From the Gotham Gazette:
Of course, in any huge operation like the census there are bound to be glitches here and there, but the Census Bureau established built in checks to assure accuracy. Most of the error and difficulties encountered by some census enumerators will be corrected during the quality control phase. For instance, any home that had an incorrect address was rechecked, and a small sample of households is being reinterviewed to validate the accuracy of the original count. Checks also exist to eliminate duplicate responses for those who filled out the form in more than one residence.
Despite the relative success and the efforts at quality control, some groups are very difficult to interview, leaving some missed homes and some incorrect or inconsistent information. To address this, the Census Bureau has a number of procedures it follows when it cannot get an interview or the response is either not complete or inconsistent with other responses from the household. These technique, based on statistical methods, come under the heading "imputation and allocation. "
There are two types of imputation and allocation:
Whole unit imputation comes into play when there is no answer for a unit known to be occupied. In such cases, the census applies what is known as the "hot deck" procedure and determines -- or imputes -- the characteristics of the household from those of other nearby households.
The term "hot deck" harkens back to the time when computers used punched cards. The hot deck contained information on those who were recently tabulated from the same neighborhood living in the same type of unit. Such data still exists -- even if the hot deck does not -- and is used to fill in information on nearby, missing with similar known characteristics.
The other method is "item and person allocation," where a household has provided answers, but they either incomplete answers or include inconsistencies. Here the information on the occupants that do exist is used to impute, much as the "hot deck'" was for the vacant unit So one can impute sex by using first name, if it is known. A Committee of the National Academy of Sciences discussed and reviewed these census procedures.
In the past, New York City has relied more on imputed data than the U.S. as whole has. The table below presents information on the imputation of basic characteristics in the city in 2000. The census reported that 4.7 percent of the housing units' vacancy statuses were imputed. In 8.9 percent, it imputed whether or not the unit was owned or rented. This compares with 2.6 percent for vacancy and 4.8 percent for owned or rented nationally.
As to imputation for persons, some 20 percent of New York City's 2000 population had at least one item imputed, compared with 10.8 percent for the United States as a while For example, in New York City 11.3 percent of people had their race imputed, versus 5.2 percent for the United States. In the Bronx, the figures are 26.4 percent for any imputation, and 16.0 percent for race..
Imputation varies widely throughout the city, as the table and map show. Areas with high proportions of poor and minority households, such as much of the Bronx, Bedford Stuyvesant, Flatbush and Harlem, as well as heavily immigrant area like Northwest Queens have high levels of imputation. In most of Staten Island and parts of Queens, and of Manhattan levels of imputation are low.