Social researchers frequently are cautioned about using retrospective selfreports because of the likelihood of inaccurate reporting. To determine whether the responses to recall items differ from self-reports obtained at an earlier point in time, data were examined from a panel study of adult men. In 1964, respondents were questioned about their health, income, family, work, and general attitudes. Ten years later, all who could be located were reinterviewed and asked to recall what their situations had been a decade before. In most areas, recall responses presented respondents in a more favorable light than did information obtained in the initial interview. A second concern was the effect of recall items on tests of association. These data indicate that recall techniques have little utility for descriptive purposes but may be used cautiously in correlational studies. Edward A. Powers is a Professor, Willis J. Goudy is an Associate Professor, and Pat M. Keith is an Associate Professor, Department of Sociology and Anthropology, Iowa State University, Ames, Iowa. A version of this article was presented at the 1976 Gerontological Society meetings. Journal Paper No. J-8707 of the Iowa Agriculture and Home Economics Experiment Station, Ames, Iowa. Project No. 2010. This investigation was supported in part through grant number 10-P-57495/7-01 from the Social Security Administration, Washington, D.C. Public Opinion Quarterly ? 1978 by The Trustees of Columbia University Published by Elsevier North-Holland, Inc. 0033-362X178/0042-0380/$1.25 This content downloaded from 207.46.13.128 on Tue, 06 Sep 2016 06:07:51 UTC All use subject to http://about.jstor.org/terms CONGRUENCE BETWEEN PANEL AND RECALL DATA 381 1970; Zody, 1969), the same age group is studied over time although the same individuals need not be reinterviewed. Typically, in a cohort study age-specific data from a series of cross-sectional investigations are compared. Even retrospective self-reports have been identified by some as a useful longitudinal design (Garfinkel, 1968; Goldfarb, 1960). In recall studies, respondents are asked about events, situations, and attitudes that occurred in the past; recalled factors are then compared with a later measure of the same variable. In this article, we will consider two of the most frequently used longitudinal methods-panel and recall designs. Both have limitations. When the same persons are repeatedly interviewed as in panel studies, previous responses can be remembered or change may occur simply because of earlier interviews-the interview effect (Baltes, 1968; Goldfarb, 1960; Parnes, 1972; Riegel et al., 1967; Riley et al., 1972). Also, with selective survival, refusals, or the loss of subjects through mobility, the reinterviewed subjects may be quite different from the original sample (Baltes, 1968; Parnes, 1972; Powers and Bultena, 1972; Riegel et al., 1967; Riley et al., 1972; Streib, 1963, 1966). Finally, there is great financial and personnel investment in initiating and continuing panel investigations (Streib, 1966; Maddox, 1962). Thus there are barriers to maintaining a panel study and serious questions about the generalizability of findings. Recall studies are not subject to many of the problems of panel designs. But when retrospective self-reports are used, there is the possibility of inaccurate reporting either from unintentional memory lapse or a conscious misrepresentation of the past (Babbie, 1973; Cahalan, 1968-1969; Campbell and Katona, 1966; Cannell and Kahn, 1966; Maccoby and Maccoby, 1954; Som, 1973; Sudman and Bradburn, 1974). Although it is assumed that individuals honestly report their status when facts are unambigious, when respondents are requested to remember a previous situation they may reconstruct the past to be more congruent with self-perceptions, may present an image more in line with what they feel to have occurred, or may give a more socially desirable response (Cahalan, 1968-1969). Nevertheless, after the usual qualifications for possible inaccuracy in respondent recall, retrospective self-reports frequently are accepted as a measure of a previous time.