Politicizing the Census

ith the economy hum- ming merrily along, public dissatisfaction with the federal bureaucracy has lessened markedly. Recent poll-ing shows that most departments and agencies command approval ratings of more than 60 percent, up from about 50 percent in a comparable 1995 survey.

But never underestimate the ability of government's political overseers to undermine the image of its career professionals. Just consider the decennial census--the most egalitarian federal enterprise of all, in which each and every one of us expects to be counted.

It now appears--thanks to a 5-4 January Supreme Court decision and continuing partisan warfare in Congress--that there may be two separate census counts next year: one for reapportioning the House of Representatives and another for allocating some $180 billion in federal aid and redrawing legislative districts within state boundaries.

In the event that funds are made available for such a "dual count" strategy, the Census Bureau, a 13,000-person agency highly regarded for its statistical expertise, will end up in the awkward position of tallying the population by two methods; one that it believes in and one that it doesn't.

"Our position is that there is only one best number," the bureau's director, Kenneth Prewitt, told reporters after assuming the office last November. But now he's under pressure to produce more than one. The court demands a traditional head count while the Clinton administration continues to press for a statistically adjusted enumeration.

Prewitt, a political scientist who has headed a think tank and run an opinion research center, strongly favors the statistical approach and takes offense at Republican rhetoric suggesting that sampling would permit the census to be politically manipulated. "There is no way to make it happen," he argues, adding that such "far-fetched" charges discredit the professionalism of the Census Bureau.

Without sampling, the 1990 census missed 8.4 million people and double counted about 4 million, for a net undercount equal to 1.8 percent of the population. It was a step backward from the roundly criticized 1980 census that came up short by 1.2 percent of the population. Congress responded by demanding better in 2000. But when the Census Bureau devised a plan for a more accurate head count employing statistical extrapolations, legislators went into parox- ysms of second thoughts. Census appropriations have been delayed and the bureau's managers have been hampered by endless oversight inquiries.

What's nominally at issue is the proposed use of statistical sampling to correct the differential undercount that historically has resulted from the omission of people who are difficult to find and the double-counting of members of families with more than one residence.

What's really at issue, of course, is raw politics. The people that are most likely to go uncounted tend to be low-income minorities--a largely urban constituency important to Democrats. The people most likely to be counted twice are affluent whites who add to Republicans' suburban base. Sometimes opposition to correcting the undercount is bipartisan.

To many, objections to sampling appear ludicrous, coming as they do from elected officials who rely heavily on public opinion polls that draw sweeping conclusions from small samples. The Census Bureau's strategy, by contrast, would make marginal corrections on the basis of massive data. Indeed, the Supreme Court's ruling notes that Congress, in amending the census law in 1976, "changed a provision that permitted the use of sampling for purposes other than apportionment into one that required that sampling be used for such purposes, 'if feasible.' " (The court reasoned, however, that the provision precludes sampling for reapportioning congressional seats.)

The bureau's proposal called for an actual count--through mail responses and door-to-door interviews--of no fewer than 90 percent of the residents in each of the nation's 4 million census tracts. Statistical methods would then have been used to infer the remainder. More important, the initial results were to be checked and corrected through a separate sampling process involving new interviews at 750,000 households across the nation.

Last year's test run for the 2000 census confirmed that traditional counting methods missed significant numbers of minorities, children and renters. In Sacramento, Calif., for example, sampling produced a count that correlated far more closely with state population estimates that are based on growth projections and administrative records, such as utility hookups and driver's licenses.

Prewitt worries about the consequences of permitting the census to be publicly perceived as a political football.

He clearly does not relish being put into the foolish position of conducting a census that produces two different results. At stake, he suggests, is not just "the quality and integrity of the decennial census, but confidence in the government itself. If the government can't do this well, what can it do well?"

Dick Kirschten can be reached at dkirschten@govexec.com.