The series of posts on quality of a census all made the point that there are various ways to measure how good a census is – all of them imperfect.

I think it important to note that, in my judgment, the ideal of “count every resident once, and only once, and in the right place” is the correct target. I also must admit that the US Census never has, and likely never will, achieve that goal.

There are inherent limits.

The population of the United States is changing minute to minute. Every hour there are about 490 births in the country and about 280 deaths. Roughly 12.5% of the households move each year (that’s about 1600 per hour), and some people leave the country and others enter the country each hour. In short, the US population never stops changing. The U.S. has chosen April 1, 2010, for the reference date of the census. However, given birth, deaths, and migration, the population was changing throughout the day. Strictly speaking, there is no single right answer of what the population on April 1, 2010 was.

This implies that the reapportionment of the House would theoretically change with such changes. Indeed, taken to the extreme, the reapportionment in December 2010 is based on an April 2010 count that is somewhat out-of-date. The country has changed somewhat between April and December.

Practical compromises are needed. For reapportionment purposes the US Constitution specifies that once every ten years is the appropriate “fixing” of the proportional representation in the House of Representatives. There have been no initiatives to try to make the distribution of House seats based on more timely data.

Countries vary on how timely the census data must be for their purposes. Some, like Canada, are not satisfied with every 10 year data and mount a full census every 5 years.

Perfection is impossible; expectations of quality might come from comparisons across countries. A common method of estimating undercounts is the post-enumeration survey technique we use here as well. (The designs of these are somewhat variable across the countries; estimates are somewhat incomparable.) The undercount in the 2001 UK Census is estimated to be 5.7% of the total population. This compares to 2.8% in Australia’s 2006 Census. The 2006 Canadian Census had an estimated 2.7% net undercount. The undercounts are not necessarily getting lower each census. For example, the 1996 Canadian Census estimated a 2.4% net undercount.

In 1980 we had 12 undercount estimates centering around 1.4%. In 1990 our undercount estimate was 1.6 %. The comparable figure for the 2000 U.S. Census is a 0.5% overcount. The average over the three decades’ censuses is an undercount of about 1%. That sounds small, but on a population of about 310 million, it translates to 3.1 million people.

Those are net figures. Most professionals would agree that the key issue is the “gross” not “net” coverage, separately accounting for those enumerated more than once (double counted) and those not counted at all. We don’t have many comparable estimates from other countries on this matter.

Each country, for each census, with a combination of professional and media commentary makes a judgment about whether the census counts are “good enough” for key uses.

Acceptable results are attained short of perfection. I could find no country that has stated an acceptable target for net undercounts. (I’d be interested if anyone knows of such an example.) Most strive for complete and uniform coverage of the population. Most acknowledge the impossibility of that goal. Most evaluate how close they come to such a goal, as an act of transparency to the society that must use the census figures.

Much attention is paid to the acceptability of efforts to achieve complete and uniform coverage, at the inception of the census, in order to gain acceptance of the final results. Throughout the decades in most countries, even though the weaknesses of censuses have been acknowledged, the societies have used them as the best uniformly available for key purposes.

In that regard, census-taking is not unlike many human endeavors that have nearly unachievable, but deeply valuable goals – eradicating childhood diseases, attaining universal voter registration, or eliminating traffic accidents. All efforts that effectively get us closer to those ideal states merit consideration. Even without achieving the goal, however, continuing our efforts toward the goal is important.

In the end, each of us must make a decision whether each census, having failed perfection, is suitable for our uses. This can best be done if the Census Bureau is completely transparent about its methods.