HD38 - Study on Methodology Used to Project Virginia's Prison Population


Executive Summary:
Virginia has been utilizing some form of computer projection model for inmate forecasting since 1977. Over the years there have been numerous revisions to existing systems or scraping of old systems as newer, more advanced and accurate modeling became available. Virginia's use of the existing National Council on Crime and Delinquency (NCCD) model began in 1988 by the Department of Corrections.

Until 1994, an informal group of people from various state agencies and branches had been meeting to review computer model projections and "refine" them based on their expertise and knowledge of many of the factors that produce the projections.

In 1994, the Secretary of Public Safety developed a new process utilizing the staff of the Department of Criminal Justice Services. This process was not subject to as much scrutiny as had the previous process. The resultant forecast was used by the 1994 special session and the 1995 regular session of the General Assembly in funding decisions.

During 1995, the "traditional" process was re-instituted by the Secretary. It involves members from all three branches of government, private research organizations, faculty from leading universities, and local and state law enforcement. They revisited the 10-year forecast that had been done the previous year. This new projection revealed that the 1994 group had overestimated the number of prison beds that would be required, the worst being for FY 98 where the 1994 estimate that had been used was projected to be more than 4,200 beds high.

Virginia's current system has an accumulated average error rate for state responsible offenders for FY95-96 of 0.05%. Virginia utilizes an extensive consensus-building process to generate the final projections. This consensus-building concept is strongly recommended by virtually all experts in the field.

Staff was unable to locate any other state that has a better system, that produces lower error rates than the one Virginia currently uses. Many states use "canned" packages, others have developed sophisticated models of their own. Almost all states surveyed use consensus-building in some form. The bottom line is, the Virginia process works, and with low error rates. Its success is largely due to the professionals who serve on the Policy and Technical Committees.

Recommendation: The process, in its current configuration, is generating reliable information with a low error rate. The process should be maintained, as is, and codified until a new method with a proven track record for accuracy is available.