MNEMONIC ADVANTAGES OF THE SELF
M weighted effect size (d÷)
0.50 0.45-0.54 451.40"** 34 (26%) 0.45
95% CI Homogeneity of effect sizes (Q)b n outliers removed to achieve homogeneity M weighted d÷ excluding outliers
95% CI M unweighted d 95% CI
0.39-0.50 0.53 0.42-0.62
Table 1 (continued)
Variables and class
All (k = 129)
Effect size summary
SR- semantic (k = 60)
11 (18%) 0.59 0.52-0.66 0.72 0.58-0.86
SR- OR (k = 69)
14 (20%) 0.26 0.19-0.33 0.35 0.22-0.45
For the purposes of these analyses, each effect size represents a separate study. C
I = confidence
interval; OR = other reference; SR = self-reference; SRE = self-reference effect.
aDavis (1979) *** p < .001.
participants, and all three of these studies also used SR versus semantic tasks. However, model tests reveal that studies with adults as participants obtained significantly larger mean SREs than studies with children as participants; in both classes, the mean SRE was significant. In addition, whereas effect sizes in the children class were homogeneous, effect sizes in the adult class were quite heterogeneous.
Exploratory analyses reveal, first, that
type of SR task related significantly to SRE. 1 Specifically, self-descriptiveness,
the magnitude of the autobiographical, and
differ in promoting
size for the association with nouns from the other tasks (see Table 5).
Moreover, the mean SRE for studies using nouns was not sig- nificant. (This pattern contrasts with the SR-semantic manipu- lation class in which results reveal significant mean SREs in both the traits and nouns classes.) However, both classes' effect sizes were quite heterogeneous.
Also as expected, model tests reveal that familiarity of the rated target other did not predict the magnitude of the SRE, whereas intimacy did. Specifically, as Table 6 shows, studies in which participants rated highly intimate targets obtained a significantly smaller mean SRE than studies in which target others were low in intimacy. 2 However, although the SRE was smaller when highly intimate others were rated, SR still tended to result in better memory.
As Table 5 shows next, a significant model test examining the SRE as a function of dependent variable used shows that the mean SRE for studies using recognition was not significant but that the mean SREs for studies using free and cued recall were significant. Study findings were quite inconsistent within the free recall class but consistent in the other two classes.
Two other model tests were significant. First, the model for mode of stimulus presentation shows that the use of projectors resulted in the largest SREs for this class, but mean effect sizes were not significantly different from studies that used index cards or booklets to present stimuli. Second, the model for par- ticipant population reveals that studies that tested undergradu- ates as participants obtained significantly larger SREs than stud- ies that tested participants who were not undergraduates,
In contrast to the usual expectation that within-subjects de- signs are more powerful statistically than between-subjects de- signs, type of experimental design did not affect the magnitude of the SRE. Finally--as Table 4 details--as the memory load induced in the studies increased, so did the magnitude of the semantic-SR effect sizes (a pattern that did not appear in the S R - O R class).
Moderators of S R - O R Effect Size Magnitude
Model tests also show that studies using self-descriptiveness tasks produced a significantly larger mean SRE than did studies using imagery tasks, as hypothesized and compared with other tasks (see Table 6). In addition, use of SR-imagery tasks within the SR- versus OR-manipulation class did not result in a signifi- cant mean SRE, as evidenced by the CI for that class. Within all three classes of SR tasks, study effects were inconsistent.
Exploratory moderators. Exploratory model tests for OR task used reveal that studies in the descriptiveness class had a significantly larger mean SRE than did the imagery, nonspecific- other, and other classes. Moreover, the mean SRE for studies using imagery tasks was not significant, as its CI shows. Model tests of dependent variable used reveals a significantly larger SRE when recognition versus free recall was used. This pattern is exactly opposite from that obtained for studies in the SR versus semantic manipulation class, which obtained a larger mean SRE when recall was used rather than recognition (see Table 5).
(text continueson page 384)
l In this case, type of SR task was considered to be exploratory because we felt that we could not make a priori predictions about the patterns in the literature. This is in contrast to the same model test for the SR-OR manipulation class; the literature predicts that imagery task studies should obtain smaller SREs than other kinds of tasks.
As Table 6 shows,
within the S R -
OR class and, consistent with expectations, the mean SRE was larger (and significant) in studies using traits rather than nouns.
2For both the familiarity and intimacy classes, studies in the unable to rate class tended to use nonspecific target others. It was difficult to classify these targets as either high or low in familiarity or intimacy.