PREDICTION OF THE DETERMINANTS OF THE NUMBER OF ANTENATAL CARE VISITS IN NFHS IV SURVEY OF INDIA: MODELING EXCESS ZERO OF COUNT DATA

Abstract

Poisson and negative binomial regression models are most widely used to model count data. Count data models are used to predict the factors that affect the number of antenatal care visits (ANC) received by pregnant women throughout their ninth and tenth months of pregnancy. It is imperative to take into account the presence of excess zero ANC visits which could be either structural or sampling zeros. Simulated datasets were produced using negative binomial distributions with a range of dispersion values (20, 40, and 120) and different zero percentages (22, 40, 60, and 70 percent). The purpose of this study was to quantify the bias and poor fit introduced by the fitting Least Square Regression (LSR), Poisson regression (PR), negative binomial regression (NBR), zero-inflated Poisson regression (ZIPR), zero-inflated negative binomial regression (ZINBR), hurdle Poisson regression (HPR) and hurdle negative binomial regression (HNBR) models. To assess zero-inflated practices and look into the relationship between the number of antenatal visits and the socio-demographic profile of women in their ninth and tenth months of pregnancy, the maternal health data from the National Family Health Survey (NFHS IV) study were subsequently examined. The relative quality of the regression models was assessed using Vuong tests and Akaike information criterion (AIC) values. The zero-inflated negative binomial regression (ZINBR) model performed better than the other models, according to the simulation study’s findings and with the application of ANC visits, which indicated that they had decreased AIC values under all zero inflation and over-dispersion scenarios.

Received: February 24, 2023 
Revised: March 17, 2023 
Accepted: April 8, 2023