Cross‐Validation as a Bandwidth Criterion for Geographically Weighted PCA: An Ethics of Ambiguity

ABSTRACT Geographically weighted principal component analysis (GWPCA) summarizes spatial multivariate structure through local eigen decompositions whose output depends strongly on a bandwidth, conventionally chosen by leave‐one‐out cross‐validation (CV). While locating the bandwidth that minimizes CV, several competing minima are frequently observed. We focus on the implications of multiple ambiguous minima as the object of study. We formalize it through the CV‐equivalent set () of bandwidths that the data do not distinguish from the CV minimum () and show that bandwidths within this set differ materially in their bias–variance composition and loading structure. Simulations across eight data‐generating processes show that wherever the covariance subspace varied locally, the equivalent set spanned roughly a quarter of the candidate range grid, and the oracle bandwidth typically lay inside the set while differing from . In applications to the Dublin voter turnout and the Kola geochemical survey, the same ambiguity appeared, and within‐set disagreement between local loadings concentrated where the local eigengap was small. We offer these findings as the first in‐depth study of the characteristics of CV as used in GWPCA, and recommend reporting standards to include the equivalent set, and mapping the local eigengap.

Authors

Institutions

Publication Details

Journal
Geographical Analysis
Published
2026-11-17
DOI
https://doi.org/10.1111/gean.70071
Primary Topic
Spatial and Panel Data Analysis
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Cross‐Validation as a Bandwidth Criterion for Geographically Weighted PCA: An Ethics of Ambiguity

Andrew Kampfschulte
Geographical Analysis
Spatial and Panel Data Analysis
article

Cross‐Validation as a Bandwidth Criterion for Geographically Weighted PCA: An Ethics of Ambiguity

Andrew Kampfschulte
article en

Abstract

ABSTRACT Geographically weighted principal component analysis (GWPCA) summarizes spatial multivariate structure through local eigen decompositions whose output depends strongly on a bandwidth, conventionally chosen by leave‐one‐out cross‐validation (CV). While locating the bandwidth that minimizes CV, several competing minima are frequently observed. We focus on the implications of multiple ambiguous minima as the object of study. We formalize it through the CV‐equivalent set () of bandwidths that the data do not distinguish from the CV minimum () and show that bandwidths within this set differ materially in their bias–variance composition and loading structure. Simulations across eight data‐generating processes show that wherever the covariance subspace varied locally, the equivalent set spanned roughly a quarter of the candidate range grid, and the oracle bandwidth typically lay inside the set while differing from . In applications to the Dublin voter turnout and the Kola geochemical survey, the same ambiguity appeared, and within‐set disagreement between local loadings concentrated where the local eigengap was small. We offer these findings as the first in‐depth study of the characteristics of CV as used in GWPCA, and recommend reporting standards to include the equivalent set, and mapping the local eigengap.

Geographical AnalysisVol. 59(1)
University of Southern California (US), Children's Hospital of Philadelphia (US)
Spatial and Panel Data Analysis
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon