SYMONS AND JOHNSON
they also showed that, because relational processing is naturally promoted already, an encoding task that promotes item-specific processing is more beneficial for recall. Thus, relational pro- cessing is more likely to help recall of small categories than large ones.
These points may help to explain why SR promotes recall better than semantic encoding. On the one hand, SR has the potential (a) to provide self-relevant category labels, which may be helpful to facilitate subsequent recall because it may facilitate the reinstatement of encoding conditions at retrieval; (b) to help with the efficient processing of trait adjectives, despite the size of such a category, because of frequent processing of that do- main; as well as (c) because it arguably promotes item-specific as well as relational processing, thus complementing the rela- tional processing that is already present to create optimal recall conditions. On the other hand, without specific attention drawn to categories, the typical semantic-processing task is at a dis- advantage on several counts because (a) for most people, pro- cessing trait adjectives (a large category) semantically is not a frequent or well-practiced task; consequently, (b) category la- bels may not be as readily available; moreover, (c) although the nature of most semantic tasks in the literature is to promote item-specific versus relational processing, to the extent that SR promotes both, it is the more versatile of the two tasks.
in memory is and the more accessible the person category is. One important consequence of this idea is the potential for reinstatement of conditions at retrieval that are compatible with those that were present when the information was encoded. Thus, SR should pose an advantage in processing over other tasks, as our findings suggest.
Pragmatic Concerns for the Design of SRE Studies
Certain task variations used in SRE studies appear to affect the size of the SRE. These moderating variables suggest practi- cal concerns that should be confronted by researchers when they design studies to investigate the SRE. It is interesting that, in some cases, moderators that have a significant effect on the magnitude of the SRE have apparently been ignored as factors in SRE study designs. The first of these moderators is memory load. Our results show that, as memory load increases, the mag- nitude of the SRE also increases but that this pattern exists only for SR-semantic comparisons. This model implies that, when there is a great deal to remember, SR is a relatively efficient processing strategy when compared with semantic encoding. One practical implication of this finding is that memory load may be an important consideration for researchers who wish to maximize (or minimize) the advantage of SR in their studies.
One notable exception to this conclusion concerns the desir- ability task. We believe that the pattern we found in our meta- analysis lends support to the idea that, although pleasantness- desirability ratings have been traditionally used in the cognitive literature as a task thought to promote elaboration (e.g., Hunt & Einstein, 1981), in the typical SRE paradigm task, it is likely to promote a certain degree of relational processing as well (Klein & Kihlstrom, 1986). It fits the criteria for relational processing in that (a) there is potential for recognition of a category label and (b) such processing draws attention to the fact that, in the stimulus list words, there are words that are related in the sense that they are either desirable or not. This assumption, of course, represents an empirical question that could be addressed in further investigations.
Along this same line of reasoning, person reference may gen- erally promote more memory than a semantic task when list words are adjectives. It is logical to assume that some of the same mechanisms that govern an SR task may operate in any person-reference task. A person-reference task probably pro- vides a potential for recognition of an obvious category label, a task that is frequently practiced, and the potential for the development of an organized domain in memory around that person because the task is frequently practiced. As an application of this logic, the difference between SR and other person refer- ence is, of course, one of degrees. In other words, information about certain specific people (your mother, best friend, or worst enemy) is more frequently processed than information about other people (Johnny Carson or the experimenter at your study). People who are more often part of the information-processing environment are likely to be more accessible. Certainly, a partic- ipant who has engaged in an encoding task involving questions about himself or herself and about another person still has that information accessible in memory when asked to retrieve it. However, the more well known the person referenced is, the more organized and elaborated the information about the person
From a theoretical standpoint, this finding is a bit puzzling. Earlier, we discussed the effects of category size (Hunt & Seta, 1984): Large categories automatically promote relational pro- cessing, thus an item-specific processing task should facilitate memory more than a relational-processing task. Because most semantic tasks involve item-specific processing, why would the SRE increase as the memory load increases? To answer the question, we looked at the types of stimulus items that were processed. When traits are used as stimulus items, the SRE does increase as memory load increases, k = 47,/3 = .26, p < .01, demonstrating an advantage for person reference when pro- cessing trait items, a finding that is in-line theoretically with the patterns we predicted earlier. However, when the stimulus items are nouns, the pattern reverses itself, k = 11, /3 = -.47, p < .01. In this case, as memory load increases, the SRE gets smaller; this is a case where semantic processing is the normal processing mode for these kind of stimuli, thus posing an advan- tage because nouns are often elaborated and organized throtigh semantic processing.
It is interesting that the choice of experimental design had no impact on the magnitude of the SRE. Our data show that the choice of a within- or between-subjects design made no differ- ence for studies across the literature (or in either manipulation class). This is a curious finding because one would expect that controlling individual differences in processing would make a marked difference in the kinds of processes investigated in our meta-analysis. Thus, the SRE should have been larger in both manipulation classes when researchers used within- versus be- tween-subjects designs. We speculate that interference may oc- cur between encoding conditions in a within-subjects design. For example, Aron et al. (1991) have suggested that intimate others may actually overlap the self. Consequently, there is a confounding that arises that may render SR less advantageous than it is when a less intimate other is referenced. Similarly, participants may sometimes spontaneously consider whether a