These two cases concern a Freedom of Information Act (FOIA) request for records from the Internal Revenue Service's Taxpayer Compliance Measurement Program (TCMP). In the first case, Susan B. Long and Philip H. Long appeal a district court decision denying release of paper records and ZIP Code data related to TCMP records already released by the IRS. In the second case, the Longs appeal a district court ruling that unreleased TCMP records are exempt from disclosure under Section 6103(b)(2) of the Internal Revenue Code because disclosure would pose a substantial risk of impairing tax collection, tax assessment, and enforcement of the tax laws. The district court reached its decision in the second case after a de novo review of the IRS Commissioner's decision denying disclosure, as mandated by this court in Long v. United States Internal Revenue Service, 742 F.2d 1173, 1182 (9th Cir.1984).

The two cases were consolidated on appeal. We affirm in part, reverse in part, and remand to the district court for further proceedings.

* The Taxpayer Compliance Measurement Program is a series of statistical studies measuring levels of compliance with the federal tax laws. TCMP source data is recorded on paper check sheets and computer data tapes. Each check sheet contains an individual taxpayer's name, address, and social security number, all financial data reported on the taxpayer's tax return, and corrected financial data arrived at by auditing the return. The computer data tapes contain the same information as the check sheets, with the exception of the taxpayer's name and address.

The first of these combined cases concerns the adequacy of TCMP records already disclosed by the IRS. In Long v. United States Internal Revenue Service, 596 F.2d 362, 367 (9th Cir.1979), cert. denied, 446 U.S. 917, 100 S. Ct. 1851, 64 L. Ed. 2d 271 (1980), we held that the FOIA required the IRS to disclose edited TCMP source data unless disclosure would create "a significant risk of indirect identification" of particular taxpayers. We therefore vacated a district court order denying disclosure of TCMP records and remanded the case for further proceedings. In 1981, after the Longs had moved for partial summary judgment on remand, the IRS produced edited computer tapes containing source data for TCMP Phase II, Cycles 1-3.1 In 1983, the IRS agreed to produce an edited tape containing data from TCMP Phase III, Cycles 1, 4, and 5.2 The IRS has not released any of the check sheets corresponding to the computer tapes it has produced.

After a trial to the bench following the remand, the district court ruled that ZIP Code information deleted from the already-released Phase II and Phase III computer tapes was exempt from disclosure under Section (b)(3) of the FOIA, 5 U.S.C. Sec. 552(b)(3)3 and Section 6103 of the Internal Revenue Code, 26 U.S.C. Sec. 6103.4 The court also stated that "defendants have fulfilled their responsibilities under the FOIA [and] the plaintiffs' other claims are denied."

The Longs contend that the district court erred in ruling that the IRS has fulfilled its responsibilities under the FOIA. They argue that the IRS was required to release edited check sheets as well as computer tapes for Phase II of the TCMP, at least where the check sheets are needed to interpret the tapes. The Longs also contend that the court erred in ruling that partial ZIP Code information is exempt from disclosure.

The question whether the IRS has fulfilled its responsibilities under the FOIA is a question of law subject to de novo review. See United States v. McConney, 728 F.2d 1195, 1201 (9th Cir.) (en banc), cert. denied, 469 U.S. 824, 105 S. Ct. 101, 83 L. Ed. 2d 46 (1984). In Long v. United States Internal Revenue Service, 596 F.2d at 367, we held that the IRS was required to disclose all TCMP source data unless disclosure would create a significant risk of identification of particular taxpayers. We specifically noted that the Longs sought disclosure of check sheets "where there is a problem with interpretation of the tapes." 596 F.2d at 366.

After our remand, it was undisputed at trial that there were problems with the interpretation of all the tapes for Phase II of the TCMP. It was also undisputed that the IRS failed to provide the Longs with three computer tapes containing Phase II source data. We therefore hold that the district court erred in ruling that the IRS has fulfilled its responsibilities under the FOIA, and we reverse this part of the district court's decision and remand for further proceedings. The Longs are entitled under the FOIA and our decision in Long v. United States Internal Revenue Service to copies of all existing check sheets for Phase II of the TCMP. These check sheets must, of course, be edited to delete taxpayer identifying information.

The question whether Phase II and Phase III ZIP Code information is exempt from disclosure is a mixed question of fact and law that involves an essentially factual inquiry. We review such questions under the clearly erroneous standard. United States v. McConney, 728 F.2d at 1203.

Under Section (b)(3) of the FOIA, 5 U.S.C. Sec. 552(b)(3), information need not be disclosed if it is "specifically exempted from disclosure by statute." Section 6103(a) of the Internal Revenue Code, 26 U.S.C. Sec. 6103(a), prohibits disclosure of tax returns and tax return information except in certain specified circumstances. The statutory definition of "return information" specifically excludes "data in a form which cannot be associated with, or otherwise identify, directly or indirectly, a particular taxpayer." 26 U.S.C. Sec. 6103(b)(2).

In the present case, the district court found that disclosure of either full or partial ZIP Code information for Phase II returns, in combination with already released data such as type of business, number of employees, and business starting date, would create a significant risk of identification of particular taxpayers. This finding is not clearly erroneous. Even if Phase II ZIP Code data were edited to reveal only the first three digits of a taxpayer's ZIP Code, it is reasonable to conclude that some three-digit ZIP Code areas are so small and some of the non-farm businesses surveyed so distinctive that disclosure would be likely to result in taxpayer identification. We therefore hold that the district court did not clearly err in ruling that ZIP Code data from Phase II of the TCMP are exempt from disclosure under Section (b)(3) of the FOIA and Section 6103 of the Internal Revenue Code. We accordingly affirm this part of the district court's decision.

The district court did not explicitly find that revealing ZIP Code information from Phase III individual returns would pose a significant risk of taxpayer identification. Such a finding is implicit in the district court's decision, however. We affirm the court's ruling that ZIP Code data from Phase III of the TCMP are likewise exempt from disclosure. There is evidence in the record that a person with access to ZIP Code information from TCMP records would have a greater chance of identifying all taxpayers, individuals as well as businesses. We might have arrived at a different conclusion, but we cannot say that the district court clearly erred in ruling that Phase III ZIP Code information is exempt from disclosure. We therefore affirm this part of the district court's decision as well.

The IRS uses data collected through TCMP surveys to create discriminant function (DIF) formulas. DIF formulas are designed to detect combinations of reported tax return line items and taxpayer reporting characteristics that are correlated with a high dollar value of total tax change after audit. A high DIF score for a particular return indicates that an audit of that return is likely to find substantial reporting errors. The IRS uses DIF formulas to select tax returns for audit.

In the second case, the Longs seek unreleased TCMP data from Phase III, Cycles 1-5 and Phase IV, Cycles 1-2.5 The IRS has used this data to develop DIF formulas. In 1981, Congress amended the Economic Recovery Tax Act (ERTA) to provide that no federal law shall be construed to require the disclosure of data used to determine standards for selecting returns for audit "if the Secretary determines that such disclosure will seriously impair assessment, collection, or enforcement under the internal revenue laws." 26 U.S.C. Sec. 6103(b)(2). The present case was remanded to the district court in light of the ERTA amendment,6 and that court ruled that the amendment barred disclosure of TCMP records used in the development of DIF formulas. The court declined to undertake an independent examination of whether disclosure would impair assessment, collection, or enforcement of the tax laws, ruling instead that the amendment applied to the requested records and that the Commissioner of the IRS had in fact determined that disclosure would seriously impair the functioning of the tax laws.

In Long v. United States Internal Revenue Service, 742 F.2d 1173 (9th Cir.1984), we reversed the district court's ruling that gave conclusive effect to the Commissioner's determination of harm. We held that the court had applied the wrong standard of review and that "the Commissioner's determination that disclosure of TCMP data would seriously impair the assessment, collection, or enforcement of the tax laws is subject to de novo review by the district court." 742 F.2d at 1182. We therefore remanded the case to the district court for a de novo review of the record.

After a hearing following the remand, the district court ruled again that the IRS Commissioner was correct in his determination. The court also denied the Longs' motion to appoint a panel of experts to establish an empirical test for the IRS claim of harm. Finally, the court declined to address the question whether the IRS was required to disclose Phase III, Cycle 5 panel data related to TCMP data tapes released in 1983. We affirm all three of the district court's rulings.

The question whether the TCMP records at issue in this case are exempt from FOIA disclosure involves an essentially factual inquiry into whether disclosure would impair the functioning of the tax laws. We review such questions under the clearly erroneous standard. United States v. McConney, 728 F.2d at 1204; see also Van Bourg, Allen, Weinberg & Roger v. NLRB, 751 F.2d 982, 984 (9th Cir.1985) (district court finding that document is exempt from FOIA disclosure is reviewed under clearly erroneous standard).

The Longs suggest three ways in which TCMP records could be edited so that disclosure would not pose a substantial risk of impairing the functioning of the tax system: (1) by releasing only corrected line-by-line information for each return, with no taxpayer-reported information or total tax change information; (2) by deleting both total tax change and most of the reported items on a rotating basis for each tax return; or (3) by releasing only a small subset of the TCMP records in each cycle, so that the remaining data would be insufficient for insight into audit selection formulas. The Longs argue that some form of partial disclosure can be accomplished without substantial risk of impairing the audit selection process, and they contend that the district court clearly erred in ruling to the contrary.

We disagree. First, the district court did not clearly err in finding that partial disclosure of TCMP data "could easily allow educated insight into [DIF] formulae, sufficient to risk serious harm to the enforcement mechanisms." The Longs did present affidavits supporting their position that any one of the three plans for partial disclosure could be accomplished without substantial risk, but the court found that the Longs' materials were speculative and conclusory. The court's account of the evidence "is plausible in light of the record viewed in its entirety," and we therefore cannot say that the court's conclusion was clearly erroneous. See Anderson v. Bessemer City, 470 U.S. 564, 574, 105 S. Ct. 1504, 1512, 84 L. Ed. 2d 518 (1985).

Second, all three of the Longs' partial disclosure plans would involve editing so extensive as to amount to the creation of new records. The law is clear that the FOIA "does not obligate agencies to create or retain documents; it only obligates them to provide access to those which it in fact has created and retained." Kissinger v. Reporters Committee for Freedom of the Press, 445 U.S. 136, 152, 100 S. Ct. 960, 969, 63 L. Ed. 2d 267 (1980). In Long v. United States Internal Revenue Service, 596 F.2d at 365-66, we held that deleting names and addresses did not amount to the production of a new record because the remaining information was "reasonably segregable." See also 5 U.S.C. Sec. 552(b). In the present case, partial disclosure of TCMP data would require the IRS to do much more than simply remove taxpayer names and addresses. The Longs' first proposal would involve deletion of all reported financial data and total tax change information for each return. The second proposal would require deletion of all total tax change information and most reported information on a rotating basis for each return. The third proposal would require deletion of all but a few TCMP records in each cycle. We conclude that the editing required for partial disclosure of TCMP records is so extensive that the remaining information is not reasonably segregable, and that all three of the Longs' proposals would require the IRS to create new records. We therefore affirm the district court's ruling that the Phase III and Phase IV data at issue in this part of the case are exempt from disclosure under Section 6103(b)(2).

The Longs' other contentions are without merit. First, the district court declined to appoint a panel of experts to examine the likelihood of harm from disclosure on the grounds that "the subject matter of the dispute is [not] beyond the capacity of the District Court" and that the court "is well acquainted with even the statistical intricacies provided by the parties." The admission and exclusion of expert testimony is within the broad discretion of the district court, and we will sustain a district court's decision unless it is manifestly erroneous. Taylor v. Burlington N.R.R., 787 F.2d 1309, 1315 (9th Cir.1986). We think it is clear that the there was no manifest error in the court's refusal to appoint an expert panel in the present case. Second, the question whether the IRS was required to release sample codes and weights for Phase III, Cycle 5 panel data was simply beyond our mandate to the district court in Long v. United States Internal Revenue Service, 742 F.2d 1173 (9th Cir.1984). We therefore affirm this part of the district court's ruling as well.

In No. 84-4117, we affirm the district court's ruling that ZIP Code information from Phase II and Phase III of the TCMP is exempt from disclosure. We reverse the district court's ruling that the IRS has fulfilled its responsibilities under the FOIA, and we remand the case to the district court for further proceedings limited to the question of disclosure of existing check sheets for Phase II of the TCMP. In Nos. 86-3678 and 86-3679, we affirm the district court's determination that the unreleased Phase III and Phase IV TCMP data sought by the Longs are exempt from disclosure under Section 6103(b)(2). We also affirm the court's denial of the Longs' motion to appoint an expert panel and the court's refusal to address the question whether the IRS was required to disclose Phase III, Cycle 5 panel data related to TCMP data tapes released in 1983. Each party will bear its own costs.

AFFIRMED IN PART, REVERSED IN PART, AND REMANDED.

CANBY, Circuit Judge, concurring in part and dissenting in part:

I agree with the court's opinion in all but two respects. With regard to appeal No. 84-4117, I would require the IRS to release three-digit Zip Code data relating to individual taxpayer returns. The testimony is undisputed that three-digit Zip Code areas include approximately 250,000 persons. In the face of this fact and other undisputed evidence in the record, there is nothing to support a finding that there would be a serious risk of taxpayer identification if the partial Codes were released.

In appeals No. 86-3678 and 86-3679, I would order the IRS to supply the line-by-line corrected information for each return in the sample. The evidence is conclusive that the release of such information, unaccompanied by taxpayer-reported line-by-line information or by any tax change information, could not possibly lead to the construction or approximation of DIF formulae. With respect to the other requested information, however, I cannot say that the district court clearly erred in finding that disclosure would seriously impair enforcement.

The court's opinion holds that modifying the tapes to reflect only the corrected information would amount to the creation of a new record. I disagree. The corrected line-by-line information is part of existing records of the TCMP. To remove the taxpayer-reported information and tax change information that is recorded with the corrected information would simply amount to redaction or deletion. It can be accomplished by a computer program. We ordered deletions of greater magnitude in Long v. Internal Revenue Service, 596 F.2d 362, 365-67 (9th Cir.1979). I would order them here.

With these two exceptions, I join the court's opinion.