Los Angeles Times

Census; A Bad Test for Bush's Outreach

Remember the census? The complaints last April about the intrusive long form? The angry denunciations of questions about race and ethnicity? Compared with the Florida recount battle, the census may seem a mere blip on the political radar screen. Yet, that may be about to change.

On Dec. 28, the U.S. Census Bureau announced the initial results of its 2000 head count, as required by law. The state totals are necessary for apportioning the House of Representatives. The U.S. Supreme Court has ruled that these numbers may not be statistically adjusted to compensate for minority undercounts. But the court left open the possibility of adjusting the subsequent, more detailed numbers that are used for redistricting within each state. The Clinton administration has been preparing to make both adjusted and unadjusted numbers available for this purpose.

The question of whether to use adjusted numbers for redistricting will fall to the new Bush administration, as it did 10 years ago to President George Bush. But unlike his father, President-elect George W. Bush will be under heightened pressure to "do the right thing" by minorities. In the aftermath of reports that some minorities were deprived of their vote in Florida, the public will be more receptive than ever to the idea that basic fairness dictates that all Americans be included in the census.

Census adjustment will be the ideal issue for liberals to pressure the new administration into demonstrating that it is indeed pragmatic and eager to reach out to minorities. The reasonableness of adjustment will be made all the more evident by the vociferous opposition of Republicans, whose already thin House majority was narrowed by November's results. But politics aside, the new president should reject adjustment as bad policy. To see why, look again at election 2000.

Advocates of adjustment draw a parallel between getting counted in the census and voting. But this does not withstand scrutiny. While there is a right to vote, there is no such right to be counted. On the contrary, cooperating with the census is a legal obligation imposed on all residents of the United States. Furthermore, while it is true that federal grant programs rely heavily on census data, the formulas determining who gets funded are shaped by the political clout that comes from voting, not from being counted in the census.

Adjustment is also urged as a purely scientific undertaking based on well-established methodologies, such as survey sampling, that nonexperts have come to understand and accept. But here, too, recent events have something to teach. However "scientific" the exit surveys on which they were based, predictions of which presidential candidate won Florida on election night were clearly mistaken. So, too, were the results of numerous preelection polls.

To be sure, professionals who designed these surveys reminded us that predictions about such a close election were squarely within the margin of sampling error. In theory, we understood this. But listen to one savvy political commentator venting her exasperation on election eve: "I have gone through 25 years as a political junkie under the impression that the familiar 'margin of error' referred to the spread between the candidates and not to each candidate's percentage."

Census adjustment would be worse. Indeed, the process of adjustment is massively more complicated than the simple "sampling" referred to by advocates. Yes, the first step is a huge post-census sample survey, undertaken several months after National Census Day on April 1. But unlike conventional samples (such as exit surveys and preelection polls), this one is not used to make inferences about the larger population from which it is drawn. Instead, the results of the post-census survey are laboriously matched against those of the original head count, for the purpose of determining who was missed.

At least, this is the theory. The practice is another matter. Census adjustment does not rely on white-coated technicians working under controlled laboratory conditions. Quite the opposite. Like the census itself, adjustment relies on human beings going out and asking other human beings questions. Enumerators do not always do their jobs conscientiously, as was evident in 1990, when heat waves in Atlanta and Denver resulted in many falsified responses to the post-census survey.

Nor are Americans easy to count. Many hard-to-count individuals actively evade the census. Others happen to move between census day and the day of the survey, and because the census cannot rely on "identifiers" like Social Security numbers, it has no reliable means of tracking such moves. All this creates tremendous matching problems, like those in 1990 that resulted in 1 million individuals being mistakenly added to the undercount estimate. Had the Bush administration used those results to adjust the census, only to have the error discovered a year later, the result would have been massive confusion.

And cynicism. The Census Bureau has acknowledged that adjustment would introduce new sources of error into the process. Just as in Florida, the effort to refine such huge numbers can actually make things worse, especially when the differences that matter are relatively small—like minority undercounts or the recent election results.

More than a decade ago, Harvard demographer Nathan Keyfitz characterized census adjustment as a "gamble." In the current political environment, that gamble looks riskier than ever.