Summary

The Resources and Workload Management System was begun in 1986 to improve management of the Internal Revenue Service's (IRS) inventory of taxpayer delinquencies and maximize collections. Using various formulas, the system assigns scores to delinquencies that relate to probable collections when returns have been filed or to probable amounts due when nonfilers are involved. IRS has recently begun revising current formulas and developing new ones to improve the relationship between scores and probable collections or amounts owed. Regression analysis will be used to test hundreds of potentially relevant variables. This report provides GAO's views on how IRS could improve the system and its future evaluation of formulas.

GAO found that IRS: (1) uses 30 formulas to compute RWMS scores, but the formulas are primarily based on the outcomes and characteristics of 1982 delinquencies; (2) plans to revise RWMS so that it can use more current information to make the relationship between scores and probable collections and probable amounts due more precise; (3) plans to create new formulas for generating substitute returns for nonfilers and employment tax delinquencies; (4) has developed a list of hundreds of variables to test in its revision efforts, but has not sought the experience of private industry in identifying potentially relevant variables for assessing collectibility; (5) emphasis on amounts due rather than on collection probability may not lead to collection maximization, since it may not be able to collect a significant amount of the taxes due; (6) does not assign RWMS scores to the prevention and identification components of revenue officers' workloads; (7) evaluations of RWMS formulas show that scores are generally related to actual collections or amounts due, but there is only limited analysis of the relationship between scores and collections or amounts due; and (8) used a sampling methodology to evaluate its formulas, but did not measure sampling error or confidence level and could not assess the validity or reliability of sample results.