Gender Differences in Object Location Memory: A Meta-Analysis
Voyer, Daniel, Postma, Albert, Brake, Brandy, Imperato-McGinley, Julianne, Psychonomic Bulletin & Review
The goal of the present study was to quantify the magnitude of gender differences in object location memory tasks. A total of 123 effect sizes (d) drawn from 36 studies were included in a meta-analysis using a hierarchical approach. Object identity memory (37 effect sizes) and object location memory (86 effect sizes) tasks were analyzed separately. Object identity memory task showed significant gender differences that were homogeneous and in favor of women. For the object location memory tasks, effect sizes had to be partitioned by age (younger than 13, between 13 and 18, older than 18), object type (common, uncommon, gender neutral, geometric, masculine, feminine), scoring method (accuracy, time, distance), and type of measure (recall, recognition) to achieve homogeneity. Significant gender differences in favor of females were obtained in all clusters above the age of 13, with the exception of feminine, uncommon, and gender-neutral objects. Masculine objects and measures of distance produced significant effects in favor of males. Implications of these results for future work and for theoretical interpretations are discussed.
When considering the question of differences in cognitive abilities between men and women, there are specific areas that come to mind quite readily. For a number of years, researchers have been aware that the domains of spatial and mathematical abilities seem especially relevant to the question of cognitive gender differences since they yield marked differences in favor of males (Benbow, 1988; Hedges &Nowell, 1995; Hyde, Fennema, & Lamon, 1990; Linn & Petersen, 1985; Voyer, Voyer, & Bryden, 1995). Although verbal abilities were long believed to be in favor of women (see, e.g., Maccoby & Jacklin, 1974), more recent reviews suggest that they are either weakly in favor of women, not significant at all, or even in favor of men, depending on the task considered (Hedges & Nowell, 1995; Hyde & Linn, 1988). At the time Maccoby and Jacklin (1974) wrote their landmark book, many believed that spatial, mathematical, and verbal abilities were the only areas where cognitive gender differences occurred.
More recent work, however, suggests that an area of cognitive abilities that was not mentioned by Maccoby and Jacklin (1974) offers potential for emerging gender differences. Specifically, in the early 1990s, a number of researchers reported gender differences in favor of females in tasks requiring memory for the location of objects (see, e.g., Crook, Youngjohn, & Larrabee, 1990; Sharps, WeIton, & Price, 1993; Silverman & EaIs, 1992). In what is probably the most cited of these studies, Silverman and EaIs (1992) developed a task that they claimed reflected a yet unexplored aspect of spatial abilities, namely spatial memory. In this task, participants were presented with a randomly organized array of line drawings on a single piece of paper. Participants studied this array for 1 min, followed by a new piece of paper on which they were required to identify which objects were old (found in the study array) and which ones had been added. This task is labeled as the objects identity memory task hereafter, and it was presented by Silverman and EaIs as a measure of memory that is independent of location. Finally, a third array was presented in which some of the objects that had been found in the study array had been moved, whereas others remained where they were (unmoved). Participants were required to identify the moved and unmoved objects. This task is labeled as the conventional object location memory task hereafter, and it was presented as a measure of object location memory by Silverman and Eals. The authors found that women performed better than men in both tasks. Since then, the results obtained in this initial study (Study 1 of the Silverman & Eals, 1992, article) have been replicated several times by various researchers with the conventional object location memory task. Note that Crook et al. (1990) reported gender differences in an object location memory task before the publication of Silverman and Eals' report. However, it appears that this finding gained more prominence with the latter authors' work, possibly because they replicated their own results several times in the same paper and framed them within a theoretically broad context.
Interpretation of Gender Differences in Object Location Memory
The interest that was generated by the Silverman and EaIs (1992) study is somewhat puzzling considering that they made a number of questionable methodological, statistical, and conceptual claims. In reality, what likely attracted the interest of many researchers is that Silverman and Eals (1992) called upon evolutionary explanations of cognitive gender differences to account for their findings of gender differences in favor of females on the conventional object location memory task. Specifically, two basic models were contrasted. First, the sexual selection for range-size model was made on the basis of the notion that in polygynous species, sexual selection for homing range occurs, typically favoring males (Gaulin & Fitzgerald, 1986, 1989). This advantage presumably occurs because good navigation skills are required in males to allow them to look for potential mates or to find resources in order to attract mates. Although this claim has been applied to nonhuman animals, Sherry, Jacobs, and Gaulin (1992) suggested that humans might have a recent evolutionary history of polygyny. Consequently, specific brain structures as well as specific cognitive abilities (in particular, spatial ability) would have shown a sex-specific evolution, resulting in significant differences between men and women. However, Silverman and Eals pointed out that this explanation might be too simple to account for human gender differences. Specifically, mis explanation provides a relatively plausible account of gender differences in spatial abilities, but it does not clearly account for gender differences in object location memory. This issue led Silverman and Eals to propose the hunter-gatherer hypothesis as an alternative (Eals & Silverman, 1994; Silverman & Eals, 1992). This approach is based on the notion that archeological and paleontological data strongly suggest the presence of task division in males and females such that the former were generally the hunters, whereas the latter were gatherers. Considering that skills that promote successful hunting (abilities to orient oneself in relation to objects and places or across distances, perform mental transformation of objects, etc.) are generally in favor of males in modern humans, Silverman and Eals argued that modern gender differences on spatial tasks can be accounted for by the hunter role held by males. However, these authors also claimed that men outperform women only on those aspects of spatial abilities that are relevant to hunting. Other aspects of spatial abilities relevant to foraging-such as object location memory-should be in favor of women. Of course, their initial work supported this prediction.
Cognitive abilities are not the only aspect of human behavior that can presumably be explained by evolutionary mechanisms. Buller (2005) discussed how the evolutionary psychology paradigm has been used to explain virtually every human behavior. However, he cautioned readers that adaptation occurs continually at the population and individual levels. A correlate of this statement is that the mind is not a static structure that has been preprogrammed to respond in specific ways since the Pleistocene era, as is implied in the hunter-gatherer model. Experiential factors should therefore not be overlooked.
Buller (2005) also made the point that the hypotheses brought forth by evolutionary models are untestable in practice even though they might be testable in principle (see Buller, 2005, p. 88, for elaborations). What Buller means is that we know what kind of evidence is needed to test these hypotheses, and this makes them testable in principle. Unfortunately, we do not have access to the data that are required for this test (such as direct observation of our Pleistocene ancestors), and this makes these hypotheses untestable in practice. This issue is further complicated by the fact that the research methodology for testing the relevant hypotheses uses by definition a quasi-experiment, and this type of design does not allow causal conclusions (Furlong, Lovelace, & Lovelace, 2000). Another point to consider is that data that are compatible with predictions based on evolutionary models support the plausibility of these models, but do not prove their validity or allow one to reject other models (Lewontin, 1998; but see Ketelaar & Ellis, 2000, for a contrary view). The argument that was made by Cornell (1997) is especially illuminating in this respect. Specifically, Cornell demonstrated how an evolutionary model could account "plausibly" and with similar arguments for a pattern of gender differences that is the opposite of what we observe today. This demonstration emphasizes Cornell's view that any evolutionary explanation is by definition a post-hoc explanation, not a prediction. In summary, the hunter-gatherer model (among others) appears to provide circular arguments that seem untestable in practice. Although it is possible that the hunter-gatherer model is true, we cannot prove its conclusions. Note also that on the basis of cross-species evidence, Jones, Braithwaite, and Healey (2003) found little support for the hunter-gatherer hypothesis. Despite the problems inherent in an examination of evolutionary explanations and their controversial status, speculation about the theoretical significance of any observed gender difference remains worthwhile.
Alternative Interpretations and Findings
The initial findings observed by Silverman and Eals (1992) were reported in Study 1 of their article. Study 2 examined whether their findings-initially obtained with drawings-would replicate in a naturalistic setting (which they did). Study 3 was mostly concerned with determining whether gender differences would appear when one was informed that he or she had to remember object location (explicit encoding) rather than uninformed of this purpose (incidental encoding). The presence of gender differences in both the explicit and incidental contexts provided a positive answer to this question. Silverman and EaIs also implemented manipulations that they claimed demonstrated that gender differences in object location memory could not be accounted for by gender differences in object identity memory. Finally, Study 4 purported to show that gender differences in favor of females on object location memory appear at puberty. Silverman and EaIs interpreted their findings as reflecting the influence of changes in hormonal status with puberty on cognitive gender differences. However, one could argue that this conclusion was made on the basis of simple main effects that were computed despite the absence of a significant interaction. These obtained findings thus have a high probability of reflecting a Type I error. Nevertheless, despite relatively weak evidence, Silverman and Eals concluded that their findings strongly supported the hunter-gatherer hypothesis.
Most of the research conducted since the publication of Silverman and Eals's (1992) article aimed at either examining alternative explanations or providing evidence for or against the hunter-gatherer hypothesis. For example, to refute the argument that women might have used verbal labels to produce better location memory scores than men in their task, Eals and Silverman (1994) used uncommon objects that could not be verbally labeled in a variation of their conventional task. Their finding of gender differences in favor of women on this task discounted the verbal mediation interpretation of their earlier findings. However, work conducted by other researchers provided mixed evidence concerning the generalizability of gender differences in object location memory. In one such study, James and Kimura (1997) noted that women were significantly better than men at tasks involving object exchanges (when objects change location with each other, as in the conventional object location memory task), but not tasks involving objects mat were shifted from their original location to occupy those that were previously unoccupied. Arguably, the former predominantly assesses a form of topological memory, whereas the latter measures a more precise aspect of location memory, possibly calling on a coordinate component. In contrast, Dabbs, Chang, Strong, and Milun ( 1998) and Epting and Overman ( 1998) did not observe any difference between men and women in object location memory. Finally, in three recent studies, Postma and colleagues (Postma, Izendoorn, & De Haan, 1998; Postma, Jager, Kessels, Koppeschaar, & van Honk, 2004; Postma, Winkel, Tuiten, & van Honk, 1999) reported a male advantage rather than a female advantage for object location memory.
Considering the focus of research on supporting or refuting the hunter-gatherer hypothesis, one can plausibly argue that the consideration of this broad picture has taken researchers away from an examination of task components that might account for the gender differences. Specifically, researchers might be focusing so closely on the broad explanation that they are disregarding other, more parsimonious accounts of the gender differences.
As a starting point, recent research suggests that object location memory may comprise three major components: (1) object processing, (2) spatial-location processing, and (3) binding objects to locations (Postma et al., 2004). A simplified account of these components within the context of the conventional object location memory task suggests that they are clearly involved and might affect the observed gender differences. Specifically, object processing requires one to recognize the objects, thereby encoding their identity. This aspect would be tapped by the object identity memory task. Spatial location processing can proceed both by means of exact coordinate location codes and by relative, categorical, or topological codes. The processing of spatial location is not necessary to object identity memory. However, it is required to achieve the third step, when the object is bound to its location. Essentially, this requirement suggests that the last operation in object location memory is the establishment of a connection between object identity and location. This discussion supports the notion that object identity memory and object location memory involve interdependent but separate processes. From this perspective, we likely identify an object first, locate it, and then bind it to a location. This process raises the possibility that gender differences in object location memory might relate to gender differences in object identity memory. The present study examined this possibility by investigating gender differences in object identity and location separately. Consideration of the magnitude of the effect on each measure should provide some clues concerning possible precedence effects.
The above findings and discussion suggest that gender differences in object location memory might not be as clear-cut as Silverman and EaIs (1992) claimed. Accordingly, in light of the apparent contradiction in findings relevant to gender differences in object location memory, the goal of the present study was to give a more conclusive overview of the existence of gender differences in this cognitive domain as well as in the conditions under which they emerge. As was already pointed out, the literature on gender differences in object location memory appears quite diffuse. The various relevant studies will not be discussed at length because they form pan of the database presented later. However, elaborations on possible reasons for contradictory results are likely to provide clues concerning clusters of homogeneous effect sizes that might emerge through meta-analysis. The first class of reasons is methodological and statistical, including factors such as sample size, the file drawer problem (which will be discussed shortly), scoring procedure, the type of objects used, the setting of the task, encoding context, and age of the sample. The second is theoretically based, comprising the exact cognitive subcomponents underlying object location memory.
Methodological and statistical factors accounting for mixed results. Starting with the first class, an important methodological issue concerns limitations in group size. In general, men and women overlap considerably in cognitive performance (see, e.g., Kail, Carter, & Pellegrino, 1979). Hence, relatively large group sizes are needed to obtain significant statistical differences. When comparing homogeneous groups (e.g., college students), significant effects might be obtained with only 20 participants in each group, but clearly this depends on the sensitivity of the test. One goal of the present meta-analysis was to minimize this problem by combining several samples of various sizes while increasing the importance of results obtained with large sample studies through the use of effect sizes that were weighted as a function of sample size (Hedges & Becker, 1986).
A second methodological concern relates to the wellknown file drawer problem (Rosenthal, 1979). Studies yielding no effects or unexpected effects may not be published, thereby obscuring the true pattern of gender differences and the validity of the meta-analytic approach. One can plausibly speculate that since the dominant view regarding gender differences in object location memory currently favors women, several studies showing no gender difference or a male advantage may not have been published and have vanished in file drawers, whereas those showing the expected effect were published. One of the advantages a meta-analysis offers is that it allows an estimate of how many studies with nonsignificant or contradictory findings would be required to offset the significance of a given effect size. Thus, the present study examined the resistance of gender differences in object location memory to the file drawer problem. In addition, efforts were expended to include unpublished studies in the analysis.
Another methodological aspect that deserves consideration concerns the scoring procedure used in a specific experiment. Voyer et al. (1995) reported that on a number of tests, the scoring procedure had a significant effect on the magnitude of gender differences in spatial abilities. It is a fact that various scoring techniques have been used in object location memory tasks. Specifically, the typical score used by Silverman and Eals (1992) is accuracy of response (i.e., the object is either in the correct location or not). However, others (e.g., Postma et al., 2004) have used the distance between actual object location and the placement made by participants as a score. Completion time has also been used as the measure of performance (Tottenham, Saucier, Elias, & Gutwin, 2003). Such variations in the scoring procedure are likely to affect the results of studies using object location memory tasks as they do for spatial tasks. This possibility also has theoretical implications (see below), and as such it was taken into account here. In addition, recent reports suggest that men show a greater propensity to guess than do women (Voyer, Rodgers, & McCormick, 2004; Voyer & Saunders, 2004). It is thus possible that test scores that include a correction for guessing could produce a different magnitude of gender differences in object location memory, as was observed by Voyer et al. (1995) for spatial tasks. In the object location memory task, a correction for guessing is typically applied by subtracting the number of objects incorrectly identified as moved and unmoved from the number that are correctly identified as moved and unmoved. The present analysis will thus examine whether this type of correction affects the magnitude of gender differences in the task.
The use of uncommon objects by Eals and Silverman (1994) was mentioned earlier as part of their attempt to examine the verbal mediation explanation of their previous findings. However, others have speculated that the type of objects that are used might influence the magnitude of gender differences in object location memory through their relevance to the participants. For example, Cherney and Ryalls (1999) speculated that their participants would be more likely to remember the location of genderrelevant toys than gender-irrelevant toys. Their finding confirmed this prediction while also demonstrating that gender relevance nullified overall gender differences in favor of females in object location memory. Findings of this type suggest that it is very important to consider the type of object used in the studies retrieved for the present analysis.
Additional factors that require consideration relate to the ecological validity of the tasks used. Specifically, considering that the hunter-gatherer interpretation of gender differences is based on the notion that the female advantage arose as a result of a long evolutionary process that occurred in the natural environment, one can plausibly expect that gender differences should be found in studies that recreate this natural setting as well as in those where artificial conditions are used. Thus, the females' advantage over males should be found when the task involves a naturalistic setting as well as when drawings of objects or computer tasks are used. The hunter-gatherer hypothesis also leads to the prediction that women should excel particularly under incidental learning conditions, because these conditions are the most closely connected to their original gatherer role (Silverman & Eals, 1992). That is, it might be a natural automatic tendency of gatherers to store locations of relevant items into memory even when there is no direct usage for those items at present. Eals and Silverman's (1994) finding that women had an advantage when participants were not informed that they had to remember uncommon object location (incidental encoding), but that men had an advantage when they were forewarned (explicit encoding) supported this point. On the basis of their statement that "for unfamiliar objects, the female advantage occurs solely for location memory in incidental learning conditions" (Eals & Silverman, 1994, p. 103), these authors concluded that females might have evolved an attentional style that is more inclusive of the environment than that of males. This might lead one to expect that gender differences in object location memory should be more pronounced under incidental rather than explicit encoding conditions, at least for uncommon objects. This question will be examined in the present analysis.
Another methodological factor that requires consideration is the type of measure used. For example, Rahman, Wilson, and Abrahams (2003) reported gender differences in favor of females in object recall, but not in recognition. Considering that the conventional object location memory task is always administered as a recognition measure, this could mask gender differences of even larger magnitude. Accordingly, the present study took into account the type of measure in retrieving studies for inclusion in the meta-analysis.
Age of the sample is a variable that should not be overlooked when investigating gender differences in cognitive abilities. In fact, Linn and Petersen ( 1985) and Voyer et al. (1995) reported age effects in the examination of spatial abilities. Linn and Petersen proposed a categorical division of age groups: under 13, 13 to 18, or over 18, and this categorization was utilized by Voyer et al. (1995). In both meta-analyses, the data suggested that gender differences emerge at puberty or around cognitive events such as the emergence of concrete-operational thoughts, around the age of 7 (Piaget, 1952). Thus, the age factor is likely to provide important insights concerning the possible role of neuroendocrinological and cognitive mechanisms, as was suggested by Silverman and Eals (1992).
Conceptual and theoretical factors accounting for mixed results. The second class of reasons that might account for mixed findings regarding gender differences in object location memory relates to the notion that object location memory is a multicomponent process. The importance of this point lies in the fact that various subcomponents might be differentially sensitive to gender differences. A consideration of the levels of spatial information required in object location memory suggests that spatial location processing entails two forms that are distinguished by the grain of the spatial code. One form consists of exact metric coding of the position of an object. Kosslyn et al. (1989) labeled this component as reflecting coordinate spatial coding. Alternatively, we may use a more global, coarse, or relative code to describe an object's position. Such relative or categorical spatial codes define an equivalent class of spatial locations, which captures the invariant or abstract aspects of an item's location (Jager & Postma, 2003). Both the categorical and coordinate components are likely to be useful when one attempts to locate an object. For example, when looking for our eyeglasses, a categorical code ("the eyeglasses are on the table") or an exact position code ("the eyeglasses are 35 mm away from the upper left corner of the table") may be used. The former suffices for finding and remembering the object's location. The latter is necessary when we want to pick it up.
The distinction between coordinate and categorical spatial coding has been applied to the issue of gender differences in spatial processing as well. Some have suggested than men are better than women in coordinate spatial relation processing, whereas gender differences favor women for categorical relation processing. While the evidence for these expectations remains weak in the field of spatial perception (Hellige, et al., 1994; Hellige & Mitchimata, 1989; Jager & Postma, 2003; Rybash & Hoyer, 1992), results from spatial memory studies appear more promising (Alexander, Packard, & Peterson, 2002; Postma et al., 2004). A simple methodological approach to distinguish these two types of spatial coding in object location memory is to consider the distance between actual object location and its placement by participants in a recall task as a reflection of the coordinate component. In contrast, a simple measure of accuracy where participants indicate moved and unmoved objects would reflect a categorical component. Although it is clear from the discussion above that the categorical-coordinate distinction is not new, it would appear that the operationalization of accuracy and distance measures in object location memory to reflect this underlying distinction is novel. In addition, this operationalization allows clear predictions concerning the outcome of a meta-analysis examining gender difference in clusters of studies that differ in scoring procedure. Specifically, a distance measure should produce gender differences in favor of males, whereas an accuracy measure should favor females.
In summary, the present study examined the magnitude of gender differences in object location memory. As was the case in similar efforts in the areas of spatial, verbal, and mathematical abilities, partitioning of effect sizes in homogeneous clusters was expected to contribute to an identification of factors that are critical to gender differences in object location memory. A systematic examination of variables relevant to a test of the hunter-gatherer hypothesis as well as alternative explanations are thus expected to provide evidence concerning the conditions under which gender differences in object location memory are observed.
Selection Criteria for Inclusion in the Meta-Analysis
The present meta-analysis includes published and unpublished studies presenting results that were obtained with different versions of the object location memory task. In many of the studies, the object identity memory task was also included as a distractor or control task. Considering that this measure is clearly different from object location memory (Silverman & Eals, 1992) and that the present analysis focuses on object location memory, object identity memory data were analyzed separately from object location memory data. Note also that studies that used only an object identity memory task were not included in the analysis.
PsycInfo searches were conducted as a starting point for the retrieval of studies, and the reference list of retrieved papers was searched for more relevant studies. Researchers who had previously published in this area were also contacted with a request for published and unpublished data. This request produced a reply from 11 of the 24 researchers contacted (a 45.8% response rate). The studies selection procedure resulted in the sampling of 123 effect sizes (37 for object identity memory and 86 for object location memory) drawn from 36 studies, 3 of which were unpublished or had not been accepted for publication at the time of data entry (11 effect sizes). Note that although it was presented at a professional meeting, the study conducted by Robert and Ecuyer-Dab (2000) was counted as unpublished because it did not appear in a refereed journal. Only two of the sampled studies, those of Janowsky, Chavez, Zamboni, and Orwoll (1998) and Sharps and Gollin ( 1987), had most of the information required for meta-analysis, but simply stated that no gender differences were found without reporting a test of significance or relevant means and standard deviations. In these cases, the authors were contacted in an attempt to obtain more complete data. However, only Janowsky et al. still had access to their data. Accordingly, only those were included in the analysis, and the study by Sharps and Gollin was excluded.
In theory, only effect sizes that are independent from each other (i.e., come from different samples) for each type of task should be included in the final sample (Hedges & Becker, 1986). However, to assess the influence of some variables (e.g., object type), nonindependent effect sizes had to be considered. Fortunately, similar to the approach used by Voyer et al. ( 1995), the assumption of nonindependence was only violated for the overall analysis and can generally be discounted as a factor in the various effect size partitions. However, in a number of cases (see, e.g., McBurney, Gaulin, Devineni, & Adams, 1997; Robert & Savoie, 2006; Vecchi & Girelli, 1998), the authors presented several relevant effect sizes from the same sample. When this occurred, nonindependent effect sizes were combined in an attempt to include as many effect sizes as possible. Rosenthal and Rubin (1986)-among others-proposed an approach that allows one to combine nonindependent effect sizes. However, this method requires knowledge of the correlation between measures. Since this information was not available in most cases, homogeneity of the correlated effect sizes was assumed, and the simple arithmetic mean was used. Although this method has typically been found to produce a slight overestimation of the actual effect size (Marín-Martínez & Sánchez-Meca, 1999; Rosenthal & Rubin, 1986), it is suitable when homogeneity of the nonindependent effect sizes is assumed (Marín-Martínez & Sánchez-Meca, 1999).
Note that coding of the included studies was performed by the first author and verified by a research assistant. Points of disagreement concerning coding or study inclusion were discussed until agreement was reached.
A further issue relevant to classification of the tasks concerns the distinction between the conventional task and versions of this task. In this respect, a strict criterion was utilized. Specifically, use of the same drawings of common objects that were utilized by Silverman and Eals (1992) in a recognition task where objects exchanged location was required for classification as the conventional task. This requirement provided a clear and objective criterion for task labeling. It resulted in the classification of variations to this approach with uncommon objects (see, e.g., Eals & Silverman, 1994) or with location shift (e.g., James & Kimura, 1997) under the "other versions" label. Following this classification, studies entered in the analysis are found in Tables 1 (object identity), 2 (conventional task), and 3 (other tasks). They are also marked with an asterisk in the reference list.
Considering that the present analysis included mostly published studies, it is possible that published studies are a biased sample of the studies that are actually carried out, because it is presumed that only experiments with significant results are published (see Rosenthal, 1979). This issue (previously discussed and called the file drawer problem by Rosenthal, 1979), is likely to produce an overestimation of the effect sizes. In this situation, the fail-safe number (Rosenthal, 1991)-that is, the number of studies averaging null results that is necessary to offset the significance of the findings at the .05 level-is typically computed. This value was calculated in the present study in order to estimate the resistance of the metaanalytic results to the file drawer problem. The larger the fail-safe value, the more confidence one can have in the obtained results. Rosenthal (1991) suggested the use of 5k + 10, where k equals the number of sampled studies, as a criterion to evaluate the significance of the fail-safe number. Values larger than this criterion are deemed resistant to the file drawer problem.
Cohen's d was used as the measure of effect size (Cohen, 1977). This index represents the standardized difference between the mean of females and males in the present study. Effect sizes were computed by using the formula presented by Cohen (1977) when means and standard deviations were available, or by using the formulae presented by Wolf (1986) when only the t, χ^sup 2^, p, or F statistic was available. A positive effect size reflected gender differences in favor of women, and a negative effect size indicated gender differences in favor of men. However, this measure is considered a biased estimate of effect sizes (Hedges & Becker, 1986). Accordingly, it was corrected based on the approach presented by Hedges and Becker (1986) to obtain an unbiased estimate, which was then used in further analyses.
The analysis followed the procedure presented by Hedges and Becker (1986). These authors developed meta-analytic techniques that were designed for both the assessment of cognitive gender differences and the evaluation of the homogeneity of the effect sizes. Homogeneity of the effect sizes allows the conclusion that the effect sizes in a specific sample of studies were drawn from the same population. In other words, homogeneity of the effect sizes indicates that the studies included in a specific meta-analysis can be considered replications of each other, and that a pooled estimate of effect size provides a valid summary of the results from the sample of studies. However, when heterogeneity is detected, it is likely that the pooled estimate is not representative of the state of affairs in a sample. One purpose of this approach is to identify variables that have a significant impact on the magnitude of effect sizes. Since there is actually no predetermined way for deciding which variables arc important in a meta-analysis (Wolf, 1986), the Hedges and Becker (1986) approach allows one to examine whether a given variable has a significant effect on the magnitude of effect sizes. Specifically, at each step a test is calculated to determine whether the partitioning that was applied to the data had a significant effect on the magnitude of effect sizes. Such a test examines whether the difference between the heterogeneity for the whole sample (total heterogeneity) and that for the sum of the partitions (within-group heterogeneity) results in a significant amount of between-group heterogeneity. This approach can thus be interpreted as a test of whether the specific variable used in partitioning produced significant between-group heterogeneity. This is essentially the same as determining whether a factor produces significant group differences in the context of an ANOVA (Hedges & Becker, 1986).
From this perspective, the present meta-analysis followed a hierarchical approach. Thus, an overall analysis examining the magnitude and the homogeneity of gender differences among the effect sizes that were selected in each of the separate analyses (object identity memory and object location memory) was first conducted. In an attempt to determine what variables have a significant influence on the magnitude of effect sizes, a number of possible partitions were systematically examined, following the method recommended by Wolf (1986). Variables considered as part of this exploratory analysis were: age of the sample (under 13, 13 to 18, over 18), as defined by Linn and Petersen (1985), type of measure (recall or recognition), object type (common, uncommon, gender neutral, geometric, masculine,