cs.AIMar 19, 2026

Spatial disparities in simulated patrol detection: a reproducible generative-model audit across Baltimore and Chicago

Authors: Pronob Kumar Barman, Pronoy Kumar Barman

Organizations: University of Maryland, Baltimore County, Department of Information Systems, Baltimore, MD, USA · Jagannath University, Department of Statistics, Dhaka, Bangladesh

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Auditing predictive policing requires distinguishing recorded crime incidents, simulated patrol observations, and the demographic characteristics of places. We present a reproducible spatial simulation audit combining a generative adversarial network (GAN) for patrol allocation with a probabilistic detection model. Using frozen archives of 145,178 Baltimore records (2017-2019) and 234,882 Chicago records (2022), we rebuild geographic covariates from Census tract boundaries and American Community Survey data. We compare GAN allocation with empirical-location sampling and uniform-area deployment under alternative observation streams. Twenty random seeds per condition yield 560 trajectories, each covering February through December. Under detection-only training, mean detection-rate differences between Black-majority and White-majority areas range from -2.03 to 0.75 percentage points across city-years, while total coverage ranges from 1.78% to 7.50%. Paired comparisons with full-record reference trajectories assess observation feedback conditional on the recorded incident archives. In an experiment using a conditional tabular generative model for balanced synthetic replacement, geographic consistency requirements fail in 107 of 220 seed-month fits, leaving no complete annual synthetic-replacement trajectories. Simulated disparities vary with the observation process, allocation policy, training budget, and random seed. These results support reproducible comparisons of modeled scenarios but do not estimate individual racial disparities in police contact or the distribution of unrecorded crime. We recommend reporting absolute detection differences, coverage, synthesis feasibility, and simulation uncertainty alongside ratio-based fairness metrics.

Figures & tables

CityYearArchivedRetainedFeb.–Dec.Unknown
Baltimore2017515115149846950330
Baltimore2018478564784344129328
Baltimore2019458114580142133269
Chicago20222348822340732188141433
Table 1: Archived and retained incident records. Retained counts include January initialization. Unknown counts cover the full retained year; they are included in total coverage but excluded from Black/White rate denominators.
City-yearPolicyTraining streamGap (pp)Coverage (%)
Balt. 2017GenerativeDetected0.75 [-0.69, 2.26]6.41 [6.12, 6.69]
Balt. 2017GenerativeMixed0.66 [0.49, 0.86]4.70 [4.52, 4.88]
Balt. 2017GenerativeReference0.75 [0.44, 1.01]4.73 [4.60, 4.88]
Balt. 2017EmpiricalDetected0.06 [-0.94, 1.03]8.79 [8.56, 8.99]
Balt. 2017EmpiricalMixed0.50 [0.12, 0.86]8.50 [8.42, 8.58]
Balt. 2017EmpiricalReference0.18 [-0.12, 0.48]8.26 [8.17, 8.35]
Table 2: Main simulation outcomes by city, policy, and observation stream. Main results: means and 95% whole-trajectory bootstrap intervals over 20 seeds. Gap is Black-area minus White-area detection in pp; coverage is detected records as a percentage of all archived analysis incidents. Conditions are paired by seed.
City-yearTraining streamAbsolute-gap change (pp)Coverage change (pp)
Balt. 2017Detected2.02 [1.12, 2.96]1.67 [1.42, 1.94]
Balt. 2017Mixed-0.23 [-0.49, 0.03]-0.03 [-0.23, 0.17]
Balt. 2018Detected1.53 [0.72, 2.38]2.34 [1.98, 2.68]
Balt. 2018Mixed-0.08 [-0.44, 0.30]-0.09 [-0.36, 0.17]
Balt. 2019Detected2.11 [1.25, 3.09]2.63 [2.34, 2.90]
Balt. 2019Mixed0.09 [-0.26, 0.39]0.18 [-0.03, 0.39]
Table 3: Paired feedback contrasts for the generative policy. Generative-policy feedback contrasts against the same policy’s full-record reference. Values are mean paired differences and 95% whole-pair bootstrap intervals. Absolute gap uses each seed’s magnitude before averaging.
ConditionFeasible fitsComplete seedsGap (pp)Coverage (%)
Raw220/22020/200.54 [0.24, 0.88]4.87 [4.72, 5.03]
Simple replacement220/22020/200.15 [-0.20, 0.49]5.12 [4.92, 5.34]
CTGAN replacement113/2200/20Not estimableNot estimable
Table 4: Synthetic-replacement feasibility and outcomes. Baltimore 2019 rolling-reference replacement experiment. Annual estimates require complete 11-month trajectories. Failure to synthesize valid samples is not imputed or replaced by a selected-month estimate.
CityTraining streamOdds ratio [interval]TractsTract-months
BaltimoreDetected0.866 [0.773, 1.014]1966456
BaltimoreMixed0.973 [0.938, 1.038]1966456
ChicagoDetected0.999 [0.941, 1.071]8178703
ChicagoMixed0.997 [0.969, 1.025]8178703
Table 5: Exploratory tract-level demographic associations. Exploratory Black-share association per 10 pp in actual tract-month simulated detection. Odds ratios and 95% seed/spatial-block resampling intervals use 3 km blocks, 250 resamples, and covariate-adjusted fractional-binomial models. These are not individual-race or causal effects.
Figure 1: Detection gaps across allocation and observation conditions. Mean pooled detection gap in each city-year; corresponding coverage estimates appear in Table 2 . Points compare the generative and empirical policies under patrol-detected, mixed, and full-record reference training. Error bars are 95% whole-trajectory bootstrap intervals over 20 seeds. Dotted horizontal lines show uniform-area means. The gap denominator is recorded incidents in each area category, not residents.
Figure 2: Monthly detection gaps for the generative policy. Monthly generative-policy detection gaps. Lines are means across 20 trajectories, with pointwise 95% whole-trajectory bootstrap bands. The trajectories share a January initialization but evolve separately by observation stream. A positive value indicates greater modeled detection of recorded incidents in Black-majority tracts.
Figure 3: Feedback contrasts in disparity magnitude and coverage. Paired differences between observed-input generative trajectories and their full-record references. Positive absolute-gap changes mean greater disparity magnitude within this simulation. Coverage differences are percentage points. Error bars resample whole paired trajectories; these are not real-world causal effects.
Figure 4: Replacement outcomes and incomplete synthetic-generation trajectories. Rolling-reference replacement experiment in Baltimore 2019. Annual outcomes are plotted only for complete 11-month seed trajectories. Conditional-generative synthesis failures are recorded in Table 4 ; missing annual estimates are not replaced by results from selected feasible months. Error bars describe trajectory bootstrap variation.
More in this topic

Algorithmic Bias

Algorithmic bias studies how automated systems produce or amplify unfair outcomes across demographic groups, and how to measure and reduce them.

See all 100 papers

Explore similar work

CardsList
Apr 19, 2026

FASE : A Fairness-Aware Spatiotemporal Event Graph Framework for Predictive Policing

Predictive policing systems that allocate patrol resources based solely on predicted crime risk can unintentionally amplify racial disparities through feedback driven data bias. We present FASE, a Fairness Aware Spatiotemporal Event Graph framework, which integrates spatiotemporal crime prediction with fairness constrained patrol allocation and a closed loop deployment feedback simulator. We model Baltimore as a graph of 25 ZIP Code Tabulation Areas and use 139,982 Part 1 crime incidents from 2017 to 2019 at hourly resolution, producing a sparse feature tensor. The prediction module combines a spatiotemporal graph neural network with a multivariate Hawkes process to capture spatial dependencies and self exciting temporal dynamics. Outputs are modeled using a Zero Inflated Negative Binomial distribution, suitable for overdispersed and zero heavy crime counts. The model achieves a validation loss of 0.4800 and a test loss of 0.4857. Patrol allocation is formulated as a fairness constrained linear optimization problem that maximizes risk weighted coverage while enforcing a Demographic Impact Ratio constraint with deviation bounded by 0.05. Across six simulated deployment cycles, fairness remains within 0.9928 to 1.0262, and coverage ranges from 0.876 to 0.936. However, a persistent detection rate gap of approximately 3.5 percentage points remains between minority and non minority areas. This result shows that allocation level fairness constraints alone do not eliminate feedback induced bias in retraining data, highlighting the need for fairness interventions across the full pipeline.

Pronob Kumar Barman, Pronoy Kumar Barman, Plaban Kumar Barman +1

Department of Information Systems University of Maryland, Baltimore County Baltimore, MD, USA · Department of Statistics Jagannath University Dhaka, Bangladesh · MBBS Shaheed M. Monsur Ali Medical College Sirajganj, Bangladesh

Algorithmic BiasGroup FairnessSpatiotemporal ForecastingHawkes Processes
AskPDF
Feb 27, 2026

Predictive Hotspot Mapping for Data-driven Crime Prediction

Predictive hotspot mapping is an important problem in crime prediction and control. An accurate hotspot mapping helps in appropriately targeting the available resources to manage crime in cities. With an aim to make data-driven decisions and automate policing and patrolling operations, police departments across the world are moving towards predictive approaches relying on historical data. In this paper, we create a non-parametric model using a spatio-temporal kernel density formulation for the purpose of crime prediction based on historical data. The proposed approach is also able to incorporate expert inputs coming from humans through alternate sources. The approach has been extensively evaluated in a real-world setting by collaborating with the Delhi police department to make crime predictions that would help in effective assignment of patrol vehicles to control street crime. The results obtained in the paper are promising and can be easily applied in other settings. We release the algorithm and the dataset (masked) used in our study to support future research that will be useful in achieving further improvements.

Karthik Sriram, Ankur Sinha, Suvashis Choudhary

Operations and Decision Sciences, Indian Institute of Management Ahmedabad, 380015, Gujarat, India · Indian Police Service (Retired), India

Spatiotemporal Forecasting
AskPDF
Oct 13, 2026

Fairness under Prediction Error in Predict-then-Optimize Pipelines

Fairness constraints in predict-then-optimize pipelines are necessarily enforced on predicted outcomes y^\hat{y}, since true outcomes yy are unobserved at decision time. We study whether a demographic parity constraint enforced this way continues to hold against yy, and how the resulting violation scales with prediction error. Using a budget-constrained allocation problem solved as a linear program, we show empirically that the true-outcome fairness gap grows with prediction error σσ and saturates, well fit by c(1−e−σ/τ)c(1 - e^{-σ/τ}) with c=0.1545±0.0103c = 0.1545 \pm 0.0103 and τ=3.9045±0.3745τ= 3.9045 \pm 0.3745; its mean is invariant to sample size while its variance decays at the standard O(1/n)O(1/\sqrt{n}) rate. We derive a two-part theoretical bound: a closed-form local sensitivity result showing the gap's first-order response to prediction error depends on only two individuals per group, validated against the solver to within three significant figures; and a universal, error-independent ceiling following from the boundedness of the decision variable, confirmed consistent with but looser than the empirical saturation level. On US Census income data, enforcing demographic parity on a trained predictor's output reduces an unconstrained disparity in true income to roughly one-third of its original size, leaving 34.4% ±\pm 3.3% (California) and 35.7% ±\pm 1.8% (Texas) of the disparity intact and undetectable from the predictions alone -- consistent not only in mechanism but in approximate magnitude across two states with different demographics and income distributions, and stable across ten independent train/test splits in each. A fairness constraint that appears fully satisfied at decision time can mask a substantial, systematic violation of the outcome it was meant to protect.

Hilda Adwubi Osei

Department of Industrial Engineering, Kwame Nkrumah University of Science and Technology, Kumasi, Ghana

Algorithmic FairnessGroup FairnessPredict-Then-Optimize
AskPDF