Between- class effect (QB)
Mean weighted effect size (d+)
95% CI for d÷
Within-class homogeneity (Qwi)a
SYMONS AND JOHNSON
Table 3 Cross-Literature Models for SRE Magnitude
Note. Mean effect sizes not sharing the same subscript significantly differed (p < .05, a priori). CI = confidence interval; OR = other reference; SR = self-reference; SRE = self-reference effect. a Significance indicates rejection of the hypothesis of homogeneity. *** p < .001.
Variable and class
Manipulation class SR vs. semantic SR vs. OR
Type of processing indu by comparison task Relational Item specific Both relational and item specific
Several models emerge that implicate task restrictions used in SRE studies that mediate the SRE. It is interesting that these restrictions were not found to affect the magnitude of SREs in the SR-semantic manipulation class. First, results show that the magnitude of the SRE was related to the use of distractor tasks. Specifically, within both the distractor-present and the distractor-absent classes, significant mean SREs resulted. How- ever, the mean SRE for the distractor-present class was signifi- cantly larger than the mean SRE for the distractor-absent class. Further analyses reveal that distractor tasks had no influence on the magnitude of the SRE for studies that used high-intimacy targets (but did for low-intimacy targets). Second, results show that for studies in which participants did not expect a memory test, there was a significantly larger mean SRE than for studies in which participants did expect a test. Moreover, the mean SRE for studies in which participants expected a test was not significant. Further analyses reveal that an expectation of a test had no effect on the magnitude of the SRE for studies that used high-intimacy targets (but it did for low-intimacy targets).
Finally, two other exploratory models were significant for this manipulation class. The model test for mode of stimulus
presentation shows that when researchers used a projector to present stimulus materials, they obtained a significantly smaller mean SRE than those who used monitors or presented stimuli orally. Moreover, studies in the projector class obtained a non- significant mean SRE. In each case, study findings were heterogeneous.
The model test for participant population reveals that studies that tested undergraduates as participants obtained a smaller mean SRE than studies that tested other populations. (Note that this pattern is opposite of that found for SR-semantic studies.) However, both mean SREs were significant; in both cases, study findings are inconsistent.
The SRE has been of interest to researchers because of the assumption that it could tell them something about the self in memory and its relationship to other kinds of encoding pro- cesses. Most of the researchers cited in the literature have at- tempted to investigate this relationship through manipulation of various task parameters or with various populations. Judging by
Length of stimulus presentation
Memory load Time between encoding and
(k = 129)
(k = 60)
(k = 69)
Table 4 Continuous Models for Study Effect Sizes Across the Literature and Within Manipulation Class
Note. Models are least-square regressions with weights equivalent to
each effect *p < .05.
size. k was smaller for some models.
reciprocal of the = self-reference.