Detectable and Recoverable UI Overpayments in FY 2001

A crude comparison of the overpayments the Benefit Accuracy Measurement (BAM) system estimates with overpayments the states actually establish through their Benefit Payment Control (BPC) activities seems to reflect poorly on BPC. In FY 2001, for example, states reported BPC establishments of about $636 million, or a bit over a quarter of the approximately $2.3 billion of total overpayments BAM estimated. This brief paper quantifies the extent that overpayments are subject to recovery, and reviews BPC detection techniques to determine what overpayments states are most likely to be able to detect, and thus to establish and recover. It argues that the estimate of "detectable and recoverable" overpayments is appropriate basis for judging establishment and recovery activities.

Addressing Overpayments: Prevention vs. Recovery

UI, like every other similar program, addresses overpayments in two ways. It tries to prevent overpayments by making correct eligibility determinations; if they occur, it tries to identify them and recover them if possible.

Prevention. Preventing UI mispayments, particularly overpayments, is difficult for a number of reasons. State UI programs have complex eligibility rules; and there are often numerous players--employers, the claimant, agency staff, often third parties-in certain eligibility decisions. The biggest reason, however, is information. To make correct decisions, agency staff must have all the relevant facts (that doesn't guarantee a correct determination, but it's the essential starting point). Unfortunately, the agency staff who must make the eligibility decision do not automatically have all the facts they need. Claimants--the very parties who stand to gain from providing inaccurate information-often have a near-monopoly on this information. For example, only they know for sure why they quit a job, or whether they are able, available and searching for work. It takes a BQC/BAM investigator about 5-8 hours to complete the investigation of a week's eligibility for a single claimant, and even after that effort certain aspects often remain inconclusive. Agencies have a tiny fraction of that time to administer the average claim. To this can be added the difficulties in receiving information from businesses, and the complexities of UI processes and eligibility conditions. The system is thus biased toward allowing overpayment errors. It is not surprising that BAM estimates that such errors average about 8-9% of benefit payments.

Every BAM case determined to be in error is coded with a "prior agency action" code. It gives the BAM investigator's judgment about the status of the claim at the time of the BAM investigation and what kind of information the agency had at the time it was making the decision to pay. The investigator codes each error with one of the following:

• Existing agency procedures could not have detected the error (key information was missing that standard procedures could have been expected to provide);
• The agency drew the wrong conclusion from information adequate to reach a correct determination (information was sufficient, but mishandled);
• The agency had adequate information to identify that an issue existed, but failed to follow through to determine the claim was potentially in error (the agency failed to follow through on information sufficient to detect the issue);
• The agency failed to follow procedures that would have detected the error (information was insufficient to detect the issue because adequate procedures were not followed); and
• The agency was in the process of reaching a correct decision when BAM did its review (the case was in error when assigned; critical information came in later).

Here is a summary of the prior agency action codings of 3,393 individual UI overpayment errors (some cases involve multiple errors) during FY 2001:

Thus, according to BAM investigators, about 85% of the errors, and 73% of the dollars, occurred because existing agency procedures could not detect the error. For another 3%, the agency didn't have the necessary information to make a correct eligibility determination but could have, had it followed its procedures. Only 8% of the errors, but 17% of dollars overpaid, resulted from the agency mishandling adequate information to make a correct decision or detect an issue. About 3-4% of errors were timing artifacts.

Table 1 shows the errors by cause code and dollar impact, and indicates, using the same Prior Agency Action codes, what percentage were not detectable by existing procedures. The range is from about 36% for the ES Registration and miscellaneous "other causes" category to 98% for Benefit Year Earnings errors. Inability of existing agency procedures to obtain accurate information was at the root of about 90% of the two largest single categories of errors-Benefit Year Earnings and Work Search violations. For all categories, only $618 million out of $2.3 billion in overpayment errors involved cases in which the agencies had, or should have expected to have, the information to prevent them.

Recovering Errors After They've Occurred. Tables 2 and 3 use other BAM codes to provide an estimate of the recoverable portion of overpayments. On the basis of their state's law, BAM investigators determine whether the overpayments they detect involve fraud or not, and also whether they are the type of non-fraud error that the state's BPC operation might be able to recover. Consistent with previous years' findings, fraud overpayments constituted about 2% of benefits in FY 2001. Nonfraud recoverable payments were about 4.8% of benefits paid. Together those major categories were about $1.8 billion out of the total of $2.3 billion. Of this, existing benefit payment procedures could have detected on the front end and thus prevented only about $338 million, or a little less than one fifth. In FY 2001 state BPC units established about one third of the $1.8 billion that BAM deemed to have been recoverable. (Most of the overpayments BPC established in FY 2001 probably occurred 6 to 12 months earlier.)

Non-Recoverable Overpayments: Preventable Only. In FY 2001 BAM estimates that approximately $480 million worth of overpayments were "non-fraud, nonrecoverable" (Table 3). These overpayments are "non recoverable" because State law prohibits collection due to the type of error (e.g., made by agency staff), finality provisions, or features such as formal warnings. If the agency cannot prevent them through front-end procedures that detect the error issue, the money is gone forever. Fortunately, because current claims administration processes can detect a higher percentage of non-recoverable than recoverable overpayments-58% vs. 19%--these are at least in theory more preventable than the recoverable payments identified above.

What Overpayments Can BPC Detect?

BAM calculations using recoverability codes indicate that in FY 2001, about $1.8 billion of the total $2.3 billion of estimated overpayments were, in concept, recoverable. However, BPC staff has the same problem as their front-line colleagues who make the initial eligibility decisions: they must detect overpayments to recover them. Using the extensive explanatory information in the Handbook 401 description and guidance about the ETA 227 report, this section reviews current BPC procedures to identify the types (based on BAM cause codes) of overpayments which are most and least likely for BPC to detect, and thus to establish for recovery.

An examination of the following table, constructed by combining the Handbook 401 information about BPC detection methods and 227 data for CY 2000 (see Table 4, below) on cases and dollars established, shows the following:

• Most pro-active BPC detection methods (called "controllable") are designed primarily, if not exclusively, to detect overpayments due to Benefit Year Earnings (BYE) violations: crossmatches, border checks, low earnings verifications, accession/new hire reports, industry studies. BYE violations involve claimants who continue to claim benefits despite having excessive earnings for the claimed week, usually because they have returned to work. ETA 227 data for CY 2000 suggest that BYE violations were at least 40%, and perhaps as high as 49%, of overpaid dollars established for recovery.
• Controllable techniques also focus on detecting separation issues, dependents' allowances, duplicate claims filed in a contiguous state, illegal aliens and payments due to use of invalid Social Security numbers. Together, these comprised no more than 11% of total establishments in 2000, and may have involved considerably less because some of the methods also pick up BYE violations. There is no way to tell how intensively states use some of the methods; some may no longer be pertinent (e.g., those designed for use in wage-request states.)
• Together, these categories, detected through controllable means, accounted for slightly over half of overpayments established.
• In CY2000 states established fewer than 50% of overpayments through passive or "noncontrollable" methods.
  • Employer protests, which detect an unknown mix of BYE and separation violations, are the only identifiable means with a known type of overpayment. Those represented about 9% of overpayments in 2000.
  • Tips and leads (2.4%) could detect any kind of violation, but BYE and A&A are highly probable.
  • 40% of reported establishments were detected through the catch-all category called "other noncontrollable."
    • Local office complaints probably involve largely A&A and separation issues.
    • According to the Handbook 401 description, "regular claims control" activities, involve largely the review of agency records. If agency staff detect what BAM investigators detected through agency records in FY 2001, this category involves the following breakdown of types of overpayments:
      • BYE: 25%
      • Separation Issues: 24%
      • Weekly Eligibility (mostly A&A and "other"): 26%
      • All other categories: 25%
    • The 227 shows that $75 million in nonfraud overpayments in 2000--12% of all establishments--resulted from "Java reversals." These are payments initially made to meet time lapse requirements that subsequent information showed should have been denials, usually because of separation issues.

To summarize, the following table shows, in approximate order of importance, the major BAM cause categories of overpayments that BPC operations are most likely to be able to detect and thus to establish, and notes their detectability:

The Base for Measuring BPC Establishments. The total of Fraud and NonFraud-Recoverable overpayments for the six categories above probably represent the most reasonable base against which to compare actual BPC recoveries. For FY 2001, these totaled as follows:

Benefit year earnings      $ 615.9 million
Separation Issues            380.7
A&A issues                   140.2
Social Security Violations    32.4
Illegal Aliens                28.2
Dependency violations         16.6
Severance/Vacation Pay        20.9
------------------------------------------
Total                     $1,234.9 million

Actual amounts established for recovery in FY 2001, $636 million, were a little over half of the total for these key categories.

Conclusions

This analysis supplements others we have produced recently, based on BAM estimates, which quantify what fraction of their overpayments states can now detect with current procedures.

The following chart depicts the results:

Estimated Dollars Overpaid by Detectability and Recoverability, FY 2001

• The portion detectable by the agency is quite small: BAM estimates that in FY 2001 states’ normal procedures could have detected only about $618 million of the $2.3 billion in overpayments, or about 27%. Thus, $1.7 billion was overpaid because states’ procedures could not provide the information that would detect the error.

  • The states were in the process of correcting about one ninth of detectable dollars overpaid (about 3% of all dollars in error) when BAM investigated the case.

  • Nearly 90% of the $618 million in detectable errors occurred because state staff drew the wrong conclusion from adequate information, failed to follow through to detect an issue, or failed to follow procedures which would have detected the error.

    • This segment of errors—detectable and not in the process of being corrected--is still a sizeable amount. In FY 2001 it was approximately $550 million. This is perhaps the first place states should begin to focus prevention activities.

• We have divided overpayments into three categories that relate to the states’ ability to recover them once they have occurred: Fraud, Non-fraud recoverable, and Nonrecoverable.

• • Non-recoverable. BAM estimates that due to finality, formal warnings, and other features of state law, some $480 million in overpayments could not be recovered through BPC or other systems. A higher than average proportion of these--$280 million or 58%--was detectable, however, so the potential for preventive action is higher. If not prevented, however, this money is gone.

  • Recoverable. This is the sum of fraud and non-fraud recoverable. In FY 2001, BAM estimated that it was about $1.7 billion. Of this, it also estimates that states could have detected about $337 million, or 19%.

• Recoverable overpayments can only be recovered if they are detectable. An examination of current active BPC techniques (termed “controllable”) and experience with “uncontrollable” methods, indicate that states focus most efforts on detecting overpayments due to Benefit Year Earnings and Separation violations. Other prime categories of overpayments most often detected and established include A&A issues, Social Security and Illegal Alien Dependency benefit and vacation pay/severance pay violations.

  • These categories sum to about $1.2 billion or over two thirds of the $1.7 recoverable.

  • BYE and Separation issues account for 80%, or just about $1 billion, of the $1.2 billion “most detectable” overpayments.

  • States reported about $636 million in BPC establishments in FY 2001, or about 51% of the “most detectable” overpayments base.

    • The base is most properly measured for a previous period because of the normal lag between occurrence (when BAM measures overpayments) and detection and establishment. A lag of about six months typically occurs between the time an overpayment occurs and when it is detectable through the wage-record crossmatch, for example. Payment amounts, and overpayment amounts, have been rising since 2000; thus a base measured earlier would have been slightly smaller, and suggests that the “true” BPC establishment ratio should be somewhat higher than 51%.