Paper

Abstract

This study tests whether a small collection of publicly observable signals can forecast weekly United States public mass-shooting incidence. The outcome is the number of Mother Jones / Codeholics-defined incidents in each Monday–Sunday ISO week: a public, indiscriminate shooting with at least three fatalities. We build a 522-week panel from 2014-W01 through 2023-W52 and reserve the final 104 weeks, 2022-W01 through 2023-W52, as a fixed test set. Predictors include trailing incident counts, seasonality, holiday and school-session indicators, lagged FBI NICS counts, annual population, and a completed-month incident-frequency proxy for public attention. We compare a week-of-year seasonal mean, a lag-only negative-binomial (NB) model, a regularized full NB model, and regularized logistic models for "any incident" and "at least two incidents."

The full NB obtained test RMSE 0.450, slightly below the lag-only NB at 0.452 and the seasonal baseline at 0.465, but had worse MAE. For the at-least-one task, logistic average precision was 0.271 versus 0.237 for the NB probability baseline. Only one at-least-two week occurred in the test set, making that discrimination estimate extremely unstable. These are weak predictive signals, not evidence of causality or a deployable warning system.

Headline TEST-set metrics (104 weeks, held out)

The full NB dominates upper-quantile loss (pinball@0.8 / pinball@0.9) where mass-shooting alerting actually fires; the dedicated logistic wins precision-recall

• calibration for the "any incident" task. The at-least-two task had only 1 positive in 104 weeks — too rare for stable evaluation.

Full manuscript

The 11-section, 2,355-word manuscript is here as static markdown + BibTeX download:

• manuscript.md — full paper
• references.bib — 7 BibTeX entries (Mother Jones / Follman, Codeholics enrichment, Lankford & Madfis, GVA methodology, BuzzFeed News NICS, FBI NICS, World Bank)

Sections

1. Abstract
2. Research question and contribution
3. Outcome definition and the GVA gap
4. Data sources and the public-attention substitution
5. Temporal design and leakage controls
6. Predictive features
7. Models
8. Evaluation measures
9. Held-out results
10. Negative findings, limitations, and ethics
11. Reproducibility and conclusion

Reproduction

git clone https://git.catalystgroup.tech/herman/msf-forecast-pipeline
cd msf-forecast-pipeline
pip install -r requirements.txt   # numpy, pandas, matplotlib, scipy, statsmodels, pytest
make all                          # ≈ 2 minutes on a single CPU

make all writes results/features.csv, results/predictions_test.csv, results/metrics_{count,any,high}.csv, results/cv_metrics.csv, results/calibration_any.png, and results/figures/*.png. Six pytest cases cover loaders, features, and metrics.

Ethics statement

Forecast association must not be interpreted as a causal effect of firearm checks, media attention, holidays, or schools. National weekly probabilities should not be used to profile people or communities, allocate punitive policing, or claim that a specific attack is predictable. Responsible use is limited to auditable methodological research and aggregate preparedness discussion.