City of New York v. United States Department of Commerce

822 F. Supp. 906, 1993 U.S. Dist. LEXIS 4818

Opinion

MEMORANDUM AND ORDER

McLAUGHLIN, Circuit Judge

Footnotes

•

*.

Plaintiffs — states, cities, citizens’ groups, and individual citizens and taxpayers — seek a judgment: (1) vacating former Secretary of Commerce Robert Mosbacher’s July 15,1991 decision that the 1990 census would not be statistically adjusted; (2) ordering that such an adjustment be made; and (3) allowing plaintiffs to use and publicize certain data generated by the Census Bureau, and already produced, subject to a protective order, to the plaintiffs during this litigation. For the reasons set forth below, the Court holds that the decision against adjustment shall not be disturbed, but grants the plaintiffs’ request to use and publish the Census Bureau data. The following constitute the Court’s findings of fact and conclusions of law in accordance with Federal Rule of Civil Procedure 52.

FACTS

Just to recount the facts of this case is arduous, given its four-year history, the number of parties involved, and the complicated statistical evidence lying at the core of the dispute. Many of the material facts have been set forth in two prior published opinions — City of New York v. United States Dep’t of Commerce, 713 F.Supp. 48 (E.D.N.Y.1989) (“City of New York I”), and City of New York v. United States Dep’t of Commerce, 739 F.Supp. 761 (E.D.N.Y.1990) (“City of New York II ”) — some familiarity with which is assumed.

Census Background

The Constitution requires a decennial census. Article I, Section 2, Clause 3 states that “[t]he actual enumeration shall be made [every ten years], in such manner as [the Congress] shall by Law direct.” Congress has, in turn, delegated to the Secretary of Commerce the duty of taking the census “in such form and content as he may determine, including the use of sampling procedures and special surveys.” 13 U.S.C. § 141(a) (1982). The Bureau of the Census, an agency within the Department of Commerce, actually conducts the census. See 13 U.S.C. § 2 (1982).

The results of the census are used for a galaxy of purposes. The federal government uses them to calculate how to dispense program funds among the states. States use the counts for political redistrieting. Sociologists and historians study them for more esoteric purposes. None of this obscures the central truth that the “basic constitutional purpose” of the census is “to determine the apportionment of Representatives among the States.” Carey v. Klutznick, 653 F.2d 732, 736 (2d Cir.1981).

The first census of the American population was in 1790. Thomas Jefferson, who was in charge of it, complained of an under-count. There have been 20 subsequent censuses. Each of them has also resulted in an undercount. More troubling than the under-count itself, however, is that racial and ethnic minorities are undercounted to a greater degree than the population as a whole. This problem, known antiseptically as the “differential undercount,” has skewed every census since the Bureau started measuring it in 1940.

Because the counts are used to calculate the political representation and financial aid to be afforded to a given area, the fear that the census may be perpetuating a system in which those most in need of representation and aid are deprived of both is a major concern. With that in mind, the Census Bureau began, after the 1980 census, to develop a method by which both the under-count of the entire population and the differential undercount could be reduced through a statistical adjustment employing a “post-enumeration survey” (“PES”). This method (and the Department of Commerce’s reaction to it) lie at the heart of this case.

The 1990 Census

Taking the census has always been a daunting task, and the 1990 count was no exception. The Bureau began preparing in 1983, seeking to improve the techniques that it had used in prior censuses. Among other things, it consulted with state and local governments, planned an extensive advertising campaign, designed a more ethnically inclusive census questionnaire, and increased the amount of automation used, including the use of an automated geographic control system, which assured accurate and timely maps and geographic files for the 1990 census. While the parties may disagree on the quality of the census counts achieved in 1990, the four-step procedure used to conduct the census is largely undisputed.

First: an address list of housing units was compiled. This list was crucial because it indicated every household in the nation to which the Bureau would send questionnaires. Since the Bureau relies on the mail return of those questionnaires to count a majority of the population, an accurate and comprehensive list was vitally important. In constructing the list, the Bureau relied primarily on commercial mailing lists, supplemented by extensive field research and collaboration with the Postal Service. Then, numerous quality controls were instituted to improve the accuracy of the list.

Second: census questionnaires were mailed to each housing unit. Householders were asked to complete and return the questionnaires to the local census district office on or before April 1,1990.¹ This is called the “mail out/mail back” phase. The effort to get individuals to participate in the mail out/ mail back phase was extensive. In addition to the Census Bureau’s general advertising campaign, it also conducted campaigns specifically targeted at African-Americans, Asians, Hispanics, and Native Americans. In addition, the Bureau published specialized, foreign-language brochures encouraging public participation in the census. It also maintained a set of toll-free numbers (in eight languages) for anyone who had questions regarding the census questionnaire, and every census form advised Spanish speakers that they could call a toll-free 800 number to obtain a census form in Spanish. Finally, the Census Bureau employed different methods in areas where it was believed that the normal procedure would be particularly ineffective. See Secretary of the Department of Commerce, Decision on Whether or Not a Statistical Adjustment of the 1990 Decennial Census of Population Should be Made for Coverage Deficiencies Resulting in an Over-count or Undercount of the Population, July 15, 1991 (the "Decision”), at 4-5-4-6.

Third: because the return rate of census questionnaires is obviously never 100%, and in 1990 was only 63%, see Transcript of Trial (“Tr.”) at 1823, the Census Bureau embarked on an extensive follow-up campaign. Second mailings were sent to households that failed to return the initial form, and in census districts with particularly low return rates, the Bureau remailed census forms to all residents. Tr. at 1730-31.

Fourth: when steps 1-3 did not produce a census return from a particular household, the Census Bureau engaged in “non-response follow-up,” the final stage of the enumeration. During this phase, each non-responding housing unit was assigned to a “census enumerator,” an employee who was directed to make up to six attempts to contact a household member to obtain the information necessary to complete a census form. If this also proved fruitless, the enumerator was then required to try to obtain basic information on the missing housing unit from a reliable source, such as a neighbor or build ing manager. Decision at 4-7. Once 95% of a district’s operations were completed, a final phase of non-response follow-up required enumerators to make one last-ditch attempt to visit each remaining unresolved household to obtain as complete an interview as possible.

After the enumeration was completed, post-enumeration “Coverage Improvement Programs”² were implemented, with the result that 5.4 million people were added to the counts. Decision at 4-7. The result of all of these efforts was that 249,632,692 people were counted during the 1990 census. Decision at 4-2.

The Differential Undercount

Despite the herculean efforts of the Census Bureau, it is undisputed that the 1990 Census was not — and could not realistically be — successful in its goal of achieving an exact count of the nation’s population. Given the nature of the task, it is not surprising that the census fails to count some individuals (“omissions”) and also adds persons into the count erroneously (“erroneous enumerations”). Tr. at 80-82.

The “net undercount” is the difference between omissions and erroneous enumerations. It is undisputed that the 1990 census, like all previous censuses, resulted in a net national undercount. Decision at 1-1. It is similarly uncontroverted that African-Americans and other minorities have been persistently undercounted to a greater degree than non-Hispanic whites in all censuses since 1940 when the Bureau began measuring such differences, and that this anomaly is perpetuated in the 1990 census. The difference between the undercount rate for non-Hispanic whites and that for minority populations is known as the “differential under-count.” Tr. at 91-92. According to the Secretary, “Blacks appear to have been under-counted in the 1990 census by 4.8%, Hispanics by 5.2%, Asian-Pacific Islanders by 3.1%, and American Indians by 5.0%, while non-Blacks appear to have been undercounted by 1.7%.” Decision at 1-1.

The Possibility of Statistical Adjustment

The Census Bureau has been aware of the existence of a differential undercount since the 1950’s. The intractable problem has been how to fix it. Following the 1980 census, concerns over the persistence of the differential undercount, its deleterious effects on the accuracy of census counts, and the unfair results arising from such inaccuracy, prompted the Bureau to start a research program aimed at developing statistical techniques to ameliorate the problem in the 1990 census.³ Tr. at 525, 1291-92. By 1984, the Bureau had developed a timetable for internal Bureau research that would ultimately lead to a decision whether to adjust the 1990 census statistically in an effort to reduce the differential undercount. Two task forces were created to consider the undercount problem as it related to the upcoming 1990 census; The Undercount Steering Committee (“USC”) was responsible for planning undercount research and policy development. The Undercount Research Staff (“URS”) conducted the actual research. Other divisions at the Bureau also conducted research on the undereount and the possibility of adjustment. Tr. at 517-25, 1292-93. In addition, the Bureau sought the opinions of outside experts and organizations, such as the American Statistical Association and the Na tional Academy of Science, regarding the possibilities for adjustment.

After considering the alternatives, the Bureau settled upon the PES as the best tool to statistically adjust the census through the use of “dual system estimation” (“DSE”). Tr. at 559-61. Dual system estimation or, in more pedestrian terms, “capture/recapture,” is, as relevant here, an approach that uses a second measurement to ascertain the quality of the estimate obtained by an initial measurement, and then uses that information to provide a purportedly more accurate, dual system estimate.⁴ Here, the original enumeration, the census, was followed by a second measurement, the PES, which attempted to measure the rate at which people were omitted and erroneously enumerated by the census, in order to determine a net under-count rate.

While the Bureau has used post-enumeration surveys in a variety of ways since 1950, it has never statistically adjusted based on DSE. The Bureau worked throughout the 1980’s to design the PES to make it an effective tool for census adjustment. Tr. at 572. For example, correlation bias, which may occur when residents become confused by an overlap between the census and the PES, was addressed by distinctly separating the two procedures. Tr. at 578-82. Another species of correlation bias, which arises when individuals who have different probabilities of being counted (“capture probabilities”) in the census are grouped together in the PES, was reduced by the use of “poststratification.” Tr. at 205-208.⁵ In addition, statistical “smoothing” was chosen to address anomalous results in the .PES.⁶

By the Spring of 1987, after much testing and fine-tuning, the Census Director, John Keane, had decided that the Bureau should proceed with plans to adjust the 1990 census data through the use of DSE, if the PES results met a certain quality standard. Dr. Keane met with his superior, Robert Ortner, the Under Secretary of the Department of Commerce, to tell him that such a decision had been made and that a press conference to that effect was imminent. Six days later, Keane met again with Ortner and other Commerce Department officials, who informed Keane that they had decided against adjustment. Shortly thereafter, Commerce Department officials instructed their Census Bureau officials not to disclose that a decision had been made. Tr. 629-30, 1330. On October 30, 1987 the Department of Commerce announced its decision against adjustment, and this lawsuit was born.

History of This Litigation

In November, 1988, plaintiffs sued to enjoin the 1990 census, challenging the methodology by which it would be taken, and seeking to reverse the decision against adjustment. Defendants — the Department of Commerce, its Secretary, President Bush, and other officials within the Department of Commerce and its subsidiary, the Bureau of the Census — moved to dismiss the application for the injunction. This Court denied the dismissal motion, holding that the plaintiffs had standing to challenge the census on constitutional grounds;⁷ the Court also ruled that it would consider the Commerce Department’s decision against adjustment under the “arbitrary and capricious” standard of review of the Administrative Procedure Act, 5 U.S.C. § 706(2)(A) (1982) (the “APA”). City of New York I, 713 F.Supp. at 54.

When the dismissal motion was denied, a hearing was scheduled on the injunction. It was set to go forward in the Summer of 1989, when, at the eleventh hour, the parties entered into a stipulation (the “Stipulation” or the “Stip.”). The Stipulation vacated the Commerce Department’s 1987 decision against adjustment and agreed that the new Commerce Secretary, Robert Mosbacher, would consider de novo and “with an open mind,” whether adjustment was warranted. Stip. at 2-3. The Stipulation also agreed that the program to gather the statistical data necessary for adjustment would proceed, that the Secretary would decide whether to adjust by July 15, 1991, and that his decision would be consistent with certain procedures, including the promulgation of “guidelines” articulating what the defendants believed to be the relevant technical and policy, considerations affecting the decision. It also mandated the creation of an eight-member Special Advisory Panel (the “Panel”) ⁸ of statistical and demographic experts to advise the Secretary on whether to adjust. Stip. at 4-5.

The defendants adopted and promulgated the required guidelines, but the plaintiffs challenged them as inadequate, and they also sought a declaratory judgment that a statistical adjustment would not violate the Constitution or any federal statute. Defendants countered that the plaintiffs’ challenge to the census presented a non-justiciable political question. This Court rejected the defendants’ political question claim, and concluded that- statistical adjustment,- per se, would not violate either the Constitution or the laws of the United States. City of New York II, 739 F.Supp. at 767-68. This Court noted that, while the guidelines were vague, they did satisfy, albeit just barely, the defendants’ obligations under the Stipulation. Id. at 770.

The Post-Enumeration Survey

Following this Court’s decision in City of New York I, the Bureau resumed work on its plans to implement the PES, and implemented it in 1990. In the first step of the PES, the Bureau methodically selected approximately 5000 blocks⁹ in an effort to attain an appropriate sample size for each poststratum. In February 1990, Bureau employees visited each sample block and listed all the housing units they found, identifying approximately 170,000 households. In July 1990, Census Bureau interviewers returned to each address to obtain information regarding the residency status of those households on Census Day. The Bureau found that those blocks contained approximately 400,000 people.. Tr. at 208. After collecting the PES data, the Bureau matched it to the information collected in the original enumeration for those same sample blocks. From this matching, the Bureau endeavored to estimate, for each poststratum, rates of omission and erroneous enumeration, and from these calculated a net undercount rate for each poststratum. Tr. at 221.

The Bureau used these results to develop an “adjustment factor” for each poststratum, i.e., the number by which the population count as indicated by the census had to be multiplied so that the entire census would reflect the variations found in the PES. Ac cordingly, the 1,392 poststrata resulted in 1,392 corresponding adjustment factors. One further statistical twist to the use of the PES was the employment of “smoothing.”¹⁰ After smoothing, the Bureau used the smoothed adjustment factors to produce adjusted counts down to the block level, which were then aggregated to provide population estimates for cities, counties, states, and the nation. Tr. at 224-25; Decision at 4-18.

A number of quality control checks were made to test the results of the PES. First, the Bureau conducted or commissioned more than twenty formal research projects, called “P-Studies,” to study the potential sources of error within the PES. The results of these P-Studies regarding particular sources of error were then combined in the “total error model” that summarized the overall quality of the PES data. Tr. at 652-59.¹¹

The final result of the PES was that the census enumeration was estimated to have undercounted the population by 5,269,917, or 2.07%. In terms of the differential under-count, the PES indicated that the census undercounted Hispanics by 5.2%, African-Americans by 4.8% and Asian/Pacific Islanders by 3.1%. The PES-calculated under-count for non-African-Americans was 1.7% and 1.2% for non-Hispanic whites, with a total national undercount of 2.1%.¹²

The Bureau also conducted a number of “loss function analyses” to compare the quality of enumeration counts to the adjusted counts. A loss function analysis is a systematic way of assessing the consequences flowing from a particular decision. In the context of the adjustment decision, the Bureau used loss function analysis to determine whether the adjusted data were expected to be more accurate than the unadjusted data. Tr. at 1941-42. This Court is satisfied that for most purposes the PES resulted in a more accurate — or to be statistically fashionable, a less inaccurate — count than the original census.

The Secretary’s Decision and The Trial

Prior to reaching his decision, Secretary Mosbacher received the recommendations of the eight Panel members. Perhaps not surprisingly, the Panel was deadlocked: the four members selected from the plaintiffs’ list recommended in favor of adjustment, while the four members chosen unilaterally by the Secretary recommended against it. Decision at 1-3. The USC voted 7-2 in favor of adjustment. Id. The Under Secretary of Commerce for Economic Affairs and the Administrator of the Economics and Statistics Ad ministration voted against adjustment. Defendants’ Exhibit 1 at 898. Finally, the Director of the Census, Dr. Barbara Bryant recommended in favor of adjustment, but acknowledged that “[tjhere is no perfect truth as to the size and distribution of the population,” and that “[ajdjustment is an issue about which reasonable men and women and the best statisticians and demographers can disagree. The minority viewpoint expressed in the Census Bureau’s report ... illustrates this.” Id. at 1118-19.¹³

On July 15, 1991, in accordance with the Stipulation, Secretary Mosbacher went on national television to announce his decision not to adjust. Simultaneously, he produced the Decision, a 178-page report giving the reasons for his decision. The decision revitalized the case and discovery resumed. Claiming that the Secretary’s decision violated the Constitution, the APA, and the Stipulation, the plaintiffs requested a trial. More specifically, they alleged that the administrative record proffered by the Secretary as the basis of his decision is a self-serving, post-hoc compilation of documents assembled for the purpose of strengthening the defendants’ litigation position and that the Secretary’s decision was tainted by partisan political influence. Over the defendants’ objection, this Court ordered a trial, which consisted almost exclusively of expert testimony in the fields of demographics and statistics, and continued for thirteen trial days.¹⁴

The expert witnesses expressed their opinions as to whether the Secretary considered all the factors specified in the guidelines in making his decision, and also analyzed at length the conclusions that the Secretary reached in the Decision. Plaintiffs’ direct case consisted of the testimony of nine witnesses, including all four of the plaintiffs’ designees to the Panel. It also included the introduction of hundreds of exhibits and numerous deposition transcripts from other witnesses.

Defendants’ evidence was similarly grand in scope. They presented five expert witnesses, including one Panel member. They also introduced the deposition transcripts of other witnesses and numerous exhibits. Of these, Exhibit 1, denominated as the Administrative Record by the defendants, and skeptically dubbed “the so-called Administrative Record” by the plaintiffs, contains over 12,-000 documents and occupies 18,000 pages. The trial transcript exceeds 2,600 pages.

DISCUSSION

Plaintiffs allege that the Secretary’s decision not to adjust the census count violates the APA, the Constitution, and the Stipulation.¹⁵ They also argue that the process the Secretary used to make his decision was a sham.¹⁶ They seek an order directing the Secretary of Commerce to make the adjustment and they ask for permission to use Census Bureau data provided to them by the defendants during the course of this litigation under a protective order, and to release that data to the public.

I. The APA Standard of Review — Finality

The standard by which the Court reviews the Secretary’s decision not to adjust should be stated at the threshold. At a previous stage in this litigation, this Court announced that “the arbitrary and capricious standard as set forth in § 706 of the APA will guide my review of the Secretary’s determination.” City of New York I, 713 F.Supp. at 54.

Defendants now contend that the plaintiffs’ claim under the APA and, with it, this Court’s decision to review the Secretary’s decision under the arbitrary and capricious standard, have been vitiated by the Supreme Court’s recent decision in Franklin v. Massachusetts, -U.S. -, 112 S.Ct. 2767, 120 L.Ed.2d 636 (1992). There, Massachusetts challenged the Census Bureau’s method for counting federal employees serving overseas, alleging that it was arbitrary and capricious, and, as such, a violation of the APA. Massachusetts also asserted that the method violated the constitutional requirements for conducting a decennial census and damaged it because it changed the congressional apportionment, moving one representative from Massachusetts to Washington. Id. at-, 112 S.Ct. at 2770.

Refusing to address the APA claim, the Supreme Court concluded that the Secretary’s determination was not “final” because, in the context of apportionment, the Secretary simply reports the results of the census to the President, who in turn transmits the apportionment for each state in the House of Representatives to the Clerk of the House. The Court reasoned that because “there is no statute that rules out an instruction by the President to the Secretary to reform the census, even after the data is [sic] submitted to him,” the Secretary’s decision as to how foreign federal employees are counted is “like the ruling of a subordinate official,” and, therefore, not final for purposes of APA review. Id. at-, 112 S.Ct. at 2774 (citation omitted).

Defendants believe that the same rationale that led the Supreme Court to reject the APA claim in Franklin v. Massachusetts, an apportionment case, applies with equal vigor here. I disagree. The Supreme Court held that the Secretary’s acts in conducting the census and reporting the counts to the President were not “final,” for purposes of challenging apportionment. That case did not involve a situation where, as here, plaintiffs challenge the counts as they are used for intra-state redistricting and for federal fund allocation. See City of New York I, 713 F.Supp. at 50. Neither of these purposes requires the Secretary to transmit the counts to the President before publishing them or transmitting them to census data users.¹⁷ The Secretary’s reporting of the counts for those purposes, accordingly, is final agency action for purposes of APA review. As Justice Stevens explained in Franklin:

Even in the Court’s view, the Secretary’s report of census information to recipients other than the President would certainly constitute “final agency action.” ' The Court’s decision thus appears to amount to a pleading requirement. To avoid the bar to APA review that the Court imposes today, litigants need only join their apportionment challenges to other census-related claims. Notwithstanding the Court’s novel reading of the statute, in view of the Secretary’s insistence on unitary census data, relief on any census claim would yield relief on all other claims-.

Franklin, — U.S. at-, n.-14, 112 S.Ct. at 2783, n. 14 (Stevens, J., concurring).

Accordingly, I adhere to my earlier decision that the APA governs the Secretary’s decision. Hence, the question for review is, as the plaintiffs have pithily stated, “whether the Secretary’s application of the decision guidelines, as construed in light of constitutional requirements, to reject the [adjusted] counts is arbitrary and capricious.”¹⁸ Plaintiffs’ Brief at 148.

II. The Constitutional Requirements

In Franklin v. Massachusetts', 'the Supreme Court reminded us that in making decisions regarding the census, “the Secretary’s interpretation [of Art. I, § 2, cl. 3] [must be] consistent with the constitutional language and the constitutional goal of equal representation.” — U.S. at-, 112 S.Ct. at 2777. The language of the Constitution is beguilingly simple: “The actual enumeration shall be made ... in such manner as [the Congress] shall by Law direct.”

' While the defendants contend that the phrase “actual enumeration” bars adjustment, I have previously concluded “that because Article I, § 2 requires the census to be as accurate as practicable, the Constitution is not a bar to statistical adjustment.” City of New York II, 739 F.Supp. at 767; cf. Kirkpatrick v. Preisler, 394 U.S. 526, 530, 89 S.Ct. 1225, 1228, 22 L.Ed.2d 519 (1969) (“[t]he whole thrust of the ‘as nearly as practicable’ approach is inconsistent with adoption of fixed numerical standards which excuse population variances”); Wesberry v. Sanders, 376 U.S. 1, 7-8, 84 S.Ct. 526, 530, 11 L.Ed.2d 481 (1964) (“as nearly as is practicable one man’s vote in a congressional election is to be worth as much as another’s”).

The defendants also claim that judicial scrutiny of the Secretary’s decision for accuracy is inappropriate after the Supreme Court’s recent decision in United States Dep’t of Commerce v. Montana, — U.S. -, 112 S.Ct. 1415, 118 L.Ed.2d 87 (1992). In Montana, the state challenged a federal statute governing the method by which Representatives are allocated to the states because it resulted in giving Montana only one congressional seat, although its population was significantly higher than that of the av erage congressional district in the nation. The Court rejected the challenge, noting that “although common sense supports a test requiring a good faith effort to achieve precise mathematical equality within each State, the constraints imposed by Article I, § 2, itself make that goal illusory for the nation as a whole.” Id. at-, 112 S.Ct. at 1429 (emphasis in original) (citation omitted). The specific constraints making mathematical precision illusory on the national level were “[t]he constitutional guarantee of a minimum of one Representative for each State,” and “the need to allocate a fixed number of indivisible Representatives among 50 states of varying populations.” Id.

I reject the government’s argument that Montana mandates a departure from my earlier conclusion that the Secretary of Commerce must conduct the census in a manner to render it as accurate as practicable. First, the constitutional constraints that warranted departure from that standard in Montana are not present here. Second, in Montana, the Court noted that Art. I, § 8, cl. 18, of the Constitution “expressly authorizes Congress to enact legislation that ‘shall be necessary and proper’ to carry out its delegated responsibilities.” Id. Here, no constitutional provision requires similar deference to the Secretary’s decision. Finally, the Montana case involved a challenge to a census procedure only as it related to apportionment, not as it related to intra-state redistricting. Here, by contrast, the decision on whether to adjust the 1990 census had profound effects on intra-state redistricting because the adjusted counts would change not only national and state population figures, but the counts for political subdivisions within states, such as cities and counties. Because the implications of the Secretary’s decision at issue here are fundamentally different from the federal statute at issue in Montana, I adhere to my earlier conclusion that the Secretary must assure that the census be as accurate as practicable.

The conclusion that the Secretary must provide the most accurate census practicable however, does not, lead inexorably to the conclusion that a decision against adjustment is therefore unconstitutional. In deciding whether the Secretary’s decision was arbitrary and capricious in light of the requirement that the decision provide the most accurate census practicable, the Court must turn to the Secretary’s consideration of the guidelines, which help to illuminate the meaning of both “accuracy” and “practicability.”

III. The Guidelines

An agency decision is arbitrary and capricious “if the agency has relied on factors which Congress has not intended it to consider, entirely failed to consider an important aspect of the problem, offered an explanation for its decision that runs counter to the evidence before the agency, or is so implausible that it could not be ascribed to a difference in view or the product of agency expertise.” Motor Vehicle Mfrs. Ass’n v. State Farm, Mut. Auto. Ins. Co., 463 U.S. 29, 43, 103 S.Ct. 2856, 2867, 77 L.Ed.2d 443 (1983). Here, the analytical scaffolding for review of the Secretary’s decision is established by the guidelines promulgated in accordance with the Stipulation.¹⁹

The Stipulation provided that the Secretary retained all authority and decision-making power, “including without limitation the decision whether or not to adjust the 1990 Decennial Census.” Stip. at 1. It also required the defendants to “develop and adopt guidelines articulating what defendants believe are the relevant technical and nontechnical statistical and policy grounds for decision on whether to adjust the 1990 Decennial Census population courts.” Stip. at 3. Accordingly, the defendants promulgated the following eight final guidelines to serve as the grid against which the Secretary’s decision must be measured:

1. The Census shall be considered the most accurate 'count of the population of the United States, at the national, state, and local level, unless an adjusted count is shown to be more accurate. The criteria for accuracy shall follow accepted statis tieal practice and shall require the highest level of professional judgment from the Bureau of the Census. No statistical or inferential procedure may be used as a substitute for the Census. Such procedures may only be used as supplements to the Census.

2. The 1990 Census may be adjusted if the adjusted counts are consistent and complete across all jurisdictional levels: national, state, local, and census block. The resulting counts must be of sufficient quality and level of detail to be usable for Congressional reapportionment and legislative redistricting, and for all other purposes and at all levels for which census counts are published.

3. The 1990 Census may be adjusted if the estimates generated from the pre-specified procedures that will lead to an adjustment decision are shown to be more accurate than the census enumeration. In particular, these estimates must be shown to be robust to variations in reasonable alternatives to the production procedures, and to variations in the statistical models used to generate the adjusted figures.

4. The decision whether or not to adjust the 1990 Census should take into account the effects such a decision might have on future census efforts.

5. Any adjustment of the 1990 Census may not violate the United States Constitution or Federal statutes.

6. There will be a determination whether to adjust the 1990 Census when sufficient data are available, and when analysis of the data is complete enough to make such a determination. If sufficient data and analysis of the data are not available in time to publish adjusted counts by July 15, 1991, a determination will be made not to adjust the 1990 Census.

7. The decision whether or not to adjust the 1990 Census shall take into account the potential disruption of the process of the orderly transfer of political representation likely to be caused by either course of action.

8. The ability to articulate clearly the basis and implications of the decision whether or not to adjust shall be a factor in the decision. The general rationale for the decision will be clearly stated. The technical documentation lying behind the adjustment decision shall be in keeping with professional standards of the statistical community.

City of New York II, 739 F.Supp. at 769 (emphasis in original).

“Most of these guidelines are embroidered with an accompanying ‘explanation.’” Id. The Decision discussed each of the guidelines in detail, and concluded that numbers 1, 2, 3, 4 and 7 militated against an adjustment, while numbers 5, 6, and 8 did not tilt either way. The plaintiffs argue that the conclusions reached by the Secretary with respect to guidelines 1, 2, 3, 4 and 7 are the result of implausible assumptions, unwarranted speculation, and misuse, misstatement, and disregard of the evidence.

Guideline One

Guideline One, establishes the point of departure for analysis of the adjustment question. It mandates that the actual count be considered the most accurate count of the population “at the national state and local level, unless an adjusted count is shown to be more accurate.”

To test the accuracy of the adjusted counts against the actual enumeration, the Decision referred to a population measurement technique that the Census Bureau had used, Demographic Analysis (“DA”). Decision at 2-9. DA estimates the population, and the sub-populations of particular groups, through administrative records such as birth and death certificates, and immigration statistics. Id. The Secretary conceded “that the PES and DA estimates are not far apart in a statistical sense,” but found “some important and puzzling differences,” which “lead to sharply different conclusions” and raise some “question” as to “whether the adjusted figures are more accurate than the census count even at the national level.” Id. at 2-10, 2-35.

Among the specific problems that the Secretary noted were: (1) that the overall undercount rate inferred from comparing the actual enumeration to DA (1.85%) is smaller than that inferred from the PES (2.07%), a result contrary to intuitive expectation;²⁰ (2) that, at the national level, there were instances where a PES-based adjustment would move sub-population totals in the opposite direction from that indicated by DA;²¹ (3) that PES would add 1,055,826 more females than DA indicates should be added; and (4) “that all groups of black males (except those aged 10-19) are substantially undercounted by the PES relative to DA.” Decision at 2-10.

In addition to a comparison with DA, the Secretary also discussed a number of other statistical techniques that were used to gauge the accuracy of the PES counts when compared to the census results. He conceded that the PES-adjusted estimates might reflect more accurately the total population, and the racial and ethnic subpopulations of the country, and that “[a]t the State and local level____the adjusted figures tend to be too high, but generally closer in numeric terms to the true population than the census counts which tend to be too low.” Id. at 2-1. He concluded, however, that “[t]he loss function analysis and hypothesis tests that have been prepared by the Census Bureau to date, although of uncertain reliability, do support the superior accuracy of the census counts versus the adjusted figures when we consider distributive accuracy — or fairness — and use reasonable estimates of the error variance of the alternative [PES-based adjustment].” Id. at 2-2.

He also expressed concern that there was little or no direct evidence that the adjusted counts led to greater distributive accuracy at local levels. On that basis, the Secretary concluded that Guideline One militated against adjustment because “acceptance of adjusted counts as more accurate requires not only that the counts themselves be shown to be more accurate, but that the distribution of those counts across the United States reflect more accurately the distribution of the population.” Id. at 2-8 (emphasis in original).

As support for these concerns, the Secretary discussed a Census Bureau loss function analysis that measured the number of individual states whose population would be made less accurate by adjustment than by using the census count. As conducted by the Bureau, the loss function indicated that 21 states’ population shares would be made less accurate by adjustment. However, when the Secretary employed a statistical variance toward the low end of the acceptable range envisioned by the USC, he found that the proportional shares of 28 or 29 states would be worsened by adjustment. Id. at 2-30. The Secretary similarly expressed his trepidation that there was insufficient evidence to support the greater distributive accuracy of the adjusted counts at the local level. Id.

In his consideration of Guideline One, the Secretary also expressed serious concern over the methodology by which the PES was taken, and the manner in which the adjusted counts were tabulated.²² He was particularly discomfited by the manner in which unresolved cases in the PES were treated because the Bureau had to determine whether people found in the PES were also found in the census in order to compute dual system estimates for the poststrata. Such determinations were made by “matching” census forms to PES forms for the same household. A household survey in the PES that was “matched” to the census record of that residence meant that there was no error in the census enumeration of that household. A non-match meant an undercount. Decision at 2-16. Because there were cases where incomplete census and PES forms made such matching impossible, the Bureau was forced to employ a mathematical model to impute enough missing characteristics to enable it to make a match determination. Even after that imputation was complete, there were people found in the PES for whom it was impossible to determine whether they matched people counted in the census, and vice-versa. In those cases a different set of formulas was used to impute match status.

The Secretary concluded that, “[i]n general, missing data were not found to be a serious problem,” but identified several areas of concern with the imputation process. First, he noted that while the rates of imputation in the P and E samples²³ were low— 1.7% and 2.1% respectively — weighted up to the national population they represented almost nine million people, a number almost twice as large as the net national undercount. Decision at 2-16. Second, in noting the high correlation between imputation ratios and undercount ratios, he stated that “the strata for which there is more doubt about the quality of the adjusted data because of imputation tend to be the same strata for which an adjustment would result in large increases in the population.” Id. Finally, the Secretary noted that the assumptions in the imputation models were largely untested. Decision at 2-17. His concern was exacerbated by his respect for the research of Panel member Kenneth Wachter, which indicated that flaws in the imputation model could render the adjusted counts “significantly in error.”

Based on all the foregoing, the Secretary concluded that there was simply not enough convincing evidence to support a finding that the adjusted counts would lead to greater distributive accuracy than the census counts, and therefore that the guideline “weighted] in favor of a decision not to adjust.” Id. at 2-36.

The plaintiffs assail this conclusion on several grounds. First, they argue that the Secretary misused DA because that technique has- historically been most accurate as a “yardstick of the ‘census in terms of national undercount ánd as a measure of differential undercount between demographic groups,” but “is much less reliable in its ability to estimate an undercount rate for a specific group in a particular census,” as the Secretary attempted to use it. They contend that the Secretary’s focus on discrepancies between PES and DA undercount rates for certain specific, groups is an attempt to obfuscate the fact that, as the Secretary himself admits, the “detailed analysis shows that the PES and DA estimates are not far apart in a statistical sense.” Decision at 2-10.

With respect to the Secretary’s professed concern over distributive accuracy, the plaintiffs contend that the Secretary’s invocation of a loss function that merely counted up the number of states whose populations would be made less accurate, regardless of the greater aggregate accuracy of the adjustment, and without reference to the extent that counts are made less accurate, is statistically insupportable. They also argue that the Secretary’s rejection of numerous loss function analyses performed by the Bureau supporting the superior accuracy of the adjusted counts, and his putative concern with the technical aspects of the PES are irrational at best, and disingenuous at worst.

I have reviewed in some detail the Secretary’s conclusion that Guideline One militated against adjustment and the plaintiffs’ arguments to the contrary. While the plaintiffs have made a compelling attack on the Decision, and the Secretary has conceded that the objective criteria used to measure the adjusted counts show a greater numeric accuracy at the national level and that the Census Bureau estimates of distributive accuracy marginally favor the adjusted counts, I find that Secretary’s conclusion under Guideline One was neither arbitrary nor capricious.

The Secretary’s decision to focus on distributive, rather than numeric, accuracy was consonant with the constitutional goal of assuring the most accurate census practicable, given the census’s function as a standard by which to distribute political representation, and economic benefits. In that regard, I find that the Secretary’s use of a loss function that considered the number of states whose populations would be made less accurate by adjustment to be appropriate. Similarly, the Secretary’s concern that “[w]ith respect to places under 100,000 population, there is no direct evidence .that adjusted counts are moré accurate” was legitimate, given Guideline One’s requirement that the adjusted counts be shown to be more accurate at the local level. Decision at 2-30.

Plaintiffs’ attack on the Secretary for subjecting the tests favoring adjustment to unrealistically rigorous scrutiny misconstrues Guideline One, which clearly states that “[t]he Census shall be considered the most accurate count of the population of the' United States, at the national, state, and local levels, unless an adjusted count is shown to be more accurate.’’ City of New York II, 739 F.Supp. at 769 (emphasis added). Thus, plaintiffs’ failure to illustrate affirmatively the superior accuracy of the adjusted counts either (1) at any level mentioned in Guideline One, or (2) for any reasonable definition of accuracy, is sufficient to support a finding that Guideline One favors use of the original census counts.

Turning to the Secretary’s focus on the mechanics of the PES and the use of imputation, I find this consideration appropriate. Because the PES, like any sample survey,²⁴ rests on an assumption to begin with — that the portion sampled is identical to the population as a whole — placement of additional assumptions into the model caused by imputation is a fair basis for escalating skepticism. While the logic of the Secretary’s conclusion regarding Guideline One is not overpowering, neither can it be characterized as arbitrary or capricious.

Guideline Two

Guideline Two states that adjustment may be made only if the adjusted counts are “consistent and complete across all jurisdictional levels: national, state, local and census block.” The guideline also requires the adjusted counts to “be of sufficient quality and level of detail to be usable for ... all ... purposes and at all levels for which census counts are published.” City of New York II, 739 F.Supp. at 769 (emphasis -in original). The Secretary recognized that “[t]he adjusted figures .,. are consistent across all jurisdictional levels and of sufficient detail for all purposes,” but nevertheless concluded that Guideline Two militated against adjustment because of the questionable quality of the adjusted counts. Decision at 2-45.

In surmising that the counts were of debatable quality, the Secretary homed in on the “homogeneity assumption” in the construct of the 1,392 poststrata. He was troubled that the “adjustment process rests on the assumption that persons in each poststratum are homogeneous with respect to their probability' of being missed by the census, i.e., their capture probability.” Decision at 2-39. Conceding that many experts did not find this assumption problematic, and that at broad levels such as the national and state levels the assumption caused no serious problems, the Secretary ultimately concluded that “ ‘local heterogeneity is a serious problem for adjusting the 1990 census’ ”, and that “ ‘[the] evidence indicates that a substantial portion, possibly a majority, of relative counts for district-size units can be made worse off by adjustment.’ ” Decision at 2-42 (quoting Report of Special Advisory Panel Member Kenneth Wackier, at 26).

In reaching this conclusion, the Secretary worried that because members of an individual poststratum might have a different likelihood of being undercounted, thereby debunking the homogeneity assumption, generalizing the undercount rate of those counted in the PES to all members of that poststratum might seriously interfere with the accuracy of the count for some census purposes, including redistricting. He discussed two studies conducted by the Bureau that addressed the homogeneity assumption, and which the Bureau had relied on in concluding that individuals within each poststratum were sufficiently uniform to warrant such an assumption, dubbed the “P12” and “P15” studies. Decision at 2-38-2-40. The Secretary opined that the Bureau’s evidence from those studies. was “mixed.” . , .

He was also concerned about the- adverse consequences that a failure in the homogeneity assumption could have on adjustments at local levels, noting that because there were only 5000 sample blocks, most jurisdictions would be adjusted based on data gathered elsewhere. Decision at 2-43.

The plaintiffs brand the Secretary’s concern about heterogeneity as “unreasonable.” They assert that because perfect homogeneity is utterly unattainable in the world of survey sampling, the relevant question is whether a departure from the homogeneity assumption has an important impact on the measurement. They contend that because the pertinent Census Bureau studies supported the homogeneity assumption, and particularly because the P12 study confirmed that the populátioñ subgroups defined for the PES are sufficiently uniform tó' be usable for adjustment, there was sufficient homogeneity to warrant the conclusion that the adjusted counts lead to improvement.

Plaintiffs’ argument is rejected.' While they have made a strong showing that the adjusted counts are more accurate than the original counts for most purposes for which the census is used, the Secretary’s concern that heterogeneity may lead to less accurate counts at local levels used for redistricting appears reasonable. Plaintiffs’ contention that the Secretary was effectively required to bite the bullet and ignore the problem that-residual heterogeneity posed, once the Bureau had concluded that there was sufficient evidence to support the homogeneity assumption,, ignores the guidelines’ mandate that the Secretary determine that the adjusted counts be usable for all purposes for which census counts are published. Clearly, there is some likelihood that residual heterogeneity will have an adverse effect on the census counts when used for redistricting. This is enough to support the Secretary’s conclusion that Guideline Two militates against adjustment. Accordingly, I find that his conclusion was not arbitrary or capricious.

Guideline Three

Guideline Three requires that the PES and other adjustment procedures be “pre-specified” and that the estimates they generate be “shown to be robust to variations in reasonable alternatives to the production procedures, and to variations in the statistical models used to generate the adjusted figures.” City of New York II, 739 F.Supp. at 769. The Secretary advanced two arguments as the basis for his conclusion that Guideline Three militated against adjustment: (1) that the actual conduct of the DSE did not proceed sufficiently in accordance with a pre-specified plan; and (2) that certain statistical techniques and assumptions were not sufficiently “robust” to support adjustment.²⁵ Decision at 2-54-2-55.

On the first point, the Secretary recounted various decisions that Bureau employees made after the pre-specification of the PES, including choices about the selection of carrier variables during the regression analysis in the smoothing process. Decision at 2-47. He noted that one member of the Panel who voted for adjustment had conceded that certain pre-specified procedures had changed during the enumeration process and had affected the PES. Id. (citing Report of Panel Member Wolter, pp. 9-10). The Secretary agreed with Wolter’s ultimate conclusion that the decisions to change pre-specified procedures made during the enumeration and the PES were treated with a high degree of professionalism and also acknowledged that the PES could not have been completely pre specified, but expressed his discomfort with the deviations as follows:

Although I believe that the decisions [to deviate from pre-specified procedures] were made for sound professional reasons in the 1990 census, using these adjustment mechanisms opens the possibility for manipulation of future post enumeration surveys in ways that are unavailable in' traditional census procedures. This weighs heavily against an adjustment of the census.

Decision at 2-48 (emphasis in original).

With respect to the robustness of the results when subjected to alternative statistical models required by Guideline Three, the Secretary concluded that “[t]he results of the adjustment procedure are broadly robust at an aggregate, national level.” Id. at 2-54. However, he found three questionable areas where the adjustment methods concerned him: (1) imputation; (2) poststratification; and (3) the use of smoothing procedures.

The Secretary concluded that the imputation was statistically robust, but expressed a fear that variations of the assumptions underlying the imputation could have an effect on the apportionment of the House of Representatives. Id. at 2-48-2-49. With respect to poststratification, the Secretary observed that if poststratification had recognized the state of residence rather than the census division of residence as a factor, three states would have had significantly different counts. Id. at 2-49. Finally, moving to the robustness of smoothing, the Secretary concluded that the numerous decisions and techniques involved in the two-stage process, including the , discretionary selection of carrier variables, led to an impermissibly high level of uncertainty to employ the adjusted counts as a basis for reapportionment. Id. at 2-49-2-54. In short, the Secretary stated that the lack of comprehensive pre-specification, the possibility it raised for future political manipulation, and the uncertainty associated with the use of extensive statistical assumptions in the adjustment process led him to find that Guideline Three militated against adjustment.

In their attack on this conclusion, the plaintiffs first argue that the Secretary’s concern over political manipulation of future censuses because of the lack of pre-specification is an inappropriate basis for making a determination under Guideline Three. While I tend to agree with that argument,-1 read the Secretary’s discussion of future political manipulation as merely an explanatory note, underscoring why he thought that pre-specification was so significant. Because Guideline Three clearly mandated pre-specification, the Secretary’s well-supported conclusion that the procedures were not adequately pre-specified supported his conclusion under this guideline.

Plaintiffs also argue that the Secretary required an impossible degree of pre-specification because some of the decisions to be made, including decisions relating to the smoothing process, were highly dependent on data to be collected during the PES, and therefore could not have been completely pre-specified. This argument ignores the fact that certain techniques were pre-specified and then changed later. See Report of Special Advisory Panel Member Kirk M. Wotter at 10. The Secretary’s conclusion that pre-specification did not occur as contemplated by the guideline was justified.

Plaintiffs also belittle the Secretary’s concern that even small changes in any of the assumptions underlying the statistical procedures of the adjustment could lead to a different apportionment of the House of Representatives. Deprecating his conclusion that this seriously compromised use of the adjusted counts, they contend that because the Secretary has conceded that small changes in the census methodology can move House seats as readily as small changes in the PES methodology, his concern under Guideline Three is illusory.

I disagree. The plaintiffs’ reliance on the imperfections in the census to blink at similar uncertainties in the adjustment procedure misses the point that, under the rubric of the guidelines, the adjusted counts must satisfy certain criteria, regardless of whether the original enumeration could survive exposure to similar criteria. It must be remembered that under Guideline One, the presumption of accuracy runs in favor of the original census count. Because the Secretary’s concerns over pre-specification and. the robustness of adjustment data were legitimate, I find that the Secretary’s conclusion under Guideline Three was not arbitrary or capricious. Guideline Four

Guideline Four counsels that “[t]he decision whether or not to adjust the 1990 Census should take into account the. effects such a decision might have on future census efforts.” City of New York II, 739 F.Supp. at 769. With this in mind, the Secretary stated that he “d[id] not find compelling evidence in either direction regarding the effects of a decision on future individual motivations.” Decision at 2-58. Weighing the effects that an adjustment might have on the efforts of state, community,' civic, and interest group leaders, the Secretary was concerned that “an adjustment [would] remove the incentive that these public officials and groups currently have to provide active support in achieving a complete count.” Id. at 2-59. The Secretary found “unpersuasive” the contention that, even with an adjustment, local officials would retain a strong incentive to gather data,' and “[found] no evidence indicating that local support would decrease as a result of a decision not to adjust the census.” Decision at 2-59.

He went on to conclude that a decision to adjust could hinder the operations of the census in other ways, including disincentives for Congress to provide funding, and for enumerators to pursue their task energetically, and the possibility that adjustment could be. distorted for partisan political purposes in future censuses. Id. at 2-60. Balancing all these fears, the Secretary conclüded “that an adjustment would adversely affect future census efforts to a greater extent than any adverse effects of a decision not to adjust.” Id. at 2-61.

Plaintiffs argue that it is futile to fret over censuses in the year 2000 and beyond in considering whether or not to adjust the 1990 census. This argument blithely ignores the express mandate of Guideline Four that the effect of the Secretary’s decision on future censuses be considered. While I recognize that Guideline Four creates a potential tension with the constitutional requirement that the census be as accurate as practicable, under the circumstances of this case, that tension is minimal. Accordingly, I find that the Secretary’s conclusion regarding Guideline Four was neither arbitrary nor capricious.

Guidelines Five' & Six

Because the Secretary’s conclusions based on Guidelines Five and Six are not challenged by the plaintiffs, I will only say that the conclusions reached by the Secretary in the Decision sufficiently considered those guidelines.²⁶

Guideline Seven

Guideline Seven provides that “[t]he decision whether or not t.o adjust the 1990 Census shall take into account the potential disruption - of. the process of the orderly transfer of political representation likely to be caused by either course of action." City of New York II, 739 F.Supp. at 769. At an earlier stage of this litigation, I rejected a request to vacate Guideline Seven, finding that it, and Guideline Eight, might, “in a constructive fashion, help define the meaning of ‘the most accurate census practicable,’ ” and concluded that, at least to that extent, they were permissible factors. Id. at 771.

In his consideration of Guideline Seven, the Secretary noted that the Clerk of the United States House of Representatives had officially certified to each of the fifty states the number of seats allotted to that state for the 103rd Congress (convened in January 1993) based on census figures released on December 26,1990, and that, as of May 1991, “some 20 states had already enacted either or ‘both of their Congressional and State legislative redistricting plans.” Decision at 2-71. The Secretary then went on to outline the disruption and delay that an adjustment would cause, particularly in those states where adjustment would change their allotted number of seats in the House of Representatives.

It should be remembered that Congress decreed in 1912 there be only 435 seats in the House of Representatives. We are, therefore, dealing with a zero-sum game; when one state gains a seat, another 'must lose one. If the adjustment were made, Cali-’ fornia and Arizona, for example, would each gain one seat in the House, while Pennsylvania and Wisconsin would each lose one. Id. at 2-72. The Secretary envisioned massive litigation over such a decision.

Ultimately, the Secretary concluded that Guideline Seven favored adherence to the census counts. He rejected the argument that non-adjustment is “inherently disruptive,” as based on the question-begging premise that the adjusted - counts are more accurate. He also concluded that, even if it were true that adjustment would result in a fairer distribution of funds, this consideration would pale in comparison to the disruption of political representation that would ensue from a decision, to adjust, because “adjustment would not result in significant shifts in those funds.” Id. at 2-75.

Plaintiffs assail the Secretary’s' Guideline Seven conclusion on two distinct grounds. First, they suggest that it is disingenuous for the Secretary to rely on the fact that the unadjusted counts were already being used for reapportionment and redistricting purposes, when the Stipulation required that any release of the unadjusted data before the Secretary’s decision be accompanied by a notice advising recipients that they used the data at their own risk. None of this however, detracts from the fact that Guideline Seven explicitly required the Secretary to consider such disruption in deciding whether or not to adjust. Nor does it contradict the simple logic of the Secretary’s argument that-a decision in favor of adjustment on July 15, 1991, would have disrupted the reapportionment and redistricting that was then ongoing.

Plaintiffs also observe that the Secretary’s conclusion that adjustment would not result in significant shifts in federal funds contradicts an earlier sentence in the Decision that city and state population “shares are very important because they determine ... how large a ‘slice of the pie’ of federal funds go to each city and state.” Decision at 1-3-1-4. Plaintiffs are right. This, however, does not render the Secretary’s decision invalid under Guideline Seven, because it involves a matter — the allocation of federal funds — only tangentially related to Guideline Seven, the basic thrust of which is the effect of a decision to adjust “on the orderly transfer of political representation.” Accordingly, while there is an obvious inconsistency in the discussion accompanying the result, the plaintiffs have failed to show that the Secretary’s conclusion under Guideline Seven was arbitrary. or capricious.

Guideline Eight

Guideline Eight requires the Secretary to articulate the factors relied upon in reaching his decision, and also requires that “[t]he technical documentation lying behind [his] decision shall be in keeping with professional standards of the statistical community.” City of New York II, 739 F.Supp. at 769. Because the plaintiffs do not specifically attack the. Secretary’s decision under this guideline, and because the Secretary concluded that application of this guideline neither favored nor militated against adjustment, I find that the Secretary’s Decision complied with Guideline Eight.

Having thus parsed the guidelines, the Court concludes that the Secretary’s conclusions under each guideline and his ultimate decision against adjustment cannot be characterized as arbitrary or capricious. The breadth of the guidelines left the Secretary enormous discretion. Plaintiffs have made a powerful case that discretion would have been more wisely employed in favor of adjustment. Indeed, were this Court called upon to decide this issue de novo, I would probably have ordered the adjustment.²⁷ However, it is not within my province to make such determinations. The question is whether the Secretary’s decision not to adjust is so beyond the pale of reason as to be arbitrary or capricious. That far I cannot go.

One of the central tenets of our founding fathers was that the role of the judiciary should be carefully delineated, especially when the controversy related to the management of the government. As Hamilton wrote:

The administration of government, in its largest sense, comprehends all the operations of the body politic, whether legislative, executive, or judiciary; but in its most usual and perhaps in its most precise signification, it is limited to executive details, and falls peculiarly within the province of the executive department.

The Federalist No. 72 at 450 (Henry Cabot Lodge, ed., 1888).

The writings of Montesquieu and Locke bristle with the notion of separation of powers. But nowhere is it articulated more succinctly than in the Massachusetts Constitution:

In the government of this commonwealth, the legislative department shall never exercise the executive and judicial powers, or either of them: the executive shall never exercise the legislative and judicial powers, or either of them: the judicial shall never exercise the legislative and executive powers, or either of them:' to the end it may be a government of laws and not of men.

Mass. Const, pt. 1, art. 30 (1780).

True, the APA sanctions judicial intervention when the parties feel aggrieved by a final administrative ruling. But the APA tightly cabins judicial oversight, permitting judicial intrusion only when the administrative decision abuses reason. It is essential to the maintenance of judicial integrity that courts reviewing such determinations zealously adhere to the arbitrary and capricious standard of review. Citizens to Preserve Overton Park, Inc. v. Volpe, 401 U.S. 402, 416, 91 S.Ct. 814, 824, 28 L.Ed.2d 136 (1971) (When reviewing agency action under the arbitrary and capricious standard of review, “the ultimate standard of review is a narrow one. The court is not empowered to substitute its judgment for that of the agency.”); Hudson Transit Lines v. United States ICC, 765 F.2d 329, 336 (2d Cir.1985) (“while a reviewing court may not supply the basis for the agency’s decision, lest it interfere with matters that Congress entrusted to the executive agency, it will uphold a decision of less than ideal clarity if the ‘path which [the agency] followed can be discerned’ ”) (quoting Colorado Interstate Gas Co. v. FPC, 324 U.S. 581, 595, 65 S.Ct. 829, 836, 89 L.Ed. 1206 (1945)); Connecticut Dep’t of Children & Youth Servs. v. Department of Health & Human Servs., 788 F.Supp. 573, 577 (D.D.C. 1992) (“Under this standard, the Court is not free to substitute its own judgment, but is limited to determining whether the agency has considered all relevant factors and whether the agency’s decision is reasonable and in accordance with the relevant statute. Under the [APA], the standard of review is highly deferential to the agency.”). As Cardozo has reminded us, “[t]he judge, even when he is free, is still not wholly free. He is not to innovate at pleasure. He is not a knight-errant, roaming at will in pursuit of his own ideal of beauty or of goodness. He is to draw his inspiration from consecrated principles.” Benjamin N. Cardozo, Nature of the Judicial Process 141 (Yale Univ.Press, 1921).

Midst all the shirm und drang, after all is said and done, the question before the court distills to this: did the Secretary act reasonably? . This, of course, depends mainly upon the evidence he had before him. In his testimony, Dr. Robert E. Fay, one of the principal statisticians at the Census Bureau (who, incidently, voted to adjust) pierced right to the heart of the case: “I told the Secretary that ... reasonable statisticians could differ on this conclusion.” Tr. at 1909. The Court agrees, and therefore, concludes that the Secretary’s decision not to adjust the 1990 census count was neither arbitrary nor capricious.

The PES Tapes

Plaintiffs also move to vacate a protective order, issued by Magistrate Judge Ross, governing certain computer tapes they got from the Government during discovery in this case. These tapes contain the adjusted census data at the block level and are the material that would have been released to the states if the Secretary had decided to adjust.²⁸ Plaintiffs argue that the Court should vacate the protective order because: (1) plaintiffs already possess the tapes, and, thus, release of the data would not violate any institutional confidence; and (2) release of the tapes is appropriate under 13 U.S.C. § 141(c), which requires the Secretary of Commerce to provide the states with data to be used in redistricting.

When they opposed production of these tapes before Magistrate Judge Ross, the defendants asserted the “deliberative process” privilege as a basis for their refusal. The “deliberative process” privilege “protects from disclosure those agency documents which reflect ‘advisory opinions, recommendations and deliberations comprising part of a process by which governmental decisions and policies are formulated.’ ” Mobil Oil Corp. v. Department of Energy, 102 F.R.D. 1, 5 (N.D.N.Y.1983) (quoting Mobil Oil Corp. v. Department of Energy, 520 F.Supp. 414, 416 (N.D.N.Y.1981)).

The privilege is a qualified or discretionary one that turns on a balance of competing policy claims. See In re Franklin Nat’l Bank Sec. Litig., 478 F.Supp. 577, 582 (E.D.N.Y.1979). The privilege does not protect purely factual material. Id. at 581. Magistrate Judge Ross assumed arguendo, that the redistricting tapes reflected certain advisory opinions, but found that the benefit to be derived from protecting such information was outweighed by the benefit to accurate judicial fact-finding that would follow upon production of the tapes. Accordingly, she ordered the defendants to produce them, but, at the request of the defendants, also entered a protective order forbidding public disclosure. The Magistrate Judge was aware at the time of what she described as the “hotly contested Ninth Circuit litigation concerning disclosure of these same tapes.”

I now have the result of that “hotly contested” Ninth Circuit case. Assembly of California v. United States Dep’t of Commerce, 968 F.2d 916 (9th Cir.1992). There, the Department of Commerce was asked to release computer tapes containing all the block-level census data for California pursuant to a claim by the California State Assembly under the Freedom of Information Act, 5 U.S.C. § 552(a) (“FOIA”). The Department of Commerce argued, as it does here, that the data should not be disclosed because of their pre-decisional and deliberative nature.

The district court rejected that argument and ordered the Department of Commerce to produce the data. Assembly of California v. United States Dep’t of Commerce, 797 F.Supp. 1554 (E.D.Cal.1992). The Commerce Department appealed. Agreeing with the district court’s findings that the data were neither pre-decisional nor deliberative, the Ninth Circuit affirmed the order that the tapes be released. 968 F.2d at 923.

A directly contrary result was reached by the Eleventh Circuit in Florida House of Representatives v. United States Dep’t of Commerce, 961 F.2d 941 (11th Cir.1992). There, the Florida House of Representatives brought a FOIA action to compel the Department of Commerce to release all the adjusted block-level data for Florida. The district court granted summary judgment for Florida, and the Department of Commerce appealed. The Eleventh Circuit reversed, finding that “[b]ecause the adjusted census block level data are a subordinate’s opinion and reflect the give-and-take of the deliberative process ... the data are deliberative, and in turn, within the scope of the deliberative process privilege.” Id. at 950.

Recognizing this split in the circuits, and assuming arguendo, that the tapes reflect certain aspects of the deliberative process, I believe the Ninth Circuit has the better of the argument. Whatever interest the Department of Commerce may have had in the confidentiality of the block level counts, that interest was seriously. diluted when the Secretary released one half, of all the data to Congress; and, whatever privacy survived as to the block level data for California was lost following the Ninth Circuit’s decision.

Balanced against the slight residuary interest that the defendant's may have in the confidentiality of the block level data is the public’s interest in full access to judicial proceedings, especially where, as here, the dispute has sparked so much public interest. Because I believe that the balance weighs heavily in favor of disclosure under these circumstances, I vacate the protective order and permit the plaintiffs to use and to release to the public the computer tapes containing the adjusted block-level counts.

CONCLUSION

To sum up I: (1) find that the Secretary’s decision not to adjust the 1990 census does not violate the APA, the Constitution, the Stipulation, or any statute; and (2) vacate the protective order governing the plaintiffs’ use of the computer tapes containing the adjusted block-level counts.

While plaintiffs’ counsel has illustrated that adjustment is statistically feasible, and would improve the quality of the counts for most purposes while ameliorating the profoundly disturbing problem'of differential undercount, the Court cannot, on the record before it, supplant the Secretary’s decision.

Finally, I would be remiss if I did not note the magnificent contribution that several law firms representing the plaintiffs have made in presenting this case to the Court. These firms, Cravath Swaine, and Moore, Arnold & Porter, and Stein, Zauderer, Ellenhorn, Frischer and Sharp have devoted unusual talent and resources to this case on a pro bono basis. This is in the highest tradition, of the Bar. I commend them.

SO ORDERED.

Sitting by designation

1

. April 1, 1990 day is officially entitled "Census Day,” and is the precise date as of which the Census Bureau seeks to count the population.

2

. These coverage improvement programs included: (1) a 100 percent re-check of vacant, uninhabitable, or nonexistent units; (2) the “Were you counted?” advertising campaign to reach people who thought they might have been missed by the census; (3) a parolee and probationer check, to set the names and Census Day addresses of those people and add them to the census if they had not already been counted; (4) the housing coverage check, in which the Census Bureau recanvassed select blocks based on evidence flushed out by the automated management information system; and (5) the local government review program, which provided local governments with the opportunity to challenge census counts for their areas. Decision at 4-7-4-9.

3

. Discomfiture over the persistent pattern of differential undercount had prompted the Bureau to conduct a Post-Enumeration Program (the "PEP”) in 1980, a survey designed to evaluate the quality of the 1980 census and to estimate the undercount, including the differential under-count, at both national and subnational levels. A lawsuit to have the 1980 census adjusted statistically by use of the PEP or another statistical technique was unsuccessful. See Cuomo v. Baldrige, 674 F.Supp. 1089 (S.D.N.Y.1987).

4

. At trial, the parties explained capture/recapture in terms of determining the number of fish in a lake. First you capture 1000 fish, tag them and throw them back. Then, you catch another 100. If 90 of those have tags, it suggests that 90 percent of all the fish in the lake are tagged. If so, then the 1000 fish initially tagged represent 90% of all the fish in the lake. Doing the algebra, the total population of fish in the lake is therefore 1,111. Tr. at 41-42.

5

. Poststratification grouped all individuals with a similar likelihood of being counted in the census. These groups, labeled "poststrata”, were defined by age, sex, race, Hispanic origin, housing tenure (i.e. whether the individual owned or rented a residence), type of place (i.e., central city, suburb, outside metropolitan area), and geographic region. Tr. at 513. This categorization resulted in a total of 1,392 exhaustive and mutually exclusive poststrata. Tr. at 206-07. In other words, each resident of the United States fits into one, and only one, poststratum.

6

. For an explanation of smoothing, see infra note 10.

7

. While the defendants continued to argue during pretrial proceedings that this case presented a non-justiciable political question, the Supreme Court has now rejected this argument, holding that constitutional challenges to the census methods employed to arrive at the apportionment are justiciable. United States Dep’t of Commerce v. Montana, - U.S. -, -, 112 S.Ct. 1415, 1424-26, 118 L.Ed.2d 87 (1992).

8

. By agreement, the Secretary chose four members of the Panel from a list of seven candidates submitted by the plaintiffs, and chose the four remaining Panel members himself. The Stipulation required the Panel members to be "of such knowledge, judgment and probity that their judgment and advice shall be entitled to the utmost respect by defendants.” Stip. at 5. The four panel members chosen from the plaintiffs’ list were Eugene P. Ericksen, Leobardo F. Estrada, John W. Tukey, and Kirk M. Wolter. The four panel members chosen unilaterally by the Secretary were William Kruskal, Michael McGeehee, V. Lance Tarrance Jr. and Kenneth M. Wachter. As required by the Stipulation, the Panel members submitted recommendations to the Secretary regarding the decision on whether to adjust. Stip. at 5.

9

. As used here, "block” means a square block; that is, all the buildings on four streets forming a square. Tr. at 209. The Census Bureau made a list of the more than 5,000,000 blocks in the United States and then selected approximately 5,000 that they believed fairly contained representative samples of the 1,392 poststrata. Tr. at 208; Decision at 4-11--4-12.

10

. Smoothing is a statistical procedure used to reduce the effects of sampling error. More particularly, it seeks to reduce the difference between the results from the PES sample and the results one would receive if one were able to survey the entire population. Smoothing in the 1990 census took place as follows. First, the 1,392 raw adjustment factors with corresponding raw variances (measures of sampling error) were compiled. The Bureau then employed presmoothing, or "modelling the variance," in an attempt to improve the accuracy of the estimates of the variances of the raw adjustment factors. Tr. at 796. Once modelling the variance was completed for each raw adjustment factor, a regression was performed. This regression moved the raw adjustment factor for each poststratum towards a typical value by an amount depending on the sampling error associated with that particular poststratum. Thus, where a particular raw adjustment factor had a small variance (i.e., where the sample was very large), it would be moved only a small amount, whereas raw adjustment factors with larger variances tended to be moved more. Carrier variables relating to raw adjustment factors were selected in an effort to give the best estimate of the typical value. Tr. 807. These carrier variables included the same characteristics that defined the post-strata, such as age, sex, race, owner/renter, and other charac- . teristics such as mail return rate. The end result is that the 1,392 raw adjustment factors became 1,392 smoothed adjusted factors. The census count for each postratum group was then multiplied by its smoothed adjustment factor and adjusted census counts were produced. Tr. at 788-89; Decision at 4-17-4-18.

11

. The major potential sources of error arising from the PES included: missing data, poor quality of the reported Census Day address list, fabrication, matching error, measurement of erroneous enumerations, balancing the estimates of gross overcount and gross undercount, correlation bias, small area estimation, and late census data. Tr. at 570-73.

12

. A recent “discovery of computer errors and some statistical changes have reduced the estimates of an undercount to 1.6 percent, about the same as in 1980.” Felicity Barringer, U.S. Population Passes 265 Million, Bureau Says, N.Y. Times, December 30, 1992, at A12. As one of the witnesses testified here, "statistics is never having to say you're certain.” Tr. at 1922.

13

. Dr. Bryant's comments in a year-end interview are enlightening. See Barringer, supra note 12. In that interview "she said [that] while the statistical tools were available to make these adjustments for small geographical units, the necessary tools to double-check the findings were inadequate. In the face of legal scrutiny she said, this made a decision to adjust untenable.” She is also quoted as stating that " '[e]very number has to become defensible,' " and " '[w]hen you say — you know how to do it but you can't prove its right or wrong — then it's no longer defensible. If it weren’t for the problem that we had to defend it in court, there would have been a strong inclination to have adjusted.

Dr. Bryant also opined that she believed that an adjustment would have improved the accuracy of counts at " ‘the national and state levels, the big levels,' ” but that the PES results were " ‘very inconclusive' when used for smaller subdivisions of the population.” Id.

14

. Before trial, two other cases presenting the identical issue in this case were transferred and consolidated with this action — City of Atlanta v. Mosbacher, 92-CV-1566; Florida House of Representatives v. Franklin, 92-CV-2037.

15

. Plaintiff Hudson County, New Jersey, also claims that the decision against adjustment violated the Voting Rights Act, which provides that:

No voting qualification or prerequisite to voting or standard, practice, or procedure shall be imposed or applied by any State or political subdivision in a manner which results in a denial or abridgement of the right of any citizen of the United States to vote on account of race or color, or in contravention of the guarantees set forth in section 1973b(f)(2) of this title, as provided in subsection (b) of this section.

42 U.S.C. § 1973(a) (1982) (emphasis added). The Court rejects this claim because it is close to frivolous. By its plain language, the Voting Rights Act applies only to misconduct by states or their political subdivisions. See Senate of California v. Mosbacher, 968 F.2d 974, 979 (9th Cir.1992) (argument that Voting Rights Act contemplates suits against the federal government is "severely flawed”); Tucker v. United States Dep’t of Commerce, 958 F.2d 1411, 1414 (7th Cir.) (“The plaintiffs cannot be serious in arguing that the refusal to adjust the headcount violates the Voting Rights Act.”), cert. denied, - U.S. -, 113 S.Ct. 407, 121 L.Ed.2d 332 (1992).

16

. Plaintiffs assert that Secretary Mosbacher was closely aligned with the Republican Party and, therefore, never seriously considered adjustment in the belief that adjustment would favor Democratic politicians. They also argue that contacts made by then-White House Chief of Staff John Sununu and a member of his staff to Commerce Department officials other than Mr. Mosbacher tainted the decision. I have reviewed these allegations in detail. While it does appear that Mr. Sununu and his subordinates expressed their contempt for adjustment to Department of Commerce personnel, I cannot, on the record before me, conclude that such contacts represented improper influence. Moreover, the plaintiffs’ attack on the integrity of Mr. Mosbacher — who was never a party to these conversations — does not warrant extended discussion here.

17

. With respect to redistricting, 13 U.S.C. § 141(c) provides, in pertinent part, that:

Tabulations of population for the areas identified in any plan approved by the Secretary shall be ... reported to the Governor of the State involved and to the officers or public bodies having responsibility for legislative apportionment or districting of such State....

Id. With respect to the plaintiffs' claim based on allocation of federal funds, the following statutes provide for direct reporting of census data by the Secretary of Commerce, without the President either acting as an intermediary or retaining final discretionary authority to report the counts: 42 U.S.C. § 9831 et seq. (Head Start program); 42 U.S.C. § 702 (Maternal and Child Health Services Block Grant); 42 U.S.C. § 5632 (Juvenile Justice and Delinquency Prevention Program); 42 U.S.C. §§ 3024, 3028(b) (Programs for Older Americans); 23 U.S.C. § 104(b)(6) (Highway Planning and Construction); 49 U.S.C.App. § 1607a (Urban Mass Transportation Capital and Operating Assistance programs).

18

. Plaintiffs also contend that the Secretary's decision was arbitrary and capricious independently of the Stipulation, because it conflicted with a Department Organization Order in which the Secretary delegated authority to conduct the Census to the Director of the Census Bureau. Department of Commerce Organization Order 35-2A, August 4, 1975, as updated July 24, 1987. I find this argument unpersuasive. While the Secretary did delegate his statutory duty to take the decennial census, he also required the Director of the Census Bureau to "report and be responsible to the Assistant Secretary for Economic Affairs,” a position which subsequently became the Under Secretary for Economic Affairs. Id.; 15 U.S.C. § 1503a. The Secretary further directed the Under Secretary for Economic Affairs to "exercise policy direction and general supervision over ... the Bureau of the Census." Department of Commerce Organization Order 10-9, § 4.03, June 26, 1984. Thus, while delegating the operational responsibility necessary to prepare and conduct the census, the Commerce Department retained the authority to control policy direction, to exercise decision-making authority in significant Bureau matters, and to supervise the Bureau in the exercise of its census-taking task.

19

. A discussion of how the guidelines were formulated, considered, and ultimately promulgated may be found in the earlier opinion in which I rejected a challenge to their sufficiency. City of New York II, 739 F.Supp. at 769 & n. 9.

20

. According to the Secretary, DA would normally be expected to reveal a higher undercount rate when compared to the Census than the PES would, because the PES and Census will both miss people who are difficult to survey, while DA, which relies solely on an examination of records, will not. Decision at 2-10.

21

. Some examples of this problem cited by the Secretary are:

An adjustment based on the PES will add 180,-318 non-black males aged 10-19, while the DA indicates 136,908 should be deleted — a difference in the wrong direction of 317,226.

An adjustment based on the PES will delete 91,631 males over the age of 65, while DA indicates that 192,950 should be added — a difference in the wrong direction of 284,541 persons.

An adjustment based on the PES will delete 245,253 females over the age of 45 while DA indicates 146,255 should be added — a difference of 391,508 persons in the wrong direction.

Decision at 2-12 (footnotes omitted).

22

. Among other things, the Secretary was troubled by the effects that erroneous enumerations in the census, correlation basis, and failure in the PES total error model could have on the adjusted counts. Id. at 1-17-23. The proof at trial, however, has made it clear that these matters were peripheral to the Secretary’s conclusion under Guideline One, and therefore do not merit significant discussion here.

23

. In the jargon of the DSE, the "P sample” represented the group surveyed by the PES. The "E sample" represented the people living in the same household as the P sample as counted by the census. Decision at 4-12-4-13.

24

. See the discussion of capture/recapture in note 4, supra.

25

. In the world of survey sampling, “robustness” describes the integrity and reasonableness of the results achieved by a particular statistical technique. Robustness is determined by exposing such statistical techniques to variations in the assumptions underlying them.

26

. Guideline Five provides that adjustment cannot violate the Constitution or any federal statute. The Secretary concluded that because he had reached a decision not to adjust based on other factors, "legal considerations did not provide a basis" for his decision. Decision at 2-65.

Guideline Six mandates that if adjusted counts could not be published by July 15, 1991, a determination would be made against adjustment. Although adjusted counts were ready to be published by July 15, 1991, the Secretary had concluded not to adjust, and so this guideline became moot.

27

. Additionally, I note that in light of recent improvement in statistical tools and the practical benefits that the 1990 PES has provided, the use of adjustment in the next census is probably inevitable.

28

. On July 15, 1991, the day the Secretary announced his decision, he also released to the public the adjusted census data at the national, state, county, and city levels, but not the block level. Subsequently the Department disclosed half of the adjusted block-level data to Congress.