Publication type
Thesis/Degree/Other Honours
Author
Publication date
June 1, 2021
Summary:
Fertility projections are a key determinant of population forecasts, which are widely used by government policymakers and planners. They are also vital to anticipate demand for maternity and childcare services, as well as school places and housing. As such, models that can generate plausible fertility forecasts with appropriate uncertainty are in high demand. To this end, in this thesis we develop two distinct Bayesian fertility projection models, using multiple data sources and state-of-the-art computational methodology.
In the first approach we take an international perspective, working with population-level data indexed by age and cohort from the Human Fertility Database. In keeping with the recent literature, we propose an intuitive and transparent hierarchical Bayesian model which borrows strength across ages and cohorts. Using Hamiltonian Monte Carlo methods, we obtain forecasts for 30 countries. Quantitative assessment of the predictive accuracy using scoring rules indicates that our model predicts at a comparable level to that of the best-performing models in the current literature overall, with stronger performance for countries without a recent structural shift. Our findings support the position of hierarchical Bayesian modelling at the forefront of population forecasting methods.
Our second approach focuses on England and Wales, modelling individual-level data in the form of fertility histories and additional information collected from 18,218 women interviewed in Understanding Society. We progress the discrete-time event history analysis literature in this context by applying the smooth, flexible framework of generalized additive models (GAMs). Through fitting parity-specific logistic GAMs to the survey data, we learn about the effects of age, cohort, time since last birth and qualification on fertility. We then develop our chosen GAMs into a Bayesian projection model incorporating population-level data. Our innovative integration method enables assessment of forecast sensitivity in relation to the balance of the two datasets, thus making important advances in statistical methodology.
Subjects
Link
https://eprints.soton.ac.uk/450468/
#578285