This study is based on the NARHS, which involved a nationally representative population sample, and marks a departure from most of the previously reported studies on maternal health services utilization in Nigeria in terms of its national coverage. In addition, unlike most previous studies, we covered the three dimensions of pregnancy-related care - antenatal, delivery and postnatal services.
Our results show that the level of utilization of orthodox health care facilities for maternal care among women in Nigeria is low. Indeed utilization of maternal health care services is lower in Nigeria than in many countries in sub-Saharan Africa. For example, whereas we found that 60.3% of Nigerian mothers utilized antenatal care services during their last birth, the comparative figures were 88.0% for Benin (2006 DHS), 72.8% for Burkina Faso (2007 DHS), 83.4% for Cameroon (2004 DHS), and 91.9% for Ghana (2003 DHS) .
Similarly, the indicators of skilled assistance during delivery and use of postnatal care are considerably lower in Nigeria than in most African countries. A recent UNICEF report  shows that regarding skilled assisted delivery, only Burundi, Chad, Eritrea, Ethiopia, Niger and Somalia performed more poorly than Nigeria in sub-Saharan Africa.
The finding that utilization of antenatal services is higher than use of skilled assistance during delivery is consistent with the results of previous studies conducted in Nigeria [29, 37] and elsewhere [38–40]. One of the reasons that have often been advanced for the lower coverage of skilled and institutional delivery compared to antenatal care coverage is the unpredictable nature of the onset of labor in the face of difficulty in accessing health facilities in resource-poor environments. Many rural communities in sub-Saharan Africa are examples of such environments, with the characteristic poor road networks, limited transportation means and underserved population in terms of health facilities. Our study would support such an explanation considering that the average number of residents per PHC is a more significant predictor of use of skilled assistance for delivery than of use of antenatal care.
The poor staffing of the health facilities, particularly the primary health care facilities, which makes it difficult to guarantee 24-hour availability of services had also been reported as a factor that discourages women in Nigeria, even when they had received antenatal care services, to seek medical services when labor commences . The role of traditional and religious beliefs as well as the perception of women with regards to comparative efficacy of the medical versus traditional birth attendants may also be contributory to failure to have skilled attendants at birth. As Addai  pointed out, modern (medical) and indigenous maternal health care services coexist in most African communities, particularly in rural areas, and women may have to choose between the two options. Some previous studies had reported that many Nigerian women, particularly those in rural areas, rate the services of the traditional birth attendants (TBAs) as being of higher quality than that of medical healthcare practitioners, particularly with regards to interpersonal communications and relationships [41, 43]. TBAs have been reported to be more considerate and to provide more compassionate care. Women in rural Guatemala have similarly been reported as being less likely to deliver in medical settings because of lack of social support provided by health-care professionals compared with traditional midwives . Furthermore, Falkingham  reported that despite the fact that medical services were accessible and free of charge, women in Tajikistan prefer to deliver at home because they perceive available medical services to be of low quality and unsafe. Economic reason also ranks strongly in the preference of some Nigerian women for TBAs as their services have been reported to be more affordable. Additionally, TBAs may offer a more convenient user-charges system that allows payment to be spread over a period of time or even to be made in kind .
Our finding regarding the significant positive association between education and each of the three indicators of maternal services use agrees with previous reports [8, 12, 45, 46]. Education serves as a proxy for information, cognitive skills, and values; education exerts effect on health-seeking behavior through a number of pathways . These pathways include higher level of health awareness and greater knowledge of available health services among educated women, improved ability of educated women to afford the cost of medical health care, and their enhanced level of autonomy that results in improved ability and freedom to make health-related decisions, including choice of maternal services to use [10, 12, 48]. Educated mothers are more likely to take advantage of public health-care services than other women [49, 50]. Education may also impart feelings of self-worth and confidence as well as reduce the power differential between service providers and clients, thereby reducing the reluctance to seek care [51, 52].
The absence of a statistically significant association between the child's rank of birth and maternal services utilization among Nigerian women is surprising. Previous studies have found a strong negative association between parity and maternal services utilization [38, 46, 53, 54].
Our finding with regards to the association between ethnicity and service utilization is an interesting one. Whereas ethnicity seems to make no significant difference for use of antenatal care, it does for use of skilled assistance and postnatal care. For these two indicators of use, the Fulanis, and the Kanuris (in the north) are not statistically different from the Hausas (in the north). In contrast, the level of service utilization was significantly higher among the Igbos (in the south) and the "minority" tribes compared to the Hausas. The pattern is consistent with the general picture of wide regional disparity in health status in Nigeria's diverse and multi-ethnic setting as has been reflected, for example, in the NDHS . Perhaps more than other factors, this result reflects the influence of culture. An analysis of the social context of childbirth among the Hausas of Northern Nigeria, for example, has highlighted the strong influence of cultural beliefs and practices on childbirth and related fertility-related behaviors, and their significant contribution to the maternal morbidity and mortality picture . In addition to the fact that a high proportion of teenage girls are married out to much older men, sometimes as early as 9 or 10 years of age, based on religious/cultural beliefs, cultural norms restrict women from readily seeking health-related assistance in pregnancy and childbirth. As Wall  noted, "Kunya, or 'shame' plays an extremely important role in Hausa childbirth, particularly in the first pregnancy. The newly pregnant girl should not draw attention to her gravid state, and all mention of the pregnancy should be avoided in conversation and action. This social pressure to remain 'modest' may well prevent her from asking questions about childbirth, and creates a major barrier to her seeking skilled assistance for delivering in hospital. As Wall further note, the pregnant girl's "mother, other relatives, and a local midwife usually stay with her during labor, but her kunya and her fear may be so great that she does not say anything until labor is well advanced." (p. 353). If there is nobody immediately available, it is unlikely that the girl in labor will send for someone, as "kunya" will prevent her from saying anything. Moreover in the cultural context of the Hausas, delivering her first child alone - unattended to by anyone - is viewed with pride.
Whereas some previous Nigerian studies had reported a significant relationship between age and maternal services utilization [56, 57], others had shown no such difference [28, 58]. We found no significant relationship between age and use of skilled assistance. For the other two indicators, we found a curvilinear relationship, such that women in the middle childbearing ages are more likely to used maternal services compared to their peers in the early or late childbearing ages. This finding agrees with the report of Obermeyer and Potter  and Gage .
Expectedly, fertility-related attitudes, which are reflected in our study by attitudes towards family planning and notion about the ideal family size, have significant relationship with maternal health services utilization. Favorable attitudes towards family planning and a clear notion about what constitutes an ideal smaller family size reflect less conservative behavior and more openness to modern health-related concepts and services.
At the household level, we found socio-economic status to be a significant predictor: for each of the three indicators, use of maternal services increases steadily with socio-economic status. Studies elsewhere have also documented positive relationship between economic status and early antenatal care use [16, 58, 59], delivery in medical settings [13, 14, 16], and utilization of postnatal services [60, 61].
A major focus of this study is to go beyond individual and household factors and investigate the effects of community and state level factors on maternal care services utilization. At the community level, urban residence was consistently associated with increased odds of service utilization. This finding is in consonance with previous studies which have reported a significantly higher use of services in urban compared to rural areas in Nigeria [7, 22, 43, 62] and elsewhere [13, 16, 63, 64]. The other two community factors assessed - community media saturation and the prevalence of the small family norm in the community - are expectedly significant predictors of service utilization. Note however that the small family norm was not significant for postnatal care. The reason for this finding is not clear.
At the state level, we found that the ratio of PHC to the population was a significant predictor for use of antenatal care and skilled attendance at delivery, but not for postnatal care. The relationships are such that the larger the number of residents to a PHC the less the odds of using the services. This negative relationship is understandable since the more people a PHC serves, the more likely it is that access to the services would be difficult and the quality of services received poor.
Finally, we found that the random effects of the state of residence on each of the three indicators of maternal care service utilization are significant. Substantively, this finding shows that unmeasured factors operating at the state of residence level play a significant role in determining utilization of maternal health services beyond the influence of individual, household and community factors.
The findings from this study have implications for evidence-based programming. Collectively, the findings highlight the need for programs to adopt a multi-level approach and address the factors affecting maternal health services utilization at various levels - individual, household and community. More specifically, programs need to explore effective ways of increasing service utilization among lowly educated and poor women who are the least likely to use maternal health services. Evidence from elsewhere have shown that access to services and cost are serious barriers to service utilization among the poor [65, 66]. As Fotso et al. surmised, it is not enough to increase the availability of services, making such services affordable to the poor is a necessity .
The strong role of community-level normative factors point to the need for interventions that target social norms. For example, using the media to disseminate consistent messages promoting the use of maternal health services could help to increase discussion of these issues within the community, a relevant step towards changing prevailing negative norms. Also relevant are efforts that involve community leaders and other key persons as agents of change. The findings that the prevalence of the small family norm in the community and personal fertility-related attitudes are associated with differences in service utilization suggest that promoting the use of family planning may ultimately help to foster the utilization of other maternal health services. In other words, programs that seek to promote the small family norm and change attitudes that are unfavorable towards family planning are relevant.
The significant state-level random effects that our study found demonstrate the need to contextualize efforts aimed at promoting maternal service utilization. There are obviously some unmeasured factors at the state level that predict service utilization. An effective strategy should be state-specific and seek to identify and address state-level factors that affect service utilization.
This study has some limitations that should be noted. First, the NARHS, the source of data for our study, was based on the self-report of respondents, and provided no validation of obtained information with any objective source such as health facility cards. The validity of self-reported behavior constitutes a concern in the literature, but it is logical to assume that biases are less likely in pregnancy-related events as compared to sensitive issues such as sexual behavior and drug abuse. Social desirability bias may also be an issue in cases that women feel they need to respond in a way expected of them. The comparability of our results with that of NDHS with regards to antenatal care use, for example, suggests that such bias is not likely to have affected our findings in any significant way.
Second, some known predictors of service utilization are obviously missing from our analyses. For example, availability of maternal health services within the immediate locality of respondents and the distance of respondents to such health services could have contributed to the picture of utilization pattern. Unfortunately these variables were not available in the NARHS. Although we included the state-level measure of PHC density (the number of residents to a PHC) in our analyses, the extent to which this variable is a good proxy for individual-level variable is uncertain.
Third, the study relied on cross-sectional data with the attendant potential selectivity and endogeneity bias. There is a possibility that the relationships that we found in our study are due to the influence of unmeasured individual and community-level variables that are associated with both the dependent and independent variables in our estimated models. It is also possible that the observed relationships reflect reverse causation or are due to measurement error. There are analytic methods to adjust for endogeneity bias in cross-sectional data (e.g., propensity score matching, bivariate probit regression, multivariate probit regression and instrumental variable regression) . Nonetheless, adjusting for endogeneity is beyond the scope of this paper.