 17,385 eligible women (aged 15–49 years who were either permanent
residents of the selected households or visitors who stayed in the household the
night before the survey) were identified for individual interview; complete
interviews were conducted with 16,515 women, yielding a response rate of
95%. The 2011 EDHS obtained information on ANC coverage from responses
of women who were married in the 5 years preceding the survey. A total of
7,773 women aged 15–49 years who gave birth during the 5-year period before
the survey were included in the analysis, and data related to the most recent
birth were collected.
Data collection
Three questionnaires were used for the 2011 EDHS: the Household
Questionnaire, the Woman’s Questionnaire, and the Man’s Questionnaire.
These questionnaires were adapted from model survey instruments devel-
oped for the MEASURE DHS project and the UNICEF Multiple Indicator
Cluster Survey (MICS) to reflect the population and health issues relevant to
Ethiopia. The data on the age and sex of household members obtained in the
Household Questionnaire were used to identify women and men who were
eligible for the individual interview.
Variables of interest
The response variable of this study, Y, was the number of antenatal care visits
with Y
i
= 0, 1, 2, . . . where ireferred to the ith individual mother. Factors that
WOMEN & HEALTH 807
Downloaded by [Enyew Assefa] at 00:06 27 October 2017
were expected to be associated with the number of ANC visits by women in
Ethiopia considered in this study were woman’s age, woman’s education,
husband’s/partner’s education, woman’s employment status, wealth index of
households, woman’s exposure to mass media, place of residence, region,
religion, birth order, terminated pregnancy, and access to health services.
Working definitions of these socio-economic and demographic explanatory
variables are provided in the appendix.
Data analyses
A bivariate chi-square test was used to assess the relation of potential factors
to number of ANC visits, and all of the 12 variables considered were
significantly associated with the response variable at the 1% (p<.01) level
of significance. Three different count models (Poisson, negative binomial
(NB), and zero inflated negative binomial (ZINB)) were fitted, and the best
fitting model was selected. The likelihood-ratio test was used to determine
the significance of a model, while AIC and BIC values were used to identify
the best fitting model. SPSS (version 20), SAS (version 9.2), and Stata
(version 11) were used for data analysis.
Count regression models with covariates
The Poisson regression model is a technique used to describe count data as a
function of a set of independent variables (Parodi and Bottarelli 2006).
If over-dispersion is observed, the negative binomial regression model that
includes an unobserved specific effect ei(random term or error term) for the
parameter μi(Cameron and Trivedi 1998; Greene 2008; Long 1997; may be
considered. If we assume that exp ei
ðÞis distributed as a gamma distribution
with Eexp ei
ðÞðÞ¼1 and var exp ei
ðÞðÞ¼1=α(Dean, Lawless, and Willmot
1989; Denuit et al. 2007), the gamma mixture of the Poisson distribution
yields the negative binomial distribution of Yi.
Conversely, Greene (1994) introduced the idea of the ZINB regression
model to handle both excess zeros and over-dispersion as a result of
unobserved heterogeneity. ZINB models assessed the binary outcome of
use or non-use of ANC services as a logit model; alternatively, the count
process of the number of ANC visits was used in a NB model (Cameron
and Trivedi 1998; Greene 1994). If the weight parameter is zero, the ZINB
model reduces to a classical NB regression model. When the dispersion
parameter and the weight parameter are zero, the ZINB regression model
reduces to a classical Poisson regression model (Gupta, Gupta, and
Tripathi 1996;Long1997).
808 E. ASSEFA AND M. TADESSE
Downloaded by [Enyew Assefa] at 00:06 27 October 2017
A likelihood-ratio (LR) test was used for the over-dispersion parameter, α,
in the NB specification against the Poisson model specification (Winkelmann
2003). To test the appropriateness of using a zero-inflated model rather than
a standard model, the statistic proposed by Vuong (1989) was used.
Ethical approval
Because the data were obtained from the Central statistical agency of
Ethiopia that implemented the survey and because the study was based on
anonymous public use data with no identifiable information on survey
respondents, no further ethical approval was necessary.
Results
Antenatal care use
While 55.14% of the mothers did not use antenatal care services, 4.37%,
6.81%, 11.65%, and 7.47% of them used antenatal care services only once,
twice, thrice, and four times, respectively (Table 1); 14.62% of the women
used antenatal-care services at least five times during their pregnancy.
Moreover, more than 77% of the women used antenatal health-care services
less than four times or did not use ANC services at all. The maximum
frequency of antenatal care visits recorded for a woman was 20. The sample
mean of the number of ANC visits was 1.82, with a sample variance of 6.634.
The difference between the mean and the variance suggested over-dispersion.
The data had an excess of zero visits, was highly skewed to the right with
excess zeroes (Figure 1), and thus both NB and ZINB would be better models
of the number of antenatal care visits.
Table 1. Frequency and percentage distribution of use of antenatal care services in 2011,
Ethiopia (n= 7,737).
ANC visits Frequency % Cum. Freq Cum.%
0 4,266 55.14 4,266 55.14
1 338 4.37 4,604 59.51
2 527 6.81 5,131 66.32
3 901 11.65 6,032 77.96
4 574 7.42 6,606 85.38
51,131 14.62 7,737 100.00
Total 7,737 100.00
Minimum 0
Maximum 20
Mean 1.82
Variance 6.634
Skewness 1.648
WOMEN & HEALTH 809
Downloaded by [Enyew Assefa] at 00:06 27 October 2017
Model selection
All three models (Poisson, NB, and ZINB) considered (because our response
variable was a count variable) were significant at the 1% level (Table 2).
Because over-dispersion in the data was significant, the NB model was
favored over the Poisson model. The value of the Vuong statistic was
34.23, with p< .0001, indicating that over-dispersion due to many zero
observations and unobserved heterogeneity. Moreover, both the AIC and
BIC values also supported the conclusion that the ZINB model provided the
best fit to the frequency of ANC care use data.
Furthermore, the fitted Poisson regression model predicted that only
32.49% of the women had not received antenatal care services, while the
fitted NB model predicted that 47.29% of the women had not received
antenatal care services, and the fitted ZINB model predicted that 55.12% of
the mothers had not used ANC services, very close to the observed value of
55.14% (Table 2). Therefore, the rest of the results will be described in terms
of the fitted ZINB model.
Figure 1. A bar graph of number of antenatal care visits in Ethiopia.
Table 2. Prediction of zero observation based on the fitted Poisson, negative binomial, and zero-
inflated negative binomial regression models.
Total sample observation considered Percentage of zero observations
Sample observation 7,737 55.14
Poisson regression model 7,737 32.49
NB regression model 7,737 47.29
ZINB regression model 7,737 55.12
810 E. ASSEFA AND M. TADESSE
Downloaded by [Enyew Assefa] at 00:06 27 October 2017
ZINB regression model analysis
The ZINB model generated two separate models. First, a logit model was
generated for the “certain/structural zero,”to assess whether women would
be in this group. Then, a NB model was generated for the counts for those
women who did not have zero visits. Finally, the two models were combined.
The expected number of antenatal care visits for women from rural areas was
decreased by a factor of 0.08 compared to the expected number of antenatal
care visits for women from urban areas, while holding all other variables in
the model constant (Table 3). The model also revealed that region of resi-
dence was significantly related to the number of antenatal care visits. For
example, the expected numbers of antenatal care visits for women from
Amhara, Affar, and Somali regions were 0.9, 0.84, and 0.82 times the
expected number of antenatal care visits by women in Tigray region.
Conversely, the expected number of antenatal care visits for women from
Addis Ababa, Dire Dawa, and Harari were 1.45, 1.24, and 1.09 times the
expected number of antenatal care visits for women from Tigray, respec-
tively, controlling for other variables in the model. The expected number of
antenatal care visits for women aged 20 to 34 and 35 to 49 years were 1.10
and 1.18 times the expected number of antenatal care visits for women in the
age group 15 to 19 years, respectively.
Also, a higher level of education of women was associated with a higher
number of antenatal care visits (Table 3). The expected numbers of antenatal
care visits for women with primary, secondary, and higher education were
1.09, 1.17, and 1.2 times the expected number of antenatal care visits for
women with no education, respectively. Furthermore, the expected numbers
of antenatal care visits for women whose husband had secondary and higher
education were 1.08 and 1.13 times the expected number of antenatal care
visits for women whose husband had no education, respectively.
The results of the fitted model revealed that the expected number of
antenatal care visits for an employed woman was 1.07 times the expected
number of antenatal care visits for an unemployed woman (Table 3). The
expected number of antenatal care visits for women in the rich quintile was
1.18 times the expected number of antenatal care visits for women in the
poor quintile while holding all other variables in the model constant.
Moreover, the expected number of antenatal care visits for women exposed
to mass media was 1.04 times the expected number of antenatal care visits for
women with no mass media exposure. The expected number of antenatal
care visits for women that were followers of Protestant, Catholic, and other
religion decreased by a factor of 13%, 20%, and 37%, respectively, compared
to the expected number of antenatal care visits for Orthodox Christian
women while holding all other variables in the model constant. The expected
number of antenatal care visits for women with access to health services was
WOMEN & HEALTH 811
Downloaded by [Enyew Assefa] at 00:06 27 October 2017
Table 3. Parameter estimates of negative binomial part of the ZINB regression model for
examining the impact of socio-economic, demographic, and health-related factors on number
of antenatal care visits.
Variable and category
Categories Coef. Exp(Coef) Std. Err. zp>|z| [95% CI]
Terminated pregnancy
No (ref.)
Yes 0.010 1.010 0.029 0.330 .739 −0.048 0.067
Age of woman
15–19 (ref.)
20–34 0.097 1.100 0.048 2.000 .046** 0.002 0.191
35–49 0.165 1.180 0.056 2.970 .003*** 0.056 0.274
Residence
Urban (ref.)
Rural −0.084 0.920 0.029 −2.940 .003*** −0.141 −0.028
Region
Tigray (ref.)
Affar −0.176 0.840 0.055 −3.210 .001*** −0.283 −0.067
Amhara −0.105 0.900 0.041 −2.580 .010*** −0.184 −0.025
Oromiya 0.031 1.030 0.039 0.790 .430 −0.046 0.107
Somali −0.196 0.820 0.062 −3.140 .002*** −0.318 −0.073
Benishangul-Gumuz −0.031 0.970 0.047 −0.660 .510 −0.122 0.061
SNNP 0.054 1.060 0.043 1.250 .212 −0.031 0.139
Gambela 0.033 1.030 0.047 0.700 .486 −0.060 0.126
Harari 0.081 1.090 0.044 1.850 .064* −0.005 0.168
Addis Ababa 0.357 1.450 0.039 9.270 .000*** 0.281 0.432
Dire Dawa 0.215 1.240 0.043 4.980 .000*** 0.131 0.300
Education of woman
No education (ref.)
Primary 0.082 1.090 0.023 3.530 .000*** 0.037 0.128
Secondary 0.161 1.170 0.036 4.430 .000*** 0.090 0.232
Higher 0.214 1.240 0.046 4.660 .000*** 0.124 0.303
Education of husband
No education (ref.)
Primary 0.023 1.020 0.024 0.940 .346 −0.025 0.071
Secondary 0.076 1.080 0.034 2.230 .026** 0.0091 0.142
Higher 0.121 1.130 0.041 2.980 .003*** 0.041 0.200
Religion
Orthodox (ref.)
Catholic −0.226 0.800 0.103 −2.200 .028** −0.427 −0.025
Protestant −0.141 0.870 0.034 −4.210 .000*** −0.207 −0.076
Muslim −0.059 0.940 0.025 −2.370 .018** −0.108 −0.010
Others −0.467 0.630 0.192 −2.430 .015** −0.844 −0.090
Birth order
1 (ref.)
2–4 0.000 1.000 0.023 .000 .999 −0.045 0.045
5+ −0.026 0.980 0.033 −0.780 .438 −0.091 0.040
Wealth index
Poor (Ref.)
Middle 0.053 1.060 0.034 1.560 .119 −0.014 0.120
Rich 0.167 1.180 0.030 5.600 .000*** 0.108 0.225
(Continued)
812 E. ASSEFA AND M. TADESSE
Downloaded by [Enyew Assefa] at 00:06 27 October 2017
1.14 times the expected number of antenatal care visits for women without
an access to health services, while holding all other variables in the model
constant.
Zero-inflation part of the model
Place of residence was significantly associated with the probability of being
an excessive zero (large number of zero counts, with proportions in excess of
what is expected under the negative binomial distribution) in the ZINB
regression model (Table 4). The probability of rural women to be in the
excess zero group was 1.39 times higher than the probability of urban women
being in excess zero group, holding all other variables in the model constant.
In other words, the more women lived in rural areas, the more likely that the
zero would be due to structural zero (i.e., the zeros are observed due to some
specific structure in the data, absence of ANC services in the case of the
present data).
The results of the logit mode of the ZINB model suggested significant
regional variations in the probability of being an excessive zero. The like-
lihood that women group in the excess zero group were 2.86, 2.88, 3.40, 3.96,
4.11, and 5.53 times higher among women residing in SNNP, Amhara,
Gambela, Benishangul-Gumuz, Oromiya, and Affar regions, respectively,
compared to women in Tigray region. The model also suggested that being
excess zero was 3.94 times higher among women residing in Dire Dawa and
Harari regions compared to those in Tigray region. The model further
suggested that the likelihood of women in Addis Ababa to be in the excess
zero group was 1.90 times higher than that of women in Tigray. Women in
the Somali region had the highest zero antenatal care visits, 9.40 times higher
than women in Tigray.
Table 3. (Continued).
Variable and category
Categories Coef. Exp(Coef) Std. Err. zp>|z| [95% CI]
Woman is employed
No (ref.)
Yes 0.070 1.070 0.019 3.690 .000*** 0.033 0.108
Exposure to mass media
No (ref.)
Yes 0.042 1.040 0.023 1.820 .068* −0.003 0.086
Access to health services
No (ref.)
Yes 0.129 1.140 0.023 5.600 .000*** 0.084 0.1741
Constant 1.086 2.960 0.083 13.120 0.000*** .924 1.248
***Significant at 1%, **Significant at 5%, *Significant at 10%, ref. = reference category of the variable.
WOMEN & HEALTH 813
Downloaded by [Enyew Assefa] at 00:06 27 October 2017
Table 4. Parameter estimates of zero-inflation part of ZINB regression model for predicting a
latent binary outcome.
Variable and category
Categories Coef. Exp(Coef) Std. Err. zp>|z| [95% CI]
Terminated pregnancy
No (ref.)
Yes 0.330 1.390 0.090 3.690 .000*** .155 0.506
Age of woman
15–19 (ref.)
20–34 −2.101 0.120 0.104 −2.200 .000*** −2.306 −1.897
35–49 −1.551 0.210 0.125 −12.460 .000*** −1.795 −1.307
Residence
Urban (ref.)
Rural 0.327 1.390 0.112 2.910 .004*** 0.107 0.547
Region
Tigray (ref.)
Affar 1.709 5.530 0.169 1.130 .000*** 1.379 2.040
Amhara 1.059 2.880 0.124 8.540 .000*** .816 1.302
Oromiya 1.412 4.110 0.139 1.140 .000*** 1.139 1.685
Somali 2.240 9.400 0.181 12.370 .000*** 1.886 2.595
Benishangul-Gumuz 1.377 3.960 0.152 9.080 .000*** 1.079 1.674
SNNP 1.051 2.860 0.150 6.990 .000*** 0.757 1.346
Gambela 1.223 3.400 0.171 7.140 .000*** 0.887 1.558
Harari 1.370 3.940 0.186 7.390 .000*** 1.007 1.734
Addis Ababa .643 1.900 0.291 2.210 .027** 0.0730 1.212
Dire Dawa 1.371 3.940 0.186 7.350 .000*** 1.005 1.736
Education of woman
No education (ref.)
Primary −0.576 0.560 0.076 −7.590 .000*** −0.725 −0.428
Secondary −1.331 0.260 0.278 −4.790 .000*** −1.876 −0.787
Higher −1.109 0.330 0.437 −2.540 .011** −1.965 −0.253
Education of husband
No education (ref.)
Primary −0.420 0.660 0.070 −5.990 .000*** −0.557 −0.282
Secondary −0.282 0.750 0.143 −1.970 .049** −0.563 −0.001
Higher −0.628 0.530 0.209 −3.000 .003*** −1.038 −0.218
Religion
Orthodox (ref.)
Catholic 0.164 1.180 0.312 .530 .599 −0.447 0.774
Protestant 0.372 1.450 0.117 3.170 .002*** 0.142 0.602
Muslim −0.169 0.850 0.100 −1.690 .091* −0.364 0.027
Others 0.024 1.030 0.379 .060 .949 −0.719 0.767
Birth order
1 (ref.)
2–4 0.393 1.480 0.095 4.160 .000*** .208 0.579
5+ 0.494 1.640 0.108 4.590 .000*** 0.283 0.705
Wealth index
Poor (Ref.)
Middle −0.420 0.660 0.085 −4.930 .000*** −0.587 −0.253
Rich −0.511 0.600 0.083 −6.130 .000*** −0.675 −0.348
Woman is employed
No (ref.)
Yes −0.128 0.880 0.066 −1.940 .053* −0.258 0.0015
(Continued)
814 E. ASSEFA AND M. TADESSE
Downloaded by [Enyew Assefa] at 00:06 27 October 2017
The probability that a woman was in the excess zero group decreased by
88% and 79% for women in the age groups 20–34 and 35–49 years, respec-
tively, compared with women in the age group 15–19 years. In other words,
older women were less likely to be in the constant zero count group (i.e.,
from a distribution that generates only zero counts).
The probability that a woman was in the excess zeros group decreased by
44% when the woman had a primary education in comparison with a woman
with no education. Additionally, the probability of having a woman in the
excess zero antenatal care group decreased by 74% and 67% when the woman
had completed secondary school and higher education, respectively, in com-
parison with a woman with no education. In other words, the higher the
education level of women, the less likely that the zero would be due to
structural zero. The probability that a woman would be in the excess zeros
group decreased by 12% when the woman was employed in comparison with
an unemployed woman. In other words, for an employed woman, it is less
likely that the zero would be due to certain/structural zero. The probability
that a woman would be in the excess zero antenatal care group decreased by
34% and 40% when the woman was in middle quintile and rich quintile
groups, respectively compared with women in the poor quintile. As com-
pared to the reference category (no education), women whose husbands had
a primary, secondary, or higher education were 34%, 25%, and 47%, respec-
tively, less likely to be in the excess zero antenatal care group. In other words,
the higher the education level of husbands/partners, the less likely that the
zero would be due to structural zero.
The probability that women were in the zero group increased by 39% when
the women had terminated pregnancy compared to women with no terminated
pregnancy. In other words, the more exposure women had to terminated
pregnancy, the more likely that the zero would be due to certain/structural
zero. Further, the likelihood that women were in the excess zero antenatal care
Table 4. (Continued).
Variable and category
Categories Coef. Exp(Coef) Std. Err. zp>|z| [95% CI]
Exposure to mass media
No (ref.)
Yes −0.304 0.740 0.068 −4.470 .000*** −0.437 −0.170
Access to health services
No (ref.)
Yes −1.830 0.160 0.092 −19.980 .000*** −2.010 −1.651
Constant 0.749 2.120 0.274 2.740 .006*** .212 1.286
ln αðÞ −4.668 0.010 0.624 −7.480 .000*** −5.890 −3.445
α0.009 1.009 0.006 0.003 0.0319
***Significant at 1%, **Significant at 5%, *Significant at 10%, ref. = reference category of the variable.
WOMEN & HEALTH 815
Downloaded by [Enyew Assefa] at 00:06 27 October 2017
group decreased by 26% when women had exposure to mass media compared to
women with no media exposure before the survey. In other words, the more
women had exposure to mass media, the less likely that they would be among
those that have never used ANC services. The probability that a woman was in
an excess zero group was 1.64 and 1.48 times higher when the number of births
she had had was above five and between two and four, respectively, compared
with a woman having only one child. In other words, the more children women
had, the more likely that the zero would be due to certain/structural zero. As
compared to the reference category (Orthodox Christians), Muslim women
were 15% less likely to be in an excess zero antenatal care group, while the
probability that a woman will be in an excess zero group was 1.45 times higher if
the woman was Protestant compared to an Orthodox Christianity follower
woman. The probability that a woman was in the excess zero visits group
decreased by 84% when the woman had access to health services compared to
a woman who had no access to health services. In other words, the more women
had access to health services, the less likely that the zero would be due to certain/
structural zero.
Discussion
The ZINB regression model was the most appropriate model for identifying
the factors associated with the number of antenatal care visits by pregnant
women. Use of the ZINB model, which assumes involvement of more than
just one source of over-dispersion, provided the smallest level error. The
ZINB model was consistent for factor identification in the extent of use of
antenatal care as well as for the number of antenatal care visits.
The findings revealed that the expected number of antenatal care visits for
women from rural areas was less than the expected number of antenatal care
visits for women from urban areas. These findings agree with the findings of
Dairo and Owoyokun (2010) and Regassa (2011) that women living in urban
areas had a higher likelihood of receiving check-ups during pregnancy than
the women living in rural areas. This study showed a regional discrepancy
partly with respect to the number of antenatal health-care visits. This was
similar to the finding of the Ethiopian Society of Population Studies (2008),
which revealed that antenatal care use was not the same in all regions.
This study also revealed that the frequency of use of antenatal care services
was higher for mothers aged 35–49 years (older mothers) than mothers in the
other age groups. In 19 out of 26 developing countries, women who were 19
years or younger were reportedly less likely than older women to seek ANC from
health professionals (Reynold et al. 2006), which is consistent with the finding of
this study. Conversely, this finding contradicts the findings of previous studies
conducted by Sharma, Sawangdee, and Sirirassamee (2007), which revealed that
women over the age of 35 years were less likely to use prenatal care.
816 E. ASSEFA AND M. TADESSE
Downloaded by [Enyew Assefa] at 00:06 27 October 2017
In this study, mothers’and their partners’education were important
factors related to the usage of antenatal care services. A study by Regassa
(2011) and Ononokpono et al. (2013) also revealed that education was
significantly related to ANC visits. Consistent with this study’s results, the
Ethiopian Society of Population Studies (2008) found that use of antenatal
health-care services was higher among educated mothers than among non-
educated mothers. A study in Nepal also found that women’s education was
strongly associated with greater use of maternal health care (Furuta and
Salway 2006). Female education improves wealth, reduces gender disparity,
and empowers women (World Bank 2008).
The results of this study showed that employed mothers were more likely
to use antenatal health-care services than unemployed mothers. This finding
is consistent with the finding by Tawiah (2011) and Sharma, Sawangdee, and
Sirirassamee (2007). In addition, this study revealed that mothers from poor
households were less likely to use antenatal care services than mothers from
rich households. This finding is consistent with the study by Regassa (2011),
who showed that women in rich quintile households had better access to
antenatal care services than women in poor quintile households. Consistent
with our finding, the study by Acharya et al. (2015) reported a positive
relation of mass media to the use of antenatal care services in Nepal.
Consistent with our finding, a significant difference was found in the use
of antenatal care services among different religious groups in Nigeria
(Babalola and Fatusi 2009). The findings in this study are consistent with
other studies (Shegaw Mulu Tarekegn, Lieberman, and Giedraitis 2014),
which revealed that socio-demographic and accessibility-related factors are
major factors related to service use. Our findings are also consistent with the
study that indicated that women’s background characteristics were signifi-
cantly related to the probability that women would have zero visits (Dina and
Mohamed 2012).
Despite the contribution of the study to the literature on maternal health
care, this study had some limitations. First, it was a cross-sectional study in
which temporal relations between independent variables and use of antenatal
care could not be assessed. Moreover, most of the variables were measured at
the time of the survey, rather than at the time of birth of the child, raising the
possibility of recall bias. Additionally, data on the distance to the nearest
health facility where ANC was sought and on the price of ANC visit were not
available, introducing the possibility of residual confounding.
Conclusion
Greater age, living in an urban area, higher maternal education, higher
paternal education, being wealthy, exposure to mass media, being orthodox
Christian, and having access to health services were all positively related to
WOMEN & HEALTH 817
Downloaded by [Enyew Assefa] at 00:06 27 October 2017
the number of ANC visits. Moreover, having had a terminated pregnancy,
greater maternal age, being employed, being in the wealthiest quintile of the
household wealth index, having had fewer births, higher paternal education,
higher maternal education, exposure to mass media, living in an urban area,
and having access to health services were negatively associated with the
probability that a pregnant woman would have zero ANC visits.
The results of this study underscore the socio-economic and demographic
factors associated with the number of ANC service visits. They support
targeting non-educated or less educated and socioeconomically disadvan-
taged women as well as women in rural areas for using ANC services.
Recommendation
Our findings have important implications for the design of health policy
concerning maternal health in Ethiopia. Policies that increase the opportu-
nity for improving the level of education of women would likely be effective
in increasing the frequency of use of ANC services. Consequently, govern-
ment policies should target women in rural areas, uneducated, and econom-
ically disadvantaged women so that the frequency of ANC visits can be
increased.
References
Acharya, D., V. Khanal, J. K. Singh, M. Adhikari, and S. Gautam. 2015. Impact of mass media
on the utilization of antenatal care services among women of rural community in Nepal.
BMC Research Notes 8:345. doi:10.1186/s13104-015-1312-8.
Assfaw, Y. T. (2010). Determinants of antenatal care, institutional delivery and skilled birth
attendant utilization in Samre Saharti district, Tigray, Ethiopia. Thesis, Umeå University,
Umeå, Sweden.
Atuyambe, L., F. Mirembe, N. M. Tumwesigye, J. Annika, E. K. Kirumira, and E. Faxelid.
2008. Adolescent and adult first time mothers’health seeking practices during pregnancy
and early motherhood in Wakiso district, central Uganda. Reproductive Health 5:13.
doi:10.1186/1742-4755-5-13.
Awusi, V. O., E. B. Anyanwu, and V. Okeleke. 2009. Determinants of antenatal care services
utilization in Emevor Village, Nigeria. Benin Journal of Postgraduate Medicine 11:21–22.
Babalola, S., and A. Fatusi. 2009. Determinants of use of maternal health services in Nigeria—
Looking beyond individual and household factors. BMC Pregnancy and Childbirth 9:43.
doi:10.1186/1471-2393-9-43.
Cameron, C., and P. K. Trivedi. 1998. Regression analysis of count data. Cambridge, UK:
Cambridge University press.
Central Statistical Agency and ICF International. 2012. Ethiopia demographic and health
survey 2011. Addis Ababa, Ethiopia, and Calverton, MD: Central Statistical Agency and
ICF International.
Dairo, M. D., and K. E. Owoyokun. 2010. Factors affecting the utilization of antenatal care
services in Ibadan, Nigeria. Benin Journal of Postgraduate Medicine 12:1.
818 E. ASSEFA AND M. TADESSE
Downloaded by [Enyew Assefa] at 00:06 27 October 2017
Dean, C., J. F. Lawless, and G. E. Willmot. 1989. A mixed Poisson-inverse Gaussian regres-
sion model. The Canadian Journal of Statistics 17:171–81. doi:10.2307/3314846.
Denuit, M., X. Marechal, S. Pitrebois, and J. F. Walhin. 2007. Actuarial modeling of claim
counts: Risk classification, credibility and bonus-malus systems. London, UK: John Wiley
and Sons.
Di Mario, S. 2005. What is the effectiveness of antenatal care? (Supplement). Copenhagen,
Denmark: World Health Organization Regional Office for Europe (Health 21Evidence
Network report). http://www.euro.who.int/Document/E87997.pdf.
Dina, M., and A. Mohamed. 2012. Impact of women’s empowerment and other indicators on
antenatal health care utilization in Egypt. Cairo, Egypt: Cairo University.
Ethiopian Society of Population Studies. 2008. Maternal healthcare seeking behavior in
Ethiopia: Findings from EDHS 2005. Addis Ababa, Ethiopia: Ethiopian Society of
Population Studies.
Fausdar, R., and S. Abhishek. 2006. Is ANC effective in improving maternal health in rural
Uttar Pradesh? Evidence from district level household survey, International Institute for
Population Science, Deemed University, Mumba, India. Journal of Biosocial Science
38:433–48. doi:10.1017/S0021932005026453.
Federal Ministry of Health, Planning and Programming Department. 2005. Health Sector
Strategic Plan (HSDP-III) 2005/6-2009/10. Addis Ababa, Ethiopa: Federal Ministry of
Health.
Furuta, M., and S. Salway. 2006. Women’s position within the household as a determinant of
maternal health care use in Nepal. International Family Planning Perspectives 32 (1):17–27.
doi:10.1363/3201706.
Greene, W. 2008. Functional forms for the negative binomial model for count data.
Economics Letters 99:585–90. doi:10.1016/j.econlet.2007.10.015.
Greene, W. H. 1994. Accounting for excess zeroes and sample selection in the Poisson and
negative binomial regression models. New York, NY: New York University, Leonard N.
Stern School of Business, and Department of Economics.
Gross, K., J. A. Schellenberg, F. Kessy, C. Pfeiffer, and B. Obrist. 2011. Antenatal care in practice:
An exploratory study in antenatal care clinics in the Kilombero Valley, south-eastern Tanzania.
BioMed Central Pregnancy and Childbirth 11:36. doi:10.1186/1471-2393-11-36.
Gupta, P. L., R. C. Gupta, and R. C. Tripathi. 1996. Analysis of zero-adjusted count data.
Computational Statistics and Data Analysis 23:207–18. doi:10.1016/S0167-9473(96)00032-1.
Long, S. 1997. Regression models for categorical and limited dependent variables. Thousand
Oaks, CA: Sage Publication.
Mangeni, J. N., A. Mwangi, S. Mbugua, and V. Mukthar. 2013. Male involvement in maternal
health care as a determinant of utilization of skilled birth attendants in Kenya. DHS
Working Papers 2013 No. 93. Calverton, MD: IDF International.
Ononokpono, D. N., C. O. Odimegwu, E. Imasiku, and S. Adedini. 2013. Contextual
determinants of maternal health care service utilization in Nigeria. Women & Health 53
(7):647–68. doi:10.1080/03630242.2013.826319.
Overbosch, G., J. Nsowah-nuamah, B. Vanden, and L. Damnyag. 2002. Determinants of
antenatal care use in Ghana. Staff Working Paper 02-13. Amsterdam, The Netherlands:
Center for World Food Studies.
Parodi, S., and E. Bottarelli. 2006. Poisson regression model in epidemiology—An introduc-
tion. Annali della Facoltà di Medicina Veterinaria XXVI (2006):25–44.
Pell, C., A. Meñaca, F. Were, N. A. Afrah, S. Chatio, L. Manda-Taylor, M. J. Hamel et al.
2013. Factors affecting antenatal care attendance: Results from qualitative studies in
Ghana, Kenya and Malawi. PLoS ONE 8 (1):e53747. doi:10.1371/journal.pone.0053747.
WOMEN & HEALTH 819
Downloaded by [Enyew Assefa] at 00:06 27 October 2017
Regassa, N. 2011. Antenatal and postnatal care service utilization in southern Ethiopia: A
population-based study. African Health Sciences 11 (3):390–97.
Reynolds, H. W., E. L. Wong, and H. Tucker. 2006. Adolescents use of maternal and child
health services in developing countries. International Family Planning Perspectives 32
(1):6–16. doi:10.1363/3200606.
Sharma, S. K., Y. Sawangdee, and B. Sirirassamee. 2007. Access to health: Women’s status and
utilization of maternal health services in Nepal. Journal of Biosocial Science 39:671–92.
doi:10.1017/S0021932007001952.
Tarekegn, S. M., L. S. Lieberman, and V. Giedraitis. 2014. Determinants of maternal health
service utilization in Ethiopia: Analysis of the 2011 Ethiopian demographic and health
survey. BMC Pregnancy and Childbirth 14:161. doi:10.1186/1471-2393-14-161.
Tawiah, E. O. 2011. Maternal health care in five sub-Saharan African countries, Regional
Institute for Population Studies–University of Ghana, Legon. African Population Studies
25:1. doi:10.11564/25-1-264.
UNICEF. 2008. The State of the World’s Children 2009. Maternal and new born health.
http://www.unicef.org/sowc09/ (accessed August 17, 2016).
Vuong, Q. H. 1989. Likelihood ratio tests for model selection and non-nested hypotheses.
Econometrica 57:307–33. doi:10.2307/1912557.
WHO. 1976. New trends and approaches in the delivery of MCH services. Technical Report
Series 600 (8):17–18.
WHO. 2005. Maternal mortality fact sheets. www.who.org (accessed June 25, 2009).
WHO. 2006. Provision of focused antenatal care for pregnant women. Geneva, Switzerland:
World Health Organization.
WHO. 2008. Fact sheet. WHO/MPS/08.15. http://www.who.int/maternal_child_adolescent/
events/2008/mdg5/mdg5_vs3.pdf
WHO. 2014. Maternal mortality fact sheet No. 348. (updated May 2014). http://www.who.int/
mediacentre/factsheets/fs348/en/ (accessed on November 29, 2014).
Winkelmann, R. 2003. Econometric analysis of count data, 4th ed. Berlin, Germany: Springer.
World Bank. 2008. Girls’Education in the 21st century: Gender equality, empowerment and
economic growth. http://siteresources.worldbank.org/EDUCATION/Resources/278200-
1099079877269/547664-1099080014368/DID_Girls_edu.pdf (accessed August 17, 2016).
820 E. ASSEFA AND M. TADESSE
Downloaded by [Enyew Assefa] at 00:06 27 October 2017
Appendix
Working definitions of the socio-economic and demographic explanatory variables
Description and name Categories
Woman’s age (AGE) 0 = 15–19 (younger mothers)
1=20–34 (adult mothers)
2=35–49 (older mothers)
Woman’s education (WOMEDUC) 0 = No-education
1 = Primary
2 = Secondary
3 = Higher
Woman’s employment status (WOMEMP) 0 = Non-employed
1 = Employed
Husband’s/partner’s education (HUSEDUC) 0 = No-education
1 = Primary
2 = Secondary
3 = Higher
Wealth index of the household (WLTHINDEX) 0 = Poor
1 = Middle
2 = Rich
Mass media exposure of woman’s (MMEXP) 0 = No
1 = Yes
Place of residence (RESIDENCE) 1 = Urban
2 = Rural
Region (REGION) 1 = Tigray
2 = Affar
3 = Amhara
4 = Oromiya
5 = Somali
6 = Benishangul-Gumuz
7 = SNNP
8 = Gambela
9 = Harari
10 = Addis Ababa
11 = Dire Dawa
Religion (RELIGION) 1 = Orthodox
2 = Catholic
3 = Protestant
4 = Muslim
5 = Others
Birth order (BIRORD) 0 = 1
1=2–4
2=5+
Terminated pregnancy (TP) 0 = No
1 = Yes
Access to health services (AHS) 0 = No
1 = Yes
WOMEN & HEALTH 821
Downloaded by [Enyew Assefa] at 00:06 27 October 2017
