The 2000 census is the most important in decades. Subject to controversy and litigation, fraught with uncertainty, it is also estimated to cost an unprecedented $6-7 billion. And while the Census Bureau missed hundreds of thousands of Hispanics, American Indians, Blacks, and Asians in 1990 (see Table 1), the 2000 census could fail to count even more, leading to renewed debate over the proposed remedy of sampling and adjustment.
The Census Bureau will produce two sets of numbers for 2000: an unadjusted set based on the traditional headcount, and an adjusted set based on a huge post-census sample survey. In January 1999, the Supreme Court ruled that adjusted numbers may not be used for apportioning Congress. But they may still be used for redistricting, federal grant allocations, and other purposes. And, as in 1990, the very existence of two sets of census numbers will fuel debate.
Undercounting minority groups has symbolic significance, especially given that the U.S. Constitution originally required each slave to be counted as only three-fifths of a person. But the fiscal and political stakes in this battle have been grossly exaggerated by all parties. And while the minority undercount might in fact be getting worse, the debate over its significance has been highly distorted, even misleading.
POLICY BRIEF #56
First, the fiscal stakes. To be sure, census data are critical in the allocation of federal dollars. In fiscal year 1998, for example, federal grant programs totaling $185 billion relied on census figures. But the GAO concluded that for fifteen such programs (accounting for four-fifths of that dollar total), only $449 million (0.33 percent of allocations by these fifteen programs) would have been distributed differently among the states as a result of adjusting the 1990 census.
How can this be? Population is only one of several factors in most federal grant formulas. Many programs designed to help distressed communities actually reduce funding when population increases. Even when population gains lead to gains in funding, the critical factor is not absolute gain but gain relative to other jurisdictions. Many jurisdictions could register population increases but still end up worse off, because their increases would be less than those of other jurisdictions. And because grant programs typically have funding ceilings, adjustment would result in a fixed pie being divided into smaller pieces.
Similarly counterintuitive outcomes pertain to minorities. The 1990 post-census survey (a huge sample on which adjustment was to be based) apparently found proportionately more Hispanics in the barrios of the Southwest than blacks in the cities of the Northeast and Midwest. Hard to prove either way, such findings nonetheless cast doubt on the widespread assumption that adjustment would benefit all undercounted minorities equally.
Similarly exaggerated have been the partisan implications of adjustment for legislative redistricting. All participants in the debate have assumed that adjustment would help Democrats and hurt Republicans. Yet these outcomes are impossible to predict. The drawing of Congressional district lines in 50 different states is subject to myriad local political factors. And even beyond the demographic vagaries, no one knows how each of these political scenarios would play out under adjustment.
Table 1:
Net Undercount of Minority Groups, 1990
Net Undercount (%) Total Population 1.58 Asian or Pacific Islander 2.33 American Indian 4.52 Black 4.43 Hispanica4.96 Source: Committee on Adjustment of Postcensal Estimates, Assessment of Accuracy of Adjusted versus Unadjusted 1990 Census Base for use in Intercensal Estimates (Census Bureau, August 7, 1992).
a. Hispanics can be of any race.
As for the assumption that more individuals from minority groups would translate into more Democratic voters, it is belied by the Republican leanings of many Hispanics and by the low registration and voting rates among minorities generally. If we focus not on voters but on sheer numbers, we cannot assume any necessary benefits to Democrats. When it comes to redistricting, counting additional people from minority groups can threaten the seats of non-minority Democratic incumbents—and, in the wake of the Voting Rights Act, create more minority officeholders whose demands may alienate non-minority voters.
Conversely, Republicans should not feel threatened by additional minority people counted as a result of adjustment. During the 1990 redistricting cycle, Republicans perfected the art of packing minorities into “majority-minority” districts that had the dual effect of weakening or eliminating non-minority Democratic incumbents and creating homogeneously white, Republican-leaning districts. We should see more of this in 2000, when Republicans will almost certainly control many more state legislatures and governorships than they did in 1991.
Of course, the real redistricting battleground after the 2000 census will be California. Recalling the beating they took with a Democrat-controlled redistricting in 1980, Republicans look apprehensively at current Democratic control of both legislative chambers as well as the executive in the nation’s most populous state. Yet the situation is very different now than it was in the recent past. Even without adjustment, in 1990 California gained seven new congressional seats, some going to Republicans after a tough redistricting battle. But current projections have the state gaining only one new seat in 2000.
Would census adjustment change this scenario? It might appear so, to judge from the huge size of the California undercount: 838,000 in 1990. If the 2000 redistricting were to rely on adjusted census numbers, California’s Democrats might be able to carve out an additional congressional seat for themselves. But this would be easier said than done, because the undercount tends to be distributed remarkably evenly across districts, and because even the most skillful redrawing of lines eventually collides with the obdurate fact that high-undercount districts tend to have extremely low voter turnouts. This should come as no surprise. Many of those uncounted in the census—minority individuals, immigrants, non-citizens—do not or cannot vote. And without votes, numbers alone cannot contribute to the partisan outcomes sought by those redrawing district lines.
The main reason for Republican jitters in California is Democratic control of redistricting, not any potential Democratic windfall from adjustment. It is understandable that Republicans holding down a slim majority in Congress are not reassured by the news that there is no sure way to forecast the marginal partisan consequences of census adjustment. But their reluctance to gamble on those consequences hardly justifies the utter certainty with which virtually all commentators assert that adjustment would help Democrats and hurt Republicans.
Many Republicans oppose census adjustment because they associate it with affirmative action. But here, too, appearances can be deceptive. It is undeniable that census questions about race and ethnicity are part of the administrative apparatus of affirmative action. Judges and administrators rely on census data to determine hiring goals, for example—a fact well understood by minority leaders concerned about the undercount.
Yet this does not mean that affirmative action offers incentives for individuals to identify themselves as protected minorities on the census. Higher census counts do translate indirectly into more benefits for designated minority groups. But as individuals, the members of those groups do not benefit directly. A student may well increase her chances of admission by checking the box marked “Hispanic” on a college application. But she receives no such benefit by checking the “Hispanic” box on the census questionnaire. It is a telling irony that if there were some tangible reward for identifying as a protected minority on the census, the minority undercount would probably be lower.
The stakes of adjustment are cloudy, but the risks are clear. Of particular concern is “participation meltdown,” the prospect that (in the words of a National Academy of Sciences panel) many people would “assume that they do not need to respond by mail because the use of sampling means that their participation (in the census) makes no difference to the results.”
It is also virtually certain that adjustment would introduce new error. What is involved here is not just the familiar technique of survey sampling but rather a procedure intended to match the results of a post-census sample survey (consisting of several hundred thousand respondents) with the original census results. By comparing the responses of the individuals in the post-census survey with the responses of those same individuals in the actual census—a process known as dual-system estimation—statisticians can produce undercount estimates for various demographic groups.
Unfortunately, dual-system estimation is hugely complicated and error-prone. One big problem is tracking down the many people who have moved to a different place between Census Day on April 1 and the post-census survey a few months later, which is doubtless why in 1990 the chief opponents of adjustment within the Census Bureau were not the statisticians but those responsible for the operational side. Indeed, the Census Bureau has acknowledged that its 1990 adjustment methodology introduced substantial new error into the results.
One such error had major repercussions in 1990, when a complicated mismatching added 1 million people to the undercount estimate, a glitch that went undiscovered for more than a year. Commerce Secretary Robert Mosbacher had already decided not to use adjusted numbers. But if the erroneous figures had been used to adjust and then deemed incorrect a year later, the Bureau would have been exposed to any number of legal challenges, and, more important, public confidence in the census would have been undermined. About a similar but relatively minor incident, then Census Director Barbara Bryant recalls that “technical errors were impossible to explain to the public.”
Beyond exaggerating the stakes and neglecting the risks, the participants in the debate engage in highly misleading rhetoric. Two prominent themes sounded by advocates of adjustment are science and rights.
Taking science first, it is clear that statistical sampling has a scientific basis. But census adjustment is hardly a controlled test-tube procedure performed by white-coated technicians in a laboratory. Rather, it is a massive logistical undertaking that requires practical political judgment at every step, not just at the operational level but also at the most technical level. Scientists disagree about census adjustment in part because it involves countless decisions that, while entwined with technical and scientific issues, must be made in a realm outside the expertise of statisticians or other scientists.
An example would be the follow-up plan proposed by the Census Bureau in 1996 to achieve a more accurate count of undercounted areas in 2000. In this plan, enumerators would make personal visits to non-responding households until 90 percent of households in each county were tallied, at which point a sampling procedure would be used to estimate data for the remaining households. Known as “sampling for non-response follow-up,” or SNRFU, this plan came under criticism by black leaders who feared that in huge metropolitan counties such as Los Angeles, the 90 percent cut-off point would be reached by focusing on easy-to-count neighborhoods and bypassing minority neighborhoods with low response rates. To remedy this, minority leaders lobbied for a modified SNRFU that would obtain responses from 90 percent of households not in each county but in each census tract. The Black Congressional Caucus introduced legislation to enforce their views, and, after some resistance, the bureau met their demands.
This episode revealed two things about the “science” of census sampling. First, from a purely statistical perspective the cut-off point could have been much lower than 90 percent. The bureau chose the 90 percent figure on the basis of focus group results indicating that anything lower would undermine public trust. Second, the bureau resisted the minority leaders’ demands mainly for fiscal reasons, having proposed SNRFU in the first place as a way to reduce costs. Thus was SNRFU guided at every turn by political, not scientific, considerations.
Nor does science produce authoritative answers to the necessarily political questions in which the census is embroiled. In the controversy over adjustment of the 1990 census, responsible advocates acknowledged that adjustment would not have improved accuracy in all jurisdictions, but would have done so in most cases, on average. In fact, Census Director Bryant admitted at the time that adjustment would actually lessen accuracy in some jurisdictions.
Clearly, in the view of many statisticians, getting more accurate data on average justified adjustment scientifically. But with winners and losers, adjustment becomes more than a scientific question. One might well decide that adjustment-produced gains for minorities outweigh the losses to nonminorities. But this is hardly a statistical or scientific question. It is a political question—and one for which statisticians have no particular expertise.
This is as it should be. Indeed, the census should not be thought of as a scientific undertaking but as inherently and properly a political undertaking. This was how the Framers originally conceived of it. Embedding the census in a political architecture designed to balance interests, they intended it to be used both to apportion congressional representatives and to levy direct taxes. As explained in Federalist No. 54, “The establishment of a common measure for representation and taxation will have a very salutary effect….By extending the rule to both objects, the States will have opposite interests, which will control and balance each other, and produce the requisite impartiality.”
Despite a remarkable degree of autonomy in the twentieth century, the Census Bureau remains political—not in the sense of “cooking the numbers” but in the sense of drawing the boundaries that define the American polity. Part of this boundary-drawing involves establishing and maintaining the racial and ethnic categories into which the census asks people to place themselves. Because this boundary-drawing is unavoidably political, statisticians disagree about census issues.
This is not the impression given by adjustment advocates, the Census Bureau, and the media. They emphasize the consensus reached by three panels convened since 1990 by the National Academy of Sciences (NAS). We hear quite often that these panels, not all of whose members were statisticians or even social scientists, studied and approved of the Census Bureau’s plans to statistically adjust the 2000 census. We almost never hear that other statisticians have strenuously opposed adjustment. While policy analysts and knowledgeable insiders may agree on the desirability of census adjustment, such “expert consensus” is not at all the same as “scientific consensus”—which does not exist.
Proponents of adjustment also adopt the language of rights. Declares New York Senator Charles Schumer, “The Constitution, of course, guarantees the right of every person residing in the United States to be counted.” Nor is this any ordinary right in the view of those who share Schumer’s outlook. The clear racial dimension of the undercount leads advocates to frame the issue in terms of civil rights and minority empowerment.
Such claims do not bear scrutiny. Historically, participation in the census has been construed not as a right but as an obligation mandated by law. Even F.D.R., 160 years after the Framers tried to balance the carrot of representation with the stick of taxation, described cooperation with the census as “one of the requirements of good citizenship.”
But most misleading is the notion that being missed by the census is tantamount to being disenfranchised. Curiously, this notion is embraced on both sides of the controversy. Advocates of adjustment argue that because the uncounted are not included in the one-person-one-vote calculations on which district lines are drawn, being uncounted amounts to being deprived of the vote. Opponents adopt a similar logic, arguing that where adjustment would result in decreased counts, the enumerated individuals thereby eliminated would lose their votes in the equivalent of election fraud.
These arguments may work rhetorically, but they rest on the dubious assumption that uncounted individuals are somehow prevented from voting—which is clearly not the case. There is nothing to stop such individuals from voting or from organizing others to vote. We see here a curious notion of political power, one that reduces political muscle to sheer numbers. Obviously, numbers are important. But more important is what people do with those numbers. For any given group, factors such as age, wealth, spatial concentration, and level of organization make a crucial difference in their ability to translate raw population totals into political power.
Meanwhile, those who are enumerated but do not vote cannot be said to be enfranchised in any meaningful sense just because they were counted. Adjustment advocates in particular seem to assume that population totals translate into power independently of political effort. In an era of concern about excessively thin notions of civic obligation, this is a strikingly anemic notion of citizenship.
Looking ahead to the 2000 census and beyond, we shall have to find ways to improve the census, especially with regard to counting minorities, and be prepared to spend whatever it takes to carry out this crucial governmental function. But above all we shall have to scrutinize the ways we think about the census.
First, we need to examine the standards by which we assess undercounts. For more than thirty years, minority undercounts have been deplored without any clearly articulated criteria to evaluate them. Because we cannot reasonably expect any census data to be without error, it would be helpful for adjustment advocates to specify what degree of accuracy they would find acceptable. As in many other policy realms, race greatly complicates the application of standards and, indeed, seems in this instance to foster the implicit adoption of unrealistically high standards. Even so, it would be better if the standards, no matter how rigorous, were made explicit and the reasons for them clearly stated.
Of course, to articulate such a standard would require making a tough judgment call, something that government officials and others in positions of authority seem unable or unwilling to do. For this problem I offer no remedy. But perhaps we might begin by asking whether there is such a thing as a structural undercount, comparable to structural unemployment: a level of undercount below which we cannot expect a free society to go.
For in addition to the Census Bureau’s own failings, it must also be acknowledged that the undercount reflects the fact that some individuals simply do not want to be counted. In our individualistic society where many resist intrusions into their privacy by the state, it is neither fair nor accurate to assign all responsibility for the undercount to the Census Bureau. Nor is it entirely prudent to regard the undercount as, in the words of the New York Times editorial page, a “national injustice.”
Second, we must remind ourselves that there are real limits to the accuracy and reliability of census data on race and ethnicity. The Census Bureau’s own studies reveal that when asked to identify themselves, Americans frequently give different answers at different times. For example, only 96.2 percent of those identifying themselves as “black” on the 1990 census did so on a subsequent bureau survey. The figure for Hispanics was 92.3 percent; for Asian or Pacific Islanders 91.9 percent, and for American Indians 65.1 percent. As sociologist Stanley Lieberson puts it: “We have to live with ambiguity in our census data on ethnic and racial groups.”
Third, we must be mindful that the ambiguity Lieberson speaks of is bound to increase. Not only is immigration introducing an array of linguistic, cognitive, and attitudinal complications into the census process, but assimilation and intermarriage are leading to a growing number of individuals who can claim membership in more than one group. At the same time, race and ethnicity are coming to be understood less as social constructs denoting affiliation to one or more groups than as psychological choices denoting individual identity. In other words, group boundaries are growing fuzzier, group membership more volatile.
These changes create a dilemma for the Census Bureau and for the rest of us. Writes the sociologist Mary Waters, “There is a tension between public policy categories, the categories that we need to do the work of government, and the principle of self-enumeration that people don’t come in neat categories, people come in blended packages.”
Indeed, we must begin to consider that there may very well be a trade-off between the diversity that Americans value and the expectations that we have for census accuracy.