SYMONS AND JOHNSON
highly intimate other conceivably involves relational processing in which a highly organized domain in memory is referenced (see Sedikides, Olsen, & Reis, 1993) and item-specific pro- cessing that involves increasing degrees of elaboration as one's knowledge of the OR target increases.
Exploratory analyses. On an exploratory basis, we exam- ined several other possible moderators of the SRE, which should prove of special interest to SRE researchers who have speculated about most of these dimensions. Indeed, in coding studies for their characteristics, we included any dimension of which we were aware that other SRE researchers had found to be plausible moderators of SRE magnitude. Because many of these modera- tors have been the subject of theoretical debate and the relevant studies have yielded quite inconsistent findings, it was difficult to make meta-analytic predictions about the moderators' roles in the SRE. Nonetheless, the exploratory moderators included (a) specific aspects of the experimental situation (e.g., whether a distractor task was used and what the length of the stimulus presentation was ), (b) dependent variable (free recall vs. recog- nition), (c) design type (within vs. between subjects), and (d) memory load (i.e., amount of to-be-remembered material). We expect results for these dimensions to provide important clues regarding the basis of the SRE. Thus, for example, if the use of a distractor task has no impact on the SRE for semantic-SR comparisons but increases the SRE for S R - O R comparisons, then it may be that OR tasks benefit more from rehearsal strate- gies, as some reseachers have argued (Kuiper & Rogers, 1979).
task (synonym decision, generate a definition, fits sentence, fits category, other), (d) relatedness of stimulus words (high, low, unable to rate), (e) type of stimuli presented (trait adjectives, nouns, other), (f) famil- iarity of the other-referenttarget (high, low, unable to rate), (g) intimacy of the other-referent target (high, low, unable to rate), and (h) type of processing promoted by the comparison task (relational, item-specific processing, or both). The exploratory dimensions included (a) type of dependent measure (free recall, cued recall, recognition), (b) design of study (within or between subjects), (c) nature of timing interval for retrieval measure (fixed, unlimited; recorded time if fixed), (d) type of OR task (descriptive, biographical, imagery, associated target other with nouns, other), (e) timing of stimulus presentation (fixed, variable based on response latency, variable based on experimenter judgment, other, unknown), (f) length of stimulus presentation, (g) duration of interval between task and retrieval, (h) presence of distractor task, (i) expecta- tion of a memory task (i.e., use of an incidental vs. an intentional learning paradigm), and (j) memory load (coded as number of words in the stimulus list). The merely descriptive characteristics included (a) year of publication, (b) source of publication (journal, other publication, dissertation or master's thesis, unpublished), (c) source of participant population (undergraduate or other), (d) percentage of male participants in the sample, (e) method to test participants (individual or group), (f) geographic area in which the study was conducted, (g) mode of stimulus presentation (index cards, monitor, projector, read by experimenter, booklet, tachistoscope, other), and (i) expectation of a retrieval test by participants (i.e., incidental or intentional learning paradigm).
Computation of effect sizes. Effect sizes for the studies were re- corded, along with their significance and direction. In so far as possible, separate effect sizes (gs) were computed for each relevant manipulation (i.e., SR vs. semantic encoding and SR vs. OR encoding) with the following formula:
Criteriafor study inclusion. Our meta-analysis focuses on two com- parisons performed in the majority of SRE studies: (a) SR versus seman- tic encoding and (b) SR versus OR encoding. Studies had to meet two criteria:
1. Researchers had to manipulate encoding instructions by presenting participants with a task in which either (a) SR and semantic encoding or (b) SR and OR were compared. We generally included studies that presented traditional DOP paradigm encoding tasks; however, stud!es in which researchers did not use traditional tasks were included if we judged them to have manipulated the type of encoding in a relevant way, even though the task was not conventional (e.g., Mueller, Heesacker, & Ross, 1984). Both within- and between-subjects designs were included.
2. Acceptable dependent measures included free recall, recognition, or cued recall. Either adjusted or raw score means were acceptable; however, when both adjusted and raw score means were available, ad- justed means were used. The mere use of an SR trait descriptiveness task in a reaction-time paradigm was not sufficient for inclusion (e.g., Markus, 1977); the task had to specifically test the effects of type of encoding on some measure of recall or recognition.
Mself -- Msemantic (other)
where Ms,if is the mean recall for the SR-encoding condition, M~,,~tlc ~oth~r)is the mean recall for the semantic- (or OR-) encoding condition, and SDpoot~dis the pooled standard deviation (see Johnson, 1989). We converted the gs to ds by correcting them for the bias that occurs, especially with small sample sizes (Hedges & Olkin, 1985). Some studies yielded multiple effect sizes because (a) more than one dependent measure was used (e.g., two different recall measures) or (b) several different levels of a variable were manipulated (e.g., several different target others; Keenan & Baillet, 1980). The effect sizes were analyzed using standard meta-analytic techniques (Hedges & Olkin, 1985; Johnson, 1989).
Characteristics of studies. marizing the characteristics analysis. Table 1 shows the
We began our analyses by sum- of studies reviewed in the meta- summary of study characteristics
Study retrieval. A thorough search of the SRE literature was con- ducted to find as many studies as possible. We obtained studies through literature searches using PsycL1T(1974-1994) and by searching refer- ence lists of relevant articles. In addition, we searched for unpublished dissertations and conference papers and made specific attempts to contact researchers of unpublished studies to avoid publication bias. Only studies that were available as of June 1994 were included in the sample.
Coding. We coded studies on several dimensions that were relevant for theoretical or exploratory purposes and on many more dimensions that were descriptive only. The theoretically relevant dimensions in- cluded (a) type of comparison (SR-semantic vs. SR-OR), (b) type of SR task (self-descriptive, autobiographical, imagery, associating self with nouns e.g., body parts or boats, other), (c) type of semantic
versus tions. study
semantic manipulations and (b) SR versus OR manipula- Table 2 presents the calculated effect size (d) for each in our meta-analysis, along with its important study
attributes. As Table 1 shows
, studies in our meta-analysis (a) were
published relatively recently; (b) were primarily published re- ports; (c) tested 39 participants on average; (d) tested partici- pants primarily from college populations; (e) usually tested adults; ( f ) were primarily conducted in North America; (g) tended to test participants individually (75%), with 24% using