Application of two level count regression modeling on the determinants of fertility among married women in Ethiopia

BACKGROUND: Fertility is the element of population dynamics that has a vital contribution toward changing population size and structure over time. The global population showed a major increment from time to time due to fertility. This increment was higher in south Asia and sub-Saharan Africa including Ethiopia. So this study targeted the factors affecting fertility among married women in Ethiopia through the framework of multilevel count regression analysis using the 2016 Ethiopian Demographic and Health Survey data. METHODS: Secondary data set on the birth records were obtained from the 2016 Ethiopia Demographic and Health Survey. The survey was a population-based cross-sectional study with a two-stage stratified cluster sampling design, where stratification was achieved by separating every region into urban and rural areas except the Addis Ababa region because it is entirely urban. A two-level negative binomial regression model was fitted to spot out the determinants of fertility among married women in Ethiopia. RESULTS: Among the random sample of 6141 women in the country, 27,150 births were recorded based on the 2016 Ethiopian Demographic and Health Survey report. The histograms showed that the data has a positively skewed distribution not extremely picked at the beginning. Findings from the study revealed that the contraception method used, residence, educational level of women, women’s age at first birth, and proceeding birth interval were the major predictors of fertility among married women in Ethiopia. Moreover, the estimates from the random effect result revealed that there is more fertility variation between the enumeration areas than within the enumeration areas. CONCLUSION: Unobserved enumeration area fertility differences that cannot be addressed by a single-level approach were determined using a two-level negative binomial regression modeling approach. So, the application of standard models by ignoring this variation ought to embrace spurious results, then for such hierarchical data, multilevel modeling is recommended.