cosmic

Conditional Ordinal Stereotype Model for Incident-Level Comparison

Implements the Conditional Ordinal Stereotype Model for Incident-Level Comparison (COSMIC), a method for analyzing ordinal outcomes observed across multiple actors within shared events. The model uses a conditional likelihood to remove event-level confounding and estimate actor-specific propensities relative to their peers. Efficient computation is achieved via a dynamic programming algorithm for the Poisson-multinomial normalization term, enabling scalable estimation with Markov chain Monte Carlo. The package provides tools for data preparation, model fitting using Stan, and extraction of posterior summaries for comparative inference. Estimation of police officer propensity to escalate force is the primary motivation for the model. For more details see Ridgeway (2026) "A Conditional Ordinal Stereotype Model to Estimate Police Officers’ Propensity to Escalate Force" <doi:10.1080/01621459.2025.2597050>.

README

DATA - Conditional Ordinal Stereotype Model to Estimate Police Officers' Propensity to Escalate Force
================

`dataSPD.RData` contains the Seattle PD use-of-force data already formatted for analysis, a single data frame `d`. The data frame contains the following columns:

- `id`: use-of-force incident identifier, sequential integers from 1 to 4,821

- `idOff`: officer identifier, integers from 1 to 1,503. `idOff` is a scrambled identifier and does not link to any Seattle PD officer information

- `y`: ordinal use-of-force level, the most serious level of force that the officer used at this incident. `y` contains integers from 1 to 4, where 1 is the lowest severity and 4 is the highest severity

Incidents with no variation in `y` (e.g. single officer incidents, incidents where no officer uses force) have been removed from the data since they have no information for the parameters of interest through the conditional likelihood.

Versions across snapshots

Dependencies (latest)