Sampling Is Not the Answer to Census Flaws

Peter Skerry
Peter Skerry Former Brookings Expert, Professor of Political Science - Boston College

February 5, 1998

In the emerging debate over adjusting the 2000 census, the battle lines are clear. Advocates of adjustment argue that the U.S. Census Bureau’s long standing (and undeniable) undercounting of blacks and Latinos is intolerable and can be painlessly remedied with sophisticated statistical techniques. Opponents are playing the role assigned them: the self-interested, if not selfish, defenders of the status quo unwilling to use advanced scientific methods to reach out to the least advantaged.

Here, I argue not only against adjustment of the census count, but against the terms of this debate.

Missing the point

Advocates of census sampling wildly exaggerate the interests of the undercounted in adjustment; they similarly exaggerate the capacity of science to resolve fundamental political conflicts. Yet the partisan interests in the complex and dynamic processes that would be unleashed by statistical adjustment are not as clear as most likely believe. Moreover, all parties to the debate seriously underestimate the risks of adjustment.

Journalists tend to emphasize that the undercount affects the allocation of $59 billion in federal grants that rely on census data. Yet as a National Academy of Sciences study has noted, only .32% (about $190 million) of total federal obligations would be shifted by correcting the population count.

As for the political stakes, again they are substantial but hardly self-evident. To be sure, if the 1990 census had been adjusted, some states would have gained and some would have lost a congressional seat. But the identity of the winners and losers is not clear.

First, corrected numbers based on a complicated set of adjustment factors would have to be developed for each of 7 million census blocks in the nation. Moreover, under the complex formula used to reapportion Congress, a population increase does not necessarily increase a state’s chances of gaining a seat. Under such conditions the National Academy guestimates that adjustment would have resulted in either one extra, or one less, congressional seat for California.

But even if California’s fortunes were a certainty, would an extra congressional seat for the state be in the partisan interests of Republicans or Democrats? An answer obviously depends on a number of variables that make the question almost imponderable in the abstract—precisely where it now resides. If this is true with regard to congressional reapportionment, then the myriad lines to be redrawn under congressional, state and local redistricting render the calculation of interests under adjustment even more opaque.

Who is complaining—and why?

What, then, is all the fuss about?

First, the Census Bureau is an easy target for political entrepreneurs. The Bureau of Labor Statistics (BLS) issues unemployment numbers that have flaws arguably as critical, if not more so, to the disadvantaged as census undercounts. But unlike BLS, the Census Bureau does not have many advocates in Congress. Census data are classic public goods around which relatively few constituencies are organized. One would think that the decennial census is of intense interest to House members facing reapportionment and redistricting. So, it is—but only once every 10 years.

Second, the undercount resonates deeply with racial minorities, who recall that the Constitution originally decreed each slave to count as three-fifths of a person. What minorities do not acknowledge, which is typical of the complexities of census politics, is that this three-fifths clause was actually a defeat for slave-holding states. They had sought “full representation” for slaves, which would have afforded those states more seats in Congress.

Finally, advocates have succeeded in presenting census sampling as a scientific silver bullet that would provide one indisputably correct set of numbers and obviate all political disagreements. Yet politics necessarily pervade the census, though this seldom, if ever, leads to finagling raw data to produce desired results. The essence of census-taking is drawing boundaries around things for which there are no natural categories—whether around a metropolitan area or a racial group. Such judgments are where politics enters the equation. As a demographer at the Census Bureau once said to me, “We don’t do science around here. All we do is glorified accounting.”

Adjustment is bad public policy

What then is to be done? Census adjustment is bad public policy. But even if the case for it was much more solid, the risks of adjustment have not been adequately addressed.

I have yet to hear the “scientific” answer to the elected official who discovers that adjustment has left one of those 7,000 census blocks in his jurisdiction with fewer residents than he and his constituents know to be living there. Adjustment will provide more accurate data for undercounted groups at high levels of aggregation, but with small-area data, there will be hell to pay.

Another risk involves the possibility of an incorrect adjustment. Advocates never mention it, but in 1992, census officials discovered that the original adjusted numbers that Commerce Secretary Robert Mosbacher had rejected for reapportionment a year earlier were found to be incorrect. Just imagine the imbroglio if the 1990 reapportionment had used adjusted numbers that then needed to be readjusted.

Finally, the greatest risk of adjustment has been completely ignored by everyone. Right now, the Census Bureau is the object of vigorous lobbying by organized minorities. Such efforts have led to the creation of a specific Hispanic-origin question on the census that is not technically necessary. Once again, politics pervades the census.

So be it. But once statistical adjustment is in place, the political action will shift from minorities pressuring the bureau to reach out to their communities (and urging their communities to get themselves counted) and toward much more complicated, technical issues about sampling frames, cutoff points and standard errors.

Politics will hardly disappear but will become ever more arcane and technocratic. The process will be as open to view as a glass-partitioned, air-conditioned computing center can be. But the vast majority of us, especially the most disadvantaged and least sophisticated, will necessarily be on the outside looking in.