Pew Research Center: 22% of NRFU based on proxy interviews is bad news for accuracy
Despite yesterday’s claims by Robert M. Groves that the 2010 Census is accurate and trustworthy, the fact that 22% of NRFU interviews were done by proxies is scary. D’Vera Cohn writes the following:
As the 2010 Census information-gathering phase winds down and the Census Bureau turns to quality-checking and data-processing, Director Robert Groves offered some statistics at a recent operational briefing to assess how the national count has gone thus far. One indicator, the quality of the address list, appears to have improved since the 2000 Census. Another, the share of proxy interviews, has worsened.
The foundation of a good census is having a complete list of addresses because Americans are counted at their homes or the other places they are living. The quality of the address list is important in aiding census-takers who head out on follow-up visits to people who did not return their mailed-out questionnaires.
During the recent non-response follow-up operation, Groves said, census-takers found fewer non-existent addresses on their rounds in 2010 than their counterparts had in 2000. In 2000, 6 million non-existent addresses were deleted from the list because census-takers could not find them. In 2010, 4.1 million were deleted. During follow-up visits, census-takers also are supposed to look for addresses that are not on the official list, so they can be added. In 2010, Groves said, “we had fewer adds proportionately” compared with 2000, although he said this is not as much of a “hard quality indicator” because it could mean that census-takers did not follow procedures for including new addresses.
On another quality measure, Groves said census-takers who were trying to collect information at addresses from which census forms were not received had to rely more heavily on neighbors and building managers than was the case during the 2000 Census. In 2000, about 17% of follow-up interviews were from proxies, not from the householders themselves, compared with 22% in 2010. This is of concern because proxy data traditionally has been less accurate than information that people provide about themselves. Groves said “this fits the expectation we had with regard to the cooperation of the American public.” Some people were never home during repeated visits by census-takers; others refused to provide information about themselves.