Journal of Experimental Psychology: Learning, Memory, and Cognition 1991, Vol. 17, No. 2 ; 323-333 Copyright 199! bv the American Psychological Association, Inc. O278-7393/9I/S3.OO Deep—Deeper—Deepest? Encoding Strategies and the Recognition of Human Faces Siegfried Ludwig Sporer Universitat Marburg Marburg, Germany Various encoding strategies that supposedly promote deeper processing of human faces (e.g., character judgments) have led to better recognition than more shallow processing tasks (judging the width of the nose). However, does deeper processing actually lead to an improvement in recognition, or, conversely, does shallow processing lead to a deterioration in performance when compared with naturally employed encoding strategies? Three experiments systematically com- pared a total of 8 different encoding strategies manipulating depth of processing, amount of elaboration, and self-generation of judgmental categories. All strategies that required a scanning of the whole face were basically equivalent but no better than natural strategy controls. The consistently worst groups were the ones that rated faces along preselected physical dimensions. This can be explained by subjects' lesser task involvement as revealed by manipulation checks. Factors determining the recognition of human faces have been classified into variables operative at encoding (e.g., ex- posure time; social stereotypes), during the retention interval (e.g., effects of delay; rehearsal), and at the retrieval stage (e.g., the principle of encoding specificity; for reviews, see Ellis, 1984; Kohnken & Sporer, 1990; Shapiro & Penrod, 1986; Sporer, 1989). The present research focuses on processes of encoding. In particular, we address the question How should faces best be encoded to lead to optimal memory perform- ance? The major theoretical impetus for this research has been derived from the levels-of-processing approach developed in research on verbal learning (Craik & Lockhart, 1972; Craik & Tulving, 1975). Applying the theoretical notions of the levels-of-processing approach to human faces, Bower and Karlin (1974) reasoned that judging the inferred character of stimulus faces (e.g., honesty) would induce deeper processing of faces and hence an increase in performance, whereas judg- Data from Experiments 1 and 2 were presented at the 27th annual meeting of experimental psychologists in Wuppertal in April 1985, and a summary of these and other experiments on facial recognition at the 36th Congress of the German Psychological Association in Berlin in September 1983. The manuscript was completed while I was a research fellow at the University of Marburg. This work was made possible by a travel grant from the Deutsche Forschungsgemeinschaft (444/698/83) and by three research grants from the Deutsche Forschungsgemeinschaft (Sachbeihilfen Sp 262/ 1-1 and Sp 262/1-2 and Me 777/2-1). I thank Dr. Roy Malpass for challenging discussions about this research. I also thank Carina Lotter, Sabine Kiichler, Heidrun Engel- hardt, Michael Basten, and Gerd Mager, who helped at various stages of this project in the preparation of the stimulus materials, conduct of the experiments, data management, and typing of the manuscript. Last but not least, I would like to thank Prof. Dr. G. A. Lienert, who supported me in this endeavor for many years while I was at the University of Erlangen-Nurnberg, and John H. Mueller. Correspondence concerning this article should be addressed to Siegfried L. Sporer, School of Law/Criminology, Universitatsstr 6, 3550 Marburg, Germany. ments of superficial properties of a face (e.g., sex of the person depicted) would lead to shallower processing, and conse- quently to a decrease in recognition. Since Bower and Karlin's (1974) pioneering study, more than two dozen articles have appeared that have tested a whole range of orienting tasks designed to induce differences between shallow and deep processing of human faces (Cour- tois & Mueller, 1979; Daw & Parkin, 1981; Deffenbacher, Leu, & Brown, 1981; Devine & Malpass, 1985; Mueller, Bailis, & Goldstein, 1979; Mueller, Carlomusto, & Goldstein, 1978; Mueller & Wherry, 1980; Parkin & Hayward, 1983; Patterson & Baddeley, 1977; Smith &Winograd, 1978;Strnad & Mueller, 1977; Warrington & Ackroyd, 1975; Winograd, 1976, 1978, 1981). Presumably, judgments of isolated physi- cal features generally require only shallow processing, whereas inferential attributions about the personality induce deep processing. Overall, results have confirmed this expectation. Most stud- ies do indeed find a reliable (though modest in magnitude) increase in recognition performance as a function of these two classes of orienting instructions (see Shapiro & Penrod, 1986, for a meta-analysis of the facial recognition literature). Despite this apparent homogeneity in findings, there are some exceptions, and the theoretical explanations are not fully convincing. For example, Baddeley (1978) has criticized the levels-of-processing approach for its circularity of reasoning because depth of processing cannot be assessed independently of recognition performance. In fact, this criticism may not strike studies on facial recognition as hard as the verbal learning research because the operationalizations of encoding instructions with regard to faces can be more cleanly differ- entiated along the rule of thumb "in the face or not" (Mueller & Wherry, 1980, p. 115-116, note 1). To circumvent this problem, for the present article we prefer the term encoding strategy as a generic term for both the various orienting instructions, which are merely operationalizations of theoret- ical constructs, and for the underlying (consciously guided or unconscious) encoding operations. Along this line Baddeley and Woodhead (1983), for example, have distinguished be- 323