The Spatial Distribution of Poverty: A geographically weighted regression (GWR)

Introduction

How can we explore the spatial distribution of poverty and determine its correlates? This exercise examines data from Sri Lanka. Many quantitative studies use ordinary least squares (OLS) regression to estimate the effect of variables such as ethnicity, proximity to urban areas, elevation, and other indicators of development on poverty rates. This exercise uses a more generalized geographically weighted regression (GWR) model in addition to the OLS model to incorporate the effects of spatial clustering.

Location

Sri Lanka

Time to complete the lab

Three hours

Prerequisites

Familiarity in the use of ArcGIS 10

Understanding of OLS regression and GWR including diagnostic statistics

Understanding of spatial statistics such as Moran's I

Data used in this lab

SriLankaCaseStudy (Amarasinghe et al. 2005): A shapefile for Sri Lanka including poverty counts at the Divisional Secretariat level

Projection parameters:

• Projection: Transverse Mercator.
• Scale factor at central meridian: 0.999600
• Longitude of central meridian: 81
• Latitude of projection origin: 0.000000
• False easting: 500000.000000
• False northing: 0.000000