Geographically-weighted Regression

This report explores the use of geographically-weighted regression (GWR) in spatial analysis, demonstrating its effectiveness in capturing spatial heterogeneity that global models like Ordinary Least Squares (OLS) might miss.

By accounting for local variations, GWR allows for a more accurate understanding of relationships between variables that change over space. In this study, children's language abilities scores in Vancouver are analyzed against factors like social skills, neighborhood demographics, and economic conditions to uncover spatial patterns that would be overlooked in a global model. GWR is particularly valuable for its ability to inform targeted policy-making, optimize resource allocation, improve urban planning, and provide more precise environmental assessments.

Below are some maps and figures from this report. See more figures and the full report here.

This map shows the difference in residuals between a global model (Ordinary Least Squares, OLS) and a local model (Geographically Weighted Regression, GWR) for predicting children's language abilities in Vancouver. Areas shaded in pink and green highlight where GWR provides more accurate predictions than OLS, capturing local variations that the global model overlooks. Or, said plainly, pink areas likely represent neighborhoods where factors like income or social context have less impact on children's language abilities than a global model would predict, while green areas indicate neighborhoods where these factors have a stronger impact than expected.

GWR takes this idea further by revealing clusters of neighborhoods with similar characteristics, helping us understand why these local variations occur. The second map clusters Vancouver neighborhoods based on factors such as childcare needs, family size, income, lone-parent percentage, and recent immigrant percentage.

For example, the red cluster represents areas with high childcare needs and lower average incomes, while the blue cluster shows neighborhoods with higher incomes and fewer lone-parent families. These clusters provide context to the local differences seen in the first map, showing why certain areas respond differently to factors affecting children's language abilities. By understanding these clusters, GWR enables more precise and tailored policy-making, urban planning, and community interventions suited to the unique needs of each neighborhood.