Advertisement

Mapping and Modelling of the FGM Prevalence

  • Ngianga-Bakwin Kandala
  • Paul Nzinga Komba
Chapter

Abstract

The previous chapter provided a state-of-the-art evidence synthesis of all existing databases for FGM/C around the world. It provided accurate and quantifiable estimates of the trends within and between regions. Countries that have the biggest influence on the changes within the regions have been identified. Besides, we mapped FGM/C prevalence using the advanced spatial statistical approach. This was critically important in a bid to shed light on some unique spatial features that may advance our knowledge and understanding of the dynamics behind the FGM/C practice. This chapter provides a specific analysis on accurate and quantifiable estimates on FGM/C both at the global and in-country levels for selected states. The reason for such a move is due to the fact that previous studies have observed huge in-country variations in FGM trends without unearthing national prevalence (Kandala et al. 2009; Kandala and Komba 2015). We also highlight the spatial disparities of FGM/C within countries at the relevant sub-national levels, to the extent, that these mask the prevalence that could be compared within the framework of efforts to eradicate the practice.

3.1 Introduction

The previous chapter provided a state-of-the-art evidence synthesis of all existing databases for FGM/C around the world. It offered accurate and quantifiable estimates of the trends within and between regions. Countries that have the biggest influence on the changes within the regions have been identified. Besides, we mapped FGM/C prevalence using the advanced spatial statistical approach. This was critically important in a bid to shed light on some unique spatial features that may advance our knowledge and understanding of the dynamics behind the FGM/C practice. This chapter provides a specific analysis on accurate and quantifiable estimates on FGM/C both at the global and in-country levels for selected states. The reason for such a move is due to the fact that previous studies have observed huge in-country variations in FGM trends without unearthing national prevalence (Kandala et al. 2009; Kandala and Komba 2015). We also highlight the spatial disparities of FGM/C within countries at the relevant sub-national levels, to the extent, that these mask the prevalence that could be compared within the framework of efforts to eradicate the practice.

This chapter is structured as follows. Section 3.2 provides a thumbnail description of FGM/C in Africa. Section 3.3 maps the risks associated with FGM/C in four selected countries—Egypt, Nigeria, Senegal, Central African Republic (C.A.R) before mapping the prevalence using a Bayesian geo-additive approach (Sect. 3.4).

3.2 FGM/C Description

As noted in Chap.  2, FGM/C is predominantly practised in northern sub-Saharan African countries; in the Sahel region, the horn of Africa and Egypt, but it is also found outside Africa e.g. amongst women and families migrating to European countries and the US from these African locations. It is estimated that between 100–140 million women are thought to have undergone FGM worldwide and three million girls annually are thought to be at risk. FGM varies from a more or less ritual and symbolic genital cutting to levels of severity, including removal of the clitoris; and/or removal of the labia minora or infibulation (abrasion and stitching together) of the labia majora. This latter procedure is elaborate and leaves a small hole after which a small hole is left through which urination and sexual activity can occur. This last form requires reopening (and re-closure) for childbirth. It is worth noting that FGM is carried out on girls at different ages ranging from babies and toddlers to teenagers. Traditional practitioners frequently carry it out under unsafe hygienic conditions. This is both the result of its traditional form and its illegality in many places means that it is conducted in such conditions. Complications can include immediate urinary and genital tract infection, pain and haemorrhage, complications in childbirth and social, psychological and sexual complications. The public health burden of FGM includes both consequences for women or daughter mortality and ongoing morbidity concerns throughout their life span.

The Demographic Health Surveys (DHS) are periodic cross-sectional health surveys funded by USAID (the U.S. Agency for International Development’s) Bureau for Global Health. The DHS includes a number of modules on demographics and household affluence; fertility, reproductive health, maternal and child health; nutrition, and knowledge and practice related to HIV/AIDS. Surveys allow for an optional additional series of questions about FGM. Since the year 2000, when these optional questions were added as part of the DHS, successive waves of these surveys have been conducted in 27 countries in SSA ranging from six FGM/C surveys in Nigeria and Egypt to one in Cameroon with a prevalence of FGM/C among women ranging from 1% in Cameroon in 2004 to 92.3% in Egypt in 2014. In Egypt, the most recent survey shows a small (3%) decrease in prevalence of FGM/C from 95.5% to just over 92.3% in 2014. Prevalence data from Nigeria shows no apparent change from the prevalence of 25.1% in 1999 to 24.8% in 2013. Another valuable source of data on FGM/C includes since 2000 UNICEF Multiple Indicator Cluster Surveys (MICS). This source uses a similar module to collect information on FGM in selected countries (Yoder et al. 2004; Creel 2001).

We used surveys from four countries—namely Nigeria, Egypt, Senegal, and the Central Africa Republic—to investigate individual, household, social, community and spatial factors associated with FGM/C in those countries. These countries were selected based on the results of a state-of-the-art evidence synthesis of all existing databases of FGM/C around the world and based on this analysis; it provided analytical details showing accurate and quantifiable estimates of the trends within and between regions. Within the regions, the countries that have the biggest influence on the changes were identified to map FGM/C using the advanced statistical approach of spatial statistics to highlight some unique spatial features that will further increase our knowledge and understanding of the dynamics of FGM/C practice. In clear terms, Nigeria (West Africa) was chosen as a prototype of countries where the prevalence of FGM/C was increasing and Egypt (North Africa), Senegal (West Africa) and Central African Republic (Central Africa) were chosen as countries where the prevalence of FGM/C was decreasing, even though such decrease remains insignificant.

3.3 The Mapping of FGM/C Risks in Outline

Before undertaking a country-by-country mapping and modelling of risks associated with FGM, we first explain the methods, prevalence data and the need for modeling and FGM/C risk mapping generally. Secondly, we consider the role of spatial factors and spatial modeling in this context. Thirdly, we highlight the benefit of pursuing a spatial analysis approach to FGM/C risk in Egypt, Nigeria, Senegal and the Central African Republic.

It must be noted that risk mapping is premised on the notion that national household surveys are designed to be precise at national and regional levels and rarely at lower levels such as districts. In order to provide district level estimates of an outcome of interest and achieve low precision values we need to simply aggregate survey data. These district level estimates are more important to planners as they assist in the acceleration of policy interventions, optimising inputs and improving coverage of health interventions. To handle the problem of making reliable estimates of a variable at these areal unities in circumstances where there is insufficient information for reaching valid estimates, social scientists have resource to the so-called Small Area Estimation (SAE) methods. For our purpose, we have used hierarchical spatial and temporal SAE techniques with a fully Bayesian geo-additive regression approach (Fahrmeir and Lang 2001; Kammann and Wand 2003) to estimate girls’ FGM/C prevalence by region and county in selected countries. The data used were the household sample survey data assembled at cluster and district levels and the reported FGM/C data to prediction years, resulting in a set of spatial and temporal neighborhood information and ancillary covariate data including individual, household and community factors. Here we made a prediction of FGM/C in all age groups that now represents the important indicator for zero tolerance and necessary when computing likely impacts on FGM/C practice. Age and related covariates correction was achieved for the DHS of countries selected.

To model FGM/C prevalence a fully Bayesian geo-additive regression approach (Fahrmeir and Lang 2001; Kammann and Wand 2003) was applied. Details of model procedures are presented below. In brief, the Bayesian model was based on Markov priors and used Markov Chain Monte Carlo (MCMC) techniques for inference and model checking. For model choice, the Deviance Information Criterion (DIC), a measure of fit and model complexity, was used (Spiegelhalter et al. 2002). The analysis was carried out using version 2.0.1 of the BayesX software package (Lang and Brezger 2004), which permits Bayesian inference based on MCMC simulation techniques. Multivariate analysis was used to evaluate the significance of the predicted posterior mean determined for the non-linear effects and spatial effects. A sensitivity analysis for the choice of priors was implemented. Standard choices for the hyper-parameters based on the Jeffrey’s Non-informative prior were selected and the sensitivity for the spatial effects (Markov random field prior and geo-spline) was investigated. The post estimation command “predict” was used to predict mean FGM/C prevalence for the different years by county or district.

Two model forms were finally selected: (a) an unadjusted model without covariates and (b) a fully adjusted model with all covariates geo-spline model and weights applied as inverse proportional to the distance of the centroids of neighbouring districts (model descriptions and accuracy metrics shown below). The data-driven, modelled predictions of the proportions of all age groups with FGM/C are shown in form of maps. Sensitivity of the uncertainty associated with sub-national level (district level) predictions is also investigated in shown in maps as standard deviations of predicted means.

The findings of the global review presented in Chap.  2 provide an important contribution to global prevalence and the current apparent decline observed in FGM/C worldwide. It revealed that the prevalence of FGM/C varied significantly across countries and geographical regions in different time periods. In the African region, the prevalence is highest in North Africa with Egypt leading with Egypt 95.10% (93.46–96.75) compared to other regions. This is followed by East Africa with a pooled estimate of 57.30% (55.70–58.89). The prevalence is particularly low in Central Africa. There is also a huge variation in the pooled estimates of FGM/C within regions. The prevalence of FGM/C appeared to be less than 50%, in the West African sub-region, however, the low estimates from Ghana, Togo and Niger with pooled individual estimates of less than 6% may have diminished the sub-regional burden driven by Guinea 97.80% (95.33–98.75), Sierra Leone, 90.80% (88.35–93.25), Mali 88.88% (84.82–92.94) and a few others including Nigeria with a very high population base and geographical mobility. These rates may be consistent with the influence of different predictors including socioeconomic and cultural factors acting at individual, family and tertiary levels.

Moreover, there are also huge within country variations of FGM/C observed in previous studies; but the national prevalence rates produced by these studies tend to hide regional variations within a country (Kandala et al. 2009; Kandala and Komba 2015). We revisit the DHS data, taking into account many confounding factors of FGM/C at the individual, family, household and community levels. Such factors appear to have been overlooked by various reports on FGM/C that only present the descriptive analysis of the practice. Our examination of these reports adds fresh evidence that ought to guide all stakeholders. Put another way, our sense is that most reports relating to successes recorded in the fight against FGM practices are already in the public domain. However, these reports fall far too short of sound analytical rigour and consistency. Specifically, the state-of-the-art paper, which relied on DHS/MICS data, was largely descriptive and over-interpretative. This has resulted in the misleading claim that FGM/C has decreased significantly worldwide. Our efforts have been to revisit such a claim and proffer specific analysis on the accurate and quantifiable estimates of FGM/C globally and within country for selected countries to show the spatial disparities of FGM/C within a country at the relevant sub-national level that national prevalence masks that could be compared with the effort made to eliminate the practice.

3.3.1 FGM/C Prevalence Data, Modelling and Risk Mapping

The key data is generated by Demographic Health Surveys (DHS, which are periodic cross-sectional health surveys funded by USAID (the U.S. Agency for International Development’s) Bureau for Global Health. The DHS includes a number of modules on demographics and household affluence; fertility, reproductive health, maternal and child health; nutrition and knowledge and practice related to HIV/AIDS (DHS, 1990–2004). In these surveys, an optional additional series of questions about FGM/C were introduced in some countries (Yoder et al. 2004; Creel 2001).

The history of the development of FGM/C questionnaire has been discussed elsewhere (Yoder et al. 2004; Creel 2001). Briefly, DHS surveys collect data from nationally representative probability samples of households and from adult women (age 15–49) and men in the sampled households. The module of FGM/C questions is added to the women’s questionnaire. The questions are designed to generate information on prevalence rates and types of FGM/C of the women themselves and their daughters. Respondents’ attitudes towards FGM are also collected. Since 2000, UNICEF’s Multiple Indicator Cluster Surveys (MICS) have used a similar module to collect information on FGM/C in selected countries (Yoder et al. 2004; Creel 2001). Female respondents are asked if they have ever heard of FGM; then those who have heard of the practice are asked about their own experience of it. We use responses to these questions to calculate and model prevalence rates, trend of FGM/C at the disaggregated level of regions or states in Egypt, Central Africa Republic, Nigeria and Senegal using selected surveys based on the results of the global review in Chap.  3 as mentioned above.

With respect to the four selected countries, we analysed a sample of more than 50,000 women participating in the various successive DHS surveys since 2000. The sampling methods employed in these surveys are described in detail elsewhere in the various countries DHS reports. For instance, in the 2010–2011 Senegal DHS-MICS, a nationally representative cross-sectional survey (multistage stratified random sampling of households) of women of reproductive age (15–49 years) is selected. The resulting sample was representative of the underlying populations of the different regions of Senegal. It was carried out between October 2010 and April 2011. A nationally representative sample of 15,688 women aged between 15 and 49 years in all selected households and 4929 men aged 15–59 years in one third of selected households were interviewed with a similar number for the previous survey of 2005 (14,602 women). Response rates were over 93% for individual women and over 87% at the household level and informed consent was obtained from participants. The study protocol conforms to the ethical guidelines of the 1975 Declaration of Helsinki as reflected in a priori approval by the institution’s human research committee. The Ethics Committee of the National Statistical Office of Senegal granted ethical approval.

Datasets collected were representative at the national level, at the urban/rural level and at the level of each administrative region when tested using a number of socio-demographic and household indicators. For both surveys, there were few participants with missing data for FGM and other covariates; thus, data analysis on FGM was based on 14,602 women in 2005 and 14,228 women in the 2010 sample with a complete set of data.

Experts generally assume that women respond truthfully when asked about their own experience. If bias exists in some of the responses, it has not been documented. It is hypothetically possible that some women may not admit to having undergone FGM in countries where the practice has been forbidden, but no solid evidence of this has been found. The module on FGM included questions on whether the woman herself had undergone FGM; and whether if she had daughters they had also undergone the practice or whether she intended that they should. Therefore, we model the binary outcome whether a women has undergone FGM and to estimate the likelihood of a woman and daughter’s FGM in a given region of each country, while accounting for a number of potential covariates.

On the one hand, we studied FGM as the main outcome variable in terms of “whether a participant had had FGM performed on her”. This question was converted into a binary variable, with two categories defined as 1 if the participant was cut and 0 if the participant had no FGM performed on her. On the other hand, we sought to define the main exposure variable in the analysis. This was described as the “region of residence” (state of residence for Nigeria and region for Central Africa Republic, Senegal and Egypt), in addition to various control variables on socio-demographic factors potentially associated with FGM: sex, age, education level, wealth index, marital status, family size, place of residence (urban vs. rural). These factors were selected based on the literature and previous studies on FGM/C. The data presented here were first analysis in Kandala et al. (2009), Kandala and Komba (2015) and Kandala and Shell-Duncan (2017).

3.3.2 The Role of Spatial Factors and Spatial Modelling

Space, place and geography matter because FGM/C is practiced by people from all educational and socio-economic and religious backgrounds. To be sure, the practice is rampant amongst both Muslim and Christians living in rural and urban areas. Within most countries in Africa, a number of studies have found significant spatial patterns, separate from the effects of individual-level factors such as female education and demographic characteristics, which indicate associations with economic disadvantage and the remoteness of communities (Kandala et al. 2009). However, it is also possible to look further at community-level variables that might explain these spatial patterns (Kandala et al. 2009). A better understanding of the geographic patterns in FGM/C outcomes and their influences are necessary for programs and policies, which seek to improve the reproductive health outcomes of poor FGM/C women and girls.

The DHS data contain geographical or spatial information, such as the region, state or district of individuals in the study and the presence of non-linear effects for some covariates. In a novel approach, the spatial patterns of FGM/C and the possibly non-linear effects of other factors will be explored within a simultaneous, coherent regression framework, using a geo-additive, semi-parametric mixed model that simultaneously controls spatial dependence and possibly nonlinear or time-varying effects of covariates and the complex sampling design (more detail in statistical methods below). The spatial modelling of FGM/C will identify key drivers of factors associated with high risk groups who can be targeted for rapid intervention at district/province level to assess the potential impact of encouraging abandonment programs and upholding laws banning FGM/C.

Advanced statistical analysis using BayesX software will be undertaken and spatial mapping will be undertaken with a combination of Mapviewer, ArcView and BayesX. Historically, variations in FGM/C rates have been linked to individual socio-economic factors (such as education and religion). In contrast, geographical patterns other than rural/urban and province has been neglected. In this chapter, we begin with a simple analysis of geographical variation within Nigeria, Senegal, the Central Africa Republic and Egypt, followed by a more detailed analytical approach.

First, an analysis is carried out to document regions’ differences in the variations of the observed FGM/C rates. Maps of observed rates will be produced from the raw data. We also test the bivariate and multivariate associations of well-known socioeconomic correlates of FGM/C. From this initial analysis, we expect to identify the association between socio-economic and demographic factors and FGM/C while showing regions’ differences in the rates and variation across them in the correlates.

Second, we use flexible methods to model spatial determinants of FGM/C risks and to allocate these spatial effects to structured and unstructured (random) components. This draws on Bayesian geo-additive discrete-time survival methods of spatial statistics, taking advantage of advances in Geographic Information Systems (Hennerfeind et al. 2006; Fahrmeir and Lang 2001). The modelling of the two components is done jointly in one estimation procedure that thereby simultaneously identifies socioeconomic determinants and the spatial effects that are not explained by these socioeconomic determinants. In this way, we are able to identify regional patterns of FGM/C that are either influenced by left-out socioeconomic variables that have a clear spatial pattern or point to spatial processes that may possibly be related to cultural behaviours and their associated hazards.

We use several modelling strategies to map the observed prevalence of FGM/C and predict the prevalence based on current data. Model Based Geo-statistical (MBG) modelling is also used as these methods interpolate from observed measure of interest of known locations in space and time to provide predictions of quantities and the empirical estimates of their uncertainty at locations and times where data do not exist (Diggle and Ribeiro 2007). MBG methods fit the data where the spatial and temporal covariance is used to generate samples of the predicted posterior distribution from which point estimates and the uncertainty around these estimates are computed simultaneously using Bayesian inference (Chilés and Delfiner 1999; Diggle et al. 2002).

DHS Data are used within a Bayesian hierarchical space-time model, implemented through Bayesian software such as Integrated Nested Laplace Approximations (INLA) for inference (Rue et al. 2009) and BayesX (Belitz et al. 2012) to produce predictions of FGM/C.

3.3.3 Data Geo-Coding

Data geo-coding, defining a decimal longitude and latitude for each survey location in the selected countries were all examined in the selected DHS year. According to their spatial representation, data were classified as enumeration areas, individual villages, communities or a collection of communities within a definable area, corresponding to provinces or states in each selected country. Where possible we aimed to retain disaggregated enumeration area, “cluster” level data rather than data across a “wide-area”. Where data were reported across communities below the enumeration area we regarded these as too low a spatial resolution, with significant possible variation within the polygon of information to be included within the modeling phase. In practice, this was a difficult criterion to apply as most DHS surveys provide representative data at the region or state level not the lower sub-national level such as villages. More recent use of statistical modelling techniques (MBG) does enable a re-aggregation of households to sub-national level such as counties or districts with greater precision and is useful in maintaining 5 km grid criteria while combining clusters of small sample sizes in space. MBG also enable us to obtain estimates in each survey location where GPS coordinates were not available in space during the data collection period.

3.3.4 Why Spatial: Data Structure and Sample Design Are Complex

Nested data in survey studies is often the rule rather than the exception. Here the data structure is retrospective birth, health and survival information including FGM/C, typically about more than one child from each sampled woman. FGM/C, health and survival information of children are nested within families, the clustering of families living within districts/provinces. In fact, heterogeneity is often present and frequently the available predictor variables do not explain this heterogeneity sufficiently (see Draper 2000). With recent computational advances in statistics it is becoming increasingly straightforward to describe such heterogeneity with mixed survival models that employ unobserved predictors in a Bayesian hierarchical structure. The DHS samples are drawn through stratified clustered sampling. However, one cannot assume that the clusters selected in each province are fully representative of the country in which they are located, as the surveys only attempt to generate a fully representative sample at the province level.

Consequently, the spatial analysis will be affected by some random fluctuations. Some of this random variation can be reduced through the relaxation of the independence assumption between neighbouring provinces/districts. Such a spatial analysis should preferably be applied to census data, where the precision of the spatial analysis would be much higher. However, FGM/C information is not collected in census and census data are often not available for such analyses and household survey data are the most widely available resource for evaluating spatial variation of FGM/C in these countries.

We assess the likely impact of the neglect of hierarchical structure and geographical location in analyses of the DHS data that ignore auto-correlation structure, time-varying and nonlinear effects and dependence in the data. The neglect of the geographical location where the child lives leads to underestimation of standard errors of the fixed effects that inflates the apparent significance of the estimates (Kandala and Ghilagaber 2006; Kandala et al. 2006; Bolstad and Manda 2001). Our analysis includes this auto-correlation structure and account for the dependence of neighboring communities in the model. The model will also permit ‘borrowing strength’ from neighboring areas to obtain estimates for areas that may, on their own, have inadequate sample sizes. This gives more reliable estimates of the fixed effect standard error.

3.3.5 The Benefits of Spatial Analysis Approach

Previous research on FGM/C has been confined to the examination of socio-economic, demographic and health-related determinants in specific contexts. Furthermore, it has generally failed to incorporate spatial aspects. In these investigations that ignore the spatial dimension in the study of FGM/C, population-level socioeconomic variables and health resources have explained very little of the variation in FGM rates. On the other hand, it is well documented that aggregate levels of FGM/C rates in many developing countries mask spatial variations and that understanding these spatial patterns may lead to identification of other important determinants of child health. More importantly, FGM/C is a social norm and often leads to concentration of the population in specific regions and ethnic groups; hence spatial patterns in FGM/C are an important contribution to information for programming and policy.

Several methodological shortcomings such as exclusion of geographical location, auto-correlation in the data, non-linear and time varying effects of covariates, and small samples characterise previous work in the field of FGM. Consequently, it is doubtful whether the findings for that work can be generalised. Specifically, these studies relied on the independence of random components at the contextual level (districts or province). Most of these studies also based their conclusions on limited statistical analysis, neglecting to control for factors that may significantly affect child survival, such as physical environment where children live and the potential impact of the geographical location where the child lives. Finally, many of the findings represented in these studies provide national statistics; they cannot be extrapolated for a particular district. In our research, we will address these shortcomings by linking the rates of FGM/C with geographical locations, by using all FGM/C cases processed in each district/province of these countries, and by accounting for the influence of such important factors as non-linear and time-varying effects of covariates, dependence of random components and geographic location on outcomes.

Our approach presents huge advantages compared to existing efforts which rely on, say, logistic models with constant-fixed effects of covariates and fixed (or random) districts (provinces) effects or standard two-level multilevel modelling with unstructured spatial effects. With existing models, it is assumed that the random components at the contextual level (district/province) are mutually independent, even though, in practice, this assumption is not actually implied by these approaches, so correlated random residuals could also be specified (see Langford et al. 1999). Borgoni and Billari (2003) pointed out that the independence assumption has an inherent problem of inconsistency: if the location of the event matters, it makes sense to assume that areas close to each other are more similar than areas that are far apart. Also, the DHS/MICS data are based on random samples within provinces. That is, the structured component introduced here allows us to ‘borrow strength’ from neighbors in order to cope with the sample variation of the district or province effect and obtain estimates for areas that may have inadequate sample sizes or be un-sampled. We now seek to try several models in order to highlight the differences that can be found by adopting this approach in a spatial context and the possible bias involved with the violation of the independent assumption between aggregated spatial areas. Some of these have a spatial component and a random component that reflect spatial heterogeneity globally and relative homogeneity among neighbouring districts, while some will not. A failure to take into account the posterior uncertainty in the spatial location (district or province) would overestimate the precision of the prediction of FGM/C risks in un-sampled districts.

3.3.6 Statistical Analysis

3.3.6.1 Bayesian Geo-Additive Regression Models

Spatial analyses of FGM are often confined to using region-specific dummy variables to capture the spatial dimension. Here, we go a step further by exploring regional patterns of FGM and, possibly nonlinear, effects of other factors within a simultaneous, coherent regression framework using a geo-additive semi-parametric mixed model. Because the predictor contains usual linear terms, nonlinear effects of metrical covariates and geographic effects in additive form, such models are also called geo-additive models. Kammann and Wand (2003) proposed this type of model within an empirical Bayesian approach. Here, we apply a fully Bayesian approach as suggested in Fahrmeir and Lang (2001), which is based on Markov priors and uses MCMC techniques for inference and model checking.

Classical linear regression models of the form

$$ {y}_i={w}_i^{\prime}\gamma +{\varepsilon}_i,\kern0.875em {\varepsilon}_i\sim N\left(0,{\sigma}^2\right), $$
(1)
for observations (yi, wi),  i = 1, …, n, on a response variable y and a vector w of covariates assume that the mean E(yi|wi) can be modeled through a linear predictor \( {w}_i^{\prime}\gamma \). In our application to FGM and in many other regression situations, we are facing the following problems: First, for the continuous covariates in the data set, the assumption of a strictly linear effect on the response y may not be appropriate. In our study, such covariate is the respondent’s age. Generally, it will be difficult to model the possibly nonlinear effect of such covariates through a parametric functional form, which has to be linear in the parameters, prior to any data analysis.

Second, in addition to usual covariates, geographical small-area information was given in form of a location variable, indicating the region, department or community where individuals or units in the sample size live or come from. In our study, the regions in Senegal, for example, give this geographical information. Attempts to include such small-area information using region/department-specific dummy-variables would in our case entail more than 30 dummy-variables for the departments and nine dummies for the regions and using this approach we would not assess spatial inter-dependence. The latter problem also can’t be resolved through conventional multilevel modeling using uncorrelated random effects (Kandala et al. 2009). It is reasonable to assume that areas close to each other are more similar than areas far apart, so that spatially correlated random effects are required.

To overcome these difficulties, we replace the strictly linear predictor,

$$ {\eta}_i={x}^{\prime}\beta +{w}_i^{\prime}\gamma +{\varepsilon}_i $$
with a logit link function with dynamic and spatial effects, Pr(yi = 1|ηi)=\( {e}^{\eta_i}/\left(1+{e}^{\eta_i}\right) \), and a geo-additive semi-parametric predictor μi = hi):
$$ {\eta}_{\mathrm{i}}={f}_1\left({x}_{i1}\right)+\dots +{f}_p\left({x}_{ip}\right)+{f}_{spat}\left({s}_i\right)+{w}_i^{\prime }\ \gamma +{\varepsilon}_{\mathrm{i}} $$
(2)
Where h is a known response function with a logit link function, f1, …, fp are non-linear smoothed effects of the metrical covariates (respondent’s age), and fspat (si) is the effect of the spatial covariate si ∈ {1,...,S} labelling the region in Senegal. Covariates in \( {w}_i^{\prime } \) are usual categorical variables such as gender and urban-rural residence. Regression models with predictors as in (2) are sometimes referred to as geo-additive models. The observation model (2) may be extended by including interaction f(x)w between a continuous covariate x and a binary component of w, say, leading to so called varying coefficient models, or by adding a nonlinear interaction f1,2 (x1, x2) of two continuous covariates.

In a further step, we may split up the spatial effect fspat into a spatially correlated (structured) and an uncorrelated (unstructured) effect

$$ {f}_{spat}\left({s}_i\right)={f}_{str}\left({s}_i\right)\ {f}_{untsr}\left({s}_i\right) $$

The rationale behind this is that a spatial effect is usually a surrogate for many unobserved influences, some of which may obey a strong spatial structure and others may be present only locally. By estimating a structured and an unstructured effect, we aim at separation between the two kinds of factors.

As a side effect, we are able to assess to some extent the amount of spatial dependency in the data by observing which one of the two effects is larger. If the unstructured effect exceeds the structured effect, the spatial dependency is smaller and vice versa. It should be noted that all functions are centred about zero for identification purpose, thus fixed effects parameters automatically include an intercept term γ0.

In a Bayesian approach, unknown functions fj and parameters γ as well as the variance parameter σ2 are considered random variables and have to be supplemented with appropriate prior assumptions. In the absence of any prior knowledge we assume independent diffuse priors γj ∝ const, j=1,...,r for the parameters of fixed effects. Another common choice is highly dispersed Gaussian priors.

Several alternatives are available as smoothness priors for the unknown functions fj (xj) (see Fahrmeir and Lang (2001), Fahrmeir, Kneib, and Lang (2004)). We use Bayesian P(enalized)—Splines, introduced by Eilers and Marx in a frequentist setting. It is assumed that an unknown smooth function fj (xj) can be approximated by a polynomial spline of low degree. The usual choices are cubic splines, which are twice continuously differentiable piecewise cubic polynomials defined for a grid of k equally spaced knot p on the relevant interval [a, b] of the x-axis. Such a spline can be written in terms of a linear combination B-spline basis functions Bm(x), i.e.
$$ f(x)=\sum \limits_{m=1}^l{\beta}_m{B}_m(x) $$
(3)

These basis functions have finite support on four neighbouring intervals of the grid, and are zero elsewhere. A comparably small number of knots (usually between 10 and 40) are chosen to ensure enough flexibility in combination with a roughness penalty based on second order difference of adjacent B-spline coefficients to guarantee sufficient smoothness of the fitted curves. In our Bayesian approach this corresponds to second order random walks

$$ {\beta}_m=2{\beta}_{m-1}-{\beta}_{m-2}+{u}_m, $$
(4)
with Gaussian errors um ∼ N(0, τ2). The variance parameter τ2 controls the amount of smoothness, and is also estimated from the data. More details on Bayesian P-Splines can be found in Lang and Brezger (2004). Note that random walks are the special case of B-Splines of degree zero.

We now turn our attention to the spatial effects fstr and funstr. For the spatially correlated effect fstr (s), s = 1, …, S, we choose Markov random field priors common in spatial statistics (Besag et al. 1991). These priors reflect spatial neighbourhood relationships. For geographical data one usually assumes that two sites or regions s and r are neighbours if they share a common boundary. Then a spatial extension of random walk models leads to the conditional, spatially autoregressive specification

$$ {f}_{str}(s)\mid {f}_{str}(r),r\ne s\sim N\left(\sum \limits_{r\in {\partial}_s}{f}_{str}(r)/{N}_s,{\tau}^2/{N}_s\right) $$
(5)
Where Ns is the number of adjacent regions, and r ∈ ∂s denotes that region r is a neighbour of region s. Thus the (conditional) mean of fstr(s) is an average of function evaluations fstr(s) of neighbouring regions. Again the varianceτ2str controls the degree of smoothness.

For a spatially uncorrelated (unstructured) effect funstr a common assumption is that the parameters funstr(s) are i.i.d. Gaussian

$$ {f}_{unstr}(s)\kern1em {\tau^2}_{unstr}\sim N\left(0,{\tau^2}_{unstr}\right) $$
(6)

Variance or smoothness parameters \( {\tau}_j^2 \), j = 1,...,p, str, unstr, are also considered as unknown and estimated simultaneously with corresponding unknown functions fj. Therefore, hyper-priors are assigned to them in a second stage of the hierarchy by highly dispersed inverse gamma distributions p(\( {\tau}_j^2 \)) ~ IG(aj, bj) with known hyper-parameters aj and bj. Standard choices for the hyperparameters are a = 1 and b = 0.005 or a = b = 0.001. Jeffrey’s noninformative prior is closer to the latter choice, and since practical experience shows that regression parameters depend on the choice of hyperparameters, we have investigated in our application the sensitivity to this choice.

Since some regions in Senegal do not have many neighbours, we have investigated the sensibility of the choice of Markov Random Field (MRF) priors with other priors supported by BayesX such as Gaussian random field (GRF) priors, but the resulting maps from the two priors did not differ much. Therefore, we considered the MRF prior for the spatial effects. For model choice, we routinely used the Deviance Information Criterion (DIC) developed in Spiegelhalter et al. (2002) as a measure of fit and model complexity. Before commenting on the substantive results, it is important to point out this model had the best fit after evaluation of the fit criteria using the Deviance Information Criteria (DIC).

The model assumes that f1( ), f2( ) and fstr are nonlinear effects and spatial effects were the same in all of the country. This was confirmed by prior separate analyses of the non-linear effects in other countries, which were found to be remarkably similar. The analysis was carried out using BayesX version 0.9, software for Bayesian inference based on Markov Chain Monte Carlo simulation techniques.

Quite clearly, the methods used here can help us to identify more subtle socioeconomic and spatial influences on FGM than reliance on linear models with regional dummy variables. As such, they are useful for diagnostic purposes to identify the need to find additional variables that can account for this spatial structure. Moreover, even if the causes of spatial structures are not fully explained, one can use this spatial information for campaigns to eliminate the practice of FGM and planning purposes, which is gaining increasing importance in policy circles that attempt to focus the allocation of public resources to the most at risk population.

Multivariate Bayesian geo-additive regression models were used to evaluate the significance of the POR determined for the fixed effects and spatial effects between prevalence of FGM in Senegal. Each factor was considered separately in unadjusted models using conventional logistic regression models. Next, fully adjusted multivariate Bayesian geo-additive regressions analyses were performed to look again for a statistically significant correlation between these variables, but this time further controlling for any influence from individual (age), ethnic, educational and religious factors. A P-value of <0.05 was considered indicative of a statistically significant difference.

3.3.7 Mapping Prevalence of FGM/C in Selected Countries

On the face of the preceding discussions, we now focus attention on the mapping of FGM/C prevalence in Egypt, Nigeria, Senegal and the Central Africa Republic. As will be noted below, these countries constitute contrasting examples of the shift or variation in the prevalence of FGM. In order to aid comparison and avert repetitions, an integrative approach is used.

3.3.7.1 A-Egypt

Let us first consider the mapping of FGM/C in Egypt, noting that the data analyzed in this section comes from four nationally-representative household surveys: the 1995 Egyptian Demographic and Health Survey (EDHS), The 2000 EDHS, the 2005 EDHS and the 2008 DHS survey with questions from the Multiple Indicator Survey. The core questionnaire for households collects data from adult women (age 15–49) and men from nationally representative probability samples of households (Yoder et al. 2004; Creel 2001).

The module on FGM/C includes three sections: (1) questions on whether the woman was circumcised or not and details about that event, (2) whether one daughter was circumcised or not and details about that event and (3) a woman’s opinion about the continuation of the practice. Since 2000, UNICEF’s Multiple Indicator Cluster Surveys (MICS) have used a similar module to collect information on FGM in selected countries (Yoder et al. 2004; Creel 2001). We draw on data from the core questionnaire for households, as well as the module on FGM/C, administered to women age 15–59 years. For the 2008 survey, the FGM/C module was modified to ask questions about the FGC status of each daughter under the age of 10, rather than about only one daughter, thus providing more detailed information about recent changes in the practice. We present findings of analyses of daughter data elsewhere (Kandala and Shell-Duncan, forthcoming). Further details on the methods, objectives, organization, sample design and questionnaires used in the 2008 EDHS and 1995–2008 EDHS are described elsewhere (Yoder et al. 2004; Creel 2001). The study protocol conforms to the ethical guidelines of the 1975 Declaration of Helsinki as reflected in a priori approval by the institution’s human research committee. The Ethics Committee of the National Statistical Office of Egypt granted ethical approval.

The sampling strategy for each survey was designed to be nationally representative enough to provide information for each governorate. A two-stage sampling process was employed. In the first stage clusters were selected from a list of enumeration areas with probability proportional to size. In the second stage, a complete household listing was completed in each selected cluster, followed by the random selection of households per cluster. In each household, all women aged 15–59 were interviewed. For instance, in 2008 survey data were collected from a total of 12,008 women aged 15–59. The collective 1995–2008 surveys collected data from 61,834 women aged 15–59. Further details on sampling can be found elsewhere in the final reports of each survey year. For all surveys, there were few participants with missing data for FGM/C and other covariates; thus, data analysis on FGM/C was based on 12,008 women in 2008 and 61,834 women in the 1995–2008 samples with a complete set of data.

We studied FGM as the main outcome in terms of “whether a participant had had FGM performed on her”. This question was converted into a binary variable, with two categories defined as 1 if the participant was cut and 0 if the participant had no FGM performed on her. The main exposure variable in the analysis was the “governorate of residence” (of which there are 27 in 2008 as shown in Figs. 3.2b, 3.3, and 3.4 in 2008), in addition to various control variables on socio-demographic factors potentially associated with FGM: sex, age, education level, marital status, place of residence (urban vs. rural). Age was recorded as a continuous variable and was re-coded into a categorical variable of 5-year age cohorts in the preliminary analysis. For the modelling of the prevalence of FGM, we examine the cohort’s effect of age of the respondent as a continuous variable using a flexible nonlinear function to estimate the age-related trend of FGM risk among women in the surveys. Education level was categorized as “None”, “Primary”, “Secondary” and “Higher”.

In terms of results about Egypt, the unweighted baseline socio-demographic characteristics are shown in Table 3.1, and by FGM/C status (whether circumcised or not) in Table 3.2. The overall prevalence of FGM/C differs slightly between the surveys (91.9% in 2008 and 96.1% in 1995–2008). Before investigating factors associated with FGM/C and trends across the four surveys, we examined comparability of women sampled in each survey. The survey populations are similar in terms of the mean ages of women (for 2008 the mean age was 32.4 years and in 1995–2008 the mean age of the sample was 33.1 years). Most of the population sampled lived in rural settings (56.5% in 2008 and 56.3% in 1995–2008) and 67.6% were married in 2008 with more at 87.6% in 1995–2008. A total of 43.6% of women in the 2008 population had a secondary education while 29.8% had no education compared to 35.3% with secondary education and 36.4% no education in 1995–2008. Women with FGM/C were mostly married (64.5% vs 87.3%), with mostly secondary education in 2008 (39.9%) and no education from 1995 to 2008 (38.4%), lived in rural areas (54.2% vs 55.6%), were living in urban Lower Egypt and urban Upper Egypt (31.3%, 22.6% vs 30.9%, 24.4%) in 2008 and 1995–2008 respectively. Notably, in all surveys, women’s age group is significantly associated with FGM/C. Thus, there is evidence of change in rates of FGM/C across age cohorts.
Table 3.1

Baseline characteristics of the study population of women (Egypt DHS, 1995–2008)a

Variable

N = 12,008 (2008)

N = 61,834 (1995–2008)

Mean ageb (SD) respondent

32.6

33.1

Circumcised (%)

 Yes

91.9

96.1

 No

8.1

3.9

Place of residence (%)

 Urban

44.0

43.7

 Rural

56.0

56.3

Married (%)

 Yes

63.6

87.6

 No

36.4

12.4

Education (%)

 No education

21.6

36.4

 Primary education

14.1

18.2

 Secondary education

47.9

35.3

 Higher education

16.4

10.0

Religion (%)

 Muslim

94.9

 

 Other

5.1

 

Governorates of residence (%)

 Matrouh

12.2

 

 Alexandria

6.5

 

 Beheira

0.8

 

 Kafr el-Sheikh

0.8

 

 Dakahlia

1.3

 

 Damietta

6.6

 

 Port Said

6.1

 

 North Sinai

5.8

 

 Gharbia

4.1

 

 Monufia

6.2

 

 Qalyubia

4.6

 

 Al Sharqia

7.6

 

 Ismailia

1.2

 

 Giza

8.7

 

 Faiyum

2.8

 

 Cairo

2.6

 

 Suez

5.6

 

 South Sinai

4.6

 

 Beni Suef

4.7

 

 Minya

3.6

 

 New Valley

1.6

 

 Asyut

0.5

 

 Red Sea

0.3

 

 Sohag

0.3

 

 Qena

0.5

 

 Luxor

0.3

 

 Aswan

0.1

 

Region of residence (%)

 Urban governorate

 

19.5

 Lower Egypt—urban

 

11.8

 Lower Egypt—rural

 

31.3

 Upper Egypt—urban

 

11.7

 Upper Egypt—rural

 

24.6

 Frontier governorate

 

1.2

Year of survey (%)

 1995

 

23.9

 2000

 

25.2

 2005

 

31.5

 2008

 

19.4

aData is expressed as a mean (standard deviation) or as percentages

bAge ranges from 18 to 97 years

Table 3.2

Baseline characteristics of the women study population by circumcision status (Egypt DHS, 1995–2008)a

Variable

2008

1995–2008: 13 years

Not circumcised (N=)

Circumcised (N=6571)

P-valueb

Not circumcised (N=)

Circumcised (N=)

P-valueb

Age group (%)

  

P < 0.001

  

p < 0.001

 15–19 years

19.6

80.5

 

9.6

90.4

 

 20–24 years

13.2

86.8

 

5.5

94.5

 

 25–29 years

8.0

92.0

 

4.9

95.1

 

 30–34 years

6.3

93.7

 

4.5

95.5

 

 35–39 years

5.1

94.9

 

4.0

96.0

 

 40–44 years

4.4

95.6

 

4.1

95.9

 

 45–49 years

4.1

95.9

 

3.3

96.7

 

 50–54 years

2.8

97.3

 

4.1

95.9

 

 55–59 years

3.4

96.6

 

3.5

96.5

 

Place of residence (%)

  

p < 0.001

   

 Urban

15.4

84.6

 

7.4

92.6

 

 Rural

4.7

95.3

 

2.7

97.3

 

Married (%)

  

p < 0.001

  

p < 0.001

 Yes

5.9

94.1

 

4.3

95.7

 

 No

15.8

84.2

 

8.4

91.6

 

Education (%)

  

p < 0.001

  

p < 0.001

 No education

4.2

95.8

 

3.3

96.7

 

 Primary education

4.8

95.2

 

2.0

98.0

 

 Secondary education

9.2

90.8

 

4.4

95.6

 

 Higher education

24.8

75.2

 

18.4

81.6

 

Religion (%)

  

p < 0.001

  

p < 0.001

 Muslim

6.9

88.0

 

71.0

29.0

 

 Other

1.2

3.9

 

91.0

9.0

 

Governorates of residence (%)

  

p < 0.001

   

 Matrouh

77.9

22.1

    

 Alexandria

14.4

85.6

    

 Beheira

7.5

92.5

    

 Kafr el-Sheikh

4.9

95.1

    

 Dakahlia

10.3

89.7

    

 Damietta

14.7

85.3

    

 Port Said

26.1

73.9

    

 North Sinai

29.4

70.6

    

 Gharbia

12.1

87.8

    

 Monufia

2.8

97.1

    

 Qalyubia

2.1

97.9

    

 Al Sharqia

0.6

99.4

    

 Ismailia

5.7

94.3

    

 Giza

11.0

89.0

    

 Faiyum

5.3

94.7

    

 Cairo

10.4

89.6

    

 Suez

9.6

90.4

    

 South Sinai

19.2

80.8

    

 Beni Suef

1.3

98.7

    

 Minya

11.7

88.3

    

 New Valley

5.3

94.7

    

 Asyut

7.1

92.9

    

 Red Sea

3.0

97.0

    

 Sohag

4.1

95.9

    

 Qena

1.0

99.0

    

 Luxor

4.1

95.9

    

 Aswan

1.5

98.5

    

Region of residence (%)

     

p < 0.001

 Urban governorate

   

7.9

92.1

 

 Lower Egypt—Urban

   

4.9

95.1

 

 Lower Egypt—Rural

   

1.1

98.9

 

 Upper Egypt—Urban

   

4.5

95.5

 

 Upper Egypt—Rural

   

1.2

98.8

 

 Frontier Governorate

   

27.0

73.0

 

Year of survey (%)

     

p < 0.001

 1995

   

4.5

95.5

 

 2000

   

3.5

96.5

 

 2005

   

4.4

95.6

 

 2008

   

9.1

90.9

 

aData are expressed as mean (standard deviation) or as percentages

bCut-off significant P-value is <0.05

With respect to regression analysis results, we have presented adjusted and fully adjusted marginal odds ratios in Table 3.3. In 2008 factors associated with FGM/C in the unadjusted analysis were: 50–54 years of age (OR = 10.1, 95% CI = 5.87–17.3), rural place of residence (OR = 3.73, 95% CI = 3.03–4.59), being married (OR = 3.77, 95% CI = 3.08–4.61), with “no education” (OR = 15.7, 95% CI = 11.4–21.7) and living in Sharqia (OR = 18.1, 95% CI = 4.32–75.5) or Qina (OR = 11.3, 95% CI = 3.52–36.3). After adjusting for all other factors (fully adjusted model), the likelihood of FGM/C occurring remained statistically most significant in women 50–54 years old, living rural, being married and with no education. Women with no education were 9.5 times more likely to be circumcised than all higher educated persons (95% CI = 6.08–14.8); women living in rural communities were 3.5 times more likely to have undergone FGM/C (95% CI = 2.61–4.70) than women living in urban areas and 1.53 times more likely to be married than not. Women with FGM/C were least likely to live in Matrouh and North Sinai and most likely in Sharqia and Qina. In the Spatial analysis the place of residence, being married, with primary education and living in Sharqia and Qina were the most significant factors associated with FGM/C. Living rural meant 3.34 times more likelihood of being cut than living urban (95% CI = 2.56–4.46), being married made the women 1.51 times more likely to have undergone FGM (95% CI = 1.13–1.99) and women with primary education were now 8.38 times more at risk of cutting than those higher educated (95% CI = 4.82–14.7).
Table 3.3

Unadjusted and fully adjusted odds ratios of women’s circumcision across selected covariates (Egypt DHS, 1995–2008)

Variable

2008

1995–2008: 13 years

Unadjusted OR and 95% CIa

Fully adjusted OR and 95% CIb

Unadjusted OR and 95%CIa

Fully adjusted OR and 95% CIb

Age

 15–19 years

1.00

1.00

1.00

1.00

 20–24 years

1.69(1.31, 2.19)

2.19(1.55, 3.10)

0.93(0.80, 1.08)

1.90(1.54, 2.34)

 25–29 years

3.98(2.85, 5.56)

5.72(3.63, 8.99)

1.00

2.49(2.02, 3.07)

 30–34 years

4.76(3.20, 7.08)

5.29(3.17, 8.82)

0.94(0.80, 1.10)

2.37(1.91, 2.94)

 35–39 years

6.33(4.00, 10.0)

5.96(3.37, 10.5)

1.12(0.95, 1.33)

2.36(1.87, 2.98)

 40–44 years

6.00(3.64, 9.89)

5.50(2.98, 10.2)

1.02(0.85, 1.23)

2.51(1.99, 3.15)

 45–49 years

5.78(3.42, 9.78)

6.57(3.41, 12.7)

1.08(0.88, 1.33)

2.54(1.99, 3.24)

 50–54 years

10.1(5.87, 17.3)

11.4(6.04, 21.6)

1.28(1.06, 1.54)

3.78(2.15, 6.65)

 55–59 years

   

5.42(3.08, 9.56)

Place of residence

 Urban

1.00

1.00

1.00

1.00

 Rural

3.73(3.03, 4.59)

3.35(2.48, 4.52)

5.14(4.62, 5.72)

0.29(0.25, 0.35)

Married

 Yes

3.77(3.08, 4.61)

1.64(1.22, 2.19)

2.29(2.02, 2.60)

1.61(1.37, 1.89)

 No

1.00

1.00

1.00

1.00

Education

 No education

15.7(11.4, 21.7)

9.27(5.92, 14.5)

21.2(18.5, 24.2)

14.1(12.0, 16.5)

 Primary education

9.39(5.96, 14.8)

7.18(4.17, 12.4)

22.6(18.5, 27.8)

17.2(13.9, 21.5)

 Secondary education

3.51(2.80, 4.40)

4.47(3.33, 5.99)

5.85(5.21, 6.57)

5.51(4.85, 6.26)

 Higher education

1.00

1.00

1.00

1.00

Religion

 Muslim

3.94(2.95, 5.27)

7.27(4.83, 10.9)

4.10(2.92, 5.77)

2.52(1.61, 3.96)

 Other

1.00

1.00

1.00

1.00

Governorates of residence

 Matrouh

1.00

1.00

  

 Alexandria

0.69(0.44, 1.08)

0.46(0.27, 0.79)

  

 Beheira

0.33(0.20, 0.53)

0.21(0.12, 0.36)

  

 Kafr el-Sheikh

1.09(0.60, 2.00)

0.95(0.48, 1.86)

  

 Dakahlia

0.67(0.39, 1.16)

0.18(0.09, 0.32)

  

 Damietta

1.01(0.63, 1.62)

0.33(0.19, 0.57)

  

 Port Said

18.1(4.32, 75.5)

7.18(1.65, 31.3)

  

 North Sinai

5.42(2.28, 12.9)

2.75(0.98, 7.71)

  

 Gharbia

2.23(1.18, 4.21)

0.69(0.33, 1.42)

  

 Monufia

0.84(0.54, 1.30)

0.28(0.17, 0.48)

  

 Qalyubia

3.89(1.72, 8.77)

1.47(0.63, 3.44)

  

 Al Sharqia

1.44(0.89, 2.33)

0.26(0.14, 0.45)

  

 Ismailia

1.92(0.92, 4.00)

1.05(0.46, 2.41)

  

 Giza

0.94(0.59, 1.50)

0.56(0.32, 0.97)

  

 Faiyum

8.51(3.04, 23.9)

3.04(0.99, 9.32)

  

 Cairo

2.05(1.08, 3.87)

0.58(0.28, 1.21)

  

 Suez

0.88(0.56, 1.36)

0.23(0.13, 0.40)

  

 South Sinai

1.52(0.92, 2.52)

0.85(0.48, 1.51)

  

 Beni Suef

2.73(1.55, 4.81)

1.87(0.90, 3.91)

  

 Minya

11.3(3.52, 36.3)

7.35(2.89, 18.7)

  

 New Valley

7.76(2.39, 25.2)

5.81(1.61, 21.0)

  

 Asyut

2.70(1.20, 6.03)

1.92(0.80, 4.59)

  

 Red Sea

3.71(0.87, 15.8)

5.24(1.16, 23.8)

  

 Sohag

2.08(0.63, 6.92)

0.84(0.20, 3.48)

  

 Qena

0.03(0.02, 0.05)

0.003(0.002, 0.007)

  

 Luxor

0.28(0.17, 0.46)

0.07(0.04, 0.13)

  

 Aswan

0.49(0.17, 1.40)

0.19(0.06, 0.56)

  

Region of residence (%)

 Urban governorate

  

1.00

1.00

 Lower Egypt—urban

  

2.03(1.74, 2.36)

2.08(1.77, 2.45)

 Lower Egypt—rural

  

9.62(8.00, 11.6)

22.0(17.1, 28.5)

 Upper Egypt—urban

  

1.62(1.39, 1.89)

1.56(1.33, 1.83)

 Upper Egypt—rural

  

6.65(5.64, 7.85)

11.6(9.14, 14.8)

 Frontier governorate

  

0.26(0.24, 0.29)

0.24(0.21, 0.28)

Year of survey (%)

 1995

  

2.88(2.47, 3.36)

1.61(1.34, 1.94)

 2000

  

3.19(2.77, 3.67)

2.09(1.76, 2.49)

 2005

  

2.04(1.80, 2.31)

1.39(1.18, 1.63)

 2008

  

1.00

1.00

aUnadjusted marginal odds ratio (OR) from standard logistic regression models

bAdjusted marginal odds ratio (OR) from standard logistic regression models. Adjusted for women’s age, region and rural/urban location

In the collated surveys of 1998–2008, factors associated with FGM/C in the unadjusted analysis were: rural place of residence (OR = 5.14, 95% CI = 4.62–5.72); being married (OR = 2.29, 95% CI = 2.02–2.60); education, with the category “Primary education” highest (OR = 22.6, 95% CI = 18.5–27.8), followed by no education (OR = 21.2, 95% CI = 18.5–24.2) and secondary (OR = 5.85, 95% CI = 5.21–6.57) vs people with higher education, living in rural Lower Egypt (OR = 9.62, 95% CI = 8.00–11.6) and being interviewed in 2000 (OR = 3.19, 95% CI = 2.77–3.67). After adjusting for all other factors the likelihood of FGM/C remained statistically significant. A statistically significant effect remained for all factors: being 55–59 years old made you 5.42 times more likely to be cut than being 15–19 years of age (95% CI = 3.08–9.56), living rural suddenly had a greatly reduced likelihood of 0.29 (95% CI = 0.25–0.35), being married made FGM 1.61 times more likely (95% CI = 1.37–1.89) than not being, having primary education made women 17.2 times more likely to be cut than higher educated persons (95% CI = 13.9–21.5), living in rural Lower Egypt made it 22 times more likely (95% CI = 17.1–28.5) than living in the Urban governorate and women being interviewed in 2000 were 2.09 times more affected than in 2008. Women with FGM/C were least likely to live in the Frontier governorate and most likely to live in rural Lower Egypt.

Table 3.3 indicates the shift in the prevalence of FGM at the regional level and across age cohorts during the 13-year-period. When exploring the overall national prevalence of FGM/C between the 13-year-period, we noted only a slight decrease of 4% points from the 96.1% average between 1995 and 2008 to 91.9% in 2008 (unweighted average). What holds true here, however, is that aggregate national figures on FGM/C prevalence conceal important spatial variations at the regional level within the survey periods. Unadjusted marginal odds ratios shown in Table 3.3 indicate that in 2008 the highest risk of FGM/C was in Sharqia (OR = 18.1, 95% CI = 4.32–75.5) and Qina (OR = 11.3, 95% CI = 3.52–36.3), and the lowest risk in Matrouh (OR = 0.03, 95% CI = 0.02–0.05) and North Sinai (OR = 0.28, 95% CI = 0.17–0.46). Regarding the effect of age, the unadjusted marginal odds ratios show a rise in FGM/C risk across age cohorts of women, suggesting that there is a secular ascent in FGM/C risk.

In terms of Bayesian spatial analysis results relating to Egypt, we introduced and controlled for spatial and nonlinear factors associated with higher FGM/C risk in all years. Governorate of residence was modelled as a spatial variable in Figs. 3.2b, 3.3, and 3.4, and age of the respondent at the time of interview was modelled as a continuous variable using a flexible nonlinear curve in Fig. 3.1. The modelled covariate results confirmed what was observed in the logistic regression analysis but the patterns differ markedly with governorate of residence and age remaining significant risk factors in both surveys. Overall, results of 2008 (Fig. 3.2b) show that after accounting for (1) sampling error in the observed data; (2) relationships with covariates and the uncertainty in the form of these relationships); (3) uncertainty in the spatial autocorrelation structure of the outcome variable, the Egyptian regions with the highest FGM/C risk included South Sinai, Suez, Ismailia, Sharqia, Fayoum and Qina and significant positive spatial effects were observed in South Sinai, Ismailia, Sharqia, Fayoum and Qina but not Suez. In the adjusted Figs. 3.3 and 3.4, the highest risk regions were still South Sinai, Ismailia, Suez, Sharqia, Fayoum and Qina but there were no significant spatial effects.
Fig. 3.1

Left: Estimated non-parametric trend of women’s FGM risk by women’s age cohort in 2008. Shown is the posterior mean within 80% of credible regions (EDHS 2008)

Fig. 3.2

(a) Map of Egypt showing governorates. (b) Left: Adjusted total residual spatial effects for women’s circumcision, at governorate level in Egypt in 2008. Shown are the posterior odds ratios. Right: Corresponding posterior probabilities at 90% nominal level (EDHS 2008). Red-coloured—highrisk. Green-coloured—low risk. Black-coloured—significant positive spatial effect. White-coloured—significant negative spatial effect. Grey-coloured—no significant effect

Fig. 3.3

Left: Adjusted unstructured random residual spatial effects for female circumcision, at governorate level in Egypt in 2008. Shown are the posterior odds ratios. Right: Corresponding posterior probabilities at 90% nominal level (EDHS 2008). Red-coloured—high risk. Green-coloured—low-risk. Black-coloured—significantpositive spatial effect. White-coloured—significant negative spatial effect Grey-coloured—no significant effect

Fig. 3.4

Left: Adjusted structured random residual spatial effects for female circumcision at governorate level in Egypt in 2008. Shown are the posterior odds ratios. Right: Corresponding posterior probabilities at 90% nominal level (EDHS 2008). Red-coloured—highrisk. Green-coloured—low risk. Black-coloured—significantpositive spatial effect. White-coloured—significant negative spatial effect. Grey-coloured—no significant effect

With regard to the shift of FGM/C by regions, in both samples, the spatial analysis has captured the substantial variation in FGM/C risk across regions observed in the marginal regression analyses. The results shown in Figs. 3.2b, 3.3, and 3.4 are in other words covariate-adjusted region FGM/C spatial variation captured by the global total residual region effects (i.e. the sum of the unstructured and structured spatial effect). There is a clear pattern of regions with higher risk of FGM/C, mostly the governorates of South Sinai, Ismailia, Sharqia, Fayoum and Qina in 2008, which were associated with a higher risk of FGM/C, while states such as Damietta, Alexandria, Dakahlia, North Sinai, Qalyubia, Giza, Cairo and Asyut in 2008 were associated with a lower risk of FGM/C. These spatial patterns confirm some of the observed marginal model findings shown in Table 3.3 while running opposite to others.

Specifically, the left-hand map in Fig. 3.3 shows estimated posterior total residual region odds of FGM/C for each governorate in 2008, ranging from a lower POR of 0.03 (0.02, 0.05) in Matrouh to a higher POR of 18.1 (4.32, 75.5) in Sharqia, with red color indicating the higher risk recorded and green color denoting lower risk. The right-hand map shows the 95% posterior probability map of FGM/C, which indicates the statistical significance associated with the total excess risk. White indicates a negative spatial effect (associated with reduced risk of FGM/C prevalence), black a positive effect (an increased risk) and grey a non-significant effect. However, the total spatial residuals in Figs. 3.2b, 3.3, and 3.4 shows that much of the variation in FGM/C likelihood remains to be explained. Overall, the results indicate that across surveys, certain high prevalence regions remain “hot spots” regarding FGM/C risk. These include Sharqia, Ismailia, South Sinai, Fayoum and Qina. Risk remained non-significant in the high prevalence regions of Aswan and Menoufia.

Putting the above results in the context of discussion, we note that tracking changes in the prevalence of FGM/C on the basis of nationally representative survey data aggregate figures at the national level may mask important variations across ethnicity or region. To remedy this situation, it is necessary to disaggregate the data, and control for potentially confounding factors. Thus, we used advanced statistical methodology to analyze survey data collected with complex sampling strategies and included possible non-linear covariates. Importantly, this novel approach, developed by Kandala et al. (2009), makes it possible to simultaneously examine individual-level and spatial variability. Overall, among women aged 15–59 in Egypt, the prevalence of FGM/C has changed little over the 13-year period between successive surveys carried out between 1995 and 2008 (FGM/C prevalence of 91.9% in 2008 and 96.1% between 1995 and 2008). We find that these unadjusted figures do indeed mask important variations at both the regional and individual levels. In the multivariate Bayesian geo-additive regression analysis, we controlled for individual-level factors while simultaneously modelling the region of residence as a spatial variable.

The spatial analysis performed in this study reveals that the risk of FGM/C varies across governorates, with the highest risk across the survey periods found in Sharqia and Qina. The question is how we can now interpret these spatial findings given that certain high prevalence regions remained “hot spots” regarding FGM/C risk and others did not. Importantly, these results show that community-level effects, above and beyond individual-level effects, play a crucial role in determining the likelihood of FGM/C. In other words, the context, in which an individual woman lives, bears an important influence on whether FGM/C is practiced. This finding is consistent with social convention theory, which predicts that interdependent expectations and social norms shared by community members serve to uphold the practice, making it difficult for individuals to abandon FGM/C without experiencing adverse social sanctions (Mackie and LeJeune 2009). The theory predicts that change is most likely to come about when members of social groups have simultaneously shifted social norms pertaining to FGM/C (Mackie 2000). It may be the case that regional differences in FGM/C risk capture this shift in social norms. At the same time, theory on social norms and conventions does not rule out the possibility of individual-level factors influence on decision-making regarding FGM/C, although the social environment can constrain these choices. Indeed, in our study we find evidence for the simultaneous influence of community- and individual-level factors influencing the risk of FGM/C.

In the fully adjusted model, a number of individual-level factors were found to be associated with the likelihood of FGM/C. In the 2008 survey data these include rural residence, no education and being married. With respect to individual-level predictors of risk of FGM/C among women in Egypt, our most surprising finding concerns the age cohort. Unadjusted estimates of risk of FGM/C showed a significant rise across age cohorts. The effect of age on the likelihood of FGM/C is highest in women aged 55-59, and decreases with decreasing age. How can we understand this unexpected finding? Yoder and Wang (2013) comment on the possibility of using DHS data to assess the magnitude of reductions in FGM/C, but join several commentators in urging caution about the limitations of self-reported survey data on FGM/C status. Because of the sensitivity of the topic or illegal status, women may be unwilling to disclose having undergone FGM/C (Askew 2005; Shell-Duncan et al. 2013). Additionally, particularly when FGM/C is performed at an early age, women may be unaware of whether they have been cut or the extent of the cutting (Yoder et al. 2004; UNICEF 2013). A number of studies have attempted to determine the reliability of self-reports of FGC status by verifying them through clinical examinations, and have reported variable rates of concordance. While one study in Sudan reported complete agreement between clinical examination and women’s reports of having undergone some form of FGM/C or not (Elmusharaf et al. 2006), others report variable degrees of discrepancy. Morison and colleagues found 3% disagreement in The Gambia, whereas studies in Tanzania and Nigeria reported disagreements in more than 20% of women (Adinma 1997; Msuya et al. 2002; Klouman et al. 2005; Snow et al. 2002). A longitudinal study in Ghana afforded a unique opportunity to assess the consistency of women’s self-reports of FGM/C status over repeat surveys (Jackson et al. 2003). The data showed that a substantial number of adolescent girls who initially reported having undergone FGM/C later denied being cut. The authors concluded that denials of having undergone FGM/C were influenced by exposure to anti-FGM/C interventions, and by passage of a law banning FGM/C. In a detailed overview of methodological considerations for measuring change in FGM/C, Askew (2005, pp. 472–73) emphasized the need to consider the context in which questions of FGM/C status are being asked: “If FGC is widespread, socially acceptable and there is no well-publicized interventions causing people to question its acceptability and legalit, then self-reporting is likely to be valid. If there are reasons why it would not be attractive for respondents to declare that they are cut… then self-reported measures should be questioned and ways sought to validate the results.” With this warning in mind, we concur with Yoder et al. (2004, p. 10), who conclude “there is sufficiently strong confirmation of FGC status from women’s reports to warrant the use of survey data to calculate the prevalence of FGC” (Yoder et al. 2004, p. 10).

Yoder and Wang (2013) discuss the possibility of using DHS data to assess the impact of interventions that promote the abandonment of FGM/C. It is important to note, however, that these data do not provide information needed to complete a rigorous evaluation of any single intervention program. Askew (2005) emphasizes that such evaluations require a quasi-experimental design, with “case” and “control” communities, as well as pre- and post-intervention measures to differentiate ongoing or “natural” change from those that may arise from intervention efforts. DHS and MICS data do not provide baseline and post-intervention measures, nor are they accompanied by information on the types of intervention activities that have taken place in any region, cluster, or village selected for inclusion in the survey. Since communities may participate in numerous intervention programs, in addition to being targeted by media messages such as those on health risks or legislation banning FGM/C, it is not possible to isolate the effect of any intervention program. It is, however, possible to examine change in attitudes or FGM/C behavior that may arise from combined “natural” social change processes and some potential mixture of directed interventions.

When examining trends in FGM/C using nationally representative survey data, it is important to specify what one should expect to see. Several considerations are salient. First, since data on national prevalence provides information on the proportion of women aged 15–49 who have undergone FGM/C, this aggregate number is unlikely to change dramatically in consecutive surveys implemented 5 years apart; women who are cut will remain cut, and the national prevalence estimate changes only as an increasing proportion of uncut women age into the 15–49 group and cut women age out. A more sensitive indicator of change is to look at the prevalence along age cohorts, as we have done in this study. Second, in doing so, we need to take into account age at cutting, as women’s FGM/C status reflects an event that took place some years in the past (Yoder and Wang 2013; UNICEF 2013). In Senegal, over 60% of girls are cut by age 1, and nearly three quarters of girls are cut before the age of 5 (UNICEF 2013). Thus, when looking at rates of FGM/C in women aged 15–19, we are examining the results of an event that likely took place between 11 and 18 years prior to collection of the survey data. To detect more recent changes, it is possible to examine trends in FGM/C among daughters of the survey respondents.

3.3.7.2 B-Nigeria

The next mapping in terms of FGM/C concerns Nigeria. As is probably well known, Nigeria is located in the Sub-Saharan African (SSA) region of the world. It shares borders with Benin Republic, Niger, Chad and Cameroon. Demographically, it is the most populous country in the region with an estimated population of 140,003,542 according to the 2006 census figures. About half of the population is women; hence issues concerning women such as FGM should not be ignored. Economically, Nigeria is not very buoyant. With the low per capita income and worsening poverty situation, the country would find it difficult to bear the cost implications of the health burden arising from the complications of FGM. Hence, there is strong economic justification to stop the practice of FGM/C.

It should be noted that the World Health Organization’s FGM typology distinguished four different types of FGM: Type I or clitoridectomy: excision of the prepuce, with or without partial or total excision of the clitoris; Type II or excision: the excision of the clitoris with partial or total excision of the labia minora; Type III or infibulation: the most severe form of FGM consisting of the partial or total excision of the external genitalia and stitching or narrowing of the raw labial surfaces, leaving a small posterior opening for urinary and menstrual flow; and Type IV: a residual category of FGM consisting of pricking, piercing or incising the clitoris and/or labia, cauterization of tissues, scrapping of the vaginal orifice or cutting of the vagina.

Of the four types of FGM, the last (type IV) is rare in Nigeria. While all the first three forms are practiced throughout the country, the pattern and type varies across states. While type III has a higher incidence in the northern states, types I and II are more common in the southern states. The Yoruba ethnic group practice mainly Type II and Type I. The Hausa and Kanuri groups practice Type III. The Ibo and Ijaw groups, depending on the local community, practice any one of the three forms. The Fulani ethnic group does not practice any of these forms.

Legally, there is no federal law banning FGM/FGC in Nigeria, although the 1999 Constitution shuns any act of torture or inhuman treatment or violence against any person. However, some states have passed laws against the practice. They include Bayelsa, Cross River, Delta, Ebonyi, Edo, Ekiti, Ogun, Ondo and Rivers states. In most cases the persons convicted under the law are liable to fines and imprisonment. The enforcement of the law is not very satisfactory and this many believed is a result of poor fines and short duration of imprisonment. For instance, Edo State banned this practice in October 1999 and convicts are subject to a fine of approximately US$10 and imprisonment of only 6 months.

The socio-cultural factors are often given as reasons for the practice of FGM more than the economic reasons. Instances were however cited that those who perform the cutting get paid either in cash or kind and that the cut parts are either buried or sold. There is no known empirical evidence to support the above facts and this is why in the concluding remark, further research was suggested in that respect.1

The data for the mapping of FGM/C in Nigeria came from 38,948 women (response rate 90%) aged 15–49 years who were interviewed in the 2013 Nigeria Demographic and Health Survey (NDHS). DHS allows a comprehensive picture to be constructed of the current global prevalence rates among women and their daughters. It provides valid data on the occurrence of FGM practices at national and state levels. The survey results can also suggest associations between prevalence and ethnicity, religion or other background variables; indicate how the practice is distributed; help identify girls at risk and enable monitoring trends over time.

The DHS focuses on two types of prevalence indicators. The first addresses FGM prevalence levels among women and represents the proportion of women aged 15–49 who have undergone FGM. The second type of indicator measures the status of daughters and calculates the proportion of women aged 15–49 with at least one daughter who has undergone genital mutilation. The methods, objectives, organisation, sample design and questionnaires used in the 2013 NDHS survey for Nigeria have been described in detail together with an overview report elsewhere (National Population Commission 2004). Briefly, this was a random probability sample of households designed to provide estimates of health, nutrition, water and environmental sanitation, education and genital mutilation practices at the national level, for urban and rural areas, for the 36 states and Federal Capital Territory (FCT). The survey used a two-stage cluster sampling design to collect data on a wide range of health issues. Personal (face-to-face) interviews were conducted with participants after obtaining their consent. The questionnaire collected information on whether the participant had undergone FGM and, if so, at what age and the type of practitioner, on attitudes and beliefs about FGM, as well as on their sexual and marital history. The Institutional Review Board at ORC Macro validated the survey procedures and instruments.

In this context, the two outcomes or dependent variables we considered in these analyses were: (1) whether a woman had had FGM and (2) whether if she had daughters, any had had FGM. We define FGM in a broad way to include all sorts of FGM respondents had undergone.

In the 2013 Nigeria DHS sample, Type I and II were the most common form of FGM with 66.8% (6295/9426) compared to 42.3% (708/1673) estimate in the 2003 DHS sample. The most severe form of FGM, Type III, was reported by 5.3% (501/9426) of the women meaning a 2% increase from the 2003 survey. A total of 15.2% (1429/9426) reported not knowing the form of FGM performed or did not respond to the question.

It is important to mention that the use of DHS data has limitations. Identification of FGM depended on mother’s report (recall) as is common in retrospective surveys. However, accuracy and completeness of mother’s recall in 19 national Demographic and Health surveys found that highly educated women were more accurate in reporting and identification of illnesses (Boerma et al. 1991). One might also argue that individual women cannot forget a life changing event such as FGM. Perhaps, this bias can only affect the classification of type of FGM. To provide a consistent sample, we did not analyse the data by type of FGM.

The exposure variable investigated is the respondent geographical location (state of residence) in addition to various control variables on socio-demographic factors known to be associated with FGM. The respondent’s age at assessment was included as an indicator of the birth cohort of the participant. Other variables included were education of the respondent and of her partner (no education vs. some education), religion (Christian, Traditionalist vs. Muslim), residence (rural vs. urban), final say on health care (Respondent alone, respondent and partner, partner alone vs. someone else), marital status (single vs. married) and a principal component based assets index as an indicator of household wealth (Filmer and Pritchett 2001).

Nigeria is divided into 36 states and the Federal Capital Territory: Abuja. FGM prevalence is aggregated and known at a national level. We went a step further and accounted, simultaneously, for geographical location effects on FGM at the disaggregated level of states, thereby highlighting the spatial distribution of FGM. We recognise that the state is still a large unit but disaggregating to this level represents a considerable advance over the use of national averages and our analysis provides state-level information on FGM. On the other hand, one cannot assume that the clusters selected in each district are fully representative of the states in which they are located, as the surveys only attempted to generate a fully representative sample at the regional level. Consequently, the spatial analysis will be affected by some random fluctuations. Some of this random variation can be reduced through structured spatial effects as it includes neighbouring observations in the analysis. It should, however, be pointed out that such a spatial analysis should preferably be applied to census data, where the precision of the spatial analysis would be much higher. Sadly, however, most censuses do not collect data on FGM and complete datasets are often inconveniently unavailable for such analyses. We used geo-additive Bayesian modelling, with dynamic and spatial effects, to assess temporal and geographical variation in FGM. The model used also allows for non-linear effects of covariates on FGM.

Data were available on 38,948 women; and response rates by state were at least 90%. Women interviewed were representative of the underlying populations of the different states of Nigeria. The minimum average number of women sample in a state was 672 and the maximum was 2228. For daughter’s sample, the minimum sample in a state was 132 and the maximum was 2007. The overall prevalence of the FGM in Nigeria in 2013 was 39.3% (9615/24,473) compared to 22% in 2003. This national prevalence and the aggregated regional prevalence concealed, however, important spatial variation in the FGM rates recorded at state level. In the North Central region, for example, the overall prevalence of FGM investigated was 7.4% but the corresponding state-level prevalence varied from 6.2% (in Kogi) to 66.3% (in Kwara). 26,202 women had living daughters in the 2013 NDHS, of whom 3710 (14.2%) had a daughter with FGM/C, a 50% reduction from the 2003 estimate.

Figure 3.5 shows the weighted and unweighted prevalence estimates for the most recent three consecutive Nigeria DHS samples (2003, 2008 and 2013). While a 50% increase was observed in women who had FGM between 2003 and 2013, a 4% reduction was found in the proportion of women who had at least one daughter with FGM within the same period (Table 3.4). We compared 2003–2013 because of the definition of FGM in the 2013 Nigeria DHS that was consistent with the WHO definition.
Fig. 3.5

Weighted and unweighted estimates of Female Genital Mutilation in Nigeria women and their daughters in 2003, 2008, and 2013

Table 3.4

Descriptive summary of Female Genital Mutilation for 2003 and 2013

 

2003 NDHS

2008 NDHS

2013 NDHS

Frequency

Percent

Frequency

Percent

Frequency

Percent

Mother circumcised

 No

5888

80.42

9473

51.24

14,858

60.71

 Yes

1433

19.58

9014

48.76

9615

39.29

Total

7321

100

18,487

100

24,473

100

Daughter circumcised

 No

1766

81.45

7232

69.65

22,492

85.84

 Yes

402

18.55

3152

30.35

3710

14.16

Total

2168

100

10,384

100

26,202

100

NB: The 2008 NDHS showed a higher prevalence of female circumcision than that reported in the 2003 NDHS (30% versus 19%). However, this increase was actually due to variations in the definition of FGC used in the two surveys. In the 2008 NDHS, some of the field teams included angurya and gishiri cuts in the FGC category while others did not. This was not the case in 2003 NDHS. In the 2013 NDHS, the definition of FGC explicitly followed the WHO definition mentioned above and captured the practice of angurya and gishiri cuts. Any comparisons of FGC data from the 2013 survey with data from these earlier surveys should be made with caution (National Population Commission 2004)

Figure 3.6 shows raw or unsmoothed prevalence rate of FGM for women (left) and daughters (right). For both women and daughters, the prevalence of FGM was more clustered around the states in the southwest region of the country, a similar trend to the one observed in 2003. As expected, smoothing and the control of confounders generally uncovered the real extend of the prevalence rate that is difficult to see in the crude raw (unsmoothed) prevalence rates. This shows the importance of multiple adjustments of confounders. The general patterns seen in the unsmoothed maps remained unchanged. Figures 3.7 and 3.8 show the smoothed results after adjustment for the state of residence and other confounders. The prevalence of FGM among women and daughters was most affected by adjustment for the state of residence in women, showing clearly a north-south divide of the spatial distribution of FGM across the country, and the concentration of high prevalence states in the south-south and southwest regions became more apparent (Fig. 3.8). Adjustment for the state of residence resulted in a similar change for daughter’s sample, though to a lesser and non-significant extent (Fig. 3.7). Therefore, adjustment made a big difference to the spatial distribution of FGM for women but not for daughters. This is contrary to the trend observed in 2003 in which spatial adjustment made a significant difference in both the women and daughter sample.
Fig. 3.6

Map of Spatially unstructured (unsmoothed) prevalence of FGM for women (left) and daughters (right) in Nigeria (2013 NDHS)

Fig. 3.7

Posterior means for the likelihood of a woman having a daughter with FGM (left) and the posterior probabilities (right) of FGM in Nigeria. White-coloured—significant positive spatial effect; Black-coloured—significant negative spatial effect; Grey-coloured—no significant effect

Fig. 3.8

Posterior means for the likelihood of FGM in women (left) and the posterior probabilities (right) of FGM in Nigeria. White-coloured—significant positive spatial effect; Black-coloured—significant negative spatial effect; Grey-coloured—no significant effect

Other factors associated with higher prevalence of FGM were examined for the 2013 Nigeria DHS sample. While Urban areas and education were strong predictors of higher prevalence of FGM in 2003, Widowed and Divorced/Separated were strong predictors of higher prevalence of FGM in women, and Christianity and Higher Education are strong predictors of lower prevalence of FGM in 2013. The mean age of the women in the 2013 NDHS sample was 31.7 years; 24% of them were never (or not yet) married. The mean level of education was quite low. On average, women had only 4.5 years of education (a 3-year increase from 2003 average). Nearly 40% reported no formal education at all and 45% had completed secondary and higher education. Fifty-eight percent of participants lived in rural areas. Fifty-two percent of the sample was Muslim.

Selected characteristics of respondents and their daughters by FGM status are reported in Table 3.5. For both outcomes (i.e. women or daughters had undergone FGM), there were significant, consistent associations for several potential correlates of FGM. The results showed that age, religion, place of residence, wealth, education, state of residence and who in the household had responsibility for health care decisions (RHCD) were all significantly related to the two outcomes (p < 0.01). FGM was prevalent among respondents of older age. Urban compared to rural women were significantly more likely to have FGM but a lower likelihood of having a daughter with FGM (see Table 3.5). This was a slight deviation from the observed trend in 2003 in which urban women had a higher likelihood of FGM and higher likelihood of a daughter with FGM (see Table 3.5). There were significant differences between states with respect to the two outcome variables (see Table 3.5). The higher the level of education attained by the respondent and her partner, the significantly lower the prevalence of FGM in women and daughters (see Table 3.6). The only exception to this linear trend is the lowest prevalence rate observed in women with no education. This was consistent with the 2003 report. There were significant differences between religions for the two outcomes. The proportion of women circumcised was significantly higher among women from the traditional/animist religion (62.5%) followed by women from the Christian religion (45.1%) and lower among women from the Muslim religion (33.2%). In contrast, a different trend was observed with the outcome for circumcision among daughters. The proportion of daughters circumcised was highest in Muslim daughters (20%) followed by traditional/animist religion (13.4%) and lowest in daughters from the Christian religion (8%). This was in sharp contrast to the 2003 estimate in which women with a Christian religious background had the highest rates of FGM. Wealth index was significantly related to FGM both in women and in daughters. The richest groups among women experienced the highest rate of circumcision, whereas the women from the poorest quintile had the lowest prevalence of circumcision. However, a lower rate of circumcision of daughters was observed in women of the richest quintile whereas a higher rate was seen in women in the poorest quintile.
Table 3.5

Distribution of factors analyzed in Female Genital Mutilation study in Nigeria (2013)

Characteristics

Mother circumcised (weighted) N (%)

p-value

Daughter circumcised (weighted) N (%)

p-value

Yes

No

Yes

No

Religion

  

<0.001

  

<0.001

 Christian

5420(45.1)

6595(54.9)

 

989(7.9)

11,524(92.1)

 

 Muslim

4039(33.2)

8120(66.8)

 

2679(20.0)

10,704(80.0)

 

 Traditionalist/animist

126(62.5)

75(37.5)

 

27(13.4)

175(86.6)

 

Total

9585(39.3)

14,791(60.7)

 

3695(14.2)

22,403(85.8)

 

Place of residence

  

<0.001

  

<0.001

 Urban

5288(45.7)

6298(54.4)

 

1385(11.2)

10,990(88.8)

 

 Rural

4327(33.6)

8560(66.4)

 

2325(16.8)

11,502(83.2)

 

 Total

9615(39.3)

14,858(60.7)

 

3710(14.2)

22,492(85.8)

 

Final say on health care

  

<0.001

  

<0.001

 Respondent alone

724(56.4)

560(43.6)

 

240(17.9)

1100(82.1)

 

 Respondent and husband

2894(45.9)

3414(54.1)

 

1008(15.4)

5552(84.6)

 

 Husband/partner

3384(33.5)

6727(66.5)

 

2188(19.9)

8833(80.1)

 

 Someone else

18(49.3)

18(50.7)

 

38(17.6)

33(82.4)

 

Total

7020(39.6)

10,720(60.4)

 

3443(18.2)

15,516(81.8)

 

Asset index

  

<0.001

  

<0.001

 1st quintile

1171(28.4)

2949(71.6)

 

967(21.1)

3626(78.9)

 

 2nd quintile

1503(35.7)

2706(64.3)

 

931(20.4)

3626(79.6)

 

 3rd quintile

1753(40.7)

2559(59.3)

 

629(14.0)

3884(86.0)

 

 4th quintile

2437(45.7)

2894(54.3)

 

643(11.5)

4973(88.5)

 

 5th quintile

2752(42.3)

3750(57.7)

 

539(7.8)

6383(92.2)

 

Total

9615(39.3)

14,858(60.7)

 

3710(14.2)

22,492(85.8)

 

Mother education

  

<0.001

  

<0.001

 None

2530(29.7)

5976(70.3)

 

1997(21.3)

7393(78.7)

 

 Primary

2060(47.6)

2270(52.4)

 

777(17.3)

3718(82.7)

 

 Secondary

3994(45.4)

4810(54.6)

 

800(8.6)

8529(91.4)

 

 Higher

1031(36.4)

1802(63.6)

 

137(4.6)

2854(95.4)

 

Total

9615(39.3)

14,858(60.7)

 

3710(14.2)

22,492(85.8)

 

Partner’s education

  

<0.001

  

<0.001

 None

2057(30.3)

4735(69.7)

 

1632(21.8)

5868(78.2)

 

 Primary

1836(50.0)

1834(50.0)

 

760(19.9)

3050(80.1)

 

 Secondary

2605(46.9)

2947(53.1)

 

885(15.2)

4940(84.8)

 

 Higher

1093(37.2)

1848(62.8)

 

313(10.2)

2764(89.8)

 

Total

7591(40.0)

11,364(60.0)

 

3590(17.8)

16,623(82.2)

 

States

North Central

  

<0.001

  

0.010

 Benue

134(23.1)

447(76.9)

 

25(4.3)

561(95.7)

 

 Kwara

411(66.3)

209(33.7)

 

115(17.7)

537(82.3)

 

 Niger

48(12.7)

332(87.3)

 

25(6.6)

356(93.5)

 

 Plateau

14(10.4)

123(89.6)

 

2(1.6)

135(98.4)

 

 Kogi

16(6.2)

240(93.8)

 

4(1.4)

252(98.6)

 

 Nassarawa

69(25.6)

201(74.4)

 

15(5.3)

267(94.7)

 

 Abuja (FCT)

25(12.1)

181(87.9)

 

2(0.9)

211(99.1)

 

Regional total

718(29.3)

1734(70.7)

 

188(7.5)

2319(92.5)

 

North East

  

0.001

  

<0.001

 Bauchi

65(11.6)

492(88.4)

 

88(15.5)

480(84.5)

 

 Borno

35(3.4)

1013(96.6)

 

25(2.4)

1052(97.6)

 

 Adamawa

9(3.0)

283(97.0)

 

2(0.6)

290(99.4)

 

 Taraba

25(5.6)

427(94.4)

 

38(8.1)

430(91.9)

 

 Yobe

28(3.4)

772(96.6)

 

23(2.7)

829(97.3)

 

 Gombe

17(8.2)

193(91.8)

 

3(1.7)

207(98.3)

 

Regional total

179(5.3)

3180(94.7)

 

180(5.2)

3287(94.8)

 

North West

  

<0.001

  

<0.001

 Kaduna

430(47.7)

473(52.3)

 

247(25.4)

726(74.6)

 

 Kano

1046(49.0)

1089(51.0)

 

682(29.1)

1663(70.9)

 

 Katsina

1(0.2)

552(99.8)

 

0(0.0)

575(100.0)

 

 Sokoto

26(6.4)

384(93.6)

 

143(16.4)

727(83.6)

 

 Jigawa

428(49.3)

442(50.7)

 

366(38.8)

578(61.2)

 

 Kebbi

26(6.3)

390(93.7)

 

30(7.0)

403(93.0)

 

 Zamfara

18(2.4)

743(97.6)

 

199(23.5)

645(76.5)

 

Regional total

1976(32.7)

4073(67.3)

 

1667(23.9)

5317(76.1)

 

South East

  

<0.001

  

0.078

 Anambra

251(36.4)

439(63.6)

 

35(4.9)

687(95.1)

 

 Imo

579(72.7)

217(27.3)

 

127(15.4)

696(84.6)

 

 Abia

169(35.7)

304(64.3)

 

47(9.6)

440(90.4)

 

 Enugu

391(58.0)

284(42.0)

 

83(12.1)

600(87.9)

 

 Ebonyi

850(78.7)

230(21.3)

 

152(13.7)

955(86.3)

 

Regional total

2240(60.3)

1475(39.7)

 

443(11.6)

3378(88.4)

 

South South

  

<0.001

  

<0.001

 Akwa-Ibom

116(19.7)

476(80.3)

 

8(1.4)

586(98.6)

 

 Edo

376(51.7)

351(48.3)

 

94(11.9)

699(88.1)

 

 Cross river

277(38.3)

446(61.7)

 

14(1.9)

726(98.1)

 

 Rivers

227(20.7)

871(79.3)

 

22(1.9)

1099(98.1)

 

 Delta

488(45.9)

575(54.1)

 

50(4.7)

1018(95.3)

 

 Bayelsa

72(21.7)

259(78.3)

 

2(0.6)

330(99.4)

 

Regional total

1557(34.3)

2978(65.7)

 

190(4.1)

4460(95.9)

 

South West

  

<0.001

  

<0.001

 Lagos

643(46.0)

756(54.0)

 

100(6.5)

1443(93.5)

 

 Ogun

93(20.9)

352(79.1)

 

11(2.4)

448(97.6)

 

 Ondo

342(71.3)

137(28.7)

 

124(22.7)

424(77.3)

 

 Oyo

968(86.3)

153(13.7)

 

334(26.6)

921(73.4)

 

 Osun

551(82.6)

116(17.4)

 

110(16.0)

579(84.0)

 

 Ekiti

222(88.6)

29(11.4)

 

69(24.6)

211(75.4)

 

Regional total

2819(64.6)

1544(35.4)

 

748(15.7)

4025(84.3)

 

Total

9615(39.3)

14,858(60.7)

 

3710(14.2)

22,492(85.8)

 
 

MFGM (Mean and Std. dev.)

Daughter FGM (Mean and Std. dev.)

Current age (years)

31.7 (9.8)

29.3(9.4)

<0.001

33.1(8.1)

29.6(9.8)

<0.001

Maternal age at FGM (years)

2.5(5.8)

     

Daughter’s age at FGM (years)

   

  

Chi-square test was used for categorical data and student t-test test for continuous data in the bivariate analysis

Table 3.6

Posterior odds ratio of Female Genital Mutilation in Nigeria (DHS 2013)

 

Model 1

Model 2

Mother circumcised N (%)

Daughter circumcised N (%)

OR

95% LCI

UCI

OR

95% LCI

UCI

Religion

Christian

0.80*

0.74

0.87

0.67*

0.60

0.74

Muslim

1.00

  

1.00

  

Others

1.07

0.85

1.32

0.80

0.61

1.01

Residence

Urban

1.09

0.96

1.24

0.98

0.85

1.12

Rural

1.00

  

1.00

  

Final say on health care

Respondent alone

0.88

0.59

1.30

1.02

0.70

1.59

Respondent and husband

0.90

0.59

1.29

1.06

0.73

1.63

Husband/partner

0.83

0.55

1.21

1.05

0.71

1.61

Someone else

1.00

1.00

  

Asset index

1st quintile

1.00

1.00

2nd quintile

0.96

0.88

1.05

1.01

0.93

1.09

3rd quintile

0.94

0.84

1.04

0.99

0.89

1.10

4th quintile

0.93

0.83

1.05

0.91

0.80

1.05

5th quintile

0.94

0.83

1.07

0.91

0.78

1.07

Mother education

None

1.00

1.00

Primary

1.08

0.99

1.18

1.06

0.97

1.15

Secondary

1.01

0.92

1.11

0.91

0.83

1.00

Higher

0.88*

0.79

0.98

0.60*

0.52

0.70

Partner education

None

1.00

  

1.00

Primary

1.05

0.97

1.14

1.02

0.94

1.11

Secondary

0.96

0.96

1.13

0.98

0.90

1.06

Higher

0.91

0.82

1.01

0.89*

0.80

0.99

Marital status

Single

1.44

0.97

2.15

Married

1.00

  

1.00

  

Living with partner

1.45

0.97

1.22

0.94

0.79

1.12

Widowed

1.22*

1.05

1.43

0.92

0.60

1.42

Divorce/separated

1.32*

1.11

1.55

0.86

0.56

1.39

Random effects

Mean

SD

95% CI

Mean

SD

95% CI

 Cluster effect

0.44

0.03

0.38–0.51

0.341

0.033

0.281–0.413

 Household effect

0.01

0.00

0.00–0.01

0.003

0.002

0.000–0.009

 Spatially unstructured effect

0.33

0.31

0.00–1.04

0.465

0.262

0.012–0.986

 Spatially structured effect

2.30

1.29

0.48–5.15

0.776

0.975

0.001–3.454

Model 1 (mother fgm): N = 24,303; iterations = 40,000; burnin = 5000, DIC = 20,301.15, pd = 692.92

Model 2 (daughter fgm): N = 20,213; iterations = 40,000; burnin = 5000, DIC = 13,808.25, pd = 517.50

* Statistically significant

This proposition might be viewed as counter-intuitive since it is usually assumed that education and wealth go hand in hand. However, the question whether there is a correlation between education and wealth has given rise to a large body of studies, which rarely rises to the level of settled principle. This is the case because the literature on this issue is fragmented. We could say that generally the more educated and affluent a family is, the less likely they are to counsel FGM for their daughters. Yet the correlation tends to break down in the light of some anecdotal reports suggesting that some educated and affluent families tend to seek modern medicalized forms of the practice for their daughters. On this basis, it might not always be appropriate to consider that poorest families are invariably more likely than not to have recourse to the practice than richest families. This is because FGM itself is considered a social norm shared across all sections of society that holds strong belief in it—irrespective of education and wealth. What remains evident, however, is that education provides tools for women and their daughters to become aware of the dangers of the FGM practices and of their rights. In sum, though, there is no guarantee that education itself leads to wealth and therefore to acute awareness of dangers of FGM practices among women where FGM is embedded in some part as an education value.

Besides, we found that there were statistically significant differences between final say on own healthcare and the likelihood of FGM. The likelihood of FGM was higher if the final say on own healthcare was being made by the respondent and partner (higher likelihood was found when health decisions were made by partner alone in the 2003 NDHS). For the two outcomes, there were statistically significant differences within states. In north central, Kwara, Nassarawa and Benue were associated with a higher proportion of circumcised women and daughters. The Federal Capital Territory Abuja was among states with the lowest proportion of women circumcised. In general, for both outcomes states in the North East and North-West regions were associated with the lowest prevalence of FGM. This was similar to the observed trend in 2003, except for their higher proportion of FGM observed in Kaduna, Kano and Jigawa in 2013.

The estimates of the spatial effects of the FGM were also mapped. Before adjustment for the geographical location, which was acting as a surrogate for cultural, ethnic and environmental differences, a higher prevalence of FGM was concentrated in the south-western regions in areas around Lagos and in the south-south region part of the country (Fig. 3.6). After adjustment, the effect became more pronounced in both areas around Lagos and in the southeastern and south-south regions of the country. In Figs. 3.7 and 3.8, maps show estimated posterior means of residual spatial state effects (i.e. adjusted odds ratios after multiple adjustment of the geographical location, taking into account the auto-correlation structure in the data and other risk factors) for FGM in both women and daughters in each state, with the red colour indicating the maximum posterior means recorded (2.59 for women and 2.08 for daughters) while blue denotes a lower prevalence.

Similar to 2003 estimates, a high prevalence of FGM was concentrated in the southwest, south south and southeastern states both for women and daughters. The right-hand maps show the 95% posterior probability maps of FGM. White colour indicates a positive spatial effect (associated with increased risk of FGM) and black colour a negative effect (a reduced risk). In multivariate geo-additive regression analyses (Table 3.6), after multiple adjustments, there were several consistent significant associations in both outcomes. For women FGM, these factors are: Christian [0.80(0.74–0.87)]; higher education (respondent) [0.88(0.79–0.98)]; widowed [1.22(1.05–1.43)] and divorced/separated [1.32(1.11–1.55)]. For daughter FGM, these factors are: Christian [0.60(0.60–0.74)]; children from educated women [0.60(0.52–0.70)]; children from educated partner [0.89(0.80–0.99)]. No significant association was found between living in urban areas and likelihood of FGM as well as health care decision made by respondent and partner in the adjusted model as observed in the 2003 survey.

Figure 3.9 shows for the whole national sample the estimated non-linear (logits) effects of the respondent’s age for the two outcomes. Shown are the estimated posterior logits of the effects of the respondent’s age within the 80% and 90% credible interval. There appears to be a clear linear association between respondent’s age (left-hand panels) for FGM in women and the right-hand panel for daughters. Respondent’s age appears to be almost linearly positively related to the prevalence of FGM in women. As expected, it appears that as age increased, the likelihood of respondents circumcised per age also significantly increased. In contrast, the association between likelihood of circumcision in daughters and respondents age is non-linear (which was inconsistent with the linear trend observed in 2003). Likelihood of daughter circumcision increases with age peaking at 30 years, then gradually drops (see Fig. 3.9).
Fig. 3.9

Estimated non-linear (logits) effects of mother’s age at daughter’s FGM (left) and respondent’s current age on the likelihood of having FGM. Shown are the posterior logits within the 95% and 80% credible intervals

The above results suggest the need for quantifying the residual effects of geographic location on the prevalence of FGM in Nigeria. The spatial effects have no causal impact but careful interpretation can identify latent and unobserved factors, which directly influence the prevalence of FGM. They can also be interpreted as surrogates of social convention, ethnic, and cultural factors that might confound the observed high prevalence of FGM. For instance, by highlighting the effects of geographic location, diffusion theory (Rogers 1995) that looks at change from the perspective of groups rather than individuals might clarify community influences on FGM. Residual spatial effects of FGM have provided us to see the inherent spatial patterns of the prevalence of FGM as the variability or “noise” has been removed. A more precise spatial pattern of the prevalence of FGM emerged with the estimated residual state effects compared with the crude prevalence without the control of the geographic location effects.

Using the 2013 NDHS, we have attempted to examine the geographical distribution including a number of socio-demographic, household and individual conditions that are likely to represent important determinants of FGM in the general population. This analysis may help to better understand the complex interplay between geographical milieu or environment and traditional practices such as FGM as well as to explore potential mechanisms underlying these associations. In fact, over the past few years, an increasing number of studies have documented that FGM practice may affect the wellbeing of women and children including high risk of HIV/AIDS (Nnorom 2000).

The observed associations of both women and children with elevated risks for morbidity and mortality have generated the current move by various states to ban the practice (Nnorom 2000). However, despite the large body of data, much uncertainty remains about the true nature and causality of these associations due to the potential of confounding by other factors and the lack of robust religious or cultural evidence on plausible mechanisms by which FGM practice operates. Specifically, studies (Freymeyer and Johnson 2007) have shown that FGM habits in the general population are the result of a complex interaction involving factors of different nature (e.g., social, behavioural, psychological, environmental). The prevalence rate of FGM is higher in the southern states than in the northern states despite the fact that level of education is relatively higher in the south. This suggests that the practice of FGM is deeply rooted in the culture, which has been very difficult to change.

The result of asset (wealth) index and decision-making supports the view that women have limited power in decision-making. According to Nwakeze (2001), women’s sexuality is influenced by their limited decision-making power and the decision-making power is a function of their economic independence. Neither the effect of education nor the effects of the household socio-economic status and the religion have a strong negative impact on FGM. The result of religion did not support the view that the practice may be more prevalent in certain religions (such as Islam). These findings support previous findings in Nigeria (Freymeyer and Johnson 2007) to consider FGC as a social convention. Therefore, reconfirming that modernization (education or religion) has minimal impact on the likelihood of FGC in Nigeria. Other factors such as the influence on the social convention may play a major role on the likelihood of women having FGM.

It follows that the mapping of residual spatial effects in Nigeria points to marked variation in the prevalence of FGM at state level. The underlying feature here is that individual; education and community factors were strongly associated with FGM. FGM prevalence maps generated here could be a useful tool for policy design, monitoring and targeted intervention to eradicate this harmful practice in SSA. Also, considering the economic implications of FGM in the policy design is important. Identifying and understanding environmental factors that are associated with state differences in FGM prevalence represents an important investigation to disentangle fully the influences of community, ethnic and cultural factors on FGM. Both novel and less conventional methodologies including various data sources are required to broaden the view of environment at both individual women and community level. Understanding where an individual or community is in the process is important in program research, design and evaluation.

3.3.7.3 C—Senegal2

The third selected country for the purpose of FGM mapping is Senegal. We consider socio-economic and cultural predictors of FGM in that country. Let us first say a few words about the demographics and ethnic breakdown. Senegal had a 2010 population of more than 12.5 million, is home to more than 20 ethnic groups, each with their own language, culture and history. The country is divided into 14 administrative regions (Fig. 3.10). Three of these regions were newly created in 2008, when Kaffrine Region was split from Kaolack, Kédougou was split from Tambacounda, and Sédhiou Region was split from Kolda. A minority of the Senegalese population practices FGM/C, predominantly non-infibulating forms. FGM/C is called excision in French and by a wide variety of terms in local languages. FGM/C in Senegal is generally understood to be tied to ethnicity, with some exceptions, due in part to increasingly common inter-ethnic marriages that are FGM/C-incongruent, where one partner comes from a family that practices FGM/C, while the other does not. FGM/C is said to be “not practiced” by the Wolof and Sereer (Sylla and Helene 1990), who make up 58% of the total population. FGM/C is thought to be near universal among the Tukulor and Mandinka ethnic groups, and among the Jola and Fula, the prevalence is understood to run along lines of lineages. Generally, the practice is passed down from elders to the younger generations by serving as an obligation for acceptance and social integration (Shell-Duncan et al. 2011).
Fig. 3.10

Map of Senegal showing the 11 administrative regions

Beyond demarking group membership, ethnographic studies reveal a multitude of reasons for the practice related to constructs related to gender ideology and proper childrearing. In some groups, the practice serves as an important rite of passage that marks the transition from childhood to adulthood. In recent decades, however, there is a trend in practicing communities to uncouple FGC from initiation and perform the procedure at younger and younger ages (Shell-Duncan et al. 2010). Nonetheless, in many groups FGM/C is still widely held to be essential for the moral upbringing of girls, teaching them, among other things, sexual restraint and how to display respect to elders. Additionally, in certain groups FGM/C is associated with Islam, providing the ritual purity and cleanliness believed to be required for prayer (Dellenborg 2007; Shell-Duncan et al. 2010).

Across different communities, the cultural meanings of FGM/C are multiple, fluid, increasingly contested and negotiated by diverse groups of people who draw on local social and political movements, as well as national and international campaigns aimed at ending FGM/C. Senegal has been the site of a number of media campaigns and NGO-sponsored initiatives aimed at ending the practice of FGM/C, including, but not limited to, Tostan. At the same time, the Senegalese Government has taken a strong stance in opposition to the practice of FGM/C. In 1999 Senegal adopted a criminal law that prohibits the violation of “the integrity of the genital organs of a female person,” carrying a penalty that includes prison for 6 months to 5 years or where cutting results in death, hard labor for life. Additionally, the Government adopted a National Action Plan for 2001–2005 that articulated a clear goal of working toward the total abandonment of FGM/C by 2015 (Diop-Diagne 2008) and reiterated this goal in a second National Action Plan for 2010–2015, adopted in February 2010 (UNICEF 2010).

With a strong local and national commitment to ending FGM/C in Senegal, backed by the international community, international organizations as well as private donors, there is tremendous interest in assessing whether progress has been made toward abandonment of FGM/C. Hence, results of nationally representative statistical surveys on FGM/C have been awaited with great anticipation. Summary statistics from these surveys have been released (UNICEF 2013; Yoder and Wang 2013; Creel 2001). The data summarized in Table 3.7 describe the circumstances surrounding the practice of FGM/C in Senegal. FGM/C typically occurs at very young ages, with the majority of girls cut by age 1, and over 70% cut by age 4. The most common form of cutting is “cut/flesh removed,” which may correspond to FGM Type I (clitoridectomy) or Type II (excision). In contrast to other parts of Africa such as Egypt, Kenya and Sudan, FGM/C in Senegal has not become medicalized (performed by health professionals). Support for the continuation of FGM/C (18% in 2005 and 17% in 2010–2011) is lower than the estimated national prevalence of FGM/C (weighted estimates are 28% in 2005 and 26% in 2010) (UNICEF 2013). Overall, aggregate statistics on prevalence, unadjusted for potentially confounding factors, provide the impression that there is very little change in FGM/C among women in Senegal. In this paper we present results from multivariate analyses that highlight patterns in the data after adjusting for the effect of proximate variables. Specifically, using data from successive household surveys, we examine the spatial distribution of FGM/C, and estimate the effects of a number of socio-demographic factors that could mediate the observed prevalence of FGM/C in Senegal.
Table 3.7

Circumstances surrounding FGM/C in Senegal

 

SDHS 2005

SDHS 2010–2011

Age at cutting (%, cumulative %)

0–1

63.3

61.7

2–4

9.5(72.8)

9.5(71.2)

5–9

14.9(87.7)

13.8(85.0)

10–14

5.1(92.8)

6.0(91.0)

15+

0.9(93.7)

0.7(91.7)

Type of FGM/C (%)

Cut, no flesh removed

0

10

Cut, flesh removed

83

53

Sewn closed

12

14

Not sure/don’t know

5

24

Practitioner of FGM/C (%)

Traditional practitioner

93

100

Medical practitioner

1

0

Don’t know

7

0

Support continuation (all women, both cut and uncut) (%)

53

52

Prevalence (weighted %)

28

26

Source: UNICEF (2013), Yoder and Wang (2013), and Creel (2001)

We analyze a sample of up to an average of 50,000 women participating in the various successive Senegal DHS surveys (2005) and the 2010–2011 SDHS-MICS. The sampling methods employed in these surveys are described in detail elsewhere (Ndiaye and Ayad 2006; EDS-MICS, 2010–2011). For instance, in the 2010–2011 SDHS-MICS, a nationally representative cross-sectional survey (multistage stratified random sampling of households) of women of reproductive age (15–49 years) is selected. The resulting sample was representative of the underlying populations of the different regions of Senegal. It was carried out between October 2010 and April 2011. A nationally representative sample of 15,688 women aged between 15–49 years in all selected households and 4929 men aged 15–59 years in one third of selected households were interviewed with a similar number for the previous survey of 2005 (14,602 women). Response rates were over 93% for individual women and over 87% at the household level and informed consent was obtained from participants. The study protocol conforms to the ethical guidelines of the 1975 Declaration of Helsinki as reflected in a priori approval by the institution’s human research committee. The Ethics Committee of the National Statistical Office of Senegal granted ethical approval.

Having described the population sample, we are now in a position to provide an indication of exposure variables. The main exposure variable in the analysis was the “region of residence”, of which there are 14 as shown in Fig. 3.10 in addition to various control variables on socio-demographic factors associated with FGM. Other cultural, political and geographic factors—sex, age, education level, wealth index, marital status, family size, place of residence that may influence the outcome were also included as confounders. Age was recorded as a continuous variable, and was re-coded into a categorical variable of seven categories in the preliminary analysis, with a truncated first category for 15–19 year olds. For the modelling of the prevalence of FGM, we examine the cohort’s effect of age of the respondent as a continuous variable using a flexible nonlinear function to estimate trend of FGM in the cohort of respondent’s women in the two surveys. Education level has been categorized as “None”, “Primary”, “Secondary” and “Higher”, and wealth index was categorized as “Poorest”, “Poorer”, “Middle”, “Richer” and “Richest”. The variable for family size was re-coded into three categories of small size “1–4 children”, middle size “5–7 children” and large size “8+ children”, the ethnicity was recorded as “Wolof”, “Poular”, “Serer”, “Mandingu”, “Diola”, “Sononke”, “Not Senegalese” and “Other”.

It should be noted that Senegal had 11 regions back in 2005 and 3 more regions were created in 2010. Therefore, we are presenting the 2010 map of Senegal including 14 regions (Tambacounda was divided into Tambacounda and Kedougou; Kolda into Kolda and; Sedhiou; Kaolack into Kaolack and Kaffrine).

The main descriptive results briefly presented in Table 3.7 describe the circumstances surrounding the practice of FGM/C in Senegal. FGM/C typically occurs at very young ages, with the majority of girls cut by age 1 and over 70% cut by age 4. The most common form of cutting is “cut/flesh removed,” which may correspond to FGM Type I (clitoridectomy) or Type II (excision).3 In contrast to other parts of Africa such as Egypt, Kenya and Sudan, FGM/C in Senegal has not become medicalized (performed by health professionals). Support for the continuation of FGM/C (18% in 2005 and 17% in 2010–2011) is lower than the estimated national prevalence of FGM/C (weighted estimates are 28% in 2005 and 26% in 2010) (UNICEF 2013). Overall, aggregate statistics on prevalence, unadjusted for potentially confounding factors, provide the impression that there is very little change in FGM/C among women in Senegal. In this paper we present results from multivariate analyses that highlight patterns in the data after adjusting for the effect of proximate variables. Specifically, using data from successive household surveys, we examine the spatial distribution of FGM/C and estimate the effects of a number of sociodemographic factors that could mediate the observed prevalence of FGM/C in Senegal.

Unweighted baseline socio-demographic characteristics are shown in Table 3.7, and by FGM/C status (whether a participant underwent FGM/C or not) in Table 3.8 (unadjusted Odd ratios). The overall prevalence of FGM/C differs slightly between the two surveys (30.1% in 2005 and 28.1% in 2010–2011). Before investigating factors associated with FGM/C and trends across the two surveys, we examined comparability of women sampled in each survey. The two survey populations are similar in terms of the mean ages of women (for 2005 the mean age was 27.8 years, with a standard deviation of ±9.9 years and in 2010–2011 the mean age of the sample was 27.9 years, with a standard deviation of ±9.5 years). Most of the population sampled lived in rural settings (51.3% in 2005 and 50.7% in 2010–2011) and 73.0% were married in 2005 with slightly less at 66.5% in 2010–2011. The mean age for women’s partner was higher than that for women in both the 2005 (42.6 vs. 27.8 years) and the 2010–2011 survey (43.5 vs. 27.9 years). A total of 14.2% of women in the 2005 population had a secondary education while 59.6% had no education compared to 18.3% (secondary education) and 57.9% no education in 2010–2011. Women with FGM/C were mostly married (34.0% vs 30.6%), with no education for themselves (34.4% vs 31.4%) or their partners (36.7% vs 32.2%), were “poorest” (43.8% vs 48.7%), had a large family size (41.0% vs 35.5%), lived in rural areas (37.7% vs 35.6%), were Soninke (78.7% vs 66.3%), were Muslim (31.0% vs 29.0%) and were living in Matam and Kédougou (94.9% vs 87.7%) and Kolda and Matam (94.3% vs 89.6%) in 2005 and 2010–2011 respectively. Notably, in both surveys, women’s age is not significantly associated with FGM/C. Thus, there is no evidence of a change in rates of FGM/C across age cohorts.
Table 3.8

Baseline characteristics of the study population of women ages 15–49 (Senegal DHS, 2005; 2010)a

Variable

Women 2005 (N = 14,602)

Women 2010 (N = 15,688)

Mean ageb (SD) respondent

27.8(9.9)

27.9(9.5)

Mean ageb (SD) partner

42.6(9.9)

43.5(17.5)

FGM/C (%)

Yes

30.1

28.1

No

69.9

71.9

Place of residence (%)

Urban

48.7

49.3

Rural

51.3

50.7

Married (%)

Yes

73.0

66.5

No

27.0

33.5

Education (%)

No education

59.6

57.9

Primary education

25.2

21.8

Secondary education

14.2

18.3

Higher education

1.0

2.1

Partner’s education (%)

No education

70.3

75.4

Primary education

12.6

11.9

Secondary education

12.9

9.2

Higher education

4.3

3.5

Wealth index (%)

Poorest

16.7

16.5

Poorer

17.6

17.9

Middle

19.4

19.8

Richer

21.6

22.3

Richest

24.7

23.5

Family size (%)

Small (1–4 children)

84.6

82.1

Middle (5–7 children)

13.0

15.4

Large (8+ children)

2.5

2.6

Ethnicity (%)

Wolof

39.7

38.7

Poular

25.2

26.5

Serer

15.9

15.0

Mandingu

4.6

4.2

Diola

4.9

4.0

Soninke

2.8

2.3

Not Senegalese

1.7

2.0

Other

5.2

7.3

Religion (%)

Muslim

95.5

95.4

Other

4.5

4.6

Region of residence (%)

Dakar

26.5

26.0

Diourbel

10.6

11.8

Fatick

4.8

4.6

Kaffrine

 

3.7

Kaolack

11.3

7.5

Kédougou

 

0.7

Kolda

7.2

4.1

Louga

6.3

7.2

Matam

3.7

3.8

Saint Louis

6.5

6.6

Sédhiou

 

2.9

Tambacounda

5.8

4.6

Thiès

13.5

12.9

Ziguinch

3.9

3.7

aData are expressed as mean (standard deviation) or as percentages

bAge ranges from 15–49 year for women, and 18–97 years for partners

It is important to point out that although the prevalence of FGM between the two successive surveys was not statistically significantly different (30.1% in 2005 and 28.1% in 2010), in terms of statistical modelling strategy, the surveys were modelled separately rather than jointly with a time dummy and exploration of other time interactions because our important research question was to investigate the magnitude and change of the spatial effects at the regional level. In the exploratory analysis, we noticed that there were higher FGM prevalence regions in the 2005 survey that reduced significantly their FGM prevalence in 2010. Such regions were Kaolack and Kolda. Therefore, we modelled the two surveys separately to investigate this shuttle change in prevalence, as the two surveys have not been analyzed to investigate FGM separately in the past.

Turning now to regression analysis results, we summarise unadjusted and fully adjusted marginal odd ratios in Table 3.9. In 2005 factors associated with FGM/C in the unadjusted analysis were: rural place of residence (OR = 2.09, 95% CI = 1.94–2.25), being married (OR = 2.22, 95% CI = 1.99–2.47), with “no education” (OR = 2.05, 95% CI = 1.13–3.70), wealth index with poorer quintile in the lead (OR = 5.37, 95% CI = 4.59–6.29), being Soninke (OR = 3.70, 95% CI = 2.62–5.21) or Mandingu (OR = 2.96, 95% CI = 2.36–3.72), being Muslim (OR = 3.54, 95% CI = 2.70–4.64) and living in Matam (OR = 18.2, 95% CI = 14.0–23.7) or Kolda (OR = 16.5, 95% CI = 13.2–20.6). After adjusting for all other factors (fully adjusted model), the likelihood of FGM/C in women with only primary or secondary education became not statistically significant. The effect disappeared also for small and middle family size; the statistically significant effect remains only for women living in rural areas, with no education, partners with primary education, in all wealth quintiles, all ethnicities and the region of residence. Women with no education were 3.26 times more likely to undergo FGM/C than all higher educated persons (95% CI = 1.04–10.1); women of the poorest wealth quintile were 4.60 times more likely to be cut than those from the richest; women living in rural communities were 1.47 times more likely to have undergone FGM/C (95% CI = 1.31–1.65) than women living in urban areas, and 3.02 times more likely to be Muslim than another religion; Soninke women were 75.4 times more likely to undergo FGM/C than Wolof women. Women with FGM/C were least likely to live in Louga and Fatick and most likely in Matam and Kolda.
Table 3.9

Unadjusted and fully adjusted odds ratios of women’s circumcision across selected covariates (Senegal, DHS 2005, 2010)

Variable

Women 2005

Women 2010

Unadjusted OR and 95% CIa

Fully adjusted OR and 95% CIb

Unadjusted OR and 95% CIa

Fully adjusted OR and 95% CIb

Age

15–19 years

0.93(0.81, 1.07)

 

1.00(0.86, 1.15)

 

20–24 years

1.00(0.86, 1.15)

See Fig. 3.4 left

0.93(0.80, 1.08)

 

25–29 years

1.00

 

1.00

 

30–34 years

1.08(0.92, 1.27)

 

0.94(0.80, 1.10)

 

35–39 years

1.08(0.91, 1.27)

 

1.12(0.95, 1.33)

 

40–44 years

1.08(0.91, 1.29)

 

1.02(0.85, 1.23)

 

45–49 years

1.09(0.90, 1.32)

 

1.08(0.88, 1.33)

 

Age partner

15–30 years

1.00

See Fig. 3.4 right

1.01(0.88, 1.16)

 

31–49 years

0.80

 

1.00

 

50–60 years

0.82

 

0.98(0.86, 1.12)

 

>61 years

1.10

 

1.28(1.06, 1.54)

 

Place of residence

Urban

1.00

1.00

1.00

1.00

Rural

2.09(1.94, 2.25)

1.47(1.31, 1.65)

1.42(1.32, 1.53)

0.78(0.70, 0.87)

Married

Yes

2.22(1.99, 2.47)

 

1.47(1.33, 1.63)

 

No

1.00

 

1.00

 

Education

No education

2.05(1.13, 3.70)

3.26(1.04, 10.1)

4.10(2.57, 6.53)

1.86(0.74, 4.65)

Primary education

1.42(0.78, 2.57)

3.11(0.99, 9.73)

3.55(2.21, 5.68)

1.77(0.71, 4.45)

Secondary education

0.93(0.51, 1.71)

2.41(0.76, 7.63)

2.87(1.79, 4.61)

1.62(0.64, 4.13)

Higher education

1.00

1.00

1.00

1.00

Partner’s education

No education

1.96(1.44, 2.68)

1.30(0.87, 1.93)

1.69(1.18, 2.43)

1.21(0.81, 1.80)

Primary education

1.58(1.12, 2.22)

1.58(1.12, 2.22)

1.55(1.04, 2.30)

1.45(0.95, 2.21)

Secondary education

1.33(0.94, 1.88)

1.33(0.94, 1.88)

1.48(0.98, 2.23)

1.43(0.92, 2.22)

Higher education

1.00

1.00

1.00

1.00

Wealth index

Poorest

5.03(4.30, 5.88)

4.60(3.61, 5.86)

5.21(4.42, 6.14)

5.77(4.55, 7.33)

Poorer

5.37(4.59, 6.29)

4.52(3.57, 5.72)

3.05(2.58, 3.60)

3.35(2.64, 4.27)

Middle

3.42(2.92, 4.02)

2.44(1.94, 3.07)

2.22(1.87, 2.64)

2.16(1.70, 2.73)

Richer

1.97(1.64, 2.37)

1.79(1.37, 2.33)

1.51(1.25, 1.83)

1.37(1.05, 1.79)

Richest

1.00

1.00

1.00

1.00

Family size

Small (1–4 children)

0.60(0.50, 0.73)

0.98(0.79, 1.23)

1.00

1.00

Middle (5–7 children)

0.66(0.53, 0.83)

0.90(0.70, 1.16)

1.22(1.08, 1.37)

1.16(1.00, 1.33)

Large (8+ children)

1.00

1.00

1.46(1.19, 1.80)

1.11(0.88, 1.42)

Ethnicity

Wolof

1.00

1.00

1.00

1.00

Poular

1.79(1.67, 1.93)

23.9(20.1, 28.5)

129(98.6, 169)

19.9(16.9, 23.5)

Serer

0.02(0.01, 0.03)

0.22(0.14, 0.36)

2.56(1.70, 3.85)

0.39(0.26, 0.60)

Mandingu

2.96(2.36, 3.72)

67.3(45.4, 99.6)

478(326, 702)

89.4(55.5, 144)

Diola

1.53(1.26, 1.87)

20.3(14.3, 28.6)

125(89.6, 174)

29.4(20.0, 43.3)

Soninke

3.70(2.62, 5.21)

75.4(44.1, 129)

185(123, 280)

31.2(17.1, 56.6)

Not Senegalese

2.49(1.69, 3.67)

33.9(20.6, 55.9)

168(107, 264)

24.8(15.4, 39.9)

Other

0.64(0.54, 0.75)

11.2(8.41, 14.9)

50.1(36.8, 68.3)

9.22(6.88, 12.3)

Religion

Muslim

3.54(2.70, 4.64)

3.02(2.11, 4.34)

4.10(2.92, 5.77)

2.52(1.61, 3.96)

Other

1.00

1.00

1.00

1.00

Region of residence

Dakar

0.22(0.19, 0.25)

 

0.26(0.23, 0.31)

 

Diourbel

1.00

 

1.00

 

Fatick

0.07(0.05, 0.08)

 

0.11(0.09, 0.13)

 

Kaffrine

New state in 2010

 

0.12(0.10, 0.15)

 

Kaolack

0.14(0.12, 0.16)

See Fig. 3.2

0.08(0.06, 0.10)

 

Kédougou

New state in 2010

 

14.5(10.4, 20.4)

 

Kolda

16.5(13.2, 20.6)

 

7.10(6.02, 8.39)

 

Louga

0.05(0.04, 0.07)

 

0.05(0.04, 0.05)

 

Matam

18.2(14.0, 23.7)

 

8.65(7.12, 10.5)

 

Saint Louis

0.88(0.84, 0.93)

 

0.78(0.72, 0.85)

 

Sédhiou

New state in 2010

 

8.29(6.90, 9.94)

 

Tambacounda

6.98(5.95, 8.19)

 

7.32(6.20, 8.63)

 

Thiès

0.08(0.07, 0.10)

 

0.04(0.03, 0.05)

 

Ziguinchor

2.32(2.08, 2.58)

 

1.59(1.40, 1.80)

 

aUnadjusted marginal odds ratio (OR) from standard logistic regression models

bAdjusted marginal odds ratio (OR) from standard logistic regression models. Adjusted for women’s age, region and rural/urban location

In the following survey year of 2010–2011, factors associated with FGM/C in the unadjusted analysis were: rural place of residence (OR = 1.42, 95% CI = 1.32–1.53); being married (OR = 1.47, 95% CI = 1.33–1.63); education, with the category “no education” highest (OR = 4.10, 95% CI = 2.57–6.53), followed by primary education (OR = 3.55, 95% CI = 2.21–5.68) and secondary (OR = 2.87, 95% CI = 1.79–4.61) vs people with higher education; wealth index for women in the poorest quintile (OR = 5.21, 95% CI = 4.42–6.14), poorer women (OR = 3.05, 95% CI = 2.58–3.60), middle income (OR = 2.22, 95% CI = 1.87–2.64), and richer women (OR = 1.51, 95% CI = 1.25–1.83), as compared to richest women; family size for middle size family of 5–7 children (OR = 1.22, 95% CI = 1.08–1.37), and for large family size of 8+ children (OR = 1.46, 95% CI = 1.19–1.80) vs people with a small size family (1–4 children); ethnicity; religion and region of residence. After adjusting for all other factors, the likelihood of FGM/C in women living rural with any education became not statistically significant. The effect disappeared also for middle and large family size and partner’s education. A statistically significant effect remained only for the wealth index, ethnicity and religion. Women from the poorest quintile were 5.77 times more likely to undergo FGM/C than the richest (95% CI = 4.55–7.33), participants from the Mandingu tribe were 89.4 times more likely to be cut than people from the Wolof (95% CI = 55.5–144); women who were Muslim were 2.52 times more likely to have FGM/C than women who were not (95% CI = 1.61–3.96). Women with FGM/C were least likely to be in Thiès and Louga and most likely in Matam and Sédhiou.

Odds ratios of 129 and 478 predicted by the unadjusted models for ethnic groups of Poular and Mandingu respectively indicate a very high likelihood of undergoing FGM/C as compared to the Wolof ethnic group as predicted by the model. In terms of policy, this implies that children of women from the Poular and Mandingu ethnic groups are at higher risk of undergoing FGM compared to their counterparts in the Wolof ethnic group.

In terms of shift in the prevalence of FGM at the regional level and across age cohorts we examined the overall national prevalence of FGM/C between the 5 years period. The results suggest that there was only a slight decrease of 2% points from 30.1% in 2005 to 28.1% in 2010–2011 (unweighted average). Aggregate national figures on FGM/C prevalence, however, conceal important spatial variation at the region level within the survey periods. By the time of the 2010–2011 survey, three regions had been subdivided; to appropriately compare trends in prevalence, we reported the weighted average of subdivided regions in 2010–2011. The observed FGM/C prevalence at the regional level shown in Table 3.9 indicates that the regions in which FGM/C prevalence is lowest and below the national average in both surveys are Dakar, Diourbel, Fatick, Kaolack (along with Kaffrine in 2010–2011), Louga and Thiès. The prevalence of FGM/C was consistently higher than the national average in both surveys in Kolda (combined with Sédhiou in 2010–2011), Matam, Saint Louis, Tambacounda (combined with Kédougou in 2010–2011). Between 2005 and 2010–2011 the prevalence of FGM/C increased slightly in the low prevalence regions of Dakar (18.0 –20.9%) and Fatick (6.2–9.6%). It reduced slightly in the low prevalence regions of Diourbel (2.0–0.5%) and Thiès (7.3–3.8%). The high prevalence region of Tambacunda saw no substantial change in the prevalence of FGM/C between the 2005 and 2010–2011 surveys (87.5–88.7), while decreases were seen in Kolda (94.3–88.3), Matam (94.8–89.6), Saint Louis (46.9–44.0), and especially Ziguinchor (69.8–61.3). Unadjusted marginal odds ratios shown in Table 3.9 indicate that in 2005 the highest risk of FGM/C was in Matam (OR = 18.2, 95% CI = 14.0–23.7) and Kolda (OR = 16.5, 95% CI = 13.2–20.6), and the lowest risk in Louga (OR = 0.05, 95% CI = 0.04–0.07). In 2010–2011, the highest risk was in Kédougou, and the lowest was in Thiès and again in Louga. Regarding the effect of age, the unadjusted marginal odds ratios show no significant differences in FGM/C risk across age cohorts of women, suggesting that there is no secular decline in FGM/C risk.

As far as Bayesian spatial analysis results for Senegal are concerned, we introduced—in the context of multivariable Bayesian geo-additive regression analyses—spatial and nonlinear factors associated with higher FGM/C risk in both years. Region of residence was modelled as a spatial variable in Figs. 3.11 and 3.12, and age of the respondent at the time of interview was modelled as a continuous variable using a flexible nonlinear curve in Figs. 3.13 and 3.14. The modelled covariate results confirmed what was observed in the logistic regression analysis but the patterns differ markedly with region of residence and age remaining significant risk factors in both surveys. Overall, results of 2005 (Fig. 3.11) show that after accounting for (1) sampling error in the observed data; (2) relationships with covariates and the uncertainty in the form of these relationships); (3) uncertainty in the spatial autocorrelation structure of the outcome variable, the regions with the highest FGM/C risk included Tambacounda, Kolda, Matam and Ziguinchor, but not Saint Louis. In 2010 (Fig. 3.12), the highest risk regions included Matam, Tambacounda, Kolda, and the newly formed regions of Sédhiou and Kédougou, but again, not the high prevalence region of Saint Louis.
Fig. 3.11

Left: Adjusted total residual spatial effects for women’s circumcision, at regions level in Senegal in 2005. Shown are the posterior odds ratios. Right: Corresponding posterior probabilities at 90% nominal level (SDHS 2005). Red coloured—high risk. Black coloured—significant positive spatial effect. Green coloured—low risk. White coloured—significant negative spatial effect. Grey coloured—no significant effect

Fig. 3.12

Left: Adjusted total residual spatial effects for women circumcision, at regions level in Senegal in 2010. Shown are the posterior odds ratios. Right: Corresponding posterior probabilities at 90% nominal level (SDHS 2010). Red coloured—high risk. Black coloured—significant positive spatial effect. Green coloured—low risk. White coloured—significant negative spatial effect. Grey coloured—no significant effect

Fig. 3.13

Left: Estimated nonparametric trend of women’s FGM by respondent’s age cohort (left) and respondent partner’s age cohort in 2005 (right). Shown is the posterior mean within 80% credible regions (SDHS 2005)

Fig. 3.14

Left: Estimated nonparametric trend of women’s FGM by respondent’s age cohort (left) and respondent partner’s age cohort in 2010 (right). Shown is the posterior mean within 80% credible regions (SDHS 2010)

With regard to the shift of FGM/C by regions, in both samples, the spatial analysis has captured the substantial variation in FGM/C risk across regions observed in the marginal regression analyses. The results shown in Figs. 3.11 and 3.12 are in other words covariate-adjusted regional FGM/C spatial variations captured by the global total residual region effects (i.e. the sum of the unstructured and structured spatial effects). There is a clear pattern of regions with higher risk of FGM/C, mostly the south-eastern states of Tambacounda, Kolda and Matam in 2005, including the eastern state of Kédougou and the southern one of Sédhiou in 2010 (Figs. 3.12 and 3.13), which were associated with a higher risk of FGM/C, while states such as Louga, Thiès, Diourbel, Kaolack and Fatick in 2005 and Louga, Thiès, Diourbel, Fatick, Kaolack and Kaffrine in 2010 were associated with a lower risk of FGM/C. These spatial patterns confirm the observed marginal model findings shown in Table 3.7 with a shift observed for Kaolack region, which moved from a not significant FGM/C risk in 2005 to a very low significant FGM/C risk in 2010 (Kaffrine and Kaolack combined).

Specifically, the left-hand map in Fig. 3.11 shows estimated posterior total residual region odds of FGM/C for each region in 2005, ranging from a lower POR of 0.04(0.01, 0.18) in Diourbel to a higher POR of 27.83(7.36, 132.76) in Kolda and in 2010 the POR ranges as low as 0.02(0.00, 0.07) in Diourbel to a higher POR of 17.57(3.96, 67.30) in Matam, with red color indicating the higher risk recorded and green color denoting lower risk. The right-hand map shows the 95% posterior probability map of FGM/C, which indicates the statistical significance associated with the total excess risk. White indicates a negative spatial effect (associated with reduced risk of FGM/C prevalence), black a positive effect (an increased risk) and grey a not significant effect. However, the total spatial residuals in Figs. 3.10 and 3.11 in both surveys show that much of the variation in FGM/C likelihood remains to be explained. The spatial effects of the Kaolack region in 2005 was greatly attenuated after multiple adjustments of other risk factors indicating that perhaps the higher number of FGM/C affected women living in the state was inflated by other factors such as ethnicity, socio-economic status and education.

Overall, the results indicate that across surveys, certain high prevalence regions remain “hot spots” regarding FGM/C risk. These include Kolda (along with the newly subdivided region of Sédhiou in 2010), Tambacounda (along with the newly subdivided region of Kédougou in 2010), and Matam. Risk remained not significant in the high prevalence regions of Saint Louis and Zinguinchor, and was attenuated between 2005 and 2010–2011 in Kaolack (including the newly subdivided region of Kaffrine in 2010–2011).

Figures 3.13 and 3.14 show the estimated nonparametric trend of women’s FGM/C risk by respondent’s age cohort (left) and respondent partner’s age cohort in 2005 and 2010 (right). Shown is the posterior mean within 80% credible regions. Surprisingly, the figures show an inverse U-shape non-linear relationship between the likelihood of FGM/C and women’s age, with a higher risk of FGM for younger cohorts (under 20 years of age) for both samples. The nonlinear association between age and the likelihood of FGM/C before age 20 hardly differs in the two samples starting very high, with a gradual decrease thereafter in both surveys. At all other ages, the two surveys also show agreement in the pattern of decreasing probability of FGM until age 49. At age 40, this probability decreases quickly as age increases, although the variation in probability increases rapidly at the same time. For women over age 45, there are wide confidence intervals suggesting few observations in both samples that make it difficult to discern a consistent downward decrease in the FGM/C risk for this cohort (instability of the estimates) in both surveys. It is worth mentioning that at first glance, the figures seem to be different. However, a careful examination of the two figures reveals that in both figures the age’s effects start at an estimate of 1, with a gradual decrease thereafter. Thus, contrary to expectations, the adjusted non-parametric estimates reveal that the risk of FGM/C increases with decreasing age.

We run a stepwise regression analysis to identify which factors were responsible for the change of age effects. It turns out that, although most factors we included in the final model impacted on the effects of age, factors that have a significant impact or factors confounding the effects of respondent’s age on the likelihood of FGM were the place of residence (urban/rural), the ethnicity, religion and the region of residence.

Overall, among women aged 15–49 in Senegal, the prevalence of FGM/C has changed little over the 5-year period between successive surveys carried out in 2005 and 2010–2011 (FGM/C prevalence of 30.1% in 2005 and 28.1% in 2010–2011). We observed that these unadjusted figures did, in fact, mask important variation at both the regional and individual levels. In the multivariate Bayesian geo-additive regression analysis, we controlled for individual-level factors while simultaneously modelling the region of residence as a spatial variable. Thus, the risk of FGM/C varies across regions, with the highest risk across both survey periods found in the high-prevalence southeastern regions including Kolda (along with the newly subdivided region of Sédhiou in 2010), Tambacounda (along with the newly subdivided region of Kédougou in 2010), and Matam. In two other high prevalence regions—Saint Louis and Zinguinchor—risk of FGM/C remained not significant across both surveys. Additionally, the likelihood of FGM/C was attenuated between 2005 and 2010–2011 in Kaolack (including the newly subdivided region of Kaffrine in 2010–2011), shifting from being not significant in 2005 to very low risk of FGM/C in 2010–2011.

How can we interpret these spatial findings given that certain high prevalence regions remained “hot spots” regarding FGM/C risk and others did not? Importantly, these results show that community-level effects, above and beyond individual-level effects, play a crucial role in determining the likelihood of FGM/C. In other words, the context, in which an individual woman lives, bears an important influence on whether FGM/C is practiced, which is in line with social convention theory as mentioned earlier in the text. The theory postulates that change is most likely to come about when members of social groups have simultaneously shifted social norms pertaining to FGM/C (Mackie 2000). It may be the case that regional differences in FGM/C risk capture this shift in social norms. At the same time, theory on social norms and conventions does not rule out the possibility of individual-level factors influencing decision-making regarding FGM/C, although the social environment can constrain these choices. Indeed, in our study we find evidence for the simultaneous influence of community- and individual-level factors influencing the risk of FGM/C.

In the fully adjusted model, a number of individual-level factors were found to be associated with the likelihood of FGM/C. In the 2005 survey data these include rural residence, no education, wealth, religion, and especially ethnicity (Soninke women were 75.4 times more likely to be cut than Wolof women) and age. In 2010–2011, significant individual-level variables in the fully adjusted model included wealth, religion, and especially ethnicity (Mandingu women were 89.4 times more likely to be cut than Wolof women) and age. The consistent and robust effect of ethnicity is unsurprising given that in Senegal, like many other setting settings, FGM/C derives much of its meaning and tenacity from its intimate association with ethnic identity (Gruenbaum 2001). Where this exist a strong link between FGM/C and ethnicity, as is the case in Senegal, ethnicity may serve to signal shared expectations that hold the practice in place. In other words, ethnicity may be a proxy for shared norms concerning personhood, religion, sexual restraint or other cultural values. Hence, it is increasingly understood that programming efforts should be uniquely tailored to address these issues (UNICEF 2013).

Our most striking finding was with respect to individual-level predictors of risk of FGM/C among women in Senegal. This related to age. Unadjusted estimates of risk of FGM/C showed no significant variation across age cohorts. However, fully adjusted nonparametric estimates show that in both surveys, age is a significant risk factor for FGM/C, but not in the anticipated direction. The effect of age on the likelihood of FGM/C is highest in women aged 15–20, and declines with increasing age. How can we understand this unexpected finding? The anticipation of reductions in FGM/C detectable in survey data has been driven by several factors. These include combined efforts at the local level, most notably the holistic development program of Tostan that by 2010 culminated in more than 4000 communities participating in public declarations to abandon FGM/C (www.tostan.org). Moreover, community-based programs have been supported by a strong national framework to create an “enabling environment” for the abandonment of FGM/C, including developing a detailed national plan of action and implementing legislative reform strategies (UNICEF 2013; Diop-Diagne 2008; Shell-Duncan et al. 2013). Thus, a recent evaluation of the UNFPA-UNICEF Joint Programme on Female Genital Mutilation/Cutting concludedthat “Senegal has made concrete progress toward abandonment of FGM/C,” and further speculated that “Senegal could be free from the practice in the near future” (UNFPA-UNICEF 2013).

When examining trends in FGM/C using nationally representative survey data, we noted several salient considerations. First, since data on national prevalence provides information on the proportion of women aged 15–49 who have undergone FGM/C, this aggregate number is unlikely to change dramatically in consecutive surveys implemented 5 years apart; women who are cut will remain cut, and the national prevalence estimate changes only as an increasing proportion of uncut women age into the 15–49 group, and cut women age out. A more sensitive indicator of change is to look at the prevalence along age cohorts, as we have done in this study. Second, in doing so, we need to take into account age at cutting, as women’s FGM/C status reflects an event that took place some years in the past (Yoder and Wang 2013; UNICEF 2013). In Senegal, over 60% of girls are cut by age 1, and nearly three quarters of girls are cut before the age of 5 (UNICEF 2013). Thus, when looking at rates of FGM/C in women aged 15–19, we are examining the results of an event that likely took place between 11 and 19 years prior to collection of the survey data. Notably, the youngest cohort of women in the 2010–2011 survey were born prior to passage of the law banning FGM/C, and at a time where the Tostan program’s work on FGM/C was just beginning to scale up in Senegal (Tostan 1999). To detect more recent changes, it is possible to examine trends in FGM/C among daughters of the survey respondents. The results of our analysis of daughter data from Senegal DHS are forthcoming. Given the timing of scaled-up local intervention activities and implementation of legal reform efforts, it may not be reasonable to expect to see a dramatic decline in FGM/C risk in the youngest cohort of women in the 2010–2011 survey. However, the finding of increased risk of FGM/C in the youngest women in the 2005 and 2010–2011 samples is a puzzling finding that we cannot easily explain. Further research will be required to understand this observation, as well as to assess whether FGM/C risk has begun to decline in girls under the age of 15.

The above suggests that the complex relationship between FGM prevalence and geographical location can effectively be disaggregated using an innovative complex Bayesian modelling framework. Such framework also identifies socioeconomic determinants, and the spatial effects that are not explained by these socioeconomic determinants while accounting for the complex sampling scheme. From the analysis of the two household surveys, we are able to observe that there is an increased risk of FGM/C in the youngest women in the 2005 and 2010–2011 as the trends in FGM/C among women across Senegal and within regions are variable. Some regions managed to reduce the prevalence of FGM during the study period while in other regions the prevalence stayed either stagnant or increased. Among factors significantly changing the impact of age on FGM, we found that place of residence (urban/rural), the ethnicity, religion and the region of residence were the most significant factors. The effect of community-level factors, captured by covariate-adjusted geographic estimates is a very significant factor as it mirrors the importance of FGM/C as social norm therefore confirming our initial hypothesis of the social convention theory. The adjusting for individual- and community-level factors has given us a clearer picture of the complex interplay between FGM/C and geography, which needs to be investigated in more detail to see the extent to which this effect is causal or a mere association.

3.3.7.4 D-The Central African Republic (CAR)

Moving, finally, to the mapping of FGM/C prevalence in the Central African Republic (C.A.R), we note that this French former colony has a 2014 population of about five million, is home to more than 80 ethnic groups, each with their own language, culture and history. The country is divided into 16 administrative prefectures and an autonomous commune (Fig. 3.14). We also note that about 43% of the Central African population practices FGM/C, predominantly non-infibulating forms. FGM/C in the Central African Republic is generally understood to be tied to ethnicity, with some exceptions, due in part to increasingly common inter-ethnic marriages that are FGM/C-incongruent, where one partner comes from a family that practices FGM/C, while the other does not. Generally, the practice is passed down from elders to the younger generations by serving as an obligation for acceptance and social integration (Shell-Duncan et al. 2011).

Beyond demarking group membership, ethnographic studies reveal a multitude of reasons for the practice related to constructs related to gender ideology and proper childrearing. In some groups, the practice serves as an important rite of passage that marks the transition from childhood to adulthood. In recent decades, however, there is a trend in practicing communities to uncouple FGC from initiation, and perform the procedure at younger and younger ages (Shell-Duncan et al. 2010). Nonetheless, in many groups FGM/C is still widely held to be essential for the moral upbringing of girls, teaching them, among other things, sexual restraint and how to display respect to elders. Sixty-eight percentage of girls in the Central African Republic are married before they are 18. FGC is seems not bound to any religion in the Central African Republic, providing the purity and cleanliness believed to be required for girls.

Across different communities, the cultural meanings of FGM/C are multiple, fluid, and increasingly contested and negotiated by diverse groups of people who draw on local social and political movements, as well as national and international campaigns aimed at ending FGM/C.

The data summarized in Table 3.10 describe the circumstances surrounding the practice of FGM/C in the Central African Republic. The most common form of cutting is “cut/flesh removed,” which may correspond to FGM Type I (clitoridectomy) or Type II (excision).4
Table 3.10

Effects of the predictors from the prediction model

Predictors

Levels

OR

Crl

Residence (Urban area as reference)

Rural areas

1.61

(1.38, 1.88)

Education (No education as reference)

Primary

0.73

(0.56, 0.95)

Secondary

0.82

(0.63, 1.06)

Higher

0.61

(0.48, 0.78)

Other

0.55

(0.44, 0.69)

Religion (Roman Catholic as reference)

Protestant

1.11

(0.98, 1.25)

Muslim

1.56

(1.09, 2.22)

Animalist

1.30

(0.95, 1.78)

Ethnicity (Haoussa as reference)

Other

1.66

(0.92, 2.99)

Sara

0.46

(0.26, 0.82)

Mboum

0.26

(0.13, 0.49)

Gbaya

0.99

(0.58, 1.70)

Mandjia

1.85

(1.07, 3.19)

Banda

3.03

(1.76, 5.22)

Ngbaka-bantou

0.28

(0.16, 0.49)

Yakoma-sango

0.10

(0.05, 0.17)

Zande-nzakara

0.22

(0.12, 0.43)

Other

1.08

(0.58, 1.99)

The data analyzed in this section originates from a nationally representative household survey: the 1995 Central African Demographic and Health Survey (CADHS).

We draw on data from the core questionnaire for households, as well as the module on FGM/C, administered to women age 15–49 years. Further details on the methods, objectives, organization, sample design, and questionnaires used in the 1995 CADHS are described elsewhere in the DHS country final report).

The sampling strategy for this survey was designed to be nationally representative, and provide information for each region. As was the case with other countries, we studied FGM as the main outcome in terms of “whether a participant had had FGM performed on her”. This question was converted into a binary variable, with two categories defined as 1 if the participant was cut and 0 if the participant had no FGM performed on her. The main exposure variable in the analysis was the “ethnicity”, in addition to various control variables on socio-demographic factors potentially associated with FGM: education level and place of residence (urban vs. rural). Age was recorded as a continuous variable, and was re-coded into a categorical variable of 5-year age cohorts in the preliminary analysis. Education level was categorized as “None”, “Primary”, “Secondary”, “Higher” and “Other”. As far as the descriptive results are concerned, these are shown as per prefecture in Fig. 3.16. The overall prevalence of FGM/C differs greatly between the prefectures. Most of the population sampled lived in rural settings and half the population was illiterate. Women with FGM/C were mostly without education, lived in rural areas and were living in Bamingui-Bangora, Kemo and Vakaga.

The main regression analysis results are presented in terms of marginal odds ratios in Table 3.9. In 1995 factors associated with FGM/C in the analysis were: rural place of residence (OR = 1.61, 95% CI = 1.38–1.88), being Primary, Higher or Other educated (OR = 0.73, 95% CI = 0.56–0.95; OR = 0.61, 95% CI = 0.48–0.78; OR = 0.55, 95% CI = 0.44–0.69), being Muslim (OR = 1.56, 95% CI = 1.09–2.22), being Banda (OR = 3.03, 95% CI = 1.76–5.22) or Mandjia (OR = 1.85, 95% CI = 1.07–3.19) and living in Bamingui-Bangora or Kemo. Women with FGM/C were least likely to live in Haut-Mboumou and Ouham-Pende.

We introduced and controlled for spatial and nonlinear factors associated with higher FGM/C risk in both years. Governorate of residence was modelled as a spatial variable in Figs. 3.16, 3.17, and 3.18, and age of the respondent at the time of interview was modelled as a continuous variable using a flexible nonlinear curve in Fig. 3.15. The modelled covariates results confirmed what was observed in the logistic regression analysis. Overall, results of 1995 (Fig. 3.18) show that after accounting for (1) sampling error in the observed data; (2) relationships with covariates and the uncertainty in the form of these relationships); (3) uncertainty in the spatial autocorrelation structure of the outcome variable, the regions with the highest FGM/C risk included Bamingui-Bangoran, Haute-Kotto, Vakaga, Ouaka, Basse-Kotto, Kemo, Nana-Grebizi, Ouham and Ombella-M’Poko.
Fig. 3.15

Map of the Central African Republic showing the 17 administrative regions

With regard to the shift of FGM/C by regions, in both samples, the spatial analysis has captured the substantial variation in FGM/C risk across regions observed in the marginal regression analyses. The results shown in Figs. 3.16, 3.17, and 3.18 are in other words covariate-adjusted region FGM/C spatial variation captured by the global total residual region effects (i.e. the sum of the unstructured and structured spatial effect). There is a clear pattern of regions with higher risk of FGM/C, mostly the prefectures of Bamingui-Bangoran, Haute-Kotto, Vakaga, Ouaka, Basse-Kotto, Kemo, Nana-Grebizi, Ouham and Ombella-M’Poko in 1995, which were associated with a higher risk of FGM/C, while prefectures such as Haut-Mboumou, Ouham-Pende, Nana-Mambere and Sangha-Mbaere were associated with a lower risk of FGM/C.
Fig. 3.16

Distribution of survey clusters of the CADHS (1995)

Fig. 3.17

Residual spatial effects for women circumcision, at regions level in Central African Republic in 1995

Fig. 3.18

Observed rates per prefecture in the Central African Republic in CADHS (1995)

The mapping results for the Central African Republic indicate that across prefectures, certain high prevalence regions remain “hot spots” regarding FGM/C risk. These include Bamingui-Bangoran, Haute-Kotto, Vakaga, Ouaka, Basse-Kotto, Kemo, Nana-Grebizi, Ouham and Ombella-M’Poko. We used advanced statistical methodology to analyze survey data collected with complex sampling strategies, and including possible non-linear covariates. Overall, among women aged 15–49 in that country, the prevalence of FGM/C has changed a lot since the survey. We find that these unadjusted figures do indeed mask important variation at both the regional and individual levels. In the multivariate Bayesian geo-additive regression analysis, we controlled for individual-level factors while simultaneously modelling the region of residence as a spatial variable.

The spatial analysis reveals that the risk of FGM/C varies across regions, with the highest risk across the survey periods found in Baminui-Bangoran and Kemo. There is a sense in which it is difficult to make sense of these spatial findings given that certain high prevalence regions remained “hot spots” regarding FGM/C risk while and others did not. One possible explanation would be that community level effect, play a crucial role in determining the likelihood of FGM/C. In other words, the context, in which an individual woman lives, bears an important influence on whether FGM/C is practiced, consistent with social convention theory (Mackie and LeJeune 2009). As in Senegal, we find evidence for the simultaneous influence of community- and individual-level factors influencing the risk of FGM/C (Figs. 3.19 and 3.20).
Fig. 3.19

Estimated nonparametric trend of women’s FGM by prefecture (CADHS 1995)

Fig. 3.20

Standard deviation by prefecture (CADHS 1995)

3.4 Discussion and Implications

The main aim of presenting the findings about the above four countries was to show how the picture about FGM appears contrasting at regional and sub-regional level in each selected country. The intention was to demonstrate and contrast the cultural mix. The findings mean that in West Africa (typified by Nigeria) the FGM prevalence was on the increase whereas the situation in North Africa (e.g. Egypt) and West Africa (e.g. Senegal) and Central Africa (as exemplied by the Central African Republic) showed signs of slight decrease. The main lesson can be discussed in terms of national, sub-national as well as regional levels for each of the countries discussed. Besides lessons can also be drawn in terms of the socio-economic and cultural predictors of FGM for the four countries.

From the point of view of national trends, the overall prevalence for Egyptian women aged 15–49, for example, was estimated at 91.9% between 1995 and 2008, which differs slightly between the surveys. As pointed out in various analyses of DHS data in the four countries, behind these national averages lurk important variations within countries. The highest FGM/C prevalence just noted in the case of Egypt appears to be an indication of the situation at the national level only, but at the sub-national plane within that country the range of FGM/C varied from 22.0% in Matrouh, 70.6% in North Sinai, 99.0% in Qena and 99.9% in Al Sharqia. These within country variations were also observed in Nigeria, Senegal and C.A.R. These variations are better investigated using our spatial analysis approach.

Moving now to the sub-national and regional trends within countries, it is remarkable that between 1995 and 2014 inclusive, the FGM/C rates observed (see the spread of maps for each of the countries) the steady increase in prevalence is the key feature noted in the case of Nigeria (West Africa). But the dataset points to a different story in relation to Egypt (North Africa), Senegal (West Africa) and the C.A.R—where the prevalence has decreased but only to a lesser degree. On balance, the FGM/C practice persists in the sense that the rates remain largely similar to what they were decades ago. This appears the case despite years of investment in campaigns designed to root out the practice. This lack of any substantial reduction is a serious cause for concern and it is something that is supported by an examination of standard deviation prediction maps of the predicted posterior. In other words, the findings mean that within each country a shift has occurred from lower FGM/C rates in the previous years to higher rates and vice versa. One could legitimately ask what factors might explain this upward trend and shift. It is a question, which goes far beyond the proper scope of this chapter. But, we can note, en passant, that the shift in attitude in countries like Senegal and others demonstrates these countries willingness to effectively ban FGM/C and avert the excruciating burden that this practice imposes on these countries. By contrast, in the case of regions where the change in attitude has been slow, much could be ascribed to the weight of traditions which are difficult to reform or abandon—especially in the light of the fact that many FGM/C-prone groups continue to refer to FGM as a route to a good marriage or a part and parcel of religious instructions.

As far as the socio-economic and cultural predictors of FGM/C in the four countries are concerned, it is clear from existing studies that a slight decline has occurred in the subsequent years after 2014, although we must hasten to note the significant variations within regional and ethnic estimates. From the statistical point of view, therefore, the likelihood of FGM/C remains significantly higher among those populations with no education. This was notably in the case among respondents within some ethnic groups-such as those in Northern parts of Nigeria, which is home to rural and poorer households. Here and in similar settings-as indicated by the probability maps or standard deviations of the predicted posterior- the total residual spatial effects prove significant. For example, on the one hand, we note that the posterior probability maps of FGM/C risk are estimated at a 95 % credible interval. The areas shaded in black on the maps suggest a significantly positive spatial effect associated with higher FGM/C risk. On the other hand, we observed from the maps that the areas shaded in white denote a significantly negative spatial effect, implying lower FGM/C risk. The remainders of the areas shaded in gray suggest that no significant effect exists with respect to FGM/C risk. The overall message is that the pattern of risk associated with FGM/C can readily be observed in each country as well as at regional and sub-national levels.

The foregoing analysis implies that monitoring this practice using national proportions can mask within country variability. That approach hardly assists in the design and implementation of policies and interventions at the local level. These findings provide new evidence to support the use of GIS methods in order to model the complex phenomenon of FGM/C around the world.

Overall, the findings, through use of advanced statistical techniques and methods, were made particularly interesting and evidently novel. They are novel because traditional and conventional statistical techniques using said fixed effects variables for the location hardly capture the complexity behind the FGM/C phenomenon. What we set out to do here with spatial analysis techniques is to make sense of this complexity and to unearth what burden lays behind the reality represented by traditional statistical methods—namely that there are more behind official accounts of decrease of FGM/C prevalence than the traditional methods can uncover. Thus, through use of advanced modeling techniques, we have disentangled the within country dynamics of FGM/C, having regard to the complexity of intermingly cultures where this practice has to a varying degree been the most important social norm. Based on those techniques and methods we have shown that the existing household survey data provides misleading cross-sectional descriptive results. Such results tell us nothing about the extent to which the FGM/C prevalence is either decreasing or increasing at the sub-national and regional level.

However, one could object that where the spatial analysis demonstrates the correlation between where people live, their ethnicity, religion and whether or not they perform FGM/C, there is a danger that such a conclusion may add fuel to ethnic and religious strife by adding another dimension of ‘othering’. One response to this objection is to argue that such a finding, however sensitive, is directed to policy makers and governments responsible for addressing FGM in those regions.

Footnotes

  1. 1.

    This conclusion is part of the effort to compliment works of campaigns against FGC that have existed in Nigeria since the 1980s in organisations such as: The media (by the Nigerian Council of Women broadcasts), the Inter-African Committee of Nigeria (organise community meetings) and medical professionals. After it has been suggested that Nigeria has a strong social convention supporting FGC (Freymeyer and Johnson 2007), we examine how this social convention varies across space (states) and the role that individual factors play in influencing this relationship.

  2. 2.

    This section represents a nontechnical summary of our 2018 article, Kandala N-B and Shell-Duncan B. Trends in female genital mutilation/cutting in Senegal: What can we learn from successive household surveys in sub-Saharan African countries? Unpublished.

  3. 3.

    Yoder and Wang (2013) describe the evolution of DHS questions designed to determine type of cutting.

  4. 4.

    Yoder and Wang (2013) describe the evolution of DHS questions designed to determine type of cutting.

References

  1. Adinma, J. (1997). Current status of female circumcision among Nigerian Igbos. West African Journal of Medicine, 16, 227–231.PubMedGoogle Scholar
  2. Askew, I. (2005). Methological issues in measuring the impact of interventions against female genital cutting. Culture, Health and Sexuality, 7(5), 463–477.CrossRefPubMedGoogle Scholar
  3. Belitz, C., Brezger, A., Kneib, T., & Lang, S. (2012). BayesX software for Bayesian inference in structured additive regression models version 2.0.1.Google Scholar
  4. Boerma, J. T., Black, R. E., Sommerfelt, A. E., Rustein, S. O., & Bicego, G. T. (1991). Accuracy and completeness of mother's recall of diarrhoea occurrence in pre-school children in demographic and health surveys. International Journal of Epidemiology, 20, 1073–1080.CrossRefPubMedGoogle Scholar
  5. Bolstad, M. W., & Manda, O. S. (2001). Investigating child mortality in Malawi Using family and community random effects: A Bayesian analysis. Journal of American Statistical Associationt, 96(453), 12.CrossRefGoogle Scholar
  6. Borgoni, R., & Billari, F. C. (2003). Bayesian spatial analysis of demographic survey data: An application to contraceptive use at first sexual intercourse. Demographic Research, 8(3), 61–92.CrossRefGoogle Scholar
  7. Chilés, J. P., & Delfiner, P. (1999). Geostatistics: modeling spatial uncertainty, Wiley series in probability and statistics. New York: Wiley.CrossRefGoogle Scholar
  8. Creel, L. (2001). Abandoning female genital cutting: Prevalence, Attitudes, and efforts to end the practice. Washington, DC: Population Reference Bureau.Google Scholar
  9. Dellenborg, L. (2007). Multiple meanings of female initiation: “Circumcision” among Jola women in Lower Casamance, Senegal. Gothenburg: Gothenburg University.Google Scholar
  10. Diggle, P., Moyeed, R., Rowlingson, B., & Thomson, M. (2002). Childhood malaria in The Gambia: a case-study in model-based geostatistics. Journal of the Royal Statistical Society: Series C (Applied Statistics), 51, 493–506.CrossRefGoogle Scholar
  11. Diggle, P. J., & Ribeiro, P. J. (2007). Model-based geostatistics. New York: Springer.Google Scholar
  12. Diop-Diagne, A. (2008). Evaluation of the national plan of action for the abandonment of the practice of Female Genital Mutilation. Republic of Senegal, Minister of the Family, National Solidarity, and Feminine Entrepreneurship.Google Scholar
  13. Draper, D. (2000). Bayesian hierarchical modeling. Tutorial presented at the sixth world meeting of the international society of Bayesian analysis, Creta, May 28, 2000.Google Scholar
  14. Elmusharaf, S., Elhadi, N., & Almroth, L. (2006). Reliability of self reported form of female genital mutilation and WHO classification: cross sectional study. BMJ, 333(7559), 124–128.CrossRefPubMedPubMedCentralGoogle Scholar
  15. Fahrmeir, L., & Lang, S. (2001). Bayesian inference for generalized additive mixed models based on Markov random field priors. Journal of the Royal Statistical Society: Series C (Applied Statistics), 50, 201–220.CrossRefGoogle Scholar
  16. Fahrmeir, L., Kneib, T., & Lang, S. (2004). Penalised structured additive regression for space-time data: A Bayesian perspective. Statistica Sinica, 14, 731–761.Google Scholar
  17. Filmer, D., & Pritchett, L. (2001). Estimating wealth effects without expenditure data—Or tears: An application to educational enrollments in States of India. Demography, 38(1), 115–132.PubMedGoogle Scholar
  18. Freymeyer, R. H., & Johnson, B. E. (2007). An exploration of attitudes toward female genital cutting in Nigeria. Population Research and Policy Review, 26, 69–83.CrossRefGoogle Scholar
  19. Gruenbaum, E. (2001). The female circumcision controversy: An anthropological perspective. Philadelphia: University of Pennsylvania Press.CrossRefGoogle Scholar
  20. Hennerfeind, A., Brezger, A., & Fahrmeir, L. (2006). Geoadditive survival models. Journal of the American Statistical Association, 101(475), 1065–1075.CrossRefGoogle Scholar
  21. Jackson, E. F., Akweongo, P., Sakeah, A., Hodgson, R., Asuru, R., & Phillips, J. E. (2003). Inconsistent reporting of female genital cutting status in Northern Ghana: Explanatory factors and analytical consequences. Studies in Family Planning, 34(3), 200–210.CrossRefPubMedGoogle Scholar
  22. Kammann, E. E., & Wand, M. P. (2003). Geoadditive models. Journal of the Royal Statistical Society: Series C (Applied Statistics), 52, 1–18.CrossRefGoogle Scholar
  23. Kandala, N. B., & Ghilagaber, G. (2006). A geo-additive Bayesian discrete-time survival model and its application to spatial analysis of childhood mortality in Malawi. Quality & Quantity: International Journal of Methodology, 40(6), 935–957.CrossRefGoogle Scholar
  24. Kandala, N.-B., & Komba, P. N. (2015). Geographic variation of female genital mutilation and legal enforcement in sub-Saharan Africa: A case study of Senegal. The American Journal of Tropical Medicine and Hygiene, 92(4), 838–847.CrossRefPubMedPubMedCentralGoogle Scholar
  25. Kandala, N. B., Magadi, M. A., & Madise, N. J. (2006). An investigation of district spatial variations of childhood diarrhoea and fever morbidity in Malawi. Social Science & Medicine, 62, 1138–1152.CrossRefGoogle Scholar
  26. Kandala, N. B., Nwakeze, N., & Kandala, S. (2009). Spatial distribution of female genital mutilation in Nigeria. American Journal of Tropical Medicine and Hygiene, 81(5), 784–792.CrossRefPubMedGoogle Scholar
  27. Kandala, N.B., & Shell-Duncan, B. (2015). Trends in female genital mutilation/cutting in Senegal: What can we learn from successive household surveys in sub-Saharan African countries? Paper prepared for the Annual Meeting of the Population Association of America, Santiago, CA, April 30–May 2.Google Scholar
  28. Klouman, E., Manongi, R., & Klepp, K. I. (2005). Self-reported and observed female genital cutting in rural Tanzania: Associated demographic factors, HIV and sexually transmitted infections. Tropical Medicine and International Health, 10(1), 105–115.CrossRefPubMedGoogle Scholar
  29. Lang, S., & Brezger, A. (2004). Bayesian P-splines. Journal of Computational and Graphical Statistics, 13, 183–212.CrossRefGoogle Scholar
  30. Langford, I. H., Leyland, A. H., Rabash, J., & Goldstein, H. (1999). Multilevel modeling of the geographical distributions of diseases. Journal of the Royal Statistical Society: Series C (Applied Statistics), 48, 253–268.CrossRefGoogle Scholar
  31. Mackie, G. (2000). Female genital cutting: The beginning of the end. In B. Shell-Duncan & Y. Hernlund (Eds.), Female “circumcision” in Africa: Culture, controversy, and change (pp. 253–283). Boulder, CO: Lynne Rienner Publishers.Google Scholar
  32. Mackie, G., & LeJeune, J. (2009). Social dynamics of abandonment of harmful practices: A new look at the theory, Special series on social norms and harmful practices: Innocenti working paper no. 2009-06. Florence: UNICED Innocenti Research Centre.CrossRefGoogle Scholar
  33. Msuya, S. E., Mbizvo, E., Hussain, A., Sundby, J., Sam, N. E., & Stray-Pedersen, B. (2002). Female genital cutting in Kilimanjaro, Tanzania: Changing attitudes? Tropical Medicine and International Health, 7(2), 159–165.CrossRefPubMedGoogle Scholar
  34. National Population Commission. (2004). Nigeria demographic an health survey 2003. Caverton, MD: Author.Google Scholar
  35. Ndiaye, S., & Ayad, M. (2006). Enquête Démographique et de Santé au Sénégal 2005. Calverton, MD: Centre de Recherche pour le Développement Humain (Sénégal) et ORC Macro.Google Scholar
  36. Nnorom, C. C. P. (2005). Female Genital Mutilation practice in Nigeria: Patterns, preference and remedies, 2005, 2(1).Google Scholar
  37. Nwakeze, N. M. (2001). Economic decision-making power in the household: The case of Anambra women. Journal of Women Academics, 1(3), 124–134.Google Scholar
  38. Rogers, E. M. (1995). Diffusion of innovations (4th ed.). New York: Free Press.Google Scholar
  39. Rue, H., Martino, S., & Chopin, N. (2009). Approximate Bayesian inference for latent Gaussian models by using integrated nested Laplace approximations. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 71, 319–392.CrossRefGoogle Scholar
  40. Shell-Duncan, B., Hernlund, Y., Wander, K., & Moreau, A. (2010). Contingency and change in the practice of female genital cutting: dynamics of decision making in Senegambia. Summary report, from http://csde.washington.edu/bsd
  41. Shell-Duncan, B., Wander, K., Hernlund, Y., & Moreau, A. (2011). Dynamics of change in the practice of female genital cuttingin Senegambia: Testing predictions of social convention theory. Social Science and Medicine, 73, 1275–1283.CrossRefPubMedGoogle Scholar
  42. Shell-Duncan, B., Wander, K., Hernlund, Y., & Moreau, A. (2013). Legislating change? Responses to criminalizing female genital mutilation/cutting in Senegal. Law and Society Review, 47(4), 803–835.CrossRefPubMedGoogle Scholar
  43. Snow, R. C., Slanger, T. E., Okonofu, F. E., Oronsaye, F., & Wacker, J. (2002). Female genital cutting in sourthern urban and peri-urban Niveria: Self-reported validity, social determinants and secular decline. Tropical Medicine and International Health, 7(1), 91–100.CrossRefPubMedGoogle Scholar
  44. Spiegelhalter, D., Best, N., Carlin, B., & Van der Line, A. (2002). Bayesian measures of models complexity and fit. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 64, 1–34.CrossRefGoogle Scholar
  45. Spiegelhalter, D. J., Best, N. G., Carlin, B. P., & van der Linde, A. (2002). Bayesian measures of model complexity and fit. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 64, 583–639.CrossRefGoogle Scholar
  46. Sylla, M., & Helene, S. (1990). Excision au Sénégal. Dakar: ENDA.Google Scholar
  47. Tostan. (1999). Breakthrough in Senegal: Ending female genital cutting. TOSTAN, from www.tostan.org
  48. UNFPA-UNICEF. (2013). Joint evaluation, UNFPA-UNICEF joint programme on female genital mutilation/cutting: Accelerating change, 2008–2012. Country case study. Executive summary, Senegal.Google Scholar
  49. UNICEF. (2010). Dynamics of social change: Toward the abandonment of female genital mutilation/cutting in five African countries. Florence: Innocenti Research Center.Google Scholar
  50. UNICEF. (2013). Female genital mutilation/cutting: A statistical overview and exploration of the dynamics of change. New York: Author.Google Scholar
  51. Yoder, P. S., Abderrahim, N., & Zhuzhani, A. (2004). Female genital cutting in the demographic and health surveys: A critical comparative analysis. Calverton, MD: Macro International.Google Scholar
  52. Yoder, P. S., & Wang, S. (2013). Female genital cutting: The interpretation of recent DHS data (DHS Comparative Rep. No. 33). Calverton, MD: ICF International.Google Scholar

Copyright information

© Springer International Publishing AG, part of Springer Nature 2018

Authors and Affiliations

  • Ngianga-Bakwin Kandala
    • 1
  • Paul Nzinga Komba
    • 2
  1. 1.Department of Mathematics Physics and Electrical Engineering, Faculty of Engineering and EnvironmentNorthumbria UniversityNewcastle upon TyneUK
  2. 2.Wolfson CollegeCambridgeUK

Personalised recommendations