The Effects of Social Interactions on Female Genital Mutilation Evidence from Egypt Results

4. Results 4.1. Direct Effects The first regression results presented are those for direct effects of individual characteristics on the likelihood of FGM only (shown in Table 5); we want to examine how particular observable characteristics of households influence the FGM decision for their daughters. The first column shows the results using an OLS regression while the second is shown for a household fixed effects (FE) regression. The first thing to observe is the apparent increased likelihood of FGM, in the FE results, for daughters of the fifth or sixth birth order. We also notice, in both estimations, a significant negative likelihood of FGM for younger cohorts, suggesting a decreasing FGM trend across the country over time. In the OLS regression, we see some expected results: a significant negative likelihood of FGM for wealthier households, households that reside in urban areas, households where mothers got married at an older age, households that discussed FGM with their neighbors, and households where mothers have higher levels of education; and a significant positive likelihood for households with more daughters. A significant negative likelihood is also found for Christian households (compared to Muslim households), which is somewhat surprising considering that FGM is a practice that predates Islam in Egypt. 4.2. Social Effects In this section, we present the results for the main linear-in-means regressions, used to identify the social effects of FGM and its medicalization, on the household FGM decision— how households are influenced by other households’ FGM and medicalization decisions. In Table 6, we first consider regression results when we do not use household fixed effects, and in Table 7 we do. In Table 6, we do not see any big change in direct effects compared to column (1) in Table 5, whether as estimates or standard errors. In the first three columns, we ran OLS regressions without any instrumental variables. In these columns, we see that the peer group mean of FGM decision has a positive and significant value. In columns (2) and (3), we add the group mean for medicalization which is positive but insignificant. The interaction between FGM and medicalization group means is negative and similarly insignificant. In column (4), we try the same regression as in (1) but with instrumental variables for the FGM peer group mean, and we find that it has a strong positive effect. In column (5), we test for nonlinearity by adding a quadratic term, but then all effects become insignificant. In columns (6) and (8), we add the peer group mean medicalization, and in the latter we use an instrumental variable. Medicalization is found to have a negative effect, but is only significant in the former specification (not IV). Peer group means of FGM continue to be positive (with a higher magnitude) and significant. In columns (7) and (9), we add quadratic terms for peer group means for both FGM and its medicalization. Again in these specification all effects become insignificant. Finally, in column (10), we interact the peer group mean regressors. FGM peer group means are found to be positive and significant, while medicalization peer group means continue to be insignificant. From this table we suspect that the hypothesis that the peer group mean of FGM has a strong positive effect on a household’s FGM decision Table 5: Direct Effects Only Estimate (Standard Error) Birth Order 2 3 4 5 6 Cohort 1989 1990 1991 1992 1993 1994 1995 1996 Wealth (Rich) Residence (Urban) Marital Age Mother FGM Religion (Christian) HH Head Sex (Female) Medical Help Dist. (Big Problem) Medical Help Cost (Big Problem) Discussed FGM Received Info on FGM Number of Daughters Education Level Primary Secondary Higher (1) OLS (2) FE 0.012 (0.012) -0.008 (0.027) -0.125* (0.060) 0.109 (0.102) 0.069 (0.052) -0.001 (0.018) -0.039 (0.036) -0.067 (0.070) 0.151* (0.060) 0.229*** (0.060) -0.014 (0.021) -0.005 (0.020) -0.043* (0.020) -0.063** (0.021) -0.071** (0.022) -0.149*** (0.024) -0.208*** (0.025) -0.372*** (0.024) -0.094*** (0.023) -0.083*** (0.016) -0.005** (0.002) 0.334*** (0.029) -0.274*** (0.032) -0.021 (0.018) -0.004 (0.014) -0.007 (0.013) -0.031* (0.013) 0.005 (0.014) 0.032*** (0.008) -0.009 (0.025) 0.009 (0.024) -0.040 (0.029) -0.038 (0.032) -0.072* (0.036) -0.120** (0.040) -0.206*** (0.046) -0.319*** (0.051) 0.009 (0.015) -0.079*** (0.016) -0.192*** (0.035) N 6206 6253 R2 0.357 0.352 0.191 0.056 Adjusted R2 1 2 Asterisks indicate significance levels: ‘·’ is 10 percent, ‘*’ is 5 percent, ‘**’ is 1 percent, and ‘***’ is 0.1 percent. In the OLS regression, fixed effects for governorates were also used. is probably true, but we are unable to make any clear conclusions about medicalization. However, as stated above, because of the problem with correlated effects and the possible endogeneity due to unobservable household effects that are correlated with our regressors, we need to leverage the within information we have about households. In Table 7, we introduce household fixed effects to address some of these problems, coupled with the existing instrumental variable strategy. In columns (1) and (2), we regress only on the peer group means of FGM and we continue to find a significant positive effect, with some convexity on introducing a quadratic term. In columns (3) and (4), we introduce medicalization, and in the latter specification we use instrumental variables. We now observe a significant negative effect (almost halved in the latter specification). The effect of peer group FGM means continues to be significantly positive and increases in magnitude. In columns (5) and (6), we introduced some nonlinearity. In column (5), there is still a positive and negative effect to peer group means of FGM and medicalization, respectively—with some concavity and convexity, respectively. In column (6), the significance and direction of effect is unchanged, but the magnitudes are diminished (in this specification we see the highest level of R2 in this table). There does appear to be a reduced negative effect of medicalization as FGM peer group means increase, but is only significant at the 10% level. Table 6: FGM Endogenous Effects Regression (Pooled) Estimate (Standard Error) Wealth (Rich) Residence (Urban) Marital Age Mother FGM Religion (Christian) HH Head Sex (Female) Medical Help Distance (Big Problem) Medical Help Cost (Big Problem) Discussed FGM Received Info on FGM Number of Daughters Education Level Primary Secondary Higher f gm jt (1) OLS (2) OLS (3) OLS (4) 2SLS (5) 2SLS (6) 2SLS (7) 2SLS (8) 2SLS (9) 2SLS (10) 2SLS -0.083*** (0.023) -0.081*** (0.017) -0.005** (0.002) 0.301*** (0.031) -0.266*** (0.031) -0.018 (0.018) -0.003 (0.014) -0.005 (0.012) -0.029* (0.013) 0.008 (0.014) 0.034*** (0.009) -0.083*** (0.023) -0.081*** (0.017) -0.005** (0.002) 0.302*** (0.031) -0.266*** (0.031) -0.018 (0.018) -0.003 (0.014) -0.006 (0.012) -0.029* (0.013) 0.008 (0.014) 0.034*** (0.009) -0.082*** (0.023) -0.081*** (0.017) -0.005** (0.002) 0.301*** (0.031) -0.266*** (0.031) -0.018 (0.018) -0.003 (0.014) -0.006 (0.012) -0.029* (0.013) 0.008 (0.014) 0.034*** (0.009) -0.073** (0.024) -0.081*** (0.017) -0.006*** (0.002) 0.276*** (0.036) -0.259*** (0.031) -0.018 (0.018) -0.002 (0.014) -0.004 (0.013) -0.027* (0.013) 0.012 (0.014) 0.034*** (0.009) -0.072** (0.024) -0.079*** (0.017) -0.006*** (0.002) 0.276*** (0.036) -0.257*** (0.031) -0.019 (0.018) -0.001 (0.014) -0.003 (0.013) -0.026· (0.013) 0.012 (0.014) 0.032*** (0.009) -0.076** (0.024) -0.086*** (0.018) -0.006*** (0.002) 0.267*** (0.039) -0.261*** (0.031) -0.014 (0.019) -0.002 (0.015) 0.001 (0.013) -0.028* (0.013) 0.014 (0.015) 0.032*** (0.009) -0.077** (0.024) -0.085*** (0.018) -0.006*** (0.002) 0.274*** (0.037) -0.255*** (0.031) -0.018 (0.019) 0.000 (0.015) 0.002 (0.013) -0.027* (0.014) 0.012 (0.015) 0.026** (0.010) -0.076** (0.024) -0.087*** (0.019) -0.006*** (0.002) 0.265*** (0.039) -0.261*** (0.031) -0.014 (0.019) -0.002 (0.015) 0.002 (0.014) -0.028* (0.014) 0.015 (0.015) 0.032*** (0.009) -0.073** (0.024) -0.082*** (0.018) -0.006*** (0.002) 0.275*** (0.037) -0.256*** (0.031) -0.019 (0.019) -0.001 (0.014) -0.002 (0.013) -0.027* (0.013) 0.012 (0.015) 0.030** (0.010) -0.073** (0.024) -0.084*** (0.017) -0.006*** (0.002) 0.273*** (0.038) -0.259*** (0.031) -0.017 (0.018) -0.002 (0.014) -0.005 (0.013) -0.028* (0.013) 0.013 (0.015) 0.035*** (0.009) 0.009 (0.015) -0.076*** (0.016) -0.204*** (0.034) 0.526*** (0.040) 0.009 (0.015) -0.076*** (0.016) -0.204*** (0.034) 0.505*** (0.046) 0.009 (0.015) -0.076*** (0.016) -0.204*** (0.034) 0.521*** (0.049) 0.009 (0.015) -0.073*** (0.017) -0.215*** (0.034) 1.065*** (0.238) 0.010 (0.015) -0.074*** (0.017) -0.215*** (0.034) 0.625 (0.437) 0.414 (0.358) 0.011 (0.016) -0.070*** (0.017) -0.216*** (0.034) 1.538*** (0.406) 0.012 (0.016) -0.069*** (0.018) -0.216*** (0.035) 1.641** (0.532) 0.119 (0.101) 0.012 (0.016) -0.072*** (0.017) -0.213*** (0.035) 0.399 (0.730) 0.642 (0.581) 0.182 (1.036) -0.289 (0.827) 0.009 (0.015) -0.070*** (0.017) -0.215*** (0.034) 1.278*** (0.350) 0.028 (0.035) 0.016 (0.016) -0.070*** (0.017) -0.209*** (0.034) -0.001 (0.663) 1.143* (0.531) 0.040 (0.420) -0.469 (0.328) ( f gm jt )2 med jt -0.547* (0.230) (med jt )2 -0.672 (0.575) -0.101 (0.101) f gm jt × med jt 0.534 (0.897) -0.717 (0.944) N 6206 6206 6206 6206 6206 6206 6206 6206 6206 6206 R2 Adjusted R2 0.387 0.380 0.387 0.380 0.388 0.380 0.361 0.354 0.354 0.347 0.332 0.324 0.335 0.327 0.319 0.311 0.355 0.347 0.354 0.346 1 2 Asterisks indicate significance levels: ‘·’ is 10 percent, ‘*’ is 5 percent, ‘**’ is 1 percent, and ‘***’ is 0.1 percent. Other regressors not shown: governorate fixed effects, cohort (year of birth) fixed effects, order of birth fixed effects, ‘modernization’ proxy variables, and peer group means of exogenous household characteristics (exogenous social effects). Table 7: FGM Endogenous Effects Regression (Household Fixed Effects) Estimate (Standard Error) f gm jt (1) (2) (3) (4) (5) (6) (7) (8) 1.180*** (0.081) -0.964*** (0.184) 1.410*** (0.145) 1.328*** (0.095) 1.299*** (0.094) 0.917*** (0.088) 1.849*** (0.120) 1.939*** (0.126) -0.592*** (0.073) -0.265*** (0.066) 1.281*** (0.238) -0.433* (0.173) -2.151*** (0.246) 1.645*** (0.198) -0.493** (0.166) -0.472*** (0.096) -1.298*** (0.139) -0.125 (0.155) -0.224· (0.131) 1.856*** (0.489) 0.207 (0.181) ( f gm jt )2 med jt (med jt )2 0.265· (0.158) f gm jt × med jt med jt × Urban med jt × Medical Help Distance med jt × Urban × Medical Help Distance med jt × Medical Help Cost 0.644*** (0.140) 1.189*** (0.328) med jt × Urban × Medical Help Cost N 6229 6229 6229 6229 6229 6229 6229 6229 R2 0.185 0.054 0.171 0.050 0.195 0.057 0.191 0.056 0.161 0.047 0.213 0.062 0.145 0.042 0.132 0.038 Adjusted R2 1 2 Asterisks indicate significance levels: ‘·’ is 10 percent, ‘*’ is 5 percent, ‘**’ is 1 percent, and ‘***’ is 0.1 percent. Other regressors not shown: governorate fixed effects, cohort (year of birth) fixed effects, order of birth fixed effects, ‘modernization’ proxy variables, and peer group means of exogenous household characteristics (exogenous social effects). Table 8: Urban/Rural and Medical Help Interactions Estimate (p-value) Rural Medical Help Distance (No Problem) Medical Help Distance (Problem) Difference Medical Help Cost (No Problem) Medical Help Cost (Problem) Difference 1 Urban -0.472*** -0.597*** (0.000) (0.000) -0.696*** 1.034* (0.000) (0.024) 0.224· (0.086) -1.631*** (0.001) -1.298*** -1.091*** (0.000) (0.000) -0.655*** 0.741** (0.000) (0.006) -0.644*** (0.000) -1.832*** (0.000) Asterisks indicate significance levels: ‘·’ is 10 percent, ‘*’ is 5 percent, ‘**’ is 1 percent, and ‘***’ is 0.1 percent. Having established that there is a negative effect to medicalization, we introduce some interaction terms in the last two columns of Table 7 to further investigate how the effect might vary in different contexts, namely, urban versus rural, and with different levels of difficulty accessing medical help (either because of distance or cost). With three levels of interactions, it is easier to analyze our results by using Table 8. We see that in rural areas medicalization has a negative effect, does not seem to be affected by difficulty in accessing medical help due to distance, and does seem to have a weaker negative effect if the household has difficulty accessing medical help due to cost. In urban areas, the effect of medicalization is more nuanced; it appears to be negative for households with no problem accessing medical help (for either of the reasons considered), but has significant positive effect if they do have problems accessing medical help. 5. Conclusion As has been shown by our results, there is a strong social component, in Egypt, to households’ decision to circumcise their daughters. Households tended to follow the same behavior as other members of their community. However, in terms of medicalization, there appears to be a negative social effect (for most households with no problems accessing medical help). Households were less likely to choose FGM for their daughters the more prevalent is medicalization in their peer group. This appears to strengthens the harm reduction argument of those who call for tolerating medicalization in the interest of providing a less painful and sanitary procedure for those who opt for FGM (Shell-Duncan, 2001). These results are also a call on policy makers who are seeking to eliminate FGM, to consider the possible unintended consequences of focusing on eliminating medicalization in government clinics. Our results point to directly influencing households’ FGM decision as a possibly more fruitful policy, since we see a strong multiplier effect, since households have a very strong endogenous influence on each other. This could be done by increasing awareness in communities of the harmful nature of this practice as was carried out in Senegal (Diop and Askew, 2009), while not pushing them away from medical clinics back into the domain of traditional circumcisers, where they might be less accessible to outreach campaigns. One proposed story explaining these findings, is related to the pernicious nature of the marriage market social network in which dayas might play the role of ‘quality certifiers’. Due to the unobservable and unverifiable nature of FGM, reputation plays a critical role here. As the FGM procedure moves away from the traditional sphere into that of professional healthcare its links to the marriage market is weakened, as healthcare providers are less likely to play a role in the marriage market. We might also be observing a modernization scenario, in which the endogenous social effect of medicalization is a proxy for a move away from traditional practices across all households in a particular area, irrespective of the other observed household characteristics. As mentioned, we attempt to address this problem by the use of data on child delivery choices contemporary to the FGM decision. However, we must be cautious in making any recommendations calling for a rollback in the elimination for FGM in medical facilities. There is always the danger that such a reversal of policy might give the impression of greater legitimacy, and households that were previously indecisive about FGM would regard such a move as approving of the procedure. In fact, some of our results should make us very cautious in dealing with the problem of legitimization. We see a positive effect to medicalization in urban areas with poor access to medical help (either due to distance or cost), which could plausibly be the influence of observing increased medicalization in one’s peer group and perceiving it as evidence of its benefit, and without easy access to medical practitioners (and their mitigating influence) FGM is increasingly sought from traditional circumcisers10. 10 Similar fears were raised in response to Indonesia’s Ministry of Health publishing guidelines for conduction FGM (Ind, 2011)