Flexible Modeling of Transition Processes via Bayesian Spline Rate Models with Application to Estimating and Projecting Modern Contraceptive Use

Herbert Susmann , University of Massachusetts at Amherst
Leontine Alkema , University of Massachusetts Amherst

Many statistical models exist to generate probabilistic estimates and projections of demographic and health indicators. Some of these models, for indicators like the modern contraceptive use rate (mCPR), draw on the observation that these indicators follow a common pattern across countries characterized by a transition between stable states (for mCPR, this is the transition from low adoption of modern family planning to high adoption.) These models work by modeling the rate of change of the indicator as a function of its level. Commonly used approaches posit parametric functional forms for the relationship between rate and level that may not be true in practice. In this work, we relax this assumption by using B-splines with time-varying coefficients to model the rate vs. level relationship. We use this approach to construct estimates and projections of mCPR in Family Planning 2020 initiative countries, and compare its performance to existing models.

See extended abstract

 Presented in Session 73. Issues in health data analysis