Police Risk

CODE: GitHub
DATATYPE: police officers
SOURCE: departmental records
YEARS: 2014
SIZE: 2,000 (approx.)
SUBSET: test set
VIZTYPE: glyph histogram
FEATURES: dispatches, traffic stops
PROJECT: Wintour: Visualizing Predictive Models

This is a glyph histogram of scores assigned to police officers using a predictive model. Each glyph encodes five additional variables, one mapped to the center square and the others surrounding it. The brightness of the center square and the saturation of the surrounding squares corresponds to the variable value. Hatching indicates missing data. We can see that high values of the primary variable, mapped to the center square, are correlated with high risk, and officers with lots of missing data are generally scored low. Since missing values in this model are imputed as zero, the relevant variables must also be positively correlated with risk.