What drive financial inclusion gender gap in Cameroon? A Fairlie
decomposition approach
METHODOLOGICAL APPROACH AND DATA
3.1 Methodology:
To analyze the drivers of gender gaps in financial inclusion in Cameroon, we use a nonlinear Fairlie
decomposition. The Fairlie decomposition is an extension of the Blinder–Oaxaca decomposition that
overcomes the problems encountered in the Blinder (1973) and Oaxaca (1973) approaches, particularly
the limitations posed when the dependent variable is a dichotomous, polytomous, censored, or
truncated variable (Jann, 2008). The main advantage of applying this methodological approach is that
the coefficient estimates from a logit or probit model can be used directly in the decomposition
specification. The technique is thus useful for applications in which it is inappropriate to model the
dependent variable as a linear function of the explanatory variables (Fairlie, 2006; Fairlie & Meyer, 1999).
We can assess the contribution of each variable to the gap by observing the change in the average
probability predicted by replacing one distribution (male) with the other (female), while holding the
distribution of other variables constant.
The contribution of each variable to the gap is equal to the change in the average predicted probability
of replacing the female distribution with the male distribution of that variable while holding the
distributions of the other variable constant. The main property of this technique is that the sum of the
contributions of individual variables will be equal to the total contribution of all the variables evaluated
in the sample.
It should be noted that a potential endogeneity bias between income and financial inclusion may arise in
this work. In fact, income, for example from farmers can be an outcome of financial inclusion that
permits individuals to borrow for agricultural activities. To deal with that endogeneity issue, we follow
Naghi et al. (2022) and Freedman and Sekhon (2010) who proposed a two-step procedure to overcome
the endogeneity issue in probit model regression. In the first step, the study employs the ordinary least
squares estimator, whereas, in the second step, the Probit Maximum Likelihood Estimator was fitted. In
this step, the residuals are used as an additional regressor. The Probit estimates obtained are thus free
from endogeneity biases
Consanguineous Marriages and the Perception of Wife-Beating
Justification in Pakistan: An Application of Fairlie Decomposition
Empirical Methodology:
To begin, descriptive statistics pertaining to various socioeconomic factors in consanguineous and nonconsanguineous marriages, as well as the prevalence of women’s perspectives concerning the justifying
of IPV have been presented. The test of association is utilized to ascertain the variables that exhibit a
statistically significant correlation with a woman’s rationale for her husband’s physical abuse of her. A
logistic regression model was subsequently utilized to compute the odds ratios pertaining to each
variable.
For our main analytical part, we utilized the nonlinear decomposition method, as suggested by (Fairlie,
1999, 2005), to ascertain the extent to which various socioeconomic variables contribute to the disparity
in the justification for IPV between the aforementioned types marriages.
The endowment effect explains the disparity in the justification of IPV, which is attributed to the varying
distribution of explanatory variables between consanguineous and non-consanguineous marriages. This
implies that the justification gap can be eliminated if women in consanguineous marriages possess the
same characteristics as those in non-consanguineous marriages. The second component pertains to the
acceptance gap of IPV, which is influenced by unobservable factors that are not accounted for in the
model.
The process of nonlinear decomposition entails the estimation of the probability associated with
justifying wife-beating for each individual woman within both the consanguineous and nonconsanguineous groups. This is accomplished by taking into account the model’s explanatory variables
and by utilizing a sample that comes from one of the groups or a combined sample from the two groups.
The predicted probabilities are then used to compare women from the two groups and evaluate the
differences between them. This process is repeated multiple times (replications), and the results are
averaged to obtain the final predicted gap. Given that there are different numbers of women in both
groups (4,274 vs. 7,320), this study used the parameter estimates from the pooled sample to find the
probabilities, as recommended by Fairlie (2005). To address the issue of path dependence (the sensitivity
of the decomposition results to the order of introducing explanatory variables), we randomly ordered
the variables across replications, following Fairlie’s recommendation (1999). We conducted 1,000
replications of this procedure. During the process of estimating the model, we utilized sample weights in
order to take into consideration the intricate sampling strategy that was utilized in PDHS.