« Marc F. Bellemare† Academic Coordinator, University of California Davis, Davis, CA, 95616, Telephone: (530) 752-7252, Email: ...»
All in the Family: Explaining the Persistence of
Female Genital Cutting in The Gambia
Tara L. Steinmetz Marc F. Bellemare†
Academic Coordinator, University of California Davis, Davis, CA, 95616, Telephone: (530) 752-7252, Email:
Corresponding Author and Assistant Professor, Duke University, Durham, NC, 27708-0312, Telephone: (919) 613Email: email@example.com.
To study the persistence of female genital cutting (FGC) through the association between whether a woman has undergone FGC and she is in favor of the practice.
Methods We complete a cross-sectional study using data from The Gambia with 9,982 women aged 15-49.
Participants belonged to the Mandinka, Wolof, Jola, Pulaar, Serere, and other ethnic groups. Selection of respondent households within clusters was random. Probability weights are used to make the sample nationally representative. Our main outcome measures were the marginal impact of a woman’s own FGC status (i.e., whether she has undergone FGC) on whether (i) she would like her daughter to undergo FGC and (ii) she thinks the practice should continue.
Findings Much of the persistence of FGC can be attributed to individual- and household-level factors, which together explain 85% (80 to 89%) of the relationship between whether a woman has undergone FGC and whether she would like her daughter to undergo FGC, and 86% (81 to 91%) of the relationship between whether a woman has undergone FGC and whether she thinks the practice should continue.
Both estimated impacts are significant at less than the 1% level. Community-level factors account for 15% or less of the persistence of FGC.
Interpretation Our findings fly in the face of popular policy interventions aimed at ending FGC in West Africa, which often involve community-wide pledges to collectively abandon the practice. Policy interventions aimed at ending FGC should directly target individuals and households rather than communities in The Gambia.
Introduction More than 100 million women worldwide have had part of their external genitalia removed in a practice called female genital cutting (FGC),(1) and at least three million girls undergo the procedure every year.
(1) Female Genital Cutting can take place at any time before the age of 15,(1) with most FGC occurring between the ages of four and eight.(2) The World Health Organization distinguishes between four types of FGC, ranging from clitoridectomy, in which the clitoris is removed, to infibulation, in which the vaginal opening is narrowed by sewing the labia together.(1) While FGC is widespread throughout Africa, Asia, and the Middle East,(3-8),(4) it is also a public health concern in industrialized countries, where immigrants sometimes import the practice.(3, 9-12) Various negative health impacts are correlated with FGC.(2) Women who have undergone severe forms of FGC face higher likelihoods of reproductive health problems.(2) Some posit that FGC increases the risk of HIV transmission.(13, 14) In addition, while there are no systematic studies looking at the psychological impacts of FGC, some speculate that FGC has psychological costs,(2, 4, 8) and others argue that FGC is a violation of human rights.(15) It is worth asking why FGC, though it has declined in some countries, persists in others. We answer that question using data on a cross-section of Gambian women. The Gambia is an ideal context to study the persistence of FGC, because even though FGC has been declining in recent years in Senegal, which encloses almost all of The Gambia,(5) FGC shows no sign of decline in The Gambia. Moreover, Gambian president Yahya Jammeh has said that FGC is “part of [Gambian] culture and we should not allow anyone to dictate to us how we should conduct ourselves.”(16) To that end, we study the persistence of FGC– defined here as the association between whether a respondent has undergone FGC and she supports the practice– in The Gambia.
Methods Data and Descriptive Statistics We use the 2005-2006 Gambian Multiple Indicator Cluster Survey (MICS) data, which were collected by UNICEF and the Gambian Bureau of Statistics. Selection of households within clusters (i.e., villages) was random. Of the 6,175 households selected, 6,071 were interviewed. In the interviewed households, 10,252 women aged 15 to 49 were identified, and 9,982 were interviewed. Because we use secondary data, study size is driven by data availability. Probability weights, which reflect the likelihood that any respondent would be included in a random sample of the Gambian population of women aged 15 to 49, are provided in the MICS to make the sample nationally representative.(17) We use those weights in the empirical results below.
Our data allow incorporating increasingly refined levels of fixed effects (i.e., district, village, and household in addition to interviewer fixed effects). We can thus control for factors common to the individuals within the same district, within the same village, and within the same household, and we can control for interviewer-specific biases. This allows eliminating important sources of bias in our estimates of the relationship between whether a respondent has undergone FGC and whether she is in favor of the practice. We thus build on previous studies that used multi-level models to partially control for the heterogeneity between communities.(18) Specifically, the data include 362 households for which there is intrahousehold variation in whether respondents underwent FGC or whether they would like their daughters to undergo FGC, and 357 households for which there is intrahousehold variation in whether respondents underwent FGC or whether they think the practice to continue.
Table 1 presents descriptive statistics for our dependent variables (i.e., indicators for whether the respondent thinks the practice of FGC should continue and for whether she would like her daughter to undergo FGC), for our variable of interest (i.e., an indicator for whether the respondent has undergone FGC), and for our control variables.
The estimation sample varied for each dependent variable because of respondent willingness to answer questions regarding our dependent variables. In what follows, control variables should control for much of the variation in response rates, and the remaining variation in response rates is assumed to be random.
The descriptive statistics in Table 1 indicate there is widespread support for FGC in The Gambia: 76% of the women in our data think the practice should continue, and 73% would like their daughter to undergo FGC. Moreover, about 80% of our respondents have undergone FGC.
We do not discuss each variable retained for analysis for the sake of brevity but we provide a short discussion of those variables whose measurement requires some clarifications. Notably, a Muslim heads the majority of the households in the data; many in West Africa believe that FGC is an integral part of Islam. In fact, the Prophet Muhammad made a passing reference to FGC in a hadith whose authenticity is debated.(19) To gauge whether respondents had some public health knowledge, they were asked whether one could get HIV/AIDS through supernatural means.
A wealth score was computed by for each household by UNICEF on the basis of its ownership of specific assets.(17) The data also included five questions about domestic violence. Each question asked the respondent whether she thought a man was justified in beating his wife under specific circumstances. We split those five questions into two categories: a woman’s behavior, and whether a woman is her husband’s property. Our measure of tolerance to domestic violence related to a woman’s behavior includes questions about whether domestic violence is justified if a woman neglects her children, argues with her husband, or burns the food. Our measure of tolerance to domestic violence related to whether a woman is her husband’s property includes questions about whether domestic violence is justified if a woman goes out without telling her husband or refuses to have sex with him. Each score is equal to one if the respondent agrees with at least one of the statements and equal to zero if she disagrees with all of them.
Estimation and Identification The equation we estimate in this study is such that, (1)
outcomes of interest (i.e., a variable equal to 1 if a woman would like her daughter to undergo FGC and equal to 0 otherwise, or a variable equal to 1 if she thinks the practice should continue and equal to 0 otherwise); is our variable of interest (i.e., a variable equal to 1 if a respondent has undergone FGC and equal to 0 otherwise); is a vector of control variables (including interviewer dummy variables to control for interviewer fixed effects), is a vector of district, village, or household fixed
equation 1 by ordinary least squares (OLS) in Stata 11 (StataCorp. 2009. Stata Statistical Software:
Release 11. College Station, TX: StataCorp LP) using the probability weights provided in the data and clustering the standard errors at the village level.
Because our dependent variables are binary, our use of OLS means we estimate linear probability models (LPM). There are two advantages to estimating LPMs instead of popular alternatives like probit and logit.(20) First, the LPM is well-suited to handle fixed effects, whereas probit and logit are not because of the incidental parameters problem.(21) Second, LPM coefficients are interpretable as marginal effects, whereas probit and logit coefficients have to be transformed before they can be interpreted as such.
Though there are some disadvantages to estimating LPMs,(22) those disadvantages are irrelevant in this
, so the LPM is heteroskedastic. Our use of probability weights, however, implies standard errors robust to heteroskedasticity. In addition, though the LPM can lead to predicted values of outside of the interval, our goal is to estimate specific coefficients rather than to make outof-sample predictions.
Results The top panel of Table 2 presents a cross-tabulation of whether a respondent has undergone FGC and of whether she would like her daughter to undergo FGC, and the bottom panel of Table 2 presents a crosstabulation of whether a respondent has undergone FGC and of whether she thinks the practice should continue. Both cross-tabulations indicate a high correlation between having undergone FGC and support for the practice. Specifically, the correlation coefficient between whether a respondent has undergone FGC and whether she would like her own daughter to undergo FGC is 0.83, and the correlation coefficient between whether a respondent has undergone FGC and whether she would like the practice to continue is 0.80, with both coefficients significant at the 1% level.
Table 3 presents the determinant of whether respondents would like their daughters to be circumcised.
The most striking result relates to the relationship between a respondent’s FGC status and whether she would like her daughter to undergo FGC. The inclusion of individual- and household-level controls in Column 1 weakens the correlation between the two variables from a pairwise correlation of 0.83 to 0.75, indicating that those control variables have little explanatory power. Likewise, the inclusion of district and village fixed effects weakens the correlation from 0.75 to 0.73, and from 0.73 to 0.70, indicating that district and village-level heterogeneity also have very little explanatory power. The inclusion of household fixed effects, however, reduces the estimated coefficient for a woman’s own FGC status by almost 45%, from 0.70 to 0.40. Put another way, 37%– i.e., (0.703 - 0.397)/0.83– of the correlation between a respondent’s own FGC status and whether she would like her daughter to undergo FGC can be attributed to heterogeneity between households rather than to heterogeneity at the village or district levels. Similarly, in Table 4, 41%– i.e., (0.687 – 0.360)/0.80– of the correlation between a respondent’s own FGC status and whether she would like the practice to continue can be attributed to heterogeneity between households rather than to heterogeneity at the village or district levels.
Figure 1 summarizes the contribution of each level of variation to the persistence of FGC in Tables 3 and
4. To determine the contribution of individual-level factors in the first column of Figure 1, the amount of variation in the relationship between a respondent’s own FGC status and whether that respondent is in favor of FGC for her daughter that is due to individual-level factors (0.397, or the estimated coefficient for whether a respondent has undergone FGC in Column 4 of Table 3) is divided by the correlation between a respondent’s own FGC status and whether that respondent is in favor of FGC for her daughter (0.83), for a total of 48%. Then, to determine the contribution of household-level factors in the first column of Figure 1, the amount of variation in the relationship between a respondent’s own FGC status and whether that respondent is in favor of FGC for her daughter that is due to household-level factors (0.703 – 0.397, or the difference in estimated coefficient for whether a respondent has undergone FGC between Columns 3 and 4 of Table 3) is divided by the correlation between a respondent’s own FGC status and whether that respondent is in favor of FGC for her daughter (0.83), for a total of 37%. The remainder of Figure 1 is obtained by similar calculations.
Discussion Strengths and Limitations The main strength of this study is the within-household variation in respondents’ own FGC status and in their support for the practice, which allow holding constant district, village, and household factors to determine how much each type of factor contributes to the persistence of FGC. Another strength of this study is that the data allow controlling for interviewer-specific effects by incorporating interviewer fixed effects. Additionally, our findings are nationally representative of Gambian women aged 15 to 49. The main weakness of this study is that for all of the sources of heterogeneity our fixed effects can filter out, we cannot control for the unobserved heterogeneity between individuals due to differences between the individuals themselves rather than to differences between their households, villages, and districts.