Geographically Weighted Canonical Correlation Analysis: Local Spatial Associations Between Two Sets of Variables

This article critically assesses the utility of the classical statistical technique of Canonical Correlation Analysis (CCA) for studying spatial associations and proposes a new approach to enhance it. Unlike bivariate correlation analysis, which focuses on the relationship between two individual variables, CCA investigates associations between two sets of variables by identifying pairs of linear combinations that are maximally correlated. CCA has strong potential for uncovering complex multivariate relationships that vary across geographic space. We propose Geographically Weighted Canonical Correlation Analysis (GWCCA) as a new technique for exploring local spatial associations between two sets of variables. GWCCA localizes standard CCA by weighting each observation according to its spatial distance from a target location, thereby estimating location-specific canonical correlations. The effectiveness of GWCCA in recovering spatial structure and capturing spatial effects is evaluated using synthetic data. A case study of US county-level health outcomes and social determinants of health further demonstrates the empirical capabilities of the proposed method. The results indicate that GWCCA has broad potential applications in spatial data-intensive fields such as urban planning, environmental science, public health, and transportation, where understanding local multivariate spatial associations is critical.

💡 Research Summary

This paper introduces Geographically Weighted Canonical Correlation Analysis (GWCCA), a novel extension of classical Canonical Correlation Analysis (CCA) that explicitly incorporates spatial heterogeneity. While traditional CCA yields a single set of global canonical variates and correlation coefficients for two multivariate datasets, it assumes stationarity and cannot capture local variations that are common in geographic data. GWCCA addresses this limitation by applying a spatial weighting scheme—similar to Geographically Weighted Regression (GWR) and Geographically Weighted Principal Component Analysis (GWPCA)—to the computation of means, covariance matrices, and cross‑covariance matrices at each target location. The method uses distance‑based kernel functions (Gaussian, Exponential, Box‑car, Bi‑square, Tri‑cube) with an adaptive bandwidth defined by the distance to the k‑th nearest neighbor, allowing the analyst to control the spatial scale of local estimation.

A Residual Goodness‑of‑Fit (RGOF) statistic is proposed to guide the selection of both the optimal bandwidth and the number of canonical variates to retain. RGOF quantifies the proportion of unexplained canonical correlation after retaining a given number of components; the bandwidth search terminates when successive improvements fall below 1 % for a preset number of steps (default patience = 2). This early‑stopping rule mitigates the tendency of geographically weighted models to over‑smooth data with excessively large bandwidths, while also reducing computational burden.

The authors validate GWCCA with synthetic data that embed known spatial patterns of canonical correlation (strong correlation in the core of a grid, weaker correlation toward the periphery). Across a range of kernel choices and bandwidths, GWCCA accurately recovers the imposed patterns, outperforming existing spatial extensions such as Spatial CCA (SCCA) and Canonical Spatial Correlation Analysis (CSCA), which remain global in nature.

An empirical case study examines U.S. county‑level health outcomes (e.g., life expectancy, obesity prevalence) and a set of social determinants (income, education, employment, housing quality). Applying GWCCA to 3,142 counties reveals pronounced spatial heterogeneity: in the densely populated Eastern corridor, income and education show very high canonical correlations (≈ 0.78) with health metrics, whereas in the Midwest’s rural areas environmental variables (air quality, green space) dominate the canonical relationship (≈ 0.62). The Western coastal counties display highly variable correlations, indicating pockets where socioeconomic disparity has a weaker health impact. Local canonical loadings further identify which specific variables drive these relationships, offering actionable insight for targeted policy interventions.

The paper discusses several limitations. High‑dimensional variable sets can produce complex canonical vectors that are difficult to interpret without prior dimension reduction. Bandwidth selection remains sensitive to data density, potentially leading to over‑smoothing in sparsely sampled regions. Moreover, the current implementation assumes linear relationships; extending GWCCA to non‑linear settings (e.g., kernel CCA, deep learning) and incorporating Bayesian uncertainty quantification are promising future directions.

In conclusion, GWCCA provides a powerful framework for uncovering location‑specific multivariate associations, bridging a methodological gap in spatial data science. Its ability to generate spatially explicit canonical correlations and loadings makes it valuable for fields such as urban planning, environmental science, public health, and transportation, where understanding local multivariate dynamics is essential. The authors anticipate that further methodological refinements—multiscale bandwidths, Bayesian extensions, and real‑time GIS integration—will broaden GWCCA’s applicability and impact.