Working Towards a Sensible Census

Ever since William the Conqueror compiled the Domesday Book in eleventh century England, census officals have had to face the fact that they cannot possibly include everyone in a national roll call. The problem has continued through the ages up to today, as current government people-counters realize that the U.S. Census Bureau's decennial national head-count is in no way completely accurate.

But a method developed recently by members of the Harvard Statistics Department may make the American census, in which officials estimate more than 1.4 percent of the 1980 population was either most or undercounted, a more exact process, and in doing so, these researchers may affect the way congressional disticts are determined and the process by which federal funds for social services are allocated.

People Tagging

"It builds on the idea of the capture-recapture" statistical technique, Donald B. Rubin, chairman of the Statistics Department, says of his census modification technique. Rubin explains the new procedure as similar to the method biology researchers use to study bird and animal populations. "It's like tagging fish in a pond, releasing them, and catching them again to see how the population has changed," he says.

Members of the Statistics Department, who have been working on this project for more than a year, propose selecting population segments of 300,000 homes in various regions across the country and then intensively studying their group dynamics after a normal U.S. census is taken by mail, said Rubin, who has been at Harvard for four years. The scientists, who include Rubin, Assistant Professor of Statistics Hal Stern and four graduate students, then use the results from these segmented studies to adjust the census' total population count.

"We would take the information from the census, which people mail in regarding numbers [of persons living] in houses, and would then send people to the same houses to take a new, more exact count. Then we would compare the two counts and arrive at a different adjustment factor for each population block--such as Black males or white, elderly women. By applying those numbers to national statistics, we would adjust for the names missed in the national mailing," says Rubin, who has worked with the Census Bureau on statistical problems several times over the last 10 years.

10-Year Census

Currently, the U.S. holds a census every 10 years in which most of the people-counting is done by mail, according to Tom Belian, a graduate student working with Rubin on the project. Every American household receives a census form, which reports the number of people living at their home. "It's a huge undertaking, costing over $1 billion," Belian said. The Census Bureau has never before adjusted its population data.

Unfortunately, Rubin admits, solving the undercount problem involves more than just locating lazy census respondents. He says that the Department of Commerce realizes it misses those people who intentionally misrepresent the number of people in their households on their reports, as well as those who do not mail in any census forms at all.

"A lot of times people who live in overcrowded apartments, where there are limits on the number of residents, will say there are fewer people in their households because they think somehow Uncle Sam will notify their landlords and get them thrown out," Rubin says. "Sometimes women will also omit their husbabnds' names from the census because they think they will get more welfare money to take care of their children."

Rubin says evasive tactics such as the above examples skew population data, particularly those pertaining to minorities. Officials estimate that six percent of all Blacks were under-counted in the 1980 census, but that only 0.5 percent of all whites were missed. Rubin estimates that the census is losing track of more than 15 percent of Black males in some areas.

"We know [the uncounted citizens] are born and that they go to school, but after that they disappear for about 20 years and resurface in census counts much later," says the Princeton graduate.

Rubin says that estimating the exact number of people missed in a census is important to government and social service officials for two reasons. "One has to do with money, and the other has to do with legislation in Congress," he says.

Funding Problems

If a city's population is undercounted, the federal government will not give the metropolis enough money to provide various services for all of its residents.

"What is happening now is that those cities must care for all their citizens but they don't get a proportional amount of money [to do so]," Rubin says.

Representatives from New York City, for example, are rallying in support of a bill sponsored by Rep. Mervyn Dymally (D.-Calif.), which will force the Census Bureau to use statistical adjustments in the 2000 census. The bill is receiving such support, according to The New York Times and The Wall Street Journal, because New York City officials estimate that the 1980 census missed more than 800,000 New York residents, causing the metropolis to lose $25 million in federal funding.

In other states, such as California, where a high number of illegal immigrants refuse to respond to the national census, cities are not receiving enough federal funding to provide adequate social services for their residents, since federal money is doled out based on a falsely low federal head-count.

"The mayors feel it is to their advantage to adjust the census. Right now they are getting fewer dollars, even if the census misses by just 1 percent," says Howard Hogan, a member of the Commerce Department's statistics division.

Rubin emphasizes that the Census Bureau does not share any of its population information with other federal departments, such as welfare agencies.

District Deviance

Congressmen are arguing that statistical adjustments be made to reform current census counting problems for another reason. They say that without fair adjustment for incorrect population data, state congressional districts are inaccurately drawn, leading to misrepresentation in Congress.

In 1980, for example, the state of Indiana argued that if the census count were adjusted in a more accurate fashion, such as according to Rubin's methods, it would have received an additional seat in the House of Representatives and Florida would have lost a space.

Indiana sued the Census Bureau because of this problem seven years ago, though the case was decided in favor of the Bureau in 1983.

In addition, some Democratic congressmen argue that because districts which are not counted accurately tend to be inhabited mostly by minority populations and because those areas are thus underrepresented in Congress, the Democrats are not receiving the representation they deserve.

A Political Struggle

Efforts to readjust the 1990 census have caused much political debate in the past few months.

Last October, the Reagan Administration decided to reject Rubin's new statistical methods and will not adjust the 1990 census data. James Gorman, a spokesman for the Department of Commerce, which oversees the Census Bureau, explains, the "point was that the Census Bureau would be open to the accusation that it was manipulating figures. Also, no one was sure if [the new technique] would be operational by 1990. Some people say it's not possible to take an accurate survey of 300,000 after the 1990 census and to give it to the president, according to law, by 1991."

David Freedman, professor of statistics at the University of California at Berkeley, testified last summer to the House Subcommitee on the Census and Population that new statistical techniques are more difficult to implement in a census than it may first appear.

"The answer is another clever idea: matching sample persons against the census, to see if they were counted. That is easy to say, but not so easy to do, because the sample will have about 300,000 persons, and the population will be around 250,000,000. The practical result is that a sample person can go unmatched by error," Freedman says.

"For example, if there are two different spellings of a surname, or an address was written down wrong, or the person moved, matching becomes significantly more complicated," he says.

There are others who charge the current administration is rejecting census readjustment for political reasons, because a statistical modification of population numbers may give the Democrats more power in the House, according to a graduate student working with Rubin.

But Vincent Barabba, former director of the Census Bureau under the Nixon, Ford and Carter Administrations says that adjusting undercounted groups is generally advisable--at least for future censuses.

"I would have suggested to them that they use the research but not use it until the [1990] census was over. They need to look at the figures before proposing an adjustment, but it should be a [permanent policy in the future]," Barabba says.

In the face of such arguments, however, the Reagan Administration continues to stand firmly against census modification, and whether or not the new research will ever be implemented remains a mystery to Rubin and his cohorts.

"The census bureau says it could not do it in 1980 because it didn't have the tools," Rubin says. "Now we have the technology, and it is just a question of the timing." Population Category  1950  1960  1970  1980 Total population  3.3  2.7  2.2  0.5-1.4 Male  3.8  3.3  3.1  N.A. Female  2.8  2.2  1.4  N.A. Legally resident population  N.A.  N.A.  N.A.  0.5 Male  N.A.  N.A.  N.A.  1.5 Female  N.A.  N.A.  N.A.  -0.4 Black population  9.7  8.0  7.6  5.3 Male  11.2  9.7  10.1  8.0 Female  8.2  6.3  5.3  2.7 White and other races population  2.5  2.1  1.5  -0.2 Male  2.8  2.5  2.1  0.6 Female  2.1  1.7  0.9  -0.9

NOTE: A minus sign indicates net overcount.

N.A. = not available; difference between total and legally resident population probably negligible (except for 1980).

Lower percentage assumes presence of 2 million undocumented aliens in the estimated population; upper percentage assumes presence of 4 million undocumented aliens. The census population used in calculating the total population rates is the actual count, including an estimated 2 million undocumented aliens that were counted.

Blacks and other nonwhites.

White only.

Net undercount rates by race and sex from demographic analysis, 1950 to 1980 decennial censuses (estimated population minus census population as a percentage of estimated population).