Texas Sex Offender Registry Race/Ethnicity Audit

 Unknown macro: anchor. Click on this message for details.
The [anchor] macro is not in the list of registered macros. Verify the spelling or contact your administrator.

Overview

 On September 9, 2023, Unbiased Crime Report (@UnbiasedCrime) published an in-depth analysis of the Texas Department of Public Safety’s sex offender registry, focusing on race and ethnicity misclassifications. The thread examines a database of 101,976 registrants and highlights systemic data errors that misrepresent non-White offenders as White.

Data Collection

 1. Downloaded the full Texas Sex Offender Registry: 101,976 entries.
2. Queried the database to flag potential mismatches between surnames, self-reported race/ethnicity, and demographic patterns.
3. Used custom scripts to compute initial accuracy metrics.

Key Findings

 1. Raw Misclassification Counts
* Hispanics flagged as “Non-Hispanic White”: 11,541
* Blacks flagged as White: 92
* Asians flagged as White: 34
* Whites flagged as Black: 1

1. Accuracy Metrics
   • Initial accuracy: 89.48 %.
     
   • After manual verification: 92.27 %.
     
   • Remaining false positives to review: 2,590.
2. Notable Examples
   • Clusters of “Hernandez” surnames incorrectly marked as White.
     
   • Dequan, a documented Black offender, listed as “Non-Hispanic White.”
     
   • Louby Innocent, originally marked Black in Florida, appeared as White in Texas.

Methodology

 1. Automated surname-based queries to generate candidate mismatches.
2. Manual review of flagged entries by cross-referencing public DPS records.
3. Iterative re-runs to refine the matching algorithm and reduce false positives.

Impact and Next Steps

 1. Archived original registry entries to preserve evidence before corrections.
2. Planned release of a comprehensive spreadsheet with all verified entries and source links.
3. Recommendation: Texas DPS should implement routine data audits to ensure accurate race/ethnicity reporting.

References

 ^1
2

Figure 1: Comparison

Figure 2: Archived Examples