A theoretical review and meta-analysis of the description-identification relationship in memory for faces

Abstract

Verbal descriptions can sometimes impair (or “overshadow”) and other times facilitate subsequent attempts at perceptual identification of faces; however, understanding the relationship between these two tasks and the theoretical mechanisms that bridge this relationship has often proven difficult. Furthermore, studies that have attempted to assess the description-identification relationship have varied considerably in demonstrating significant and null results, often across a variety of paradigms and design parameters. In the present paper we review the relevant literatures and theoretical positions proposed to explain this relationship, and we present the first meta-analysis of this effect across 33 research papers and a total of 4278 participants. Our results suggest that there does appear to be a small, but significant, relationship between the description measures of accuracy, number of incorrect descriptors, and congruence with that of subsequent identification accuracy. Furthermore, certain conditions were found to strengthen the magnitude of this relationship, including the use of face recognition versus eyewitness identification paradigms and the length of delays between relevant tasks. We discuss both the theoretical and practical implications of this relationship for understanding memory for faces.

Acknowledgements

The authors would like to thank Jasmine Koestler for her assistance with coding the data.

Notes

¹Hedges and Olkin (1985) have suggested the use of standardised residuals as well as homogeneity statistics to search for outliers. Unfortunately, they only provide formulae for the effect size d. We have adjusted these procedures to Zr as effect size, calculating the residuals and homogeneity statistics for Zr analogous to the procedure adopted by Hedges and Olkin for d. The basic rationale of this procedure is to calculate adjusted mean effect sizes after removing the effect size in question and then examine the standardised residual of this particular effect size, as well as the homogeneity Q after removal of this study. Residuals larger than 2 are considered as potential outliers. Removal of one, or several, outliers should reduce the observed heterogeneity indicated by a failure to reject the null hypothesis of homogeneity for the reduced number of studies. However, removal of these studies may inadvertently lead to a drop of important “exceptions” of the general observed pattern which (through the conduct of moderator analyses) could be particularly interesting in understanding the underlying theoretical mechanisms.

²Additional figures displaying these relationships are available from the second author.