Application of Variance Components Modelling Approach to Tanzania Demographic and Health Survey Data; Identifying Determinants of Modern Contraceptive Uptake

Abstract Background: Over time, Demographic and Health Survey (DHS) data remain valuable to examine variables relating to nationally representative population for low- and middle-income countries. In Tanzania, there are very limited DHS-based studies on the uptake of Modern Contraceptive Use (MCU). Present studies were focused on individuals’ levels measurements, yet research have shown the MCU variations still exists to other levels of populations. In this study, we aimed to use the variance component modelling approach to explore variation on MCU at PSU and region levels while considering survey sample weighting. Methods: Using DHS 2016-2017 in Tanzania, we study various variance structure and the respective variation on MCU to more than 5174 Women of Reproductive Age (WRA) 15-49 years. Firstly, a single variance component was used, followed by its extension to random coefficient model and we tracked changes in the models. Results: There was an influence of random variations on MCU on the levels of populations much explained by PSU-level clustering than region. On the fixed part, age of a woman, education level, husband education level, desire to have children, exposure to media and wealth index were the important determinant for MCU among WRA in Tanzania. For example, the odds of MCU among middle ages women (20-29 and 30-39) were 29% and 90%. Also, odds of MCU increases with an increase to media exposure and primary and secondary school women had higher chances for MCU. We also found assessing effects of covariates at two different nested levels eventually resulted in different estimates.Conclusion: This study highlighted on utility of accounting for variance structures in addressing various sources of variations on MCU while using DHS national level data. Apart from MCU, the TDHS data has been widely applied to examined other variables pertaining to public health issue. Thus, this approach could be considered a better modelling technique for the DHS studies.