tionalized are tracked and interviewed subsequently if they return to a family setting.
The PSID experienced a large sample loss—24 percent—at the initial interview in 1968, but additional sample loss dropped to 8 percent of the eligible families at the second interview, and it was only 1-2 percent at each interview thereafter (Survey Research Center, 1989: Table 2a). The initial large sample loss was partly due to the PSID sample design, which originally included a national probability sample of about 2,900 families and a sample of about 1,900 low-income families drawn from the sample used for the 1967 SEO. Several factors increased the nonresponse from the SEO sample, including the requirement by the Census Bureau that SEO families sign a release allowing their names to be given to the PSID (Hill, 1992).
The extent to which attrition introduces bias into estimates from the PSID is not clear. Several studies in the 1980s found that, although cumulative sample loss was over 50 percent (52% by 1980 and 58% by 1985), there was no evidence that attrition correlated with individual characteristics in a way that would produce biased estimates. For example, Becketti et al. (1988:490) found no evidence that attrition ''has any effect on estimates of the parameters of the earnings equations that we studied." Duncan, Juster, and Morgan (1984) also found that response rates were just as high in the PSID among families in the lowest income decile as in the middle or upper income deciles (see also Curtin, Juster, and Morgan, 1989, and other studies cited in Hill, 1992). However, Duncan and Rodgers (1991) found bigger differences in poverty rates for white children between the PSID and the March CPS in 1981-1986 than in 1967-1971 (the PSID rates were lower in both periods). They attribute the finding to the fact that, as of 1986 (before the addition in 1990 of a new Hispanic sample), the PSID represented only about one-third of the Hispanic children reported in the CPS while it represented all non-Hispanic white and black children.
One indicator of data quality is that about 95 percent of heads and spouses provide "adequate responses" for labor and asset income so that the responses do not have to be edited. The percentage of adequate responses has been in the range 94-98 percent over the life of the survey (Survey Research Center, 1989: Table 5).
The PSID collects the most detailed information about family heads and, since the late 1970s, about wives and cohabitors. The core content includes
employment information—current and employment history in past year;
income sources and amounts for the head for the past calendar year