Health & Place 29 (2014) 67–78

Contents lists available at ScienceDirect

Health & Place journal homepage: www.elsevier.com/locate/healthplace

Gender inequality and the use of maternal healthcare services in rural sub-Saharan Africa Vissého Adjiwanou a,n, Thomas LeGrand b a b

Centre for Actuarial Research (CARe), University of Cape Town, South Africa Département de Démographie, Université de Montréal, Canada

art ic l e i nf o

a b s t r a c t

Available online 1 July 2014

In this study, we measure gender inequality both at individual level by women's household decision-making and at contextual level by permissive gender norms associated with tolerance of violence against women and assess their impact on maternal healthcare services utilisation in rural Africa. We apply multilevel structural equation modelling to Demographic and Health Survey (DHS) data from Ghana, Kenya, Tanzania and Uganda to gain better measure and effect of the gender norms construct. The results show that women in Ghana and Uganda, who live in areas where gender norms are relatively tolerant of violence against women, are less likely to use skilled birth attendants and timely antenatal care. In Tanzania, women who live in this type of environment are less likely to attend four or more antenatal visits. In contrast, the effects of a woman's decision-making authority on maternal health service use are less pronounced in the same countries. & 2014 Elsevier Ltd. All rights reserved.

Keywords: Sub-Saharan Africa Maternal health service use Women's autonomy Sociocultural and gender norms Multilevel and structural equation modelling

1. Introduction Reducing infant mortality by two-thirds and maternal mortality by three-quarters – Millennium Development Goals 4 and 5, respectively – may not be reached in many low income countries partly because of under-utilisation of maternal healthcare services (Bhatia and Cleland, 1995; Koblinsky, 1995; Campbell and Graham, 2006; Islam and Yoshida, 2009). Sub-Saharan Africa carries the heaviest burden of maternal and infant mortality: about a quarter of a million women die each year from pregnancy complications and four million children die before they turn five years (The Partnership for Maternal, Newborn and Child Health 2006). Moreover, the WHO (2009) estimates that, during the period 2000–2008, one in two women who gave birth did so without qualified personnel. In other words, of the 30 million women who get pregnant each year in the sub-region, roughly 18 million give birth at home without the assistance of a health professional (Lawn et al., 2005). Recently, the research agenda on maternal and infant health has shifted to consider women's autonomy or empowerment, defined as “women’s ability to make decisions which affect outcomes of importance to themselves and their families” (Malhotra and Schuler, 2005, p.5), as a fundamental factor for understanding their reproductive behaviour in developing countries (Bloom et al., 2001; Stephenson et al., 2012; Singh et al., 2013). However, most previous studies have not considered the broad context of gender inequality and the social norms under which women live, which

n

Corresponding author. Tel.: þ 27 21 650 5936. E-mail address: [email protected] (V. Adjiwanou).

http://dx.doi.org/10.1016/j.healthplace.2014.06.001 1353-8292/& 2014 Elsevier Ltd. All rights reserved.

define and structure their status (Kritz and Makinwa-Adebusoye, 1999; Dixon-Mueller and Germain, 2000; Blanc, 2001; Desai and Johnson, 2005). Many of these studies dealt with Asian countries and focused on family planning issues, limiting their value for understanding behaviour and maternal healthcare use in the African context (Fotso et al., 2009; Corroon et al., 2013). This paper examines the influence of the gender inequality on the use of antenatal care and skilled birth attendance in rural subSaharan Africa, where gender norms discouraging women's autonomy persist, and where the use of maternal health services remains low (Beninguisse et al., 2005). Conceptually, it assesses the gender inequality both in terms of individual women's decision-making power within the household, and of gender norms regarding intimate partner violence in the community. Methodologically, the study combines the structural equations modelling approach to multilevel modelling to derive an appropriate measure of the gender inequality and to assess directly and indirectly its “contextual effect” (Sampson et al., 2002; Raudenbush, 2003; Lüdtke et al., 2007). Models are estimated on data from four Demographic and Health Surveys, the results demonstrating the value of applying this approach to other health behaviour outcomes in sub-Saharan Africa.

2. Background Gender inequalities describe the disparities in roles, characteristics and behaviours between men and women that are grounded in the expectations and social norms prevailing in local society (Blanc, 2001). These norms are strengthened by sanctions that may occur

68

V. Adjiwanou, T. LeGrand / Health & Place 29 (2014) 67–78

when an individual refuses to conform to a specific acceptable behaviour, and by the fear of sanctions and shame sometimes dissuades to act (Horne, 2001). Examples of gender inequalities considered by previous research include the practices of purdah or of seclusion that occur in Bangladesh, India, and Nigeria that limit women's interaction with others or their mobility outside the household (Balk, 1997; Kritz and Makinwa-Adebusoye, 1999; Desai and Andrist, 2010). Another example reported in Ghana concerns the multiple domestic duties that the society expects from women (Amoakohene, 2004). These and similar inequalities are found throughout sub-Saharan Africa; they are reproduced informally, aim to maintain male dominance and are thought to be prejudicial to women's status (Heise, 1998; Moursund and Kravdal, 2003; Desai and Andrist, 2010; Wang, 2010). Several studies have shown that these gender inequalities have a negative effect on women's reproductive health practices and outcomes by limiting their access to information, increasing their vulnerability to gender-based violence and also by limiting their access to health services (Mason, 1987; Riley, 1998; Blanc, 2001; Dodoo and Frost, 2008; Stephenson et al., 2012). In empirical research, the operationalization of gender inequality has mainly been done at the individual level, with an emphasis placed on women's decision-making authority (Blanc, 2001). Previous research has shown that gender inequality at the household level has significant effects on infant mortality and morbidity (Caldwell and Caldwell, 1993), family size (Balk, 1994), contraceptive use (Woldemicael 2009), age at marriage (Desai and Andrist, 2010) and health services utilisation (Beegle et al., 2001; Bloom et al., 2001; Pallikadavath et al., 2004; Ahmed et al., 2010). However, limiting the analysis to the individual or household level can provide only partial explanations of the relationship between gender inequality and maternal health service use (Caldwell, 1990; Dodoo and Frost, 2008). Gender and social norms at contextual level may also shape and define the imbalance of power at individual level (Caldwell and Caldwell, 1993; Balk, 1994; Kritz and Makinwa-Adebusoye, 1999; Mumtaz and Salway, 2005; Adjiwanou, 2013). Therefore, by defining the way women should behave in society and by limiting their autonomy and decision-making authority, gender norms can indirectly influence women's use of health care services (Mason, 1987; Caldwell and Caldwell, 1993; Balk, 1994; Kritz et al., 2000; Desai and Johnson, 2005; Stephenson et al., 2006a). Gender norms may further have a direct impact on the use of maternal health service by limiting women's mobility and access to employment opportunities or resource management, lowering their self-esteem and creating an environment of fear and stress. Specific norms regarding violence are essential to consider, as they are directly related to the exercise of violence against women (Moore, 1999; Koenig et al., 2006; Sarkar, 2008; OECD, 2010; Stephenson et al., 2012) and are more susceptible to “create a climate of fear and intimidation” (Gilfus et al., 2010; Nanda et al., 2013). Intimate-partner violence (IPV) is said to be grounded in patriarchal society that gives primacy and privilege to men (Amoakohene, 2004; Kishor and Subaiya, 2008), and “serves to maintain the unequal balance of power” between men and women (Watts and Zimmerman, 2002; Wilson-Williams et al., 2008). Qualitative research by Hatcher et al. (2013) in the province of Nyanza in Kenya has shown that at times violence against women was motivated by what men and the “society” considered as women's challenging of the social norms and of men's authority.

countries (Ghana, 2003; Kenya, 2003; Tanzania, 2004/2005; and Uganda, 2006). These countries share a common history as former British colonies and, according to the Atlas of Gender and Development produced by OECD (2010), the gender context in all four countries is highly discriminatory against women. DHS data are comprehensive and cover multiple topics related to reproductive health behaviour. The information collected allows both for a direct measures of women's decision-making authority in the household, and to assess attitudes with regard to intimatepartner violence that can be used to measure gender inequality at the contextual level. The DHS are considered to be the most important source of high quality data for comparative studies in developing countries in the general social and health spheres of research (Pullum, 2008; Subramanian et al., 2011). Our analysis focuses on the behaviour of women of reproductive age 15–49 who have had a live birth in the past five years in a rural area. Limiting the study to rural residents was done for three reasons: (1) access to health care service is relatively low in those areas, (2) consequences of non-use of care are correspondingly dramatic, and (3) prevalent norms and attitudes are more homogeneous in the countryside. The primary sampling unit in the surveys are clusters and these were used to assess the local context (Montgomery and Hewett, 2005). As was done by previous studies (Franzini et al., 2005), clusters in which fewer than five women had been interviewed were omitted from the study. The number of clusters per country included in the analysis thus ranges from 237 in Ghana to 361 in Tanzania, and the average number of women aged 15–49 years old interviewed per cluster ranges from 14 in Ghana to 23 in Uganda. In all, these clusters cover 1814 women in Ghana, 2662 in Kenya, 4223 in Tanzania and 3529 in Uganda. 3.2. Method Multilevel structural equation models (MSEM) are used to measure gender norms regarding violence against women and to estimate their direct and indirect effects (acting through women's decision-making authority) on maternal health service use. Structural equation modelling (SEM) comprises two components namely, that are the measurement and the regression model. The measurement model, commonly called confirmatory factor analysis (CFA), specifies the relationships between the latent and the observable variables (indicators) that attempt to provide an approximate measure. In this study, this model is used to assess the contextual latent variable for gender norms related to the local acceptability of violence against women. The regression model estimates the relationship between the dependent and independent variables. This model has been developed and presented elsewhere (Adjiwanou, 2013). Appendix A presents and describes the main system of equations used in the present study. Overall, the model estimated can be understood as a two-level logistic regression model with two notable differences: a contextuallevel variable is measured as “latent” factor, and two dependent variables are included at level 1. These two dependent variables are related to the measure of maternal health service use and also to the variable that captures women's decision-making authority (cf. Fig. 1). The system of equations was estimated simultaneously using Mplus 6.11. Standard errors were estimated using the maximum likelihood method with the MLR estimator based on the expectation maximisation algorithm (Gottfredson et al., 2009). This is a robust estimator in cases of non-normality of dependent variables (Muthén and Muthén, 1998–2010).

3. Methodology 3.3. Variables 3.1. Data The analysis is based on Demographic and Health Survey (DHS) data from the period 2003–2006 for four sub-Saharan African

3.3.1. Dependent variables Maternal health service utilisation is captured by three separate dichotomous variables relating to the last pregnancy occurring in the

V. Adjiwanou, T. LeGrand / Health & Place 29 (2014) 67–78

69

Within

Low authority 0.10°

-0.35°

-0.032

Attitude toward wife beating

-.12

3.12***

-1.53*

Y1

Skilled birth attendance

-0.40

0.97

-0.21

.886

.34*** Y2 Y3

.879 .902 .851

Gender norms favoring violence (1)

Éducation

Revenue

Y4 .767 .36***

0.66***

-0.59***

Y5 -0.39***

Between Fig. 1. Some pathways (direct effects) from the final model for Ghana; standardized results are shown only for the measurement models. Significance at 10% (1), at 5 % (*), at 1 % (**) and at 0.1% (***). Y1. Goes out without telling him. Y2. Neglects the children. Y3. Argues with him. Y4. Refuses to have sex with him. Y5. Burns the food.

five years preceding data collection: a first antenatal care occurring in the first trimester of pregnancy, having four or more antenatal care visits, and delivery with trained medical personal (physicians, midwives, nurses, nursing assistants, or a trained birth attendant). The mediating dependent variable concerns the women's decisionmaking authority at the household. Attempts to measure it as a latent variable have not been successful due to the number of indicators, and the fact that these are all binary. It is measured by an index based on answers to the four questions about the “final say” variables in the DHS. These are related to the person in the household who was responsible for decisions concerning the woman's health (personal decision-making authority), major and daily purchases (economic decision-making authority), and visits to friends or family (mobility decision-making authority). Each response is dichotomized, taking the value 1 if the woman reports that a decision is made by her or jointly with her spouse/another person, and 0 otherwise. The index is defined by summing the four responses and took the value 0 if the woman reported taking no decisions (no decision-making authority), 1 if she reported one or two decisions (low decision-making authority), and 2 if she reported more than three decisions (high decision-making authority). We run two separate models to reduce the level of complexity in the model (when using the mediating variable with three categories): the first comparing women of low decision-making authority to those of high decision-making authority; whereas the second compared women of high decision-making authority to those who have no decision-making authority. The results commented in this article are based primarily on the first comparison (Table 2) as this two groups tended to be more homogenous than the group of women with no decision-making authority, which encompassed more healthier women and more voiceless women. Table 4 showed the results comparing women with no decision-making authority to those with high decision-making authority. In addition, in the same table we provide the same results when the mediating index variable is replaced by the specific “final say” variable.

3.3.2. Explanatory variables At the contextual level, the main independent variable of interest is the measure of the gender norms regarding violence

against women, which is assessed as a latent variable. The indicator variables used to define this latent variable were drawn from five questions on respondents' perceptions regarding violence against women, whose relevance to the issue has been demonstrated elsewhere (Rani et al., 2004; Agarwala and Lynch, 2006). These questions are formulated as follows: “Sometimes a husband is annoyed or angered by things that his wife/partner does. In your opinion, is a husband justified in hitting or beating his wife in the following situations: (1) she goes out without her husband's permission; (2) she neglects her children; (3) she argues with her husband; (4) she refuses to have sexual relations with him; or (5) she burns the food”. The indicators are measured as the proportion of women who agree with each statement within each cluster. Table 1 and Fig. 2 present the distribution of these variables for each country. The model also takes into account three variables that reflect sociocultural norms related to fertility (the proportion of households with children under five years of age, and proportions of women not using modern contraception) and to service utilisation (the proportion of women who say that “not wanting to go alone” is a big problem for them for attaining health services) (Beninguisse, 2003; Stephenson et al., 2006a; Carter, 2010; Desai and Andrist, 2010). Other contextual level variables include aspects of the clusters' socioeconomic context: the proportion of households that are among the 60% richest households in each country,1 the proportion of women with secondary-level education or higher, and the proportion of women not working in agriculture. Finally, service accessibility is measured indirectly by the proportion of women in each survey cluster who judged distance to healthcare facilities to be a serious problem for them (not asked in Kenya). The distribution of these community variables is presented in Table 1. To readily show the contextual effect, all contextual variables have their counterpart at the individual level. Especially, the variable measuring attitude toward violence is summing from the five questions and centred at the cluster level.

1

Computed from the wealth quintile of DHS.

70

V. Adjiwanou, T. LeGrand / Health & Place 29 (2014) 67–78

Table 1 Descriptive statistics (mean, linearised standard error or standard deviation) for analysis sample by country, continued. Ghana

Kenya

Ugandac

Tanzania

.323 (.019) .626 (.018) .411 (.016)

.370 (.018) .499 (.013) .093 (.008)

.391 (.015) .454 (.013) .160 (.009)

.506 (.016) .598 (.012) .133 (.008)

.237 (.016) .301 (.019) .462 (.024)

.257 (.013) .380 (.012) .362 (.013)

.201 (.012) .276 (.011) .523 (.015)

.242 (.014) .386 (.013) .372 (.014)

.566 (.020)

.516 (.017)

.596 (.014)

.593 (.018)

.477 (.022) .578 (.021)

.317 (.012) .528 (.013)

.498 (.015) .639 (.015)

.340 (.014) .467 (.015)

.574 (.022)

.520 (.014)

.615 (.016)

.507 (.015)

.050 (.006) .427 (.013) .380 (.014) .144 (.009)

.084 (.006) .519 (.011) .306 (.010) .091 (.006)

.059 (.004) .492 (.009) .349 (.009) .100 (.005)

.075 .506 .321 .099

.109 (.009) .217 (.012) .237 (.011) .196 (.011) .240 (.012)

.160 (.008) .284 (.009) .237 (.009) .166 (.008) .153 (.008)

.121 (.006) .261 (.009) .249 (.009) .187 (.008) .183 (.007)

.183 (.007) .271 (.008) .235 (.008) .161 (.006) .150 (.007)

.896 (.009)

.832 (.009)

.225 (.011)

.238 (.010)

.143 (.007)

.225 (.008)

.344 (.013) .248 (.011) .184 (.011)

.359 (.010) .223 (.008) .180 (.009)

.317 (.009) .263 (.008) .277 (.009)

.372 (.009) .223 (.007) .180 (.009)

.285 (.017) .406 (.021) .170 (.021) .139 (.016)

.249 (.017) .645 (.020) .106 (.015)

.451 (.016) .343 (.013) .098 (.010) .107 (.008)

.289 (.019) .304 (.020) .230 (.021) .177 (.025)

Women's Education – (No education) Primary Secondary and more

.469 (.025) .238 (.014) .293 (.018)

.146 (.017) .667 (.016) .187 (.013)

.250 (.013) .650 (.012) .099 (.007)

.288 (.017) .690 (.016) .022 (.003)

Spouse’s education – (primary or less) Middle Secondary and more Other, do not know or no partner (No education) Primary Secondary and more Other, do not know or no partner PERCEIVED BENEFIT/NEED Skilled attendance for previous birth (No) Yes Only one child in last five years

.461 (.026) .347 (.020) .102 (.010) .090 (.009) .107 (.014) .513 (.016) .287 (.014) .093 (.008)

.114 (.007) .657 (.011) .228 (.011)

.194 (.018) .706 (.016) .049 (.005) .051 (.005)

.281 (.015) .104 (.009) .615 (.015)

.306 (.014) .145 (.008) .549 (.013)

.418 (.012) .211 (.011) .371 (.010)

.254 (.014) .224 (.011) .522 (.013)

.186 (.016) .400 (.016) .414 (.018)

.191 (.014) .253 (.011) .556 (.015)

.213 (.010) .278 (.009) .509 (.015)

.310 (.014) .344 (.010) .346 (.013)

.577 (.020) .211 (.014) .124 (.010) .089 (.009)

.450 (.020) .167 (.010) .064 (.006) .318 (.019)

.830 .075 .043 .052

(.014) (.007) (.004) (.007)

.885 (.012) .012 (.002) .048 (.007) .055 (.006)

.727 (.018)

.491 (.016)

.685 (.013)

.861 (.012)

.105 (.010) .168 (.014)

.246 (.011) .262 (.014)

.172 (.009) .143 (.009)

.047 (.005) .093 (.009)

.367 (.023) .309 (.019) .324 (.023)

.258 .256 .233 .253

.245 (.016) .252 (.012) .217 (.011) .286 (.016)

.256 .257 .239 .248

1814

2662

3529

4223

Variables Name

Remarks

DEPENDENT VARIABLES Skilled birth attendanceb At least four antenatal visits b Antenatal care starts in the first trimesterb Women decision-making authority indexc “No” decision-making authority Low decision-making authority High decision-making authority Personal decision-making authority Final say (FS) on own health carec Economic decision-making authority FS on making large household purchasesc FS on making hh. purchases for daily needsc Mobility decision-making authority FS on visits to family or relativesc INDEPENDENT VARIABLES INDIVIDUAL LEVEL VARIABLES SOCIOCULTURAL FACTORS Women's age at survey- (15–19 y) 20–29 30–39 40–49 Women's age at pregnancy - (15–19 y) 20–24 y 25–29 y 30–34 y 35–49 y Living with partner

b

Number of living children before the index child– (0) 1 or 2 3 or 4 5 or more Religion – (Catholics) Other Christians Muslims Others

In Kenya, ‘others’ are grouped with Muslims

The information is available for the last five years.

Exposure to radio – (Never) Sometimes Everyday ECONOMIC ACCESSIBILITY Women’s Employment - (Agriculture) Sales/services Others No employment Spouse's employment – (Agriculture or no employment) Sales or services Skilled manual Wealth quintile – (Lowest) Second Middle Fourth and highest n

Re-categorised from DHS wealth quintile measure. In Ghana, Middle is quintile 4 ¼ 3

(.018) (.013) (.013) (.019)

(.005) (.009) (.008) (.006)

.866 (.008)

(.015) (.012) (.012) (.017)

V. Adjiwanou, T. LeGrand / Health & Place 29 (2014) 67–78

71

Table 1 (continued ) Variables Name

Remarks

CONTEXTUAL VARIABLESa Indicator of the latent variable (Uj) Proportion of women by cluster who said that a woman should be beaten by her husband if she : “Goes out without telling him” “Neglects the children” “Argues with him” “Refuses to have sex with him” “Burns the food” Other contextual variables (Zj) Proportion of women aged 20 or over with secondary education Proportion of households of 3 or more quintiles Proportion of women employed outside the agricultural sector Proportion of households with children under 5 years than the average Proportion of married women not using modern contraception Proportion of women saying they “do not want to go alone”d Proportion of women who think that the distance is a big problem for their health-seekingd N

Ghana

Kenya

Ugandac

Tanzania

.405 (.231) .445 (.250) .360 (.226) .245 (.194) .193 (.183)

.443 (.195) .610 (.155) .526 (.203) .348 (.182) .172 (.135)

.554 (.183) .600 (.201) .451 (.221) .351 (.164) .273 (.164)

.442 (.186) .462 (.220) .452 (.212) .308 (.166) .198 (.136)

.359 (.260)

.218 (.186)

.144 (.146)

.086 (.149)

.358 (.339)

.538 (.319)

.527 (.344)

.552 (.302)

.512 (.265)

.630 (.230)

.274 (.239)

.268 (.263)

.356 (.193)

.325 (.175)

.477 (.158)

.387 (.187)

.782 (.172)

.630 (.261)

.796 (.165)

.776 (.181)

.248 (.170)

.275 (.185)

.286 (.205)

.451 (.499)

.606 (.489)

.416 (.493)

308

361

237

262

a

Standard deviation in brackets. Dummy variables. c Questions asked only to women living with partner in Uganda. d Questions not asked in Kenya. b

The individual level independent variables are those frequently used by studies of similar topics: socio-cultural factors, the perceived benefits/needs, and the economic and physical accessibility to health care (Simkhada et al., 2008; Gabrysch and Campbell, 2009; Adjiwanou and LeGrand, 2013). All of these variables and their categories are summarised in Table 1.

than in the other three countries. The proportion of women receiving at least four antenatal visits varies from 45% in Uganda to 63% in Ghana, whereas the proportion of women who start their first antenatal care at the first trimester ranges from 9% in Kenya to 41% in Ghana. Finally, Table 1 shows the distribution of the other variables of analysis while Table 3 presents the correlation matrix of the contextual variables.

4. Results

4.1. Gender context and maternal health services use in rural Africa

Table 1 describes the distribution of the women's decisionmaking index and of their indicators (final-say variables) and Fig. 2 shows the plot of the gender norms indicators in the four countries. There were higher proportions of women with high decision-making authority in Ghana (46%) and in Uganda (52%) as compared to the other countries. In contrast, the proportion of women with high decision-making authority is 36% in Kenya and 37% in Tanzania. On the four final-say variables, the one relative to large household purchases shows the most gender inequality in the four countries. Concerning the gender norms indicators, the distribution of the variable on attitude toward wives beaten for “burning the food” departs the most from the normal curve in all countries (Fig. 2). This variable also shows the least supportive attitude from the women in all countries. In Uganda, 27% agreed that a woman should be beaten if she burns the food compared to less than 20% in the other three countries. By contrast, most women approved of the infringement in a case of child neglect. Table 1 also presents the statistics of the main dependent variables related to antenatal care and skilled birth attendance. It shows that less than 50% of the women in Ghana (32%), Kenya (37%), and Uganda (39%), and 51% in Tanzania, used skilled personnel during their last delivery. With regard to antenatal care, the frequency and the timing of the first visit are higher in Ghana

Panels A and B of Table 2 present respectively the coefficients for the effects of contextual and individual level variables on the use of maternal health services in the four countries. Panel C shows the intra-class correlations (ICC)2 for the null and final model for the three dependent variables, namely, the frequency (at least four visits) and timing (first visit in the first trimester) of antenatal care and the use of skilled birth attendants. Two observations emerge from these results. First, the initial ICC is large and significant for all three dependent variables, indicating a high level of heterogeneity across clusters in terms of practices related to seeking maternal health services. In other words, the decision to use maternal health services is strongly influenced by the cluster in which the woman lived. The range of the ICC are similar to those found by Gage (2007) in Mali. Secondly, despite substantial reductions in these correlations after taking into account the model's explanatory variables, especially for SBA, there remains unexplained variance at the contextual level, probing that other contextual variables are under consideration. The results presented in Panel A of Table 2 (and in Table 4 when comparing no decision-making authority to high decision-making 2 In the case of a model with a dichotomous dependent variable using the logit function, the intraclass correlation is obtained using the formula σ 2c =ðσ 2c þ ðπ 2 =3ÞÞ , where σ 2c is the level 2 variance.

72

V. Adjiwanou, T. LeGrand / Health & Place 29 (2014) 67–78

authority) show that gender norms regarding violence against women have a significant and negative effect on at least one dependent variable in each country, except in Kenya where no effects were found. Fig. 1 presents a few illustrative direct pathways of the final model of Ghana to help understand the results presented in Table 2. In these tables, we found that women who live in environments where gender norms regard violence against women as relatively acceptable are less likely to use skilled birth attendants or to start their antenatal care in the first trimester in Ghana and to exhibit a lower frequency of antenatal care in Tanzania. When comparing “no” decision-making authority to high decision-making authority, we also found that these gender norms have negative effects on the time of onset of antenatal care in Uganda (Table 4). Some significant effects of the gender norms also emerge in Table 4 when the women's finalsay variable is used as mediation, especially in the negative effect on skilled birth attendance in Uganda. Even though the gender norms' effects are not significant in all countries, their impact, however, is generally substantial and in the expected direction. At the individual level, the inequality of power in the household reveals little effect on maternal health service use. The results show that women with low decision-making authority are less likely to use skilled delivery at birth in both Ghana and Tanzania (Table 2) and to use prenatal care in Uganda (Table 4). This low effect of women's decision-making authority renders the indirect effect of gender norms regarding violence weak. These norms indirectly impact negatively on the use of skilled birth attendance at delivery in Ghana (Table 2) and the time of onset of antenatal care in Uganda (Table 4). With regards to the other contextual norm variables, the two variables related to fertility norms, i.e., the presence of young children and the proportion of married women not using modern contraception, show a strong and consistent effect on the use of maternal health services as they had a negative effect on at least one dependent variable in each country. For instance, the results presented in Table 2 shows that a woman living in a community with a large proportion of women not using modern contraception is less likely to have used skilled birth attendants for her last delivery in Kenya and in Uganda, to have at least four antenatal consultations in Ghana and to start antenatal care in the first trimester in Tanzania, compared to a woman living in a community with high contraceptive use. Social barriers related to service use and distance has an anticipated negative effect on maternal health service use. Environments in which a high proportion of women do not want to seek services on their own are also those where the use of skilled birth attendants is lowest, such as in Uganda. In Ghana, similar results are found in the frequency of antenatal care and for timely antenatal care use. In addition, women living in communities where a high proportion of respondents considered distance to be a major problem in accessing healthcare are found to be less likely to have used skilled birth attendants for their last delivery in Tanzania and Uganda, compared with women in communities where the proportion was low. While not significant in Ghana, the effect of this variable nevertheless was in line with those in the latter two countries. In most countries, regardless of the dependent variable considered, the results show that contextual socioeconomic variables do not have a strong influence or tend to have a contra-intuitive effect on the use of maternal health care. The results in Panel A of Table 2 show a positive, significant and direct effect of the community’s education level on the use of skilled birth attendants only in Uganda. Again, women living in a community with a high proportion of non-agricultural women are more likely to deliver with a trained person in Ghana and to have timely antenatal care in Uganda than those who live in a community with a low proportion of non-agricultural women. On the other hand, in

Kenya, women who lived in a community with a high proportion of non-agricultural women tended to under-utilize skilled birth attendants at delivery. Finally, the effects of the individual explanatory variables (Table 2, Panel B) are mostly in the expected direction. The most important effect is related to the previous service used, which may reflect both the quality of care and also the availability of services. When the influence of this variable is significantly associated with assisted delivery and frequency of antenatal care in all countries, its estimated effect on timely antenatal care visit is weak and insignificant in Kenya and Uganda. Results also show a strong association between modern contraceptive use and maternal health service in all four countries.

5. Discussion In this study, we have analysed the effects of gender inequality, measured at individual level by women's decision-making authority at the household, and at the contextual level by the permissive gender norms regarding violence against women, on the use of antenatal care and skilled birth attendance in the rural areas of four African countries. Our results showed that the presence of restrictive sociocultural and gender norms negatively affects the use of maternal health services to some degree in all four countries. We found that women who lived in areas where gender norms are relatively favourable to violence against women, are less likely to deliver with a health professional in Ghana (and to some extent in Uganda, Table 4), to have four antenatal visits in Tanzania or to start their prenatal visits in the first trimester in Ghana and in Uganda, after controlling for women's own attitude toward violence. Similar results were reported in different contexts or on other issues. Desai and Andrist (2010) in India showed that gender-related sociocultural norms negatively influenced the age of marriage. Similarly, although using a scale developed at individual level, Nanda et al. (2013) found a significant influence of women's attitude towards wife beating on the scale of women's contraceptive use in Tanzania. In contrast, a study by Stephenson et al. (2006b) in Uttar Pradesh, India, found no significant effect of the community norms tolerant to domestic violence on contraceptive uptake. The difference in the effects of the gender norms on maternal health service use in the four countries is not unexpected. The four countries may be subject to and influenced by other specific sociocultural norms that were not captured in the present study. Moreover, these countries may be drained by unequal social change, which may reduce the effect of the gender norms and explain the results found here. For instance, divergent patterns were revealed in the level of gender inequality at country level in Uganda and the women's assessment of their decision-making authority in the same country (OECD, 2010). One may also argue that the responses to the gender norms indicators are not understood in the same way by the women in each country. For this purpose, Yount et al. (2013) demonstrated in the case of Bangladesh that differences in the women’s responses to intimate partner violence attitudinal questions are influenced by the perception of transgression, with acceptance of wife beating higher in cases of wilful infringement. Nonetheless, recent studies in the same four countries covered here relativized this view by showing a strong negative effect of these gender norms on women's decision-making authority (Adjiwanou, 2013). In all cases, these results require additional studies in a sub-Saharan African context to assess more deeply the meanings of these questions in respect of gender norms. Our analytical framework strongly emphasised women's decisionmaking authority, grounded in the more general gender context, as a

V. Adjiwanou, T. LeGrand / Health & Place 29 (2014) 67–78

73

Table 2 Determinants of skilled birth attendance, of four or more antenatal visits and of the use of antenatal care in the first trimester by country, coefficient from two-level logistic structural equation modelling. INDEPENDENT VARIABLES

Sig. Ghana SBA

PANEL A: Contextual-level variables Gender norms regarding violence against women Indirect effect through women’s decision  making authority Percentage of: women aged 4 ¼ 20 with secondary education households with quintile 4 ¼three (middle) women employed outside the agricultural sector households with 0–5 year children 4the mean number in rural area married women not using modern contraception women saying that they “did not want to go to the clinic alone”e women saying that “distance is a big problem” e PANEL B: Individual level variables WOMEN'S DECISION-MAKING AUTHORITY Low decision-making authority (vs. High decisionmaking authority) OTHER BARRIERS Attitude toward violence against women Not using modern contraception “Not wanting to go alone” is a major probleme Distance is a major probleme SOCIOCULTURAL FACTORS Women’s age at pregnancy - 20–24 y (15–19 y) 25 29 y 30 34 y 35–49 y Living with partner

þ

Kenya ANC_4

ANC_d

Tanzania

SBA

ANC_4

ANC_d

SBA

ANC_4

Uganda ANC_d

SBA

ANC_4

ANC_d

 

 1.53b  0.84  1.08  0.22

 0.85b  0.81  0.25  0.23

0.05  0.04

0.203 0.07

0.21  0.20

 0.85b  0.38 0.07  0.22

 0.89  0.13

0.090  0.21

 0.67  0.39

þ þ þ 

0.97  0.21 1.29b 0.34

0.08  0.38 0.81  0.09

0.29  0.29  0.10  0.85b

0.13  0.34  0.57  0.44

 1.11 0.16 0.48  0.63

0.75 0.12 0.58  0.41

 0.30 0.13  0.12  0.48a

0.20 0.43  0.49  0.53

1.82b  0.26  0.17 0.58

 0.33  0.14 0.33  0.26

 2.15c 0.04 1.10c  0.96b

 

 0.93  0.93

 1.78c 0.32  1.07b 0.83b  1.38b  1.26b

 0.17

 0.53 0.42

 0.40 0.02

 0.89b  1.15b  0.35 0.19  0.93b  0.46



 0.54

0.46

0.23 0.48  0.85b  1.46c

 0.59a  0.01

 0.05

 0.49a  0.09

 0.47

0.04

 0.17a

0.06

 0.18

 0.05

 0.08

 0.14

 0.03  0.20a 0.06  0.17a

0.02  0.03 0.14  0.07

0.18 0.21 0.40a 0.39a

0.21 0.09 0.04 0.11

0.48

 0.35a  0.07

 0.08

 0.15

 0.03  0.48a  0.13 0.27

 0.05  0.35a  0.04  0.11

 0.06  0.25  0.27  0.00

0.01  0.04  0.40c  0.07

 0.01  0.26

 0.03  0.51d 0.12  0.07

0.05a  0.29c  0.13  0.01

0.05  0.02  0.41c  0.44c  0.03 0.17 0.11  0.21a

 0.19  0.06 0.30 0.56

 0.35  0.04 0.17 0.28

 0.17 0.36 0.44 0.16

0.25 0.34 0.87c 0.50

0.10 0.34 0.34 0.28

 0.64b  0.33  0.37 0.03

0.29a 0.52c 0.44b 0.64c

 0.19  0.15 0.20 0.11

0.01  0.13 0.35 0.47

 0.02

0.07

0.21

0.23

0.07

 0.15

 0.01

0.17

0.11

0.18

 0.99

 0.07

0.06

 0.67

0.12

 0.12

 0.40b  0.63d  0.534c

 0.99c  1.22c  0.39  0.04

 0.48  0.71b  0.08 0.06

 0.29  0.33 0.32a  0.02

 1.63d  1.69d 0.41 0.99c

 0.33  0.48a 0.38b 0.69c

 0.07  0.21 0.32 0.69

 1.00d  1.01d 0.25b 0.26

 0.05  0.38a 0.16 0.54b

 0.36  0.77b  0.11  0.25

 0.75c  0.69b 0.04 0.65c

 0.46b  0.52b 0.01 0.23

 0.58b  0.60b  0.31b  0.01

0.43a 0.49a  0.20

0.05  0.12 0.39 0.19 b  0.59 0.02 0.15 0.49 0.46

0.11  0.62a 0.37c 0.27  0.79b 0.89d  0.59a  1.29b  0.13

0.07 0.26  0.30

0.08 0.28  0.51

0.25 0.45b

0.01 0.56c

0.12 0.11

 0.58

Spouse's education – Middle (primary or less) Secondary and more Other, do not know or no partner Primary (No education) Secondary and more Other, do not know or no partner PERCEIVED BENEFIT/NEED Skilled attendance for previous birth – Yes (No) Only one child in last five years ECONOMIC ACCESSIBILITY Women’s Employment - Sales/services (agriculture) Others No employment Spouse’s employment – Sales or services (agriculture or no employment) Skilled manual Wealth quintile – Second (Lowest) Middle Fourth and highest PANEL C Intra Class Correlation – ICC for Null model Intra Class Correlation – ICC for Final Model N n

234 1314

d

d

 0.13 0.15  0.05  0.17

0.11

Number of living children before the index child – 1 or 2 (0) 3 or 4 5 or more Women’s Education – Primary (No education) Secondary and more

b

 0.03

0.60 0.24

2.61d 1.35d

1.03d 0.99d

0.86c 0.70d

1.77d 1.11d

0.48c 0.45c

0.25 0.24

2.50d 1.52d

0.43c 0.54d

0.43b 0.44b

2.18d 0.99d

0.58d 0.30c

0.06 0.06

0.18 0.12  0.32 0.23

0.12 0.34  0.42 0.73b

0.21 0.32  0.39 0.17

0.13  0.04  0.19 0.48c

0.12 0.45b  0.16 0.35b

 0.11 0.52a  0.63b 0.57b

0.25 0.31  0.13  0.03

0.19 0.15  0.14 0.28

 0.15  0.01 0.11  0.01

0.12  0.12 0.59b 0.11

0.25 0.23 0.16  0.09

 0.07 0.10 0.09 0.05

0.06 0.34 0.32

 0.13 0.10 0.18

0.08 0.26a 0.34

0.25a  0.04 0.08 0.57b

0.01  0.08 0.18 0.28

0.40a 0.20  0.09 0.34

0.15 0.24a 0.28a 0.54c

0.19 0.22b 0.09 0.16

 0.08  0.09  0.21  0.17

0.11 0.01  0.04 0.44b

 0.16  0.11  0.16  0.07

 0.16  0.39  0.15  0.04

.399d .237d

.281d .217d

.115d .081d

.389d .119c

.175d .122d

.193d .175c

.319d .173d

.097d .073d

.191d .151d

.289d .119d

.080d .076d

.155d .127d

261 1909

361 3099

308 2866

Sig.¼ Expected sign. SBA¼ Skilled birth attendant. ANC_4¼ At least four antenatal care visits. ANC_d¼ first antenatal care starts in the first trimester. References are in italic and shown in brackets. a

Significance at 10%. 5 %. c 1 %. d At 0.1 %. e Questions not asked in Kenya. b

74

V. Adjiwanou, T. LeGrand / Health & Place 29 (2014) 67–78

Ghana

.8

1

.2 .4 .6 .8 1 neglects the children

0

.2 .4 .6 .8 burns the food

2 .4 .6 .8 1 neglects the children

0

.1 .2 .3 .4 .5 burns the food

1.5

.2 .4 .6 .8 1 argues with him

Density

0

0 .2 .4 .6 .8 refuses to have sex with him

Uganda

0

.2

.4

.6

.8

.2

.4

.6

.8

1

.2

.4

.6

.8

.2

.4

.6

.8

0

1

1

Density

.5 0

.5 0 0

1

goes out without telling him

argues with him

1.5

2

2.5 1.5

Density

1

1.5 0

0

.2

.4

.6

.8

1

neglects the children

0

.2

.4

.6

.8

1

argues with him

.2

.4

.6

.8

0

.2

.4

.6

burns the food

3 2

Density

0

0

refuses to have sex with him

0

0

0

.5

1

1

1.5

Density

2

3 2 1

1

2

Density

2.5

3

4

neglects the children

1

Density

.5

.5

0

goes out without telling him

2

2.5 2

2.5 2 1.5

Density

1

1

1

1.5

0

0

0

.5

.5

1

1.5

Density

2

2

2.5

Tanzania

Density

0

1

.5 0

0

0 0 .2 .4 .6 .8 refuses to have sex with him

Density

1 .5 0

.2

4

2 1.5 1

Density

Density

3 1 0

0 .2 .4 .6 .8 1 goes out without telling him

2

4

Density

2 1

Density

2

Density

1.5 .5 0

.2 .4 .6 .8 1 argues with him 2.5

0

6

3

1

Density

1.5 .5

0

5

.6

3

.4

goes out without telling him

2

.2

0

0

0 0

1

Density

1.5 .5

1

Density

1.5 1 .5

Density

2

4

2

2

2

Kenya

0

.2

.4

.6

.8

refuses to have sex with him

0

.2

.4

.6

.8

burns the food

Fig. 2. Distribution of the indicators used to measure the gender norms regarding violence against women in the four countries.

variable mediating the effect of the gender norms regarding intimate partner violence on women's health behaviours. However, compared to the permissive gender norms, our findings did not reveal for the most part an important impact of women's decision-making authority on their decisions regarding the use of health services. Overall, we found significant effects of women's decision-making authority in the household on skilled birth attendance in Ghana and in Tanzania and on the time of onset and frequency of antenatal care visits in Uganda. These results are in line with those found by Ahmed et al. (2010) in their study, which covered the same countries as ours but which did not consider contextual factors. More generally, the ambiguous impact of women’s autonomy or decision-making authority on maternal service utilisation or contraceptive use in Africa has also been reported by other authors (Desai and Johnson, 2005; DeRose and Ezeh, 2010). There are several plausible explanations for these results, beyond the possibility that women's decision-making authority, conceptually and operationally based on the Asian context, may not be an important determinant of the use of maternal health service in these specific African contexts (Schatz and Williams, 2012; Heckert and Fabic, 2013). Conceptually, we measure women’s decision-making authority by whether they have a joint final say with their partner in the household, a model advocated by many authors (Mullany et al., 2005; Ahmed et al., 2010). However, joint decision-making may hide the true power inside the household if women's voices are not heard, or if joint decision-making means only following the husband’s/ partner’s view (Mumtaz and Salway, 2005; Shroff et al., 2009; Hadley et al., 2010). Furthermore, it may be that cross-sectional data

are not adequate for measuring a phenomenon that is, by nature, dynamic. Nevertheless, it should be noted that even if women do not always enjoy personal autonomy on their own behalf (as the results here demonstrate), studies have shown that their autonomy often at least allows them to be proactive in decisions concerning their children (Desai and Johnson, 2005). This study measured other contextual-level variables and showed that norms and practices related to protective familybuilding behaviours – contraceptive use and a high proportion of children in the community – strongly affect the use of maternal health service in the four countries. For example, the community measure of modern contraceptive use, considered in many studies as a proxy of sociocultural norms, showed a negative impact on skilled birth attendance in Kenya and Uganda, on the timing of the first antenatal care visit in Ghana and the frequency of antenatal care visits in Uganda. In addition, women who resided in communities where a high proportion of women are unwilling to go to the clinic alone are less likely to report skilled delivery for their last birth in Uganda, or to use antenatal care appropriately in Ghana. This variable may explain other sociocultural norms or the non-confidence that women have in health providers in their community. On the other hand, our results unexpectedly failed to reveal strong direct effects of the socioeconomic contextual factors on the use of maternal health care. This suggests that, once the underlying mechanisms of the contextual effects are assessed, cluster socioeconomic conditions have relatively little effect on maternal health care use (Sampson, 2003).

V. Adjiwanou, T. LeGrand / Health & Place 29 (2014) 67–78

75

Table 3 Correlation matrix of the contextual level variables. Ghana

Gender norms regarding violence against women (Z1) Percentage of women aged 4 ¼ 20 with secondary education (Z2) Percentage of households with quintile 4 ¼ three (middle) (Z3) Percentage of women employed outside the agricultural sector (Z4) Percentage of households with 0–5 year children 4 the mean number in rural area (Z5) Percentage of married women not using modern contraception (Z6) Percentage of women saying that they “did not want to go to the clinic alone” (Z7) Percentage of women saying that “distance is a big problem” (Z8) Kenya Gender norms regarding violence against women (Z1) Percentage of women aged 4 ¼ 20 with secondary education (Z2) Percentage of households with quintile 4 ¼ three (middle) (Z3) Percentage of women employed outside the agricultural sector (Z4) Percentage of households with 0–5 year children 4 the mean number in rural area (Z5) Percentage of married women not using modern contraception (Z6) Percentage of women saying that they “did not want to go to the clinic alone “ (Z7) e Percentage of women saying that “distance is a big problem” (Z8) e

Z1f

Z2

Z3

Z4

Z5

Z6

Z7

 .59d  .39d  .21c .31d .21d .09 .30d

.66d .50d  .53d  .37d  .21d  .44d

.59d  .38d  .26d  .29d  .51d

 .29d  .19c  .28d  .44d

.32d .10a .19c

.10 .25d

.42d

 .53d  .46d  .19c .34d .54d

.55d .01  .41d  .65d

.05  .46d  .61d

.05 .10

.46d

 .46d  .32d  .37d .08  .12b .28d .19d

.52d .73d  .14c .03  .16d  .26d

.52d  .25d  .16c  .15c  .24d

 .11b .05  .07  .23d

..38d  .01 .07

 .05  .08

.60d

 .25d  .42d  .36d .25d .33d .21c .07

.65d .50d  .22d  .61d  .12b  .17c

.43d  .18c  .61d .02  .17c

 .33d  .42d  .22d  .27d

.17c .08 .03

.08 .28d

.32d

Z8

Tanzania Gender norms regarding violence against women (Z1) Percentage of women aged 4 ¼ 20 with secondary education (Z2) Percentage of households with quintile 4 ¼ three (middle) (Z3) Percentage of women employed outside the agricultural sector (Z4) Percentage of households with 0–5 year children 4 the mean number in rural area (Z5) Percentage of married women not using modern contraception (Z6) Percentage of women saying that they “did not want to go to the clinic alone” (Z7) Percentage of women saying that “distance is a big problem” (Z8) Uganda Gender norms regarding violence against women (Z1) Percentage of women aged 4 ¼ 20 with secondary education (Z2) Percentage of households with quintile 4 ¼three (middle) (Z3) Percentage of women employed outside the agricultural sector (Z4) Percentage of households with 0 5 year children 4the mean number in rural area (Z5) Percentage of married women not using modern contraception (Z6) Percentage of women saying that they “did not want to go to the clinic alone” (Z7) Percentage of women saying that “distance is a big problem” (Z8) a

Significance at 10%. at 5%. c 1%. d 0.1%. e Questions not asked in Kenya. f Latent variable. b

This study raises certain theoretical and empirical questions. Despite the significant reduction in the intraclass correlation of the service utilisation variables in the full model, some unexplained variance remains at the contextual level. This is due possibly to two causes: the probable bias in the measurement of contextual variables (when using simple aggregation) and our inability to take into account other contextual variables such as the supply side factors or those related to physical context. In fact, most contextual variables used in developing countries are proxies that may not capture what we really wish to measure. For instance, contextual-level contraceptive use may be understood also as the availability of family planning services. Our attempt to measure gender norms related to violence against women by using the latent variable approach (Raudenbush, 2003) is an effort to have a more accurate and interpretable measure of contextual variables based on individual assessment. The second limit concerns our inability to consider the supply side effect – though not without some attempts to do so. This is the time to advocate again for the possibility of merging DHS and SPA (Service Provision Assessment) data, something that has not been feasible for researchers to date. The choice of the countries is primarily based on the availability of the data from the two surveys around the same period. However, the fact that gender norms related to violence against women showed some significant effects on some outcomes

of maternal health care use that are less related to supply factors (for instance antenatal care), and the broad measure of distance in this study may indicate that the results are robust in the absence of supply factors. Finally, whereas new DHS data are available for the four countries, the analysis provided here retains its significance. First, in three out of the four countries (except Uganda), the questions related to violence against women were asked for the first time in the DHS we used, which permits comparison with future research. Second, the four countries are far from reaching the Millennium Development Goals 4 and 5 and work remains to be done to understand the barriers women face (Singh et al., 2013). Finally, the methodological approach developed in this paper brings some new insights into the use of multilevel modelling and is important enough to contribute to new developments. The results of this study suggest that governments, in their efforts to improve women’s health and reduce maternal mortality, should also consider interventions that limit the negative role and the extent of sociocultural and gender norms (Balk, 1997) – interventions which appear to lead to a greater and more effective use of existing health services (Basu, 1990; Kiss et al., 2012). Policies are needed not only to strengthen the education of girls (Heaton et al., 2005), but also to combat directly gender inequality in all aspects of women's lives and at all levels (Fikree and Pasha, 2004). In terms of future research, this

76

V. Adjiwanou, T. LeGrand / Health & Place 29 (2014) 67–78

Table 4 Direct and indirect (i.e., through other dimensions of women's autonomy) effects of gender norms on skilled birth attendance, on the incidence of four or more antenatal visits and on the use of antenatal care in the first trimester by country, coefficient for two-level logistic Structural Equation Modelling. Mediating dependent variables

Ghana SBA

No decision-making authority versus High decision-making authority Gender norms regarding violence against women Indirect effect through women decision-making authority “No” decision-making authority (high) N n Personal decision-making authority Final say (FS) on “own health care” Gender norms regarding violence against women Indirect effect through FS on “own health care” No final say on own health Economic decision-making authority FS on “making large household purchases” Gender norms regarding violence against women Indirect effect through FS on “making large household purchases” No final say on large household purchases

ANC_4 ANC_d

Tanzania

Uganda SBA

ANC_4

 0.99c  0.04  0.05 0.01  0.02 0.01

 0.96 0.15 0.07 307 2538

 0.01  1.65b  0.57  0.88a  0.24b  0.38c

 1,29b  0.27  0.66a  0.88  0.54  0.24 0.16 0.03 0.04 0.01 0.11 0.08 0.65 0.19 0.06 0.09 0.03 0.26 0.03 0.28a 0.06

 1.41c  0.10 0.17 0.60 0.05 0.18

 1.13a 0.26 0.11

 0.11  0.29  0.12

 0.68  0.61b  0.26b

 1.23b  0.30

 0.56

 0.81  0.47

 0.06 0.18

 1.11c

0.05

 1.13a

 0.07

 0.71

 0.31

0.08

 0.45

 0.01  0.15

0.15

 0.23

0.02

 0.42a 0.10

 0.38a  0.33

0.03

 0.17

 0.01  0.09 0.09

 0.13

0.01

 0.24b 0.05

 0.17a

 1.27b 0.07 0.21 0.28 0.07 0.09 234 1240

 0.11 FS on “making hh. purchases for daily needs” Gender norms regarding violence against women  1.28b Indirect effect through FS on “making hh. purchases for daily 0.01 needs” No final say on making purchases for daily needs 0.01 Mobility decision-making authority FS on “visits to family or relatives” Gender norms regarding violence against women  1.30b Indirect effect through FS on “visits to family or relatives” 0.11 No final say on visit to family or relatives 0.04 N n

Kenya

 0.46 0.07 0.023

SBA

ANC_4 ANC_d SBA

ANC_4

0.05 0.43 0.12 262 1680

 0.58 0.05  0.10 0.62  0.03 0.17

0.31  0.22  0.08 359 2516

c

ANC_d

a

ANC_d

 0.15

 0.26

 0.64

 0.79  0.49

 0.07 0.16

 1.11

0.01

 1.09

 0.14

 0.77

 0.08

 0.04

0.08

0.01

 0.24

 0.01

0.02

 0.12

 0.01

 0.03

 0.05

 0.03

 0.02

0.05

0.01

 0.15

 0.01

0.02

 0.17

 0.01

 0.07

 0.11

0.02  0.10  0.07

 1.10a  0.08  0.03

 0.11  0.22  0.09

 0.76  0.19  0.07

 0.25  0.66  0.10 0.12  0.04 0.04

237 1814

 0.97  0.44  0.19 0.50a  0.20 0.57  0.09 0.25a 0.21a

0.21  1.11c  0.22a 0.04  0.16a 0.03

262 2662

361 4216

308 3522

SBA¼ Skilled birth attendant. ANC_4¼At least four antenatal care visits. ANC_d ¼ antenatal care starts in the first trimester. The models estimated are the same as those in Table 2. The shadow shows the variable used as mediation. The table presents the results of this variable but also the direct effect and indirect effect of the contextual gender norms regarding violence against women. a b c

Significance at 10% At 5 %. At 1 %.

study highlights the need for a new approach to conceptualization, measurement and methodology in order to gain better insights into the process of women's autonomy and its effects in Africa (Blanc, 2001; Schatz and Williams, 2012; Upadhyay et al., 2014).

Appendix A. Specification of the multilevel structural equations modelling The multilevel structural equations model used in this study includes a single latent variable at the cluster level, the one related to gender norms favoring violence against women (ηÞ. In matrix form, this model is summarised in a series of equations linking the different levels of analysis n ð1a : level1Þ Y ij ¼ αj þ Β1 Y ij þ Γ1 X ij þ ζ ij 8 2 2 1 3 > > α > j > 60 > 4 > > : ηj 0 

X k ¼ Λk η þ εk

0

β1j

0

β2j

0

0

3

2 1 3 αj 7 7 6 2 7 7  6 α j 7 þ Γ2 Z j þ ζ j 7 4 5 5

ð1b : level2Þ

ηj

ð1c : level2Þ

where (ζ ij ; ζ j ; εk Þ are the random variations with null covariance of the different equations; i refers to the woman; j, to the cluster; and k (¼ 5), to the number of indicators in the measurement models.

Eq. (1a) expresses the relationship between the independent variables (Xij) measured at the individual level and the two dependent variables (Yij): the use of maternal health service and women’s decision-making authority. The unidirectional relationship (from decision-making authority to use of healthcare services) between the two dependent variables is expressed by the coefficients of the square matrix B1(2x2), whose first row is null. For this reason, Eq. (1a) describes a recursive model which is, by nature, always identifiable (Bollen, 1989). In this Equation 1, we assume that the constants (one for each dependent variable, α1j and α2j ) vary from one group (cluster) to another: in multilevel terminology, this is a random intercept model. Eq. (1b) describes the relationship at the cluster level (level 2) linking these random variables with the contextual variables. The latter are of two types: the observable contextual variables Zj (Table 1) and the latent variable ηj that describes the gender norms regarding violence against women. The observable contextual variables, like the indicators for the latent variable, were calculated based on the responses provided by all women aged 15–49 years in each cluster. The β coefficients and those of the matrix Г2 with dimension (3, p), with 0 for the last row and p being the number of contextual explanatory variables, are the parameters to be estimated. Finally, Eq. (1c) expresses the relationship at the contextual level between the latent variable (ηÞ and its indicators (X k Þ. This is the measurement model that was tested separately before being introduced into the overall model. In our case, we have only one

V. Adjiwanou, T. LeGrand / Health & Place 29 (2014) 67–78

latent variable (gender norms) measured by five indicators related to women's attitude toward wife beating. The lambda matrix of coefficients (Λk) represents the saturations. The system of equations was estimated with Mplus version 6.11 along with the estimator MLR. Fig. 1 in the text shows the simplified pathways of this model. In the case of two-level logistic structural equation modelling, the system of equations can be summarised by   π 1ij 1 1 (1) logitðπ 1ij Þ ¼ log ¼ β00 þ β 1 DECMAK ij 1  π 1ij þ β k X kij þ β01 ηj þ β 0p Z pj þ ϵ1j 1

1

(2) logitðπ 2ij Þ ¼ log

1





π 2ij 2 2 2 2 ¼ β00 þ βk X kij þ β01 ηj þ β 0p Z pj þ ϵ2j 1  π 2ij

(3) X k ¼ Λk η þ εk

where π 1ij and π 2ij are, respectively, the probability that woman i of cluster j used maternal health service for her last pregnancy, or is of low decision-making authority (DECMAK). References Adjiwanou, Vissého, 2013. Contexte de genre, autonomie de la femme et utilisation des services de santé maternelle en milieu rural africain: Analyse par modèles d’équations structurelles. Université de Montréal, Montréal. Adjiwanou, Vissého, LeGrand, Thomas, 2013. Does antenatal care matter in the use of skilled birth attendance in rural Africa: a multi-country analysis. Soc. Sci. Med. 86 (0), 26–34. Agarwala, Rina, Lynch, Scott M., 2006. Refining the measurement of women's autonomy: an international application of a multi-dimensional construct. Soc. Forces 84 (4), 2077–2098. Ahmed, Saifuddin, Creanga, Andreea A., Gillespie, Duff G., Tsui, Amy Ong., 2010. Economic status, education and empowerment: implications for maternal health service utilisation in developing countries. PLoS One 5 (6), e11190. Amoakohene, Margaret Ivy, 2004. Violence against women in Ghana: a look at women's perceptions and review of policy and social responses. Soc. Sci. Med. 59 (11), 2373–2385. Balk, Deborah, 1994. Individual and community aspects of women's status and fertility in rural Bangladesh. Popul. Stud. 48 (1), 21–45. Balk, Deborah, 1997. Defying gender norms in rural Bangladesh: a social demographic analysis. Popul. Stud. 51 (2), 153–172. Basu, Alaka Malwade, 1990. Cultural influences on health care use: two regional groups in India. Stud. Family Plan. 21 (5), 275–286. Beegle Kathleen, Frankenberg Elizabeth, Duncan Thomas, 2001. Bargaining power within couples and use of prenatal and delivery care in Indonesia. Stud. Fam. Plann. 32 (2), 130–146. Beninguisse, Gervais., (Ed.) Université Catholique de Louvain. 2003. Entre tadition et modernité: Fondements sociaux de la prise en charge de la grossesse et de l'accouchement au Cameroun. Louvain-la-Neuve: Bruylant-Academia s.a. Beninguisse, Gervais, Nikiema, Béatrice, Fournier, Pierre, Haddad, Slim, 2005. L'accessibilité culturelle: une exigence de la qualité des services et soins obstétricaux en Afrique. Afr. Popul. Stud. 19, 243–266. Bhatia, J.C., Cleland, J., 1995. Determinants of maternal care in South India. Health Trans. Rev. 2, 127–142. Blanc, Ann K., 2001. The effect of power in sexual relationships on sexual and reproductive health: an examination of the evidence. Stud. Fam. Plann. 32 (3), 189–213. Bloom, Shelah, Wypij, David, Das Gupta, Monica, 2001. Dimensions of women's autonomy and the influence on maternal health care utilisation in a North Indian city. Demography 38 (1), 67–78. Bollen, Kenneth A., 1989. In: Statistics, W.S. i. P. a. M. (Ed.), Structural Equations with Latent Variables. Wiley Series in Probability and Mathematical Statistics, New York. Caldwell, John C., 1990. Cultural and social factors influencing mortality levels in developing countries. Ann. Am. Acad. Political Soc. Sci. 510 (1), 44–59. Caldwell, John C., Caldwell, Pat, 1993. Women's Position and Child Mortality and Morbidity in Less developed Countries. In: Federici, N., Mason, K.O., Sogner, S. (Eds.), Women's Position and Demographic Change. Clarendon Press, Oxford, pp. 122–139.

77

Campbell, Oona M R, Graham, Wendy J., 2006. Strategies for reducing maternal mortality: getting on with what works. Lancet 386, 1284–1299. Carter, Alexandra. 2010. Factors That Contribute to the Low Uptake of Skilled Care During Delivery in Malindi, Kenya. ISP Collection. Paper 821. Corroon, M., Speizer, I.S., Fotso, J.C., Akiode, A., Saad, A., Calhoun, L., et al., 2013. The role of gender empowerment on reproductive health outcomes in urban Nigeria. Matern. Child Health J., 1–9. DeRose, L.F., Ezeh, Alex C., 2010. Decision-making patterns and contraceptive use: evidence from Uganda. Popul. Res. Policy Rev. 29 (3), 423–439. Desai, Sonalde, Andrist, Lester, 2010. Gender scripts and age at marriage in India. Demography 47 (3), 667–687. Desai, Sonalde, Johnson, Kiersten, 2005. Women's Decisionmaking and Child Health: Familial and Social Hierarchies. In: Kishor, S. (Ed.), A Focus on Gender: Collected Papers on Gender Using DHS Data. Calverton. ORC Macro, MD, pp. 55–68. Dixon-Mueller, Ruth, Germain, Adrienne, 2000. Reproductive Health and the Demographic Imagination. In: Presser, H.B., Sen, G. (Eds.), Women's Empowerment and Demographic Processes. University Press, Oxford. Dodoo, F. Nii-Amoo, Frost, Ashley E., 2008. Gender in African population research: the fertility/reproductive health example. Ann. Rev. Sociol. 34 (1), 431–452. Fikree, Fariyal F., Pasha, Omrana, 2004. Role of gender in health disparity: the south asian context. Br. Med. J. 328 (3), 823–826. Fotso, Jean-Christophe, Ezeh, Alex, Essendi, Hildah, 2009. Maternal health in resource-poor urban settings: how does women's autonomy influence the utilisation of obstetric care services? Reprod. Health 6 (1), 9. Franzini, Luisa, Caughy, Margaret, Spears, William, Eugenia, Fernandez, Esquez, Maria, 2005. Neighborhood economic conditions, social processes, and selfrated health in low-income neighborhood in texas: a multilevel latent variables model. Soc. Sci. Med. 61, 1135–1150. Gabrysch, Sabine, Campbell, Oona, 2009. Still too far to walk: literature review of the determinants of delivery service use. BMC Pregnancy Childbirth 9 (1), 34. Gage, Anastasia J., 2007. Barriers to the utilisation of maternal health care in rural Mali. Soc. Sci. Med. 65 (8), 1666–1682. Gilfus, Mary, Trabold, Nicole, O'Brien, Patricia, Fleck-Henderson, Ann, 2010. Gender and intimate partner violence: evaluating the evidence. J. Soc. Work Educ. 46 (2), 245–263. Gottfredson, Nisha C., Panter, A.T., Daye, Charles E., Allen, Walter F., Wightman, Linda F., 2009. The effects of educational diversity in a national sample of law students: fitting multilevel latent variable models in data with categorical indicators. Multivar. Behav. Res. 44, 305–331. Hadley, Alexandra, Craig, Brewis, Pike, Ivy, 2010. Does less autonomy erode women's health? Yes. No. Maybe. Am. J. Hum. Biol. 22, 103–110. Hatcher, A.M., Romito, P., Odero, M., Bukusi, E.A., Onono, M., Turan, J.M., 2013. Social context and drivers of intimate partner violence in rural Kenya: Implications for the health of pregnant women. Cult. Health Sex. 15 (4), 404–419. Heaton, Tim B., Tina, J., Huntsman, Flake, Dallan F., 2005. The effects of status on women's autonomy in Bolivia, Peru, and Nicaragua. Popul. Res. Policy Rev. 24, 283–300. Heckert, Jessica, Fabic, Madeleine Short, 2013. Improving Data concerning women’s empowerment in Sub-Saharan Africa. Stud. Fam. Plann. 44 (3), 319–344. Heise, Lori L., 1998. Violence against women: an integrated, ecological framework. Violence Against Women 4 (3), 262–290. Horne, Christine, 2001. Sociological Perspectives on the Emergence of Social Norms. In: Hechter, M., Opp, K.-D. (Eds.), Social Norms. Russel Sage Foundation, New York, pp. 3–34. Islam, M., Yoshida, S., 2009. MDG5 : how close are we to success? BJOG Int. J. Obstet. Gynaecol. 116 (Suppl. 1), 2–5. Kishor, Sunita, Subaiya, Lekha, 2008. Understanding Women's Empowerment : A Comparative Analysis of Demographic and Health Surveys (DHS) Data. Macro International, Calverton, Maryland, USA. Kiss, Ligia, Schraiber, Lilia Blima, Heise, Lori, Zimmerman, Cathy, Gouveia, Nelson, Watts, Charlotte, 2012. Gender-based violence and socioeconomic inequalities: does living in more deprived neighbourhoods increase Women's risk of intimate partner violence? Soc. Sci. Med. 74 (8), 1172–1179. Koblinsky, Marge, 1995. On the pathway to maternal health. Results from Indonesia. Mother Care Matters Newsl. 5, 1–2. Koenig, Michael A., Stephenson, Rob, Saifuddin, Ahmed, Jejeebhoy, Shireen J., Campbell, Jacquelyn, 2006. Individual and contextual determinants of domestic violence in North India. Am. J. Public Health 96, 1. Kritz, Mary M, Makinwa-Adebusoye, Paulina, 1999. Determinants of women's decisionmaking authority in Nigeria: the ethnic dimension. Sociol. Forum 14 (3), 399–424. Kritz, Mary M, Makinwa-Adebusoye, Paulina, Gurak, Douglas T, 2000. The Role of Gender Context in Shaping Reproductive Behaviour in Nigeria. In: Presser, H.B., Sen, G. (Eds.), Women's Empowerment and Demographic Processes. University Press, Oxford, pp. 239–260. Lawn, Joy E., Simon Cousens, Jelka Zupan, 2005. 4 million neonatal deaths: When? Where? Why? The Lancet 365 (9462), 891–900. Lüdtke, Oliver, Herbert W. Marsh, Alexander Robitzsch, Ulrich Trautwein, Tihomir Asparouhov, L.K. and Muthén Muthén, B.O., et al. 2007. The multilevel latent covariate model: a new, More Reliable Approach to Group-Level Effects in Contextual Studies. 1–69. Malhotra, Anju, Schuler Sydney Ruth, 2005. Measuring Women's Empowerment as a Variable in International Development. In: Narayan, D. (Ed.), Measuring Empowerment: Cross-disciplinary Perspectives. World Bank Publications, Washington, DC, pp. 71–88.

78

V. Adjiwanou, T. LeGrand / Health & Place 29 (2014) 67–78

Mason, Karen Oppenheim, 1987. The impact of women's social position on fertility in developing countries. Sociol. Forum 2 (4), 718–745. Montgomery, Mark R., Hewett, Paul. C., 2005. Urban poverty and health in developing countries: household and neighborhood effects. Demography 42 (3), 397–425. Moore, Melissa, 1999. Reproductive health and intimate partner violence. Fam. Plann. Perspective 31 (302–306), 312. Moursund, Anne, Kravdal, Oystein, 2003. Individual and community effects of women's education and autonomy on contraceptive use in India. Popul. Stud. 57 (3), 285–301. Mullany, Britta C., Hindin, Michelle J., Becker, Stan, 2005. Can women's autonomy impede male involvement in pregnancy health in Katmandu, Nepal? Soc. Sci. Med. 61, 1993–2006. Mumtaz, Zubia, Salway, Sarah, 2005. 'I never go anywhere': extricating the links between women's mobility and uptake of reproductive health services in Pakistan. Soc. Sci. Med. 60 (8), 1751–1765. Muthén, Linda, Muthén, Bengt (Eds.), 1998–2010. Mplus User's Guide, Sixth Ed. Muthén & Muthén, Los Angeles, CA. Nanda, Geeta, Schuler, Sidney Ruth, Lenzi, Rachel, 2013. The influence of gender attitudes on contraceptive use in Tanzania: new evidence using husbands' and wives' survey data. J. Biosoc. Sci. firstview, 1–14. OECD, 2010. Atlas of Gender and Development: How Social Norms Affect Gender Equality in non-OECD Countries. OECD Publishing (Available /content/book/ 9789264077478-en). Pallikadavath, Saseendran, Foss, Mary, Stones, William R., 2004. Antenatal care: provision and inequality in rural North India. Soc. Sci. Med. 59, 1147–1158. Pullum, Thomas W., 2008. An Assessment of the Quality of Data on Health and Nutrition in the DHS Surveys, 1993–2003. Methodological Reports No. 6. Macro International Inc, Calverton, Maryland, USA. Rani, Manju, Bonu, Sekhar, Diop-Sidibe, Nafissatou, 2004. An empirical investigation of attitudes towards wife-beating among men and women in seven subSaharan African Countries. Afr. J. Reprod. Health 8 (3), 116–136. Raudenbush, Stephen, 2003. The Quantitative Assessment of neighborhood Social Environments. In: Kawachi, I., Berkman, L.F. (Eds.), Neighborhoods and Health. Oxford University Press, Oxford, pp. 113–131 (Chapter 5). Riley, Nancy E., 1998. Research on gender in demography: limitations and constraints. Popul. Res. Policy Rev. 17 (6), 521–538. Sampson, Robert J., 2003. Neighborhood-Level Context and Health: Lessons from Sociology. In: Kawachi, I., Berkman, L.F. (Eds.), Neighborhoods and Health. Oxford University Press, New York. Sampson, Robert J., Morenoff, Jeffrey D., Gannon-Rowley, Thomas, 2002. Assessing “neighborhood effects”: social processes and new directions in research. Ann. Rev. Sociol. 28, 443–478. Sarkar, N.N., 2008. The impact of intimate partner violence on women's reproductive health and pregnancy outcome. J. Obstet. Gynaecol. 28 (3), 266–271.

Schatz, Enid, Williams, Jill, 2012. Measuring gender and reproductive health in Africa using demographic and health surveys: the need for mixed-methods research. Cult. Health Sex. 14 (7), 811–826. Shroff, Monal, Griffiths, Paula, Adair, Linda, Suchindran, Chirayath, Bentley, Margaret, 2009. Maternal autonomy is inversely related to child stunting in Andhra Pradesh, India. Mater. Child Nutrit. 5, 64–74. Simkhada, Bibha, van Teijlingen, Edwin R, Porter, Maureen, Simkhada, Padam, 2008. Factors affecting the utilisation of antenatal care in developing countries: systematic review of the literature. J. Adv. Nurs. 61 (3), 244–260. Singh, Kavita, Shelah Bloom, and Paul Brodish. 2013. Gender equality as a means to improve maternal and child health in Africa, Health Care for Women International: null-null. Stephenson, R., Bartel, D., Rubardt, M., 2012. Constructs of power and equity and their association with contraceptive use among men and women in rural Ethiopia and Kenya. Global Public Health 7 (6), 618–634. Stephenson, Rob, Baschieri, A., Clements, S., Hennink, M., Madise, N., 2006a. Contextual influences on the use of health facilities for childbirth in Africa. Am. J. Public Health 96 (1), 84–93. Stephenson, Rob, Koenig, Michael A., Ahmed, Saifuddin, 2006b. Domestic violence and contraceptive adoption in Uttar Pradesh, India. Stud. Fam. Plan. 37 (2), 75–86. Subramanian, S.V., Özaltin, Emre, Finlay, Jocelyn E., 2011. Height of nations: a socioeconomic analysis of cohort differences and patterns among women in 54 low- to middle-income countries. PLoS One 6, 4. The Partnership for Maternal, Newborn and Child Health. 2006. Lawn, J. Kerber, K. (Ed.) Opportunities for Africa's newborns: practical data, policy and programmatic support for newborn care in Africa WHO on behalf of The Partnership for Maternal Newborn and Child Health. Upadhyay, Ushma D., Dworkin, Shari L., Weitz, Tracy A., Diana, Greene Foster, 2014. Development and validation of a reproductive autonomy scale. Stud. Fam. Plan. 45 (1), 19–41. Wang, Guang-Zhen, 2010. Reproductive Health and Gender Equality: Methods, Measurement, and Implications. Ashgate, Farnham, England. Watts, Charlotte, Zimmerman, Cathy, 2002. Violence against women: global scope and magnitude. Lancet 359, 1232–1237. WHO. 2009. WHO Global Database on Maternal Health Indicators, 2009 update. Geneva. Wilson-Williams, Lyndsey, Stephenson, Rob, Juvekar, Sanjay, Andes, Karen, 2008. Domestic violence and contraceptive use in a rural Indian village. Violence Against Women 14 (10), 1181–1198. Woldemicael, Gebremariam, 2009. Women's autonomy and reproductive preferences in Eritrea. J. Biosoc. Sci. 41, 161–181. Yount, KathrynM, Halim, Nafisa, Schuler, SidneyRuth, Head, Sara, 2013. A survey experiment of women’s attitudes about intimate partner violence against women in Rural Bangladesh. Demography 50 (1), 333–357.