Commission launches larceny/fraud guideline study and nonviolent‑risk assessment redevelopment using stratified sample and regression analysis

Commission staff presented a multi‑part research plan to update larceny and fraud sentencing worksheets and to redevelop the nonviolent offender risk assessment for larceny and fraud offenders.

For the larceny/fraud guideline work, staff reported they drew the study population from fiscal‑year 2021–2025 guideline records, excluding cases with primary‑offense dates before 7/1/2020 (to reflect the current $1,000 felony threshold) and excluding petty larceny third‑or‑subsequent cases removed by the 2021 General Assembly. Staff described a stratified random‑sample design that produced a study sample of 2,147 cases: fraud strata (including take‑all strata for rare primary fraud offenses) and several grand‑larceny and other larceny strata. The sample intentionally overweights fraud and non‑grand larceny cases to ensure adequate coverage for analysis; weights can be applied in analysis to reflect population proportions.

Staff said they have already requested in‑state criminal‑history data from the Virginia State Police and that FBI (out‑of‑state) clearance for additional history is pending. Planned analyses include cross‑tabulations, descriptive statistics, logistic regression for the incarceration/no‑incarceration decision, and multiple linear regression for length‑of‑incarceration recommendations. The aim is to translate statistically important factors into worksheet scoring for revised guideline instruments.

Separately, Dr. Chang Kwan outlined the plan to redevelop the nonviolent offender risk assessment specifically for larceny and fraud offenders. The approach uses propensity‑score matching (PSM) to pair current offenders (FY2022–2024 cohort) with historically similar offenders from FY2015–2018 so that follow‑up (three‑plus years) conviction outcomes can be observed. After matching and exclusions, staff reported a matched analysis cohort of 1,775 defendants; staff checked balance statistics to confirm the matched cohort resembles the current cohort on key covariates (gender, age, sentence length, primary‑offense threshold indicators). The validation measure is new felony conviction within three years that resulted in a violent conviction; the model will use multivariate regression to identify statistically significant risk and extra‑legal factors and produce a revised risk‑assessment worksheet.

Staff outlined next steps: finalize criminal‑history matching (pending FBI data), run multivariate models, convert factor effects into worksheet scores, and provide results to the commission for review. Commissioners asked about sample attrition, how plea bargains and negotiated dispositions are reflected in worksheet data, and the feasibility of completing substantive analysis before the Sept. 3 meeting. Staff said they would supply data and caveats and indicated some analyses (and refined sample weights) will depend on receipt of out‑of‑state history from the FBI.