Accuracy of self-reported weight: A meta-analysis

Robert L Bowman, Purdue University

Abstract

The debate over the ability of researchers to substitute self-reports for measured weight, or "true scores," is not settled (Schlichting, Hoiland-Carlsen, & Quaade, 1981). Many investigators believe the validity and reliability of self-reported data to be cause for concern (Black, 1990b). Studies conducted to date on the validity of self-reports of weight have not provided a clear conclusion as to whether self-reports are valid. Few studies have had sufficient numbers of subjects to explore the questions of how demographic variables such as sex or type of subject population affect the ability of researchers to avail themselves of self-reported data. These studies frequently report disparate results even when similar samples and methods are used. Studies have also differed in regard to aspects of the problem addressed and types of analysis employed. Before further studies are undertaken, an integration of the research is needed. The purpose of the present study was to organize and integrate the conflicting findings of epidemiological and treatment studies in the weight-loss field with regard to the validity of self-reported data. This integration utilized the statistical methods of meta-analysis outlined by Hedges and Olkin (1985). Twenty-four studies were located which directly compared self-reports with measured weight: 15 were used in this study. Two separate effect sizes were calculated for each study to test the validity of self-reports. The first was an effect size based on standardized mean differences. The second utilized estimates of the absolute values of the differences between self-reports and measured weight. The effect sizes calculated for the studies were combined to give an overall effect size estimate. Null hypotheses, that the effect of self-reports was not significantly different from zero, were tested for the total sample, females, males, clinical subjects, and general population subjects. All null hypotheses were rejected. This indicated that bias was a significant component of self-reports for all subject groups by either method of calculating effect sizes. The standardized mean difference effect sizes ranged from $-$.14 to $-$.80 for the five subject groups and the effect sizes based on absolute value estimations ranged from 1.02 to 1.28. Recommendations for future research and practice are given.

Degree

Ph.D.

Advisors

DeLucia, Purdue University.

Subject Area

Academic guidance counseling|Public health

COinS