Small-Area Multidimensional Poverty Estimates for Tonga 2016: Drawn from a Hierarchical Bayesian Estimator

Type Journal Article - Applied Spatial Analysis and Policy
Title Small-Area Multidimensional Poverty Estimates for Tonga 2016: Drawn from a Hierarchical Bayesian Estimator
Author(s)
Publication (Day/Month/Year) 2019
Page numbers 1-24
Publisher Springer
URL https://link.springer.com/article/10.1007/s12061-019-09304-8
Abstract
Tonga has recently adopted the consensual approach to produce its official multidimensional poverty measure. This index is computed using data from the Household’s Income and Expenditure Survey (HIES). The population of Tonga is scattered across 5 main groups of islands and high-quality spatial data is vital to inform policies. One limitation is that HIES data originate from a nationally representative survey that cannot produce reliable estimates for small areas such as constituencies, villages or blocks, and governments require highly disaggregated data to better inform policies, for example, with regard to natural disasters. This paper produces small-area estimates of poverty based on a hybrid hierarchical Bayesian (HHB) estimator, which draws on the standard hierarchical Bayes (HB) approach but uses a more efficient computation process – Hamiltonian (hybrid) Monte Carlo (HMC) – to produce the posterior distributions. The HHB estimator is then applied to Tonga’s National Population Census (2016) to present estimates down to island, constituency and block level. The results suggest that the extent of poverty is lower in Tongatapu than in Eua, Vava’u, Ha’apai and Niuas, while its prevalence is very similar (around 35%) in Eua, Vava’u and Ha’apai. Constituencies in Tongatapu show lower poverty rates than in the rest of the islands, and block-level data show a clear spatial pattern of poverty distribution in the capital Tongatapu. These are the first small-area indirect poverty estimates based on a hierarchical Bayesian model and drawn from the consensual approach (CA).