THE USE OF BAYESIAN QUALITATIVE AND LIMITED DEPENDENT VARIABLE MODELS IN EVALUATING WOMEN PARTICIPATION IN AGRICULTURAL PRODUCTION ACTIVITIES IN OIL RICH AREA OF NIGERIA

Abstract

This study examined the use of Bayesian qualitative and limited dependent variable models in women's participation in agricultural production activities in Ogoni, Rivers State, Nigeria. In specifics, the study aimed to; ascertain the socio-economic characteristics of the women farmers, and use a Bayesian approach to estimate the parameters of our proposed model and compare the Bayesian logistic regression result with the classical logistic regression. Data were analysed using the Bayesian logistic regression model as well as comparing it with the classical logistic regression. The findings of the study revealed a positive and significant relationship between selected socio-economic characteristics of the respondents and women's level of participation in agricultural production activities, except age and cooperative membership which were negatively signed. Holding other variables at fixed value, we would see an average of 110.5% increase in the odds of having high women participation in agriculture production activities for a unit increase in level of education (X2). There is an average of 100% increase in the odds of getting high women participation in agricultural production activities for a unit increase in household size (X3). Bayesian estimation performed better than the Classical Logistic Regression (maximum likelihood estimation) in terms of parameter estimates as the Bayesian logistic model had lower standard deviation and standard error of mean (SEM) compared to the Classical logistic (vis-à-vis narrower credible interval versus the confidence interval of the frequentist). However, both the Bayesian and Classical logistic models performed very well in terms of model prediction accuracy with prediction accuracy values of 0.80 and 0.8025 respectively.The leave-one-out cross validation (LOOCV) prediction accuracy was 0.78 or 78%. This is still high enough and implies that the accuracy of the model is 78% likely to hold up when used in a different data set. The study concludes that, predictions in women participation in agricultural production activities is very efficient when using Bayesian Logistic Regression using the independent Cauchy prior as proposed by Gelman et al (2008). It is therefore recommended that Bayesian inference and models can be adopted in estimating predictions in women participation in agricultural production activities particularly using Bayesian Logistic Models with Independent Cauchy Priors.