Semiparametric Bivariate Probit Model Early Breastfeeding Initiation and Exclusive Breastfeeding

Abstract

Regression analysis in which the response variable is categorical can be processed using the probit model. The probit model is based on the normal distribution, in addition to its interpretation based on marginal effect values. A probit model consisting of two response variables is called the bivariate probit model, in which the response variables each consist of two categories. The predictor variables in bivariate probit model can be either categorical and also continuous data. Bivariate probit model both response variables have a relationship. One of the developments of the bivariate probit model is the semiparametric bivariate probit model, where the bivariate probit model in which there is a parametric and a nonparametric model in this case in the form of a continuous covariate smooth function. Semiparametric bivariate probit model have the advantage of being able to address the problem of nonlinearity of undetected continuous predictor variables that can cause modeling inaccuracies that can effect the results of estimation accuracy. Parameter estimation of semiparametric bivariate probit model uses the Penalized Maximum Likelihood Estimation approach, but the equation obtained is not closed form so iteration are needed to solve it. The iteration used is Fisher Scoring. The semiparametric bivariate probit model was applied to data on early breastfeeding initiation and exclusive breastfeeding in East Java Province in 2021 with variables that affect early breastfeeding initiation being birth attendants while those affecting exclusive breastfeeding are maternal age and maternal education level.