Skip Header

Measuring Quality in a Census, Part 2

July 08, 2010
Robert Groves

An earlier post, “Quality in a Census, Part 1,” provided basic concepts of quality in population counts based on a census. This post gives the basic sets of information we use to measure those concepts, to answer the question, “How good is the 2010 Census?”

There are three common approaches to measuring quality. First is a process-oriented approach. That measures whether the steps used to create the product contained features to improve its performance. For example, we have learned that it is more likely that a successful space shuttle launch will occur when there are redundant sub-systems and test-firings of propulsion systems, and would label such as of “higher quality” than comparable shuttle designs that omit them. Another common method is to compare outcomes of different alternatives. For example, if your watch says 8:58 and the clock on the wall says 9:01, you feel safer thinking that it’s about 9 than if the clock said 9:42. Another common approach is to imagine the ideal and focus on how close the actual is to the ideal. For example, airlines use the percentage of their flights leaving on-time as a quality indicator.

For the 2010 Census, we have two of these kinds of indicators:

a. Design and process quality indicators (use of bilingual forms, use of replacement forms, additions and deletions to the address list, participation rates during nonresponse followup, percentage of cases yielding counts of persons within the household). These indicators are starting to become available in preliminary forms.

b. Comparison of the 2010 Census to other approaches (demographic estimates of population counts based on birth, death, immigration, and emigration records; use of a large sample survey matched to census records). These indicators won’t be available for some time.

We do have a notion of the “ideal” census – everyone residing in the US enumerated once and only once and assigned to their appropriate residential location. However, we don’t know what that ideal would actually achieve for counts at all levels of geography, for different demographic groups. Thus, the third type of indicator we don’t really possess.

I’ll talk about the two kinds of indicators over coming posts.

Please submit any questions pertaining to this post to

Back to Header