New York City officially had a population of eight million in 2000, according to the U.S. Census Bureau. But is this an accurate number? Can we know what the actual number was? The answer is: probably not. To complete a truly accurate census two things must happen: First, one must be able to find all the places in New York City where people live. Second, one must count all the people living in each place (and count them only once), whatever the place is (a single-family home, or a dormitory, a prison, a mental hospital, an old age home, or a convent). Census methods do not readily capture New York City's incredible diversity. Assumptions and procedures that easily count Ozzie and Harriet Nelson and their two children (or even Ozzy and Sharon Osbourne and their two children) living in a single family house in the suburbs, do not work well in much of New York City. With its large number of immigrants, households of roommates and unmarried partners, illegally converted apartments and lofts, and homeless and transient population, New York poses unique challenges for enumeration.
2000 COUNT MORE ACCURATE THAN 1990
To improve the 2000 census, the U.S. Census Bureau asked for help from local authorities. Using a variety of city lists and other methods, the NYC Department of City Planning found households that the census had overlooked. Many were concentrated in areas of high immigration, including Northwest Queens, as well as the Bronx and Brooklyn. According to Joe Salvo, director of the Department of City Planning's population division, the Census Bureau sent forms to at least 160,000 of these households. Since the average household in New York City holds 2.6 persons, that means there were 420,000 persons who otherwise would not have been counted. Such households were probably uncounted in 1990, so a large part of New York City's 680,000 population increase was probably just a result of having a more accurate count.
COUNTING HOUSEHOLDS AND MAKING UP DATA
The census uses a variety of methods to contact households and collect census data. First, they send out a form hoping that it will be mailed back. Nationally, about two-thirds are returned. The fraction is lower in the city and varies greatly by neighborhood. Follow-up methods must be used to count those households that do not return the forms. If a phone number is known, the household is called. For those that do not answer the phone, or for which the phone numbers are unknown, enumerators are sent to the household. When, after repeated attempts, no one is found at home, the enumerator contacts a landlord, a supervisor or even a neighbor to confirm that someone lives in the unit. At that point, the census substitutes for those data. Since they do not know how many people live there, they fill in the household responses by using information from a nearby unit. This is known as the "hot deck" procedure, so named because in the distant past, the data was actually processed using computer cards, and the information was drawn from a "hot deck" of recently processed households. In other words, the census makes up data for households that they cannot reach.
To give one small example of the special difficulties in using current methods to get an accurate count in New York City: A trained Census Bureau enumerator, unable to reach individual households, may look at a building that has been divided into a multi-family dwelling and count the doorbells. In one actual example that was explained to me, one house had three doorbells in one location and what appeared to be two additional doorbells in another location. It would seem obvious that a homeowner or landlord may have created additional apartments. But is the presence of five doorbells proof that there are five households? I posed this question to my students who, for the most part, live in Queens. One woman from Elmhurst lived in a single-family home that had previously been divided into three units. Her house, she said, had four doorbells: an older set of three that no longer worked and one working that she had installed. At the same time, my student also had three mailboxes. Since the basic address list that the U.S. Census Bureau uses is from the U.S. Postal Service, she received three census forms instead of just one.
The census expects one person to fill out the form for each household. Even where data are supplied by one of the occupants of a unit, difficulties may ensue. In certain cases, the degree of cooperation or knowledge about the other residents may leave something to be desired. For understandable reasons, sets of roommates, boarders, illegal immigrants, and live-in household help (who may or may not be documented) may be purposely omitted. When the long form, which is received by about one-sixth of households and includes detailed data on economic standing, occupation, education, housing cost, migration, birth place, and the like, is involved, missing data are even more likely. If characteristics of some individuals are left blank, the census bureau imputes (makes up) those data. In short, the results from every census include made-up data for entire households and for particular characteristics for particular individuals. According to preliminary results such made-up data for NYC increased greatly since 1990.
ANOTHER APPROACH THAT MAY BE MORE ACCURATE
Since 1980, demographers and survey research experts, backed by a report by the National Academy of Sciences, have advocated "statistical sampling" to get at the hard to enumerate households. By this approach, the census bureau would use completed interviews or questionnaires with residents in most households. But for those that proved difficult to reach, they would attempt to interview only a random sample, perhaps one in ten. Results from that sample would be used to estimate the number and characteristics of all of the unreached. These sample interviews would be done by well-trained enumerators, who would use their resources to "get answers" from the hard to reach households.
Though backed by most scientific opinion, the use of statistical sampling in the census is highly controversial. The GOP, perhaps fearing that it would increase the numbers and thus the power of Democratic urban areas, has consistently blocked its use. Scientific procedures to get a more accurate count have repeatedly been rejected by Congress and two Bush administrations. As late as this month, the director of the Census Bureau rejected any sort of adjustment or correction for ongoing programs of population estimates. Instead for Census 2000, Congress allowed the bureau to pour billions of dollars into coverage improvement to help reach more of the hard to count, but still required that every household be enumerated. As we begin to prepare for Census 2010 most demographers think it would make sense to follow the recommendation of a panel of the National Academy of Sciences and use statistical sampling.
Andrew A. Beveridge has taught sociology at Queens College since 1981. Since 1993 he has done demographic analyses and consulted for the New York Times. He has provided expert testimony in districting and redistricting, housing discrimination, and numerous other civil rights cases in the metropolitan area and elsewhere. The opinions expressed in Topics are his alone.