sampling (collecting answers from only a selection of people or establishments and subsequently preparing estimates of what a complete count would have shown) rather than the summarizing of complete-count information, the number of sample responses has a direct bearing on the variability of the resulting data and in the confidence data users should place in them. In subdividing larger areas such as counties into smaller entities (for example, census tracts or BNAs), it is important to keep in mind their minimum desirable population size because of the many data items that are based only on sample responses. This often means observing minimum thresholds for the number of items (persons, households, housing units, business establishments) that each small area contains. If individual pieces of geography are too small, with correspondingly few persons, households, housing units, or business establishments in the sample, the resulting statistics will have lower confidence intervals and more sampling error. For such reasons, the Census Bureau recommends that a census tract contain at least 2,500 people.
Size also can refer to the extent of the area contained within a geographic entity, usually the number of square miles or square kilometers. For some purposes, such as developing enumerator assignment areas, the areal extent of a geographic entity can be an important consideration. Physical size also can be an important factor in data analysis. To meet the latter need, the Census Bureau provides area measurements along with many of its 1990 census data tabulations.
The geographic coverage of each type of statistical area varies according to their purpose. To be of use, major regions and subregions usually must cover the entire Nation; that is, they must provide complete geographic coverage. By contrast, places and UAs, whose purpose is to distinguish more densely settled areas from the rest of the country, do not attain complete national geographic coverage. (The same is true of MAs.) On the other hand, some census tables provide data for nonmetropolitan area and outside of urbanized area, thereby suggesting the notion of complete coverage. On the local level, MCDs, CCDs, and other county
