Speaker: Dr Niamh Cahill, Maynooth University Department of Mathematics and Statistics
Title: "Bayesian Modelling Approaches for Family Planning Indicators"
Abstract: Knowledge about past and future trends in contraceptive use (modern and traditional) and the unmet need for family planning is the key to assessing progress towards meeting family planning targets and achieving goals. For over 2 decades the family planning community has relied on national-scale surveys to track progress in family planning. The family planning estimation model (FPEM) was developed to model these survey data for the purposes of producing country-specific annual estimates and projections of family planning indicators. FPEM takes survey observations of contraceptive prevalence and unmet need as input and produces estimates and projections based on combining assumed systematic trends with data driven distortions that capture how the observed data deviate from expected trends over time. Country-specific model parameters are estimated with a Bayesian hierarchical model, such that estimates are based on the data available in the country of interest, and also the sub-regional, regional, and global experience.
Reliance on large-scale surveys to inform projections in FPEM and track family planning programs can leave many countries in a difficult situation given that the time between subsequent surveys can be up to 5 years or more. In more recent years, routinely collected family planning service statistics including the following data elements (1) number of contraceptive commodities distributed to clients and/or facilities, (2) number of family planning service visits, and (3) number of current contraceptive users, have been used to provide additional information for estimating changes in modern contraceptive use at national and sub-national levels. While an advantage of using service statistics is the frequency at which they can be made available, a limitation is that none of these data elements are sufficiently accurate to be used to produce reliable stand-alone observations of modern contraceptive use that are comparable to those produced by surveys. However, defective data can still be useful if used properly. Therefore, I will present some exploratory analysis of available service statistics data and discuss some modeling options for using these routinely collected data in FPEM.