Scaling down, getting it wrong: The ecological fallacy and spatial disaggregation in crime and place research

The ecological fallacy cautions against inferences at a spatial scale finer than the data support. Despite this, researchers disaggregate areal data and redistribute census variables from larger to smaller units. Using ground-truth household and commercial property counts, we conduct a controlled round-trip aggregation experiment to quantify error introduced by two common redistribution methods: uniform redistribution and geographically weighted redistribution. Across all experiments, Spearman rank correlations between redistributed and true values range from 0.35 to 0.53, indicating that redistribution scrambles the rank ordering of spatial units. Counterintuitively, weighted redistribution, the more sophisticated commonly used approach, consistently underperforms compared to naive uniform redistribution. We demonstrate that redistributed census variables produce systematically different regression coefficients when used as predictors of crime, with approximately 27% of coefficients changing sign. Area-weighted redistribution inflates model R 2 spuriously while producing just as many sign reversals as uniform redistribution, creating misplaced confidence in results that may be systematically wrong. We conclude that, under the study conditions examined here, disaggregating areal data to finer spatial scales introduces errors substantial enough to reverse coefficient signs in approximately one quarter of cases, and caution strongly against the practice in similar research contexts.