Abstract

The knowledge of accurate population estimates is key to the development and implementation of relevant government policies designed to address myriads of developmental problems and challenges within a country. For example, in other to more effectively tackle the effect of inflation or pandemic within a country, the accurate knowledge of the number of people at risk within a given small area is usually invaluable, and such information are usually only available in population census data. However, censuses are often incomplete and outdated due to logistic issues in that several countries have not conducted census for many years ago thereby relying mostly on imperfect population projections for such planning. The advancement of data collection methods through satellite based imagery that could obtain counts of buildings and other geospatial covariates means that these could be combined in a statistically robust manner along with multiple survey data sources to predict population count within these countries. So far, the statistical methods used for this have mostly relied on the Markov chain Monte Carlo (MCMC) methods to draw samples from the posterior distribution of the model parameters. However, as with most MCMC-based methods there are often issues with convergence especially in the face of high-dimensional data. In addition, sampling from the posterior distribution of most parameters based on MCMC can be very time consuming. In this paper, we presen an alternative method for combining these satellite-based imagery and several other population. Our method which based on the integrated nested Laplace approximation (INLA) which also accounted for error within measurement errors is shown to be both faster and very efficient.