actually views an object. When imagining others, he argued, one assumes one's usual visual orientation. In contrast, SR imagery forces participants to imagine themselves, causing them to as- sume an unusual visual orientation. Because imagining oneself is not a customary visual perspective, it does not help to facilitate memory. Imagining others, however, is customary; thus imagin- ing others facilitates memory as well as SR imagery. Our results provide no evidence to contradict Lord' s explanation. We would note, however, that Lord's formulation suggests that the imagin- ing of others should promote recall superior to that of SR. This hypothesis was not supported by our data: Results of the meta- analysis do not show that the imagining of others facilitates
memory better than does SR. Models for the SR-semantic class.
Three model tests
unique to the SR-semantic class were significant: (a) age, (b) type of SR task, and (c) type of semantic task. Results from the model test for age show that the SRE was significantly larger for studies that tested adults than for those that tested children. This finding is consistent with the hypotheses of the researchers who investigated age differences (e.g., Halpin, Puff, Mason, & Marston, 1984; Pullyblank et al., 1985). It has been argued that a likely explanation for this difference is that it is related to the rate of development of the self-concept. Specifically, children may exhibit a smaller advantage of SR relative to semantic encoding because their self-concepts are in an earlier develop- mental stage than adults. Consequently, the self-concept either has not developed sufficiently to facilitate processing to the degree that the adult self-concept can or has not been sufficiently elaborated to provide a ready network of potential retrieval cues to facilitate recall.
We offer a different explanation: An examination of the tasks used in studies that compare SR in children with SR in adults suggests that, although task adaptations were made for younger participants, the tasks were still very similar to those presented to adults. Thus, our alternative explanation is that researchers may not have observed SREs because task demands were inap- propriate to participants' developmental stage, not because chil- dren lack sufficiently developed self-concepts to produce SREs. Some data are available to support this contention. Barnas and Symons (1995) compared preschool children (4-5 years old) with older children (in kindergarten and first grade). The encod- ing task was modified such that pictorial stimuli were presented with each corresponding stimulus word at encoding. Children were then given a standard free-recall memory test. Results show that not only do very young children demonstrate an SRE but also there was no interaction between encoding task and age. Thus, we suggest that conclusions about the relationship be- tween self-concept development and the SRE may be premature. More research is needed to address this issue.
The model for type of SR task shows that there was a signifi- cant difference between studies that used either self-descriptive- ness tasks or autobiographical retrieval tasks and those that used tasks involving association of the self with nouns (e.g. , 9; doctor"). professions: "Did you ever wish to be a
Studies that used tasks involving noun associations consistently did not observe SREs--a finding that may well be due to the use of nouns as stimuli, an issue that we discuss at some length below. More interesting is the finding that there is no difference between self-descriptiveness tasks and autobiographical re-
trieval tasks in terms of the magnitude of the SRE they produce. Klein and his colleagues (Klein & Loftus, 1993; Klein et al., 1989) have made convincing arguments that to engage in a self- descriptiveness judgment versus an autobiographical retrieval task may actually involve the tapping of separate memory repre- sentations. They showed, for example, that a person who en- gages in a self-descriptiveness task before engaging in an auto- biographical retrieval task does not shorten response latencies on the second task, implying that information obtained during the self-descriptivenessjudgment does not facilitate the autobio- graphical retrieval task. They showed the same pattern when autobiographical retrieval precedes the self-descriptiveness tasks. Klein et al. concluded that the two tasks tap different sources of information. Although their findings are compelling, we note that, on the basis of our meta-analysis, the differences that they observed do not seem to extend to dependent variables involving retrieval. That is, although self-descriptiveness judg- ments and autobiographical retrieval may involve accessing dif- ferent areas of memory and some researchers have argued that the two tasks should generate different levels of recall (Bellezza, 1993), the magnitude of the SRE is equivalent in the two task classes. Thus, the two tasks appear to generate equivalent levels of recall. An extension of Klein and Loftus's model to other dependent variables may help to explain the bases of the SRE.
Although the model for type of semantic task used compares several types of encoding tasks, some patterns emerge that may suggest important theoretical issues. The first pattern is that studies that used desirability ratings did not observe SREs on average, consistent with Ferguson et al.'s (1983) hypothesis. These researchers argued that the evaluative component inherent in SR tasks was a confounding variable and that, if SR is com- pared with an evaluative-judgment semantic task (i.e., judg- ments of the stimulus word's desirability), the SRE would disappear.
Our results do confirm this pattern. However, an obvious alternative explanation, given the focus of our article, is that the desirability task promotes both relational and item-specific processing, as we discussed earlier. If future research supports this conclusion, then it will explain why studies thai used desir- ability ratings obtained significantly smaller SREs than studies that used either synonymjudgments or generate-definitiontasks, which primarily promote elaborative processing (cf. Klein & Kihlstrom, 1986; Klein & Loftus, 1988). One other purely spec- ulative explanation may be offered for researchers not finding a signficant mean SRE in studies in which desirability ratings were used. It may be that, when participants judge the desirabil- ity of a trait adjective, there is a degree of SR involved, thus the same sort of interference that may result when one is processing information about an intimate other target. That is, an answer to the question, "Is this a desirable word?," may implicitly involve a judgment regarding whether "this is a desirable word for me" from the participant's perspective. Thus, ironically, it is possible that desirability and SR are confounded but not in the way Ferguson et al. (1983) hypothesized. Future researchers should specifically address the nature of the personal relevance of the to-be-processed words for the participant and how SR and desirability judgments facilitate memory. To date, very few researchers have done this, with the exception of researchers who investigated the effects of self-schematicity on the SRE