Methodology 2.0
How the Disparity Score is calculated
The model asks two separate questions and publishes a score only when the data is strong enough to support the comparison.
How it works
How CourtScope checks sentencing patterns
We do not treat every case as if it were the same. We compare outcomes across racial groups after accounting for supported case differences.
- We compare cases with similar offense severity and prior-record histories.
- We account for plea versus trial, age, and gender when supported by the data.
- We compare each racial group separately with White defendants; we do not hide differences inside one combined minority group.
- We publish “Not Enough Data” when coverage, sample size, or uncertainty does not pass the rules.
1. Build one city analysis window
For each city release, CourtScope uses a shared rolling window of up to eight years, anchored to the latest qualifying sentence date in that city. Every judge in the release is evaluated against that same window.
2. Require enough qualifying cases
A judge needs at least 50 qualifying cases. The White reference group and at least one separately reported non-White comparison group must each have enough cases to support publication. Small unsupported groups are withheld rather than merged into a single “minority” category.
3. Compare similar cases
The model accounts for supported case characteristics including offense severity, prior-record category, plea versus trial, age band, and gender. A factor is used only when the dataset contains it consistently enough to support the adjustment.
4. Measure two outcomes
Incarceration outcome
The adjusted difference in the probability that a sentence includes incarceration that must be served.
Sentence-length outcome
The adjusted difference in incarceration sentence length among cases with time to serve.
5. Keep direction visible
Each group is compared separately with White defendants. The published highlights state which group had the higher adjusted incarceration rate or longer adjusted sentence when the data supports that statement.
6. Use the larger signal
The overall Disparity Score is the larger of the incarceration component and the sentence-length component. This conservative rule prevents a large gap in one outcome from being hidden by a smaller gap in the other.
7. Publish uncertainty and limits
Every result includes a data-strength label, qualifying case count, analysis window, and uncertainty checks. CourtScope publishes Not Enough Data when publication gates fail.
Interpretation limit: The score describes aggregate statistical patterns in the analyzed records. It does not prove racism, discrimination, intent, misconduct, causation, or whether any individual sentence was appropriate.
Fixture and official-data boundary
Synthetic fixtures are used to test the system. They are labeled as fixtures and must not be presented as findings about real judges, courts, or defendants. Official city results remain unpublished until a complete authorized dataset passes the city pipeline.