Skip to content

Borrowing Strength- a collaborative software development for Small Area Estimation

StatusFinished
Period1/03/1831/07/19

Description

Small Area Estimation (SAE) is the name given to the process of calculating statistics of interest for each of a set of small areas within a larger population, for example neighbourhood average house prices. Typically, standard surveys provide insufficient data on each small area to give accurate estimates and so instead these techniques 'borrow strength' by incorporating information from other areas, measured variables, and datasets.

In this project we will develop a user-friendly statistical software toolkit that will assist with all aspects of the process of constructing small area estimates.

We will focus on a computationally intensive estimation approach known as Markov Chain Monte Carlo (MCMC) methods and will exploit the possibilities of modern computing to develop methods that use parallel processing to make MCMC a viable approach for the big datasets that are often used in SAE. MCMC methods are particularly good at accurately capturing all sources of uncertainty in the modelling and so produce better calibrated estimates. They can also be easily extended to include additional features of the data within the modelling process.

There already exist many competing approaches to SAE and so we will not only develop new approaches but also provide access to existing methods developed in the emdi software package by members of the team.

We will develop through the Stat-JR software package a common interface to both the new MCMC methods and those developed in emdi so that users can try out and compare alternative approaches through one convenient software interface. We will do this by using our statistical analysis assistant features in Stat-JR which will allow a single interface to the two alternative approaches and will enhance the statistical analysis by including details of the different steps involved in the analysis.

Structured keywords

  • SoE Centre for Multilevel Modelling

View research connections

Related faculties, schools or groups