By Dr. Martha Farnsworth Riche
[Article from Insight magazine, August 18, 1997]

Yes: A traditional head count will undercount minorities and many city dwellers.

Considering that the first census was supervised by Thomas Jefferson, read by Benjamin Franklin and delivered to President George Washington, it is clear that surveys of America--who we are, where we live, what we do--are woven into the very fabric of this nation. On April 1, 2000, this fundamental element of the republic will be renewed with census day.

Our Constitution spells out the primary role of the decennial census, which is to establish an accurate, once-a-decade count of the population. The census also places our population in a particular location as of census day so Congress can be reapportioned and the state and local governments redistricted.

During the sixties and seventies two additional political developments increased the importance of decennial census data. The Voting Rights Act of 1965 and subsequent court opinions used census data as a yardstick to implement "one-person, one vote" principles. And, the federal assistance initiatives of the seventies used decennial census data as the baseline indicator for billions of dollars of annual federal aid that still flow to state and local government.

Given these uses, it should not be surprising that the Census Bureau's plans for a more accurate census 2000 have come under attack.

In 1990 a number of troubling trends occurred regarding the census. The 1990 census undercounted approximately 4 million people, about the same number who were counted all together in the first census 200 years ago. Even more troubling, this last census was, for the first time in history, less accurate than its predecessor. The undercount of the population was 33 percent greater than the undercount in the 1980 census.

And, as before, the 1990 undercount was not uniform across the population. African-Americans, Hispanics, Asians, and American Indians were missed at a much greater rate than whites. Finally, the cost of the census escalated sharply. Even after accounting for inflation and the greater population, the 1990 census cost twice as much as the 1970 census. In large part this was due to the significant decline in the percentage of households that returned the census questionnaire and the resulting need for more extensive follow-up procedures utilizing hundreds of thousands of census-takers going door to door.

Congress concluded that the 1990 census failed on two grounds: It cost too much and measured too few people.

It's easy to figure the cost increase: you just take the total cost of the census, divide it by the number of households counted and adjust for inflation. The 1970 census cost $10 per household (in 1990 dollars); the 1990 census cost $25.

It's harder to see how far the census falls short of measuring all the people, but the Census Bureau has been doing it since 1940 using records such as birth and death certificates as an independent check. For example, in 1940, 3 percent more draft-age men showed up for the draft pool than the census found including 13 percent more black men. Although all census directors knew the census never counted everybody, this was the first measured knowledge that the undercount was higher for minorities.

In response to these developments, bipartisan legislation in Congress created a special panel of experts in 1992 at the National Academy of Sciences to study the mounting problems regarding census accuracy and cost issues. In 1994 the congressionally mandated panel, composed of nationally recognized experts in the fields of demography and statistics, reached three basic conclusions:

The panel accordingly recommended that "[e]fforts to follow up individually those who fail to return the mail questionnaire should be simplified and truncated after a reasonable effort based on several criteria ... and statistical sampling should be used to estimate the number and characteristics of the nonrespondent households that remain. In addition, evaluation surveys should be undertaken to improve the overall count and reduce the differential undercount."

The conclusions of this panel have been reaffirmed by a second panel that issued interim reports in 1992 and 1996, finding that the use of sampling techniques is "critical to the success of the year 2000 census." A decennial census that "reduces costs, reduces nonresponse bias, increases accuracy and reduces differential undercoverage can[not] be conducted" without the use of sampling, the most recent report concluded.

Moreover, failing to include sampling as an element of census 2000 would produce results worse than those obtained for the 1990 census. The panel added, "It is likely that repeating 1990 methods with the same relative level of resources to conduct the 2000 census will yield results that are of worse quality than obtained in 1990 and that have bias and undercoverage problems of unknown size and direction."

Based on the expert recommendations of the National Academy of Sciences, the Census Bureau first outlined its plans for a reengineered census 2000 in February 1996. The bureau's plan called for a simpler, less costly, more accurate census.

The bureau announced a variety of reengineered initiatives recommended by the 1994 report. These ranged from increased use of technology to tabulate better data, to better and more wide-ranging mapping and address-listing with the help of the U.S. Postal Service and local governments. The bureau also redesigned its census questionnaire to ensure that it was user-friendly and announced its plans to have the census 2000 questionnaire available in post offices and community meeting places for wider distribution. Finally, the bureau's census 2000 plans call for an intensive direct-mail campaign involving mailing and, for the first time, resending the questionnaire to every household, as well as a widely advertised toll-free number to accept response by telephone also for the first time.

However, as the academy recommended, the linchpin of the bureau's census 2000 reengineering involves scientific sampling to increase accuracy and reduce costs. The bureau's plans call for an aggressive program to count 90 percent of every neighborhood and then to account for the rest through scientific sampling techniques first utilized by the bureau in the 1940 census. The census 2000 plans also call for a nationwide, but state by state, 750,000-household quality check of the population to ensure accuracy right down to the local level and to eliminate the differential undercount.

Critics of the census 2000 plan have raised three major concerns:

Cost. Congressional critics say they are willing to write a "blank check" to cover the costs of a traditional census plan. However, the bureau estimates that the additional costs would range from $675 million to $800 million for a traditional head count over and above the $4 billion already planned for census 2000. And it still would yield a less-accurate census than the 1990 census.

Accuracy. The bureau can give no assurances that increasing its census 2000 budget dramatically to implement a 1990-style census would lead to increased accuracy. Quite the opposite, the bureau believes that accuracy at all levels, including the local level, again would decline using the old methods. As Barbara Bryant, the census director in the Bush administration says, "Throwing more money and more temporarily hired census-takers at the job of enumeration will not find the missing! After many local governments complained of undercounted blocks in 1990, we expensively sent the best-trained enumerators out to comb and recomb thousands of disputed blocks. This costly effort netted less than one-half percent addition to the 1990 census."

Constitutionality. The Department of Justice, under the Carter, Bush and Clinton administrations, has issued three opinions regarding the constitutionality and legality of sampling in the decennial census. All three opinions concluded that the Constitution and relevant statutes permit the use of sampling in the census. Every federal court that has addressed this issue has held that the Constitution and federal statutes allow sampling.

The Census Bureau has a well-deserved reputation for non-partisan, expert collection of data. The bureau's most important concern for census 2000 is accuracy. In the early 1800s federal marshals on horseback rode to the country's frontier to collect census information as best they could at county gatherings and court days. Throughout most of this century housewives and college students fanned out across America to take a census that was appropriate for rural and small-town America.

In 2000 our methods are to adapt to a largely urban population. We also must account for those in our country who come from traditionally hard-to-count populations or who follow the crops, work the third shift, refuse to answer the census or cannot be found. Scientific sampling will provide this accounting.

As the inspector general of the Department of Commerce states, "If carefully planned and implemented, sampling can be employed by the bureau in the 2000 census to produce overall more accurate results than were produced in the 1990 census, at an acceptable cost."

For census 2000 the continuing quest for an accurate and cost-effective decennial census must include scientific sampling to supplement the bureau's extensive plans for a physical enumeration of the population. Only then will the American people get the fair, accurate census they deserve.