Introduction

THE MYTH: STATISTICAL ADJUSTMENT WILL CORRECT SEVERE UNDERCOUNTS IN PREDOMINANTLY MINORITY NEIGHBORHOODS

For more than 10 years, the country has debated the use of “statistical adjustment” to resolve the problems of undercounting endemic to the decennial census – particularly in minority neighborhoods. The January 1999 Supreme Court ruling in large measure settled this debate by requiring the Census Bureau to attempt a full (100 percent) enumeration for the apportionment of seats in the U.S. House of Representatives.

The Administration and the Census Bureau, however, have insisted that statistical adjustment be used for all other purposes, including the allocation of federal and state funds for a variety of uses, from roads and schools, to health care and community development. The primary purpose of statistical adjustment is “to measure and correct overall and differential coverage of U.S. residents in Census 2000.”¹ Presumably, this includes correcting the disproportionate undercount of African Americans, Asians, Latinos and Native Americans.

However, a new study, undertaken by the Congressional members of the U.S. Census Monitoring Board, reveals the ability of statistical adjustment to correct the undercount has been wildly overstated. The evidence clearly indicates that:

Statistical adjustment will not correct large undercounts or overcounts in small areas such as blocks and neighborhoods.
Heavily undercounted areas will remain heavily undercounted.
Statistical adjustment will add people to many overcounted areas (areas where the Bureau mistakenly counts too many people).
Until the census is improved in local areas that are heavily undercounted, the differential undercount will persist at the local level.

Note, statistical adjustment may well “correct the undercount” in a global sense – that is, improve the overall, national “count.” But the practical point of the census is to apportion the count correctly. That is, “the count” is a numerical product, which may well improve through even random additions to population. But “the count” also represents real people, with real political rights and needs. A statistical adjustment that “corrects” the count nationally may do nothing palpable for the needs of actual people living in undercounted areas. Unless statistical adjustment is distributed correctly, such that people are added where and only where they are missed, the Bureau will not have remedied the failure of fairness that statistical adjustment is intended to address.

The Board’s review of Bureau data shows the local benefits of adjustment are more myth than methodology – a statistical promise to cure the problem of local undercounts which cannot be kept. Even worse, for local leaders who believe in the promise of an accurate census, the illusion of a proper adjustment could encourage a false sense of security concerning the ability of statistical adjustment to correct local undercounts.

BACKGROUND ON THE U.S. CENSUS MONITORING BOARD STUDY
Statisticians debate whether statistically adjusting the census will provide a better picture of the population at the national or state level. Many statisticians and demographers have published extensive analysis and criticism of the Census Bureau’s attempts to measure the undercount, questioning the reliability of the Bureau’s methodology even at the national or state level (Appendix E).

Everyone agrees, however, that statistical adjustment gets increasingly less accurate at lower levels of geography. Everyone also agrees that political power and money are distributed among local areas defined by geographic or political boundaries, and that large undercounts are distributed unevenly in these areas throughout the country. Many blocks and neighborhoods have large undercounts – undercounts greater than 10 percent.

The Bureau proposes to correct these undercounts through statistical methodology – adjustment. Adjusting the census using statistical methods involves two distinct operations: measuring the undercount of different groups of people, and adding people in the right places to correct such undercounts. There is strong reason to believe that both parts of the proposed methodology will fail.

This study conducted by the Board focuses on the second of these two tasks: adding people in the right places. Thus, for the purposes of this analysis, the survey’s direct measurements of undercount are treated as if they were accurate. This analysis focuses on whether statistical adjustment using the so-called “synthetic method”² would have succeeded in correcting undercounts and overcounts identified by the sample survey at the lowest levels of geography – blocks and neighborhoods.

The Congressional members of the Board requested detailed data in late 1998, which were only provided by the Bureau after four months’ delay. To our disappointment, the Bureau resisted the release of these data to public and academic review for almost a year. Only recently, after repeated requests from the Congressional members of the Board and the U.S. House of Representatives’ Subcommittee on the Census, did the Bureau agree to release most of these data.

We analyzed the 5,170 local areas surveyed nationwide in the Bureau’s 1990 post-enumeration survey (PES).³ The PES is virtually the same as the Accuracy and Coverage Evaluation (ACE) survey the Bureau plans to use for Census 2000.

This study found that statistical adjustment does not perform adequately at the local level.. Regardless of how “accurate” the national picture may appear, statistical adjustment adds far too few people to heavily undercounted local areas to assure these communities get their fair share of representation or public funding. In addition, adjustment adds people to many overcounted areas – making overcounts worse.