Newspaper Article

The Census: Why We Can't Count

Feynman., James Gleick, James Gleick, the author of ''Chaos: Making a New Science,'' is working on a biography of the physicist Richard

1990

Overview

Yet once again thousands of people chose the wrong century of birth and the final tally of centenarians proved much too large, about double what the bureau and most demographers believe to be the actual number -based on other surveys and on actual records of births and deaths. Why? The computers took care of the 130- and 140-year-olds. ''But we had no requirement that children in a household shouldn't be over 100,'' says Gregory Spencer, chief of the Population Projections branch. ''You'd look and find two 112-year-old parents with their 109-year-old child.'' Sex-Change Procedures. The bureau's understanding of what constitutes a plausible household makes some arbitrary assumptions. In this year's census, whenever both members of a gay couple fill in the ''husband/wife'' dot and the ''married'' dot and the ''male'' dot, the census computers will automatically choose one of the men at random and ''correct'' his sex. (The same for two women.) In general, sex is the category best reported by those who respond to the census; even so, in 1980, more than 3 million people had their sex chosen by computer, Dr. Spencer says. The 1900 Problem. Too many people think they were born in exactly 1900. Based on past experience, the current census is certain to show a remarkable number of people just turning 90. Teen-Age Widows. Mispunching errors can create entire categories of nonexistent people. The teen-age widows are one group famous in census lore. ''If you have a one-in-a-thousand mispunching, it can create quite a sizable fictitious population,'' says Dr. [Leslie Kish] of the University of Michigan. Confusion about Relationships. Filling in the dots under ''If a RELATIVE of Person 1,'' many householders glance at their elderly parents and mark ''son/daughter'' when they should mark ''father/mother.'' Or they look at their husband or wife, who happens to be the father or mother of their children, and mark ''father/mother'' when they should mark ''husband/wife.'' Capture-recapture has its problems. One is the sampling error that arises from sheer chance. Ten identical boats taking samples under identical circumstances will still produce 10 varying counts. Statisticians know how to calculate the expected range of deviation; the larger the sample, the less it will be. The method also produces systematic errors, errors caused by the nature of the selection itself. Some animals are ''trap-happy,'' others ''trap-shy.'' The capture and the recapture have to be far enough apart that the tagged fish can mix randomly through the population, but close enough together that the population has not changed significantly in the meantime. ''This is the hairy part,'' says Henry Horn, a Princeton ecologist, ''making the assumption that all individuals get a chance to mix completely, that the marking process itself has no effect, and that there's no tendency for the marked individuals to be recaptured since they were the easiest to capture in the first place. All those assumptions have to be made, and none of them are true.'' For serious counts, the method must be combined with modeling techniques designed to face the selection biases head on, acknowledging them and trying to measure them. To count the 10 billion stars of the Milky Way, astronomers actually count only 100,000 or so. ''You have to get basic data, make a model, and understand how your counting depends on the parameters of the model,'' says Dr. [John Bahcall] of the Institute for Advanced Study. The model is a guide to the galaxy's structure: the ages and compositions of stars, the nature of their clustering, the proportions of hard-to-count blue stars to easy-to-count red stars. Just as the model depends on the raw data produced by telescopes and computers, the accuracy of the count depends in turn on the model. ''You try the model on new data, and use the new data to refine the model,'' Dr. Bahcall says. Though the process is circular, it is effective.

• Publisher: New York Times Company
• Subject: CENSUS / GLEICK, JAMES / POPULATION / STATISTICS