larly valuable for the purposes of creating regularly published nonmarket accounts is that its information will be provided every year, not just at intervals.
During its deliberations, the panel heard from a number of people who have been involved in the development and fielding of time-use surveys. Several of them raised questions about certain features of the ATUS design, questions which raise legitimate concerns. Not all of these concerns, however, are directly relevant to the potential usefulness of the ATUS for constructing nonmarket economic accounts.
Despite the tremendous step forward that the ATUS represents, and its prospective role as a linchpin of nonmarket economic accounts for the United States, there are real concerns about its reliability for this purpose that need to be addressed. They can be summarized by the questions: Who? When? and What? The most important, the “who” concern, is engendered by the lower-than-expected response rates in the ATUS. Response rates on the precursor studies also have been low (Egerton et al., 2004), but the 59 percent response rate on the ATUS was even lower, and far below the 90-plus percent response rates in the CPS.
Because most of the ATUS nonrespondents provided CPS responses the month before, a good deal is known about their demographic and economic characteristics. It is a relatively simple matter to reweight the sample averages of time allocations to account for differential nonresponse rates across groups with different observable characteristics. The difficulty is that there is no reason to assume that nonresponse is random (relative to the CPS sampling frame) across unobservable characteristics that may be correlated with time allocations. Although there is no way to know for sure until the data can be carefully examined, it seems plausible that busier individuals, or those with more irregular schedules, might simply be less likely to participate in the survey, meaning that the survey estimates could be distorted. Additionally, these biases in the ATUS are compounded with any biases in the sampling frame related to survivorship from the CPS.
There is no simple way to adjust for nonrandom nonresponse related to unobservables. The BLS is fully aware of this difficulty and has explored various means of boosting survey response rates (Horrigan and Herz, 2005). Nonetheless, low response rates—and the possible resulting nonresponse bias—are the biggest concern with using ATUS data to construct satellite nonmarket accounts. Efforts to assess the extent of any possible bias in the survey responses—and, if necessary, to address that bias by raising response rates or making appropriate adjustments to the estimates—should be a priority.
Recommendation 2.2: The Bureau of Labor Statistics should commit resources adequate to improve response rates in the American Time Use Sur-