Department of Veterans Affairs, Petitioner, v. Federal Labor Relations Authority, Respondent, 33 F.3d 1391 (D.C. Cir. 1994)

The Portland, Oregon office of the Veterans Administration implemented new performance standards that it proposed to enforce by sampling a portion of its employees' work product. In response, Local 1974 of the National Federation of Federal Employees submitted two bargaining proposals that contemplated the use of a particular statistical methodology to ensure the accuracy of the sampling process. We agree with the Federal Labor Relations Authority that these proposals were procedural in nature and hence negotiable under the statute governing relations between a federal agency and a union representing its employees.

The Federal Service Labor Management Relations Act ("Act") requires federal agencies to bargain in good faith over the conditions of employment. 5 U.S.C. §§ 7114(a) (4), 7103(a) (12), (14) (1988). Certain management prerogatives are exempted from this requirement. Among these are the right to assign employees and direct their work:

[N]othing in [the Act] shall affect the authority of any management official of any agency ...

(A) to hire, assign, direct, layoff, and retain employees in the agency, or to suspend, remove, reduce in grade or pay, or take other disciplinary action against such employees;

(B) to assign work, to make determinations with respect to contracting out, and to determine the personnel by which agency operations shall be conducted ...

Id. Sec. 7106(a) (2). The Act, however, also provides that

[n]othing in this section shall preclude any agency and any labor organization from negotiating ... procedures which management officials of the agency will observe in exercising any authority under this section ...

Id. Sec. 7106(b) (2) (emphasis added). Accordingly, even where a management right is implicated, the Act requires an agency to negotiate the procedures employed in its exercise.

In 1992, the Veterans Administration's Portland Regional Office ("VA") announced plans to change the performance standards for certain of its employees. Those employed as "Rating Board Members (Non-Medical)" were to be evaluated under a "percentage error" standard. In order for such employees to be rated "fully successful," they must produce work that is 92 percent free of substantive errors; put another way, their rate of error may not exceed 8 percent. Because an employee in this position might handle upwards of 2,000 cases per year, the VA elected to base its evaluation of each of these employee's performance on a sample of 75 cases; and because 8 percent of 75 is six, the VA decided that it would only issue "fully successful" ratings to an employee who committed no more than six errors in the 75 cases sampled.

For other categories of employees, the agency adopted an "allowable error" standard. Instead of measuring the percentage of errors made by employees, this standard evaluates their performance on the basis of the number of errors found in a sample of work. For example, the performance standard for the Correspondence Clerk position permits an employee to commit no more than 24 errors in the sample examined, regardless of how many cases the employee has handled. If the number of errors in the original sample exceeds the number allowed, the standard permits the sample to be expanded.

In response to these new performance standards, Local 1974 ("Union") submitted several proposals for collective bargaining, two of which are relevant here. Both are concerned with the methodology used to establish conformance with the standards. The first, Proposal 3, is directed to the percentage error standard and provides:

Where quality is expressed in the standard as a percentage error rate or percentage error free rate, valid statistical methods shall be used to determine if the true error rate is within the stated standard. A three standard deviation table shall be employed.

Appendix ("App.") at 5. The second, Proposal 4, concerns the allowable error standard:

Where quality is expressed in terms of a number of allowable errors, the sample size shall be the same for each employee performing like duties. If an expanded sample is reviewed for an employee, valid statistical methods shall be employed to determine if the true error rate is within the error rate expressed in the standard for the sample size in the normal review. A three standard deviation table shall be employed.

Id.

At issue here is the Union's proposal that a "three standard deviation" ("3SD") table be used in applying these standards to percentage error employees and, in the case of allowable error employees, to those being assessed on the basis of expanded samples. A 3SD table is a statistical device that permits the "use [of] a sample to predict a population characteristic ... with a 99.7% certainty that the sample observation was not due to random chance." Affidavit of Joseph C. Williams II, Director of the Department of Veterans Affairs Regional Office, Portland, Oregon, dated June 25, 1992, App. at 59 (discussing 3SD table displayed in Appendix E to Veterans Administration's Benefits Manual M20-2).

The VA refused to bargain over these proposals, contending that they infringed upon its right to assign and direct its employees. The FLRA disagreed. It concluded that Proposals 3 and 4 were negotiable procedures within the meaning of section 7106(b) (2) and directed the VA to negotiate over them. National Fed'n of Fed. Employees, Local 1974 and U.S. Dep't of Veterans Affairs, Regional Office, Portland, Oregon, 46 F.L.R.A. 1170 (1993) ("Local 1974"). The VA filed a timely petition for review of that decision.II. DISCUSSION

Our role in reviewing a decision of the FLRA is well defined:

[The Act] provides that the Authority's orders are to be reviewed in accordance with the standards established in the Administrative Procedure Act. Thus, we must reverse the Authority's negotiability decision if it is "arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law." Moreover, because "the FLRA was intended to develop specialized expertise in its field of labor relations and to use that expertise to give content to the principles and goals set forth in the [Civil Service Reform] Act," the Authority's "interpretation of its Statute, if reasonable, is entitled to deference." Still, the Supreme Court has made clear that courts "must not rubber-stamp ... [FLRA] decisions that they deem inconsistent with a statutory mandate."

NLRB v. FLRA, 2 F.3d 1190, 1197 (D.C. Cir. 1993) (citations omitted). We also accord particular deference to the FLRA when the question involves, as it does here, "the distinction between proposals encroaching on management's nonnegotiable substantive authority and those concerning properly negotiable procedural matters...." National Treasury Employees Union v. FLRA, 691 F.2d 553, 561 (D.C. Cir. 1982).

Determining whether the FLRA correctly found Proposals 3 and 4 negotiable requires a rudimentary understanding of the statistical principles at work here. Any sampling involves a margin of error. Suppose that a jar contains 100 marbles, of which ten are red. If an individual randomly selects 50 marbles from the jar, he may have anywhere from ten red marbles to none. If he selects all ten of them, the red marbles will comprise 20 percent of the sample; but he would obviously be wrong to infer that 20 percent of all the marbles in the jar were red.

To cope with this probability of error, statisticians have developed the concept of "standard deviation" ("SD") to establish a range of numbers around the mean on which one may rely with a given degree of confidence--the wider the range, the greater the confidence. A 3SD table will reflect three times the range of numbers as a one standard deviation table, but it will yield a confidence level of 99.7 percent as opposed to 68.3 percent for the one standard deviation table. See Morris Hamburg, Basic Statistics 68 (2d ed. 1979).

Under the "8.0" percent column in the VA's 3SD table, App. 60, we find a range of 17.40 to 0.00 percent where 75 is the size of the sample. According to this table, a random sample of 75 cases yielding six errors--an error rate of 8 percent--indicates that the total population has a true error rate that might lie anywhere from slightly above zero to 17.40 percent. Or, to return to the marble analogy, in a random sample of 75 marbles from a jar, by selecting six red marbles, one can be 99.7 percent certain that no more than 17.40 percent of all marbles in the jar are red. The VA complains that if this table is used in applying its percentage error standard, it would have to grant a "fully successful" performance rating to employees committing up to 13 errors (75 cases X 17.40%) instead of limiting the award to those who have committed no more than six (75 cases X 8%). Therefore, while use of the 3SD table might ensure a 99.7 percent degree of confidence that all percentage-error-rated employees whose total work is 92 percent error free will receive their "fully successful" performance ratings, it will also ensure that a significant number with higher percentages of error will be so rated.

The VA concludes from this that Proposal 3 would effect a change in its percentage error performance standard. This argument misconstrues the nature of the proposal. To be sure, "the content of performance standards is nonnegotiable." Overseas Educ. Ass'n, Inc. v. FLRA, 872 F.2d 1032, 1034 (D.C. Cir. 1988) (proposal is nonnegotiable in that it "intrudes on management's exclusive right to determine the substance of performance standards, and is outside the agency's duty to bargain.") As the FLRA noted, however, Proposal 3 "concerns only the evaluative methodology used to measure error rate." Local 1974, 46 F.L.R.A. at 1177. It does not interfere with the VA's "discretion to establish its performance standards and to determine its evaluative methodology." Id.

We are faced with a problem that is inherent in any resort to sampling in order to ascertain the performance of a particular individual. Because of the inevitable margins of error, any methodology chosen will tend to be either over- or underinclusive. The VA's denial of a favorable rating to any percentage performance employee who has committed seven or more substantive errors in a sample of 75 cases will no doubt succeed in minimizing the number of unqualified employees who are given the favorable rating. On the other hand, the agency will withhold its "fully successful" rating if it discovers errors in seven out of a sample of 75 of the 2,000 cases an employee may have handled during the year even if those seven were the only errors that employee had committed. Thus the VA's approach will reduce the risk of giving a favorable rating to unqualified workers at the cost of excluding others who are fully qualified. By the same token, while the Union's proposal may ensure that virtually all employees who are entitled to the "fully satisfactory" rating will receive it, it will also allow a substantial number of the unqualified to slip through the door.

The fact remains, however, that Proposal 3 would affect neither the VA's choice of a 92 percent error-free standard nor its decision to rely on a 75-case sample in determining which employees will be awarded its "fully successful" rating. Compare National Fed'n of Fed. Employees, Local 1454 and Veterans Admin., 26 F.L.R.A. 848, 852 (1987) (proposals concerning use of standard deviation table were nonnegotiable because they "prescrib [ed] the particular method by which the Agency must sample employees' work"). Rather, the proposal concerns the methodology to be applied in determining whether an employee has met the VA's standard. Such a requirement is manifestly procedural and therefore negotiable under section 7106(b) (2). See American Fed'n of Gov't Employees, Local 3172 and Dep't of Health and Human Services, 35 F.L.R.A. 1276, 1284 (1990) (proposals requiring the use of "commonly accepted" statistical techniques are negotiable as they neither "interfere with management's choice of evaluation methodology nor do they preclude the Agency from taking personnel actions based on application of the selected methodology").

For the same reasons, we conclude that Proposal 4 is procedural and hence negotiable. The nature of the proposal is not affected by the fact that it would apply the 3SD table to ascertain compliance with an allowable error standard when an expanded rather than a standard sample is employed.

The VA concludes with two arguments, neither of them persuasive. First, it contends that the FLRA's decision represented an unreasoned departure from its precedent. This argument is meritless in light of American Fed'n of Gov't Employees, Local 3172, where the FLRA approved as negotiable a proposal requiring the compilation of statistical data "using commonly accepted statistical techniques." 35 F.L.R.A. at 1283. Second, the VA urges that the FLRA's ruling on Proposals 3 and 4 is inconsistent with its determination that Proposal 5 was nonnegotiable. Proposal 5 required that " [o]nly records which can be verified as accurate [may] be used to evaluate performance." Local 1974, 46 F.L.R.A. at 1179. The FLRA found it nonnegotiable because it "precludes the Agency from using certain information in evaluating employee performance." Id. at 1180. The VA finds Proposals 3 and 4 virtually indistinguishable from Proposal 5 because, as it sees it, they would deny it the use of samples of its employees' work unless "statistically validated." The simple answer is that Proposals 3 and 4 deal with methodology; they do not eliminate any information from the agency's consideration.

For the foregoing reasons, the VA's petition for review is

Denied.