DEALING WITH EXCESS ZEROS IN A DISCRETE DEPENDENT VARIABLE IN THE REPORTED NUMBER OF ANTENATAL CARE VISITS

Abstract

Data with excess zeros have often been associated with count regression models. Poisson regression and Negative Binomial regression models have been used as a standard for modelling count outcomes but these methods do not take into account the problems associated with excess zeros and over-dispersion of count data. Failure to account for these extra zeros may result to biased parameter estimates and wrong inferences while over-dispersion causes the standard error of the estimates to be underestimated. Therefore, this study was designed to evaluate the performance of the Zero Inflated Poisson and Zero Inflated Negative Binomial models in determining the factors associated with the number of antenatal care visits in Nigeria. Data for this study was obtained from the National Demographic and Health Survey (NDHS 2013). The survey made use of a cross-sectional population based study design. A sample of 31,482 women within the reproductive age of 15-49yrs who gave birth five years prior to the survey and provided information about antenatal care visits were utilised. Number of antenatal care visits was used as the dependent variable while the explanatory variables include age, region, residence, parity, educational level, religion, wealth index, employment status, husband/partners employment status and husband/partners occupation. Data were analysed using descriptive statistics, chi square test, Zero Inflated Poisson (ZIP) regression analysis and Zero Inflated Negative Binomial (ZINB) regression models, Kolmogorov-Smirnov test was used to check for over-dispersion and three test criteria (AIC, BIC and -2LogL)were used to assess model fit using SPSS version 19 and STATA version 12. Mean age of women was 29.5 ± 7.0yrs and median number of ANC visits was 4 visits. (Range = 30). Findings revealed an urban-rural differential in antenatal care utilisation; with respondents living in urban areas (78.2%) having higher proportion compared to those in rural areas (41.2%). About 53.5% had at least 4 antenatal visits while 46.5% had no ANC visit. The Zero Inflated Negative Binomial regression analysis revealed that age, region,education, wealth index, husband/partners education, respondent's employment, religion, residence and parity were significant determinants of number of antenatal care visit (p<0.001). The ZINB model fits the data better than ZIP model with AIC values of 85,707.75 and 94,635.19 respectively. Antenatal care utilisation was relatively low among women living in rural areas compared to women living in urban areas. The Zero-Inflated Negative Binomial regression model provided the better fit for the data on number of antenatal care visits. This study suggested Zero-Inflated Negative Binomial model for count data with excess zeros and over-dispersion.