Despite an extensive effort to reach the hard-to-count and the most ambitious attempt ever to enumerate the homeless, the 1990 U.S. census will miss millions of Americans, most of them poor blacks and Hispanics in large cities.
It will double-count millions of other people, often from rural and more affluent areas. And it will create records of hundreds of thousands of people who may not exist.
More Top Picks Best Atv Cargo Boxes
The Census Bureau is aware of the imperfections in its once-every-10-years national censuses, and has spent huge sums trying to achieve greater accuracy.
But errors remain, and the census process is becoming even more difficult. For example, the bureau is scrambling to cope with the surprisingly large percentage of American households (more than 35 percent) that haven`t returned their 1990 census questionnaires.
”It`s just a mean, impossible job when you are trying to count this many people in so short a period of time and have so little funds,” said Philip Klutznick, who, as commerce secretary, oversaw the bureau`s handling of the 1980 census.
Census officials acknowledge that some people are relatively easy to count: people who live in the suburbs, for example, and middle-class people.
Harder to enumerate are those who live on the margins of society. These are people who may have a hard time understanding the complex census questionnaire or who may never see a questionnaire because the Census Bureau can`t find their homes.
J. Michael McGehee, a Republican consultant serving on an eight-member panel to advise the federal government on the 1990 census, said, ”We recognize that some people won`t be counted because they don`t want to be. They don`t want to participate, or live in areas that are not interested in government.”
But others contend the census process is stacked against people outside the mainstream.
”The census is designed by middle-class people for middle-class people,” said Barbara Bailar, the former associate director for statistical standards and methodology for the Census Bureau, who resigned in a dispute over adjusting for the undercount.
”The census procedures do extremely well for middle-class populations, but not for people who have problems with basic housing, poverty and fear of government.”
The search for accuracy in counting the nation`s quarter-of-a-billion people is much more than an expression of scientific fastidiousness. It goes to the heart of American democratic institutions.
”The census is about the most fundamental political act in society,”
said U.S. Rep. William Thomas (R-Calif.).
Flaws in census data distort the distribution of power within the country. They mean that some people, groups and states get more power-and federal money-than their population entitles them to, while others get less.
Reducing those errors is a difficult logistical and statistical proposition. This year, the question is even more complicated because it is being debated not just by statisticians, but also by politicians who stand to win or lose big, depending on how decisions are made.
”This is not a statistical argument; this is a political argument,”
McGehee said. ”This is not a statistical fight; it`s a political fight.”
The central issue is whether to adjust the 1990 census results to compensate for the expected undercount of blacks and other minorities.
It is an issue that illustrates the human failings of the Census Bureau, and one that raises fundamental questions about the government`s counting methods.
Lawyers representing New York, Chicago and several other plaintiffs in a 1988 lawsuit on the undercount, a suit reactivated this month, asserted that the undercount ”is not a quirk; it is an inevitable consequence of census design.”
But the Census Bureau and the Commerce Department argue that an adjustment to compensate for the undercount would mar the purity of the census, and they have made it difficult for an adjustment of the 1990 numbers to be carried out.
”The census is a standard,” said Michael Darby, undersecretary for economic affairs at the Commerce Department, during testimony in January before the House Subcommittee on Census and Population. ”Census numbers do not have standard errors precisely because they are directly taken from a complete count.”
Indeed, the Commerce Department stated in published guidelines concerning a possible adjustment that such an effort would represent the first time the count would be adjusted. But the census process is not a pure head count and hasn`t been for decades.
”The number that is accepted as the true population of the United States is merely a scientific estimate,” wrote former Census Bureau director Vincent Barabba in 1983.
”This estimate is not a simple addition of all persons-and only those persons-directly observed by census enumerators. . . . It is dependent on complex scientific and management models and processes that are designed to insure the taking of an accurate census insofar as is humanly possible.”
This happens in a number of ways.
For example, the long census form, which includes 33 questions on topics such as income and ethnic ancestry, was sent this year to just one of every six households. The bureau will take the responses from those households and extrapolate them to cover the general population.
Another example is the way the bureau uses the techniques of imputation and allocation to fill in missing data (see box). Census officials said that, in 1980, bureau computers made up information about more than 27 million people-one of every nine Americans.
In most cases, officials said, the bureau knew something about the people and was just rounding out the picture. But for nearly 3 million people, an entire set of characteristics (age, sex, race, etc.) was constructed.
There was a final category of housing units that the bureau thought might be occupied but couldn`t be sure. After letting its computers loose, the bureau determined that 762,000 people lived in those units-although no one really knew for sure.
That final group included more than 130,000 ”people” who were imputed to live in Florida. Their addition gave Florida`s population enough of a boost to win the state an additional seat in Congress-at the expense of Indiana, where only 22,000 people were imputed in this way. When Indiana sued, a federal judge upheld the procedure.
Census Bureau officials have argued that imputation and allocation are based on at least some physical evidence, such as electrical hookups at housing units, while an adjustment for the undercount would be rooted solely on statistical discrepancies.
Despite the controversy over statistical techniques, no one disputes the existence of the undercount.
Indeed, every group is undercounted to a certain extent. But critics say some are missed more than others: blacks, Hispanics, men, the young, the poor, the less-educated and city dwellers.
The undercount problem dates to the first census in 1790, but has been growing as a political issue over the last 20 years as the federal government has placed heavier reliance on census data in determining how to distribute funds to states, cities and other local governments.
But the bureau`s counting error isn`t limited to missing people. It also involves counting others more than once or, in case of lazy enumerators, fabricating people.
More Top Picks Best Floodlight Cameras For Homes
Again, every group of Americans is affected, but, just as some are undercounted, others are overcounted, according to demographers: people who live outside urban and suburban areas, college students and families with two homes.
The political implications of the overcount are as significant as for the undercount because such duplications ”further distort population totals between cities and suburban areas,” said Bailar, the former census official, in testimony before the House Subcommittee on Census and Population.
Because it`s hard to determine what isn`t known, no one`s sure of the exact size of the overcount and undercount. But, according to some estimates, the overcount in 1980 may have been as large as 6 million people and the undercount as big as 9.2 million.
By using birth and death records, the Census Bureau is able to come up with a more definite figure on 1980`s net undercount (the total undercount minus the overcount): 3.2 million people, or 1.4 percent of the population.
The national undercount and overcount represent significant error, but what makes them much worse, according to proponents of an undercount adjustment, is that they`re not uniform.
Census Bureau studies found that blacks, who made up 12 percent of the population in 1980, accounted for more than half of the net undercount that year-1.7 million, compared to 1.5 million for whites and all other races.
The upshot was that the official figure for the number of blacks in the U.S. was nearly 6 percent below the actual number, according to Census Bureau estimates. In contrast, the official figure for whites and others was just seven-tenths of a percent under the actual number.
Undercount figures for Hispanics are much harder to determine, but one Census Bureau estimate was that 4.5 percent were missed in 1980.
Because of its reliance on clearly identifiable housing units and addresses, the Census Bureau has a much less difficult time counting people in the suburbs than in cities.
”When you`ve got a suburban area with all detached houses, it`s pretty easy to go out and find them all,” said Eugene Ericksen, a Temple University sociology professor who is also on the adjustment advisory panel.
”But, for example, in Philadelphia, there are neighborhoods with 3- and 4-story rowhouses, built around 100 years ago, and many of them have been subdivided into 3 or 4 apartments. The only way the Census Bureau could find out is to knock on every door.
”In Los Angeles, there is a substantial number of people living in garages or in makeshift housing in back yards. But you can`t check every garage.”
As a result of such problems, the general undercount in 16 major cities, including Chicago, was estimated at 4.5 percent in 1980, compared to an undercount of five-tenths of a percent in the rest of the nation, according to Ericksen. And it was even worse for black residents of those cities-9.1 percent.
In some ways, the Census Bureau is forced to choose between accuracy and completeness, as in the case of the census questionnaire.
”When I got my census form, it reminded me about the IQ test I got in grade school. You have to read it carefully,” said Peter L. Zimroth, former corporation counsel of New York City who is still representing the city and other plaintiffs in the undercount suit.
But Ericksen said, ”It`s not possible to simplify it. They`re really between a rock and a hard place. In order to get the right information, it has to be designed in a certain way.
”The more you do to make it accurate, the harder it is for people with less education and less reading ability to follow the instructions properly.” HOW CENSUS BUREAU WILL FILL IN THE GAPS
The 1990 census will rely in part on a statistical sleight-of-hand that the U.S. Census Bureau uses to develop its tables and reports without having to include a ”Not Sure,” ”No Response” or ”Unknown” category.
Most of the 1990 census data will come directly from Americans who have either filled out the census questionnaire themselves or have given the information about their household to a visiting U.S. Census Bureau enumerator. But, for one of every nine people, some or all of the necessary information will be missing.
To fill in the gaps, the Census Bureau will crank up its computers, as it has in previous years, to create characteristics (such as age or sex) and even entire statistical people through two processes: allocation and imputation.
– Allocation: When only a few characteristics are missing-children`s ages, for example-the census computer will search to find a second household as nearly identical as possible. Then, it will use the second household`s data to complete the form for the first household.
– Imputation of characteristics: When all of the characteristics are missing, the bureau will match the housing unit to a second one that is as nearly identical as possible and use the data from the first housing unit to complete the form. This occurs in situations in which the census bureau knows that a unit is occupied and the number of people living there.
– Imputation of people: Sometimes, the bureau imputes entire persons. This can happen when the bureau knows that a particular housing unit is occupied, but doesn`t know how many people live there.
It also can happen when the bureau isn`t sure if a unit is occupied or vacant. In that case, the census computer will decide, based on similar units in similar neighborhoods, whether the unit should be classified as occupied and, if so, will impute one or more people to the unit with an entire set of characteristics.
Total U.S. population in 1980 226.5 million 100.0%
All people with allocated or imputed characteristics 27.2 million 12.0%
People with some allocated characteristics 23.8 million 10.5%
People with all imputed characteristics 2.7 million 1.2%
People imputed by computer 762,000 0.3%
Source: U.S. Census Bureau.