Race and Conviction Likelihood

If we compare cases that look similar on other characteristics, does the defendant's race still correlate with whether the case ends in conviction? Three independent methods all point in the same direction.

Short answer: Yes. Across all three approaches, non-White defendants have lower adjusted conviction rates than White defendants with otherwise similar case profiles. The differences are modest but consistent. The Black–White gap is roughly 2.5–2.8 pp.

Raw Conviction Rates

Unadjusted

Raw conviction rates vary widely by race, but they mix together many confounding factors. White defendants might be concentrated in counties with higher conviction rates, or charged more often with case types that tend to result in conviction.

Raw Conviction Rate by Race

Unadjusted rates across 1.6 million arraignment records

These raw rates do not account for differences in case characteristics across groups.

Race Group	Conviction Rate	Cases
Asian	25.1%	46,882
Black	33.7%	564,258
Indian	32.5%	4,719
Other	61.7%	10,195
Unknown	59.8%	497,824
White	45.7%	485,374

Adjusted Analysis

Three Methods

To understand whether race itself is associated with outcomes, we need to compare cases that are similar in other ways. Three complementary approaches were used:

Core-Adjusted Regression

Logistic regression controlling for county, court type, arrest type, charge severity, age, ethnicity, gender, and arraignment timing.

Charge-Detail Regression

Adds specific law section, charge weight, and attempt flag on top of all core controls. Tests whether the association persists among cases with the same specific charge.

Matched-Strata Comparison

No statistical model. Groups cases into narrow strata sharing the same profile on all variables. Compares conviction rates between race groups only within shared strata.

Adjusted Conviction-Rate Gaps

vs. White Reference

Difference vs. White Defendants (percentage points)

Negative = lower conviction rate than White defendants with similar case profiles

All three methods agree in direction. Black–White gap is 2.6–2.8 pp across methods.

Race Group	Core-Adjusted	Charge-Detail	Matched Strata
Asian	−5.3 pp	−5.2 pp	−6.7 pp
Black	−2.7 pp	−2.8 pp	−2.6 pp
Indian	−0.6 pp	−1.9 pp	−5.3 pp
Other	−8.0 pp	−5.0 pp	−5.4 pp
Unknown	+1.8 pp	−5.9 pp	−3.1 pp

Odds Ratios

Regression Coefficients

Odds ratios below 1.0 mean lower odds of conviction compared to White defendants. Both regression models produce consistent estimates.

Odds Ratio vs. White (Reference = 1.0)

Both regression specifications shown. Dotted line = no difference.

Race Group	Core-Adjusted OR	Charge-Detail OR
White (reference)	1.000	1.000
Asian	0.738	0.717
Black	0.856	0.835
Indian	0.965	0.886
Other	0.628	0.727
Unknown	1.108	0.686

What the Analysis Suggests

Race is associated with conviction likelihood even after adjusting for other observable case characteristics. The association is not an artifact of one particular model specification.
The direction is consistent: non-White groups show lower adjusted conviction rates than White defendants with similar case profiles.
The Black–White gap is roughly 2.5 to 2.8 percentage points across methods. The Asian–White gap is larger, roughly 5 to 7 percentage points.
The pattern is robust to adding charge-level detail. Controlling for the specific law section and charge weight does not explain away the race association.

Limitations

Unmeasured variables: The public data lacks evidence strength, attorney quality, criminal-history detail, plea dynamics, and pretrial detention status. Any of these could partly or fully explain the adjusted gaps.

Large "Unknown" category: About 31% of cases have race recorded as "Unknown," which may distort comparisons if recording practices vary by jurisdiction.

Observational design: No adjustment fully substitutes for random assignment.

New York only, 2021–2025. Results may not generalize to other states or time periods.