Food stamp payment error rates: can state-specific performance standards be developed?

Since 1981 the Food and Nutrition Service, U.S. Department of Agriculture has imposed financial penalties on 49 states with excessive overpayment errors in the Food Stamp program, many in more than one year, for a total of $423.5 million. Almost all have appealed the sanctions, and several have taken their appeal to the courts, questioning the fairness of the financial penalties. Although the quality control (QC) system holds all states to the same standard, payment error rates (defined as the proportion of total benefits paid erroneously) may reflect not only the administrative performance of an agency, but also the difficulty of the case load served and the characteristics of the social environment in which the program operates. To investigate the feasibility of adjusting states' overpayment error rates for differences on such external factors, we developed regression models for five components of the overpayment error rate. When fitted to the case-level data from the fiscal year (FY) 1984 QC sample, these models generally accounted for little of the variation in payment error. In addition, the resulting adjustments were sensitive to specific choices of explanatory variables and to year-to-year differences in the data. Thus it does not appear feasible to adjust states' overpayment error rates on the basis of available external factors. For purposes of exploration we separated the contributions to overpayment error rates into five parts (treating an underpayment as 0): the probability of an overpayment error occurring; the amount of an overpayment error should one occur; the probability of an ineligibility error occurring; the amount of an ineligibility error should one occur; and the size of the participant's food stamp allotment. This was done because most food stamp cases involve no errors, and thus the data have a large concentration of overpayment values at 0. For each component, a regression model related the outcome measure (occurrence of an error, the logarithm of the error amount, or the allotment amount) to case and area socioeconomic characteristics. The explanatory variables used represented six types of external factors: household size, earned income, unearned income, household resources, allowable income deductions, and population density. For the data from FY 1984, only the model for the allotment amount had a high value of [R.sup.2] (.87). When the separate results were combined to yield an adjusted payment error rate for each state, the changes from the state-reported error rate ranged from a reduction of 2.9 percentage points to an increase of 2.5 percentage points, and in 15 states the size of the change is statistically different from 0. Examinalion of alternative sets of explanatory variables in the five models, however, revealed substantial differences in magnitude and direction of the adjustments for a number of states. Fitting the five models to the data from FY 1985 gave further evidence of instability. Thus these statistical procedures for using available data to adjust payment error rates are not sufficiently stable to use in calculating financial penalties. KEY WORDS: Quality control; Regression adjustment; Welfare programs.
1. INTRODUCTION The Food and Nutrition Service (FNS) of the U.S. Department of Agriculture has used the food stamp overpayment error rate as the basis for financial liabilities and incentives [...]