Beyond Sex Differences In Visu Essay, Research Paper
BEYOND SEX DIFFERENCES IN VISUO-SPATIAL PROCESSING: THE IMPACT OF GENDER TRAIT POSSESSION
Much research has emphasized the presence of sex differences in visuo-spatial processes while neglecting individual differences in performance within the two sexes (Archer, 1987). The present study looks beyond sex differences and considers the association of self-perceived gender trait possession with performance in two visuo-spatial tasks. The findings indicate that, in a 3-D mental rotation task, where a substantial sex difference occurred, gender trait possession adds significantly to the overall explanation of performance, the important gender trait variable being a measure of androgyny. With the Group Embedded Figures Task, gender trait measures were the only significant variables in differentiating performance, in this case masculinity was the important gender trait variable. The implication of such results for conventional explanations of individual differences in visuo-spatial processing is discussed
Research into the existence of sex differences in visual cognition and related areas has been extensively reviewed in the past 20 years (Archer & Lloyd, 1985; Fausto-Sterling, 1992; Halpern, 1992; Hyde, Fennema & Ryan, 1990; Linn & Petersen, 1985; Maccoby & Jacklin, 1974; McGee, 1979; Newcombe, 1982; Tracy, 1987). The general conclusion of such reviews suggests that visuo-spatial performance appears to differ significantly between the sexes (though see Fausto-Sterling, 1992, for an alternative interpretation). Figure 1 indicates the typical pattern of results in mental rotation task performance for females and males, males typically achieving significantly higher scores.
While the conventional emphasis has been made on a comparison of the two sample means (female vs. male) it is apparent from the figure that there is considerable variation in performance within the sexes and extensive overlap between the two distributions. Archer (1987) and Halpern (1992) have argued for the use of effect size in the discussion of results in order to complement significance levels and gain some indication of the extent of variance in the data which can be explained by the variable of interest. In Fig. 1 the effect size, d, associated with sex, is approximately 0.94. This is typical of the effect size found in studies employing 3-D mental rotation tasks (Halpern, 1992).
Much of this research has attempted to establish the extent to which biological factors underlie such differences. Thus, McGee (1979) considers issues such as genetic endowment, handedness and cerebral lateralization. (See also Annett, 1992 & McKeever, 1986, for a discussion of putative genetic factors underlying spatial ability.) The relevance of cerebral lateralization has also been identified by Newcombe, Dubas & Baenninger 1989. Nyborg (1988) has suggested that it is the level of estradiol that underlies the difference in visuo-spatial performance between the two sexes. Research indicating menstrual cycle impacts upon spatial performance would also support the importance of oestrogen levels, e.g. Hampson & Kimura, 1988 (see Ho, Gilger & Brink, 1986, for a qualification of this argument). Findings such as these have led some reviewers such as McGuinness (1985), Ruddisill & Morrison (1989) and Kimura (1992) to conclude that neuro-anatomical differences may account for sex differences in visuo-spatial performance and the mathematics skills which might depend upon these processes.
Research by Newcombe, Bandura & Taylor (1983) and Tracy (1990) highlighted the importance of the individuals’ interaction with the physical environment, emphasizing play with toys or activities such as sports which have a spatial component to them. Baenninger & Newcombe’s (1989) meta-analytic study identified a small, but statistically significant, relationship between spatial activity participation and spatial ability. Findings such as these have prompted the development of training programmes in visuo-spatial processing (Ben-Chaim, Lappan & Houang, 1988; Ferrini-Mundy, 1987; Kyllonen, Lohiman & Snow, 1984). The meta-analytic study by Baenninger & Newcombe (1989) concluded that training did result in an improvement for both males and females indicating that for both sexes an asymptotic level of performance had not been reached. One should also note that Casey, Colon & Goris (1992) have suggested that right-handed females with familial sinistrality are most likely to benefit from such training.
The observations of Nash (1979) and later Durkin (1987) have suggested that it is the child’s experience of the sociocultural context which is more critical in the development of sex differences. Part of the argument made by Durkin and Nash concerns the suggestion that the sex linkage of a task can have considerable impact upon the subject’s performance expectations, performance and performance evaluations. Nash predicted that sex-typed individuals would perform more effectively on tasks with the commensurate sex-type label. Thus, masculine-typed individuals would be expected to perform more efficiently on masculine-labelled tasks, i.e visuo-spatial tasks, whereas feminine individuals would perform feminine-labelled activities, such as verbal tasks, more effectively. Observations such as these led Durkin to conclude that the social cognition processes informed by the social context ‘. . . are clearly much more central to the construction and maintenance of sex differences than are fixed traits . . .’. This social cognition approach to sex differences has found support in research, e.g. evidence by Basow & Medcalf (1988) does indicate that differences in attributions concerning academic achievement exist between the sexes.
The work of Casey and her co-workers (1990, 1991, 1992, 1993) has strongly implicated biological/environmental interactions, an interpretation made earlier by Benbow (1988) when discussing mathematics ability. This interactionist approach relies on the notion of plasticity in the central nervous system and, consequently, the CNS being amenable to what Greenough, Black & Wallace (1987) have labelled experience-expectant and experience-dependent processes. The latter process is still evident in mature animals (Greenough, 1993). Halpern’s (1992) approach reflects this perspective, labelling the approach a ‘biopsychosocial perspective’. In her model of cognitive sex differences, factors such as prenatal hormones, intra-uterine environment, sex-role stereotypes, values and expectations, life experiences and rewards and preferred cognitive strategies all contribute to cognitive task performance. An important element of this perspective is that the factors are seen to be interactive. With this approach, the child’s competence in spatial processing could be the result of an interplay between CNS priming, peer reinforcement and the child’s attitude towards the task. Hence, the identification of causal relationships between putative causal factors and cognitive performance becomes extremely problematic. The difficulty in disentangling such factors is apparent in McGuinness’s observations on research into sex differences in attitudes towards mathematics (McGuinness, 1985).
A critical component of the Basow & Medcalf study discussed above was that attributions were not only compared between the sexes but also within the sexes. Here, the gender of the subject, as well as the sex of the subjects, was employed as a relevant variable. Richardson (1991) and Unger (1979) have suggested that the label sex should refer to the biological distinctions between males and females, while gender should refer to the sociocultural characteristics and behaviours associated with the two sexes. By making use of self-report gender trait measures, research, subsequent to Nash’s original observation, has attempted to identify the extent to which self-perceived gender characteristics are associated with individual differences in visuo-spatial performance.
A major meta-analytic review of the impact of sex role (measured by self-report inventories) upon cognitive functioning was carried out by Signorella & Jamison (1986). This study considered separately spatial perception, mental rotation, spatial visualization, mathematical ability and verbal task performance. For most of the spatial/mathematics tasks the higher the perceived masculine score in relation to the feminine score, the greater the cognitive performance. This supported Nash’s (1979) observations that masculine sex-typed individuals would perform well on spatial processing tasks, which would possess a masculine sex-typed label. Later research by Newcombe (Newcombe et al., 1989; Newcombe & Dubas, 1992) and Bernard, Boyle & Jackling (1990) confirmed the importance of the masculinity factor for high spatial performance level (the Newcombe studies looking at females only).
Signorella & Jamison, however, found no support for Nash’s (1979) hypothesis with verbal task performance and also observed that, in some spatial tasks, greater performance in the male subjects (only) was associated with higher levels of self-perceived femininity. This suggestion of a link between androgyny (relatively high self-rating of femininity and masculinity) and spatial performance was also found by McKeever (1986) in a sample of undergraduates undertaking a mental rotation task. (It should be noted, though, that McKeever’s method of measuring androgyny differed from earlier median split procedures (Spence & Helmreich, 1978).
An alternative or complementary impact of perceived gender trait possession upon visuo-spatial performance could be through the use of a gender schema guiding the child’s recreational and play activities (Bern, 1981). Thus, an individual with a masculine-based gender schema would choose to participate in activities that confirmed such a schema, activities which, according to the research of Newcombe et al. (1983), Baenninger & Newcombe (1989) and Tracy (1990) identified above, could allow the individual to develop the appropriate spatial skills and subsequently perform well on visuo-spatial tasks.
The aim of the present research is to identify the importance of individual differences, as identified by gender trait possession, in visuo-spatial performance. The Bem Sex Role Inventory (Bem, 1974) was employed to assess the extent to which subjects perceived themselves to possess the gender traits, instrumentality and expressiveness (Archer, 1989; Archer & Rhodes, 1989; Spence, 1985, 1991). Two different types of spatial tasks, a 3-D mental rotation task, the Phillips (1979) S & M task, and an embedded figures task, Witkin, Dyk & Paterson’s (1962) Group Embedded Figures Task, were employed. These two spatial tasks were chosen as the former is consistently found to possess a large effect size (due to sex) while embedded figures tasks show a less consistent sex difference (Halpern, 1992; Bernard et al., 1990). This provides the opportunity to look at impact of perceived gender trait possession in a task where sex is an important variable and another task where it is considered not to be so important.
One hundred and seventy-six A-level students were drawn from a variety of educational institutions in the northeast of England: two colleges of further education, two (11 18) comprehensive schools, one sixth form college and one independent school. The sample was composed of 122 females (56 from FE colleges) and 54 males (22 from FE colleges).
The mean age of the female subjects was 17.7 years (SD = 0.96, range 15.83-20.83 years) with the male subjects possessing a mean of 18.09 years (SD = 1.39, range 16.75 21.75 years).
In order to assess visuo-spatial processing two spatial tasks were employed. The first task was the Phillips (1979) S & M mental rotation task, based upon the original Shepard & Metzler (1971) stimuli. In this task 20 pairs of block stimuli are presented to the subjects who have to decide, within two minutes, whether the two are the same stimuli in different orientations or a dissimilar (mirror reflection) pair of block stimuli. Scoring was one mark for correct choice minus one mark for incorrect choice. While scoring correct/incorrect may be less sensitive than measuring reaction time, the original observations by Phillips (1979) indicated that the task could successfully differentiate male and female performance (d = 0.84). The second spatial task was the Group Embedded Figures Task (Witkin et al., 1962). With this task 18 complex stimuli are shown and respondents have to trace the lines of a simpler shape within the complex shape. Scoring is one for a correct item, zero for an omitted or wrong answer.
In order to identify self-perceived gender-related trait possession in the subjects, the Bern Sex Role Inventory (BSRI; Bem, 1974), was employed. Twenty masculine traits, 20 feminine traits and 20 social desirability traits were presented to subjects in a random order. A trait glossary was also provided for the subjects’ use. Subjects had to identify the extent to which they possessed the trait. Subsequently a mean masculinity score and femininity score was determined. Conventional methods (Bem, 1974, Spence & Helmreich, 1978) have employed the t difference or median split measures to categorize subjects as undifferentiated, androgynous, masculine or feminine. In the present study in order to facilitate a multiple regression analysis four measures were undertaken:
1. a masculine score (M) 2. a feminine score (F) 3. the difference of M and F, M-F 4. the product of M and F, M x F
The M-F measure is a basic measure of the t difference measure originally employed by Bern (1974). A problem with this measure is that it does not distinguish between individuals achieving both low masculine and feminine scores (the undifferentiated category as defined by the median split technique) and other individuals scoring relatively highly on both the masculine and feminine traits (the androgynous category). The median split technique allows these distinctions to be made. In the present study the M x F measure allows for this discrimination to be made; ‘undifferentiated’ scores would achieve a relatively small M x F product score while an ‘androgynous’ pattern of scores would produce a relatively high M x F product score (Archer & Rhodes, 1989, and Archer, personal communication, Hargreaves, 1986). One confounding feature of the M x F measure lies in the difficulty of discriminating a high masculine x low feminine pattern of scores from a low masculine x high feminine pattern. Therefore, the M-F measure is retained to do just that. With the former pattern, high masculine x low feminine scores, a positive value would be elicited, while in the low masculine x high feminine score pattern, a negative value would occur. While the BSRI is considered to be fairly complex in its structure (Ballard-Reisch & Elton, 1992; Brems & Johnson, 1990; Wong, McReary & Duffy, 1990) this tool was employed so that comparison could be made with the earlier studies (e.g. the study of Signorella & Jamison, 1986). (See the Discussion for an elaboration of the considerations in the choice of trait self-report measure.)
The subjects were assessed in a formal classroom session, where the students were informed of their right of withdrawal at all times in the procedure. All subjects were given the BSRI and S & M task but procedural/access difficulties led to the GEFT being given to only 127 of the subjects. Where the two visuo-spatial tasks were presented to the same group, the task order was counterbalanced between groups. In all cases the BSRI was given last, either in the group situation or on an individual basis. Question order of the BSRI was randomized. All subjects received an immediate debriefing.
Note that for all descriptive statistics and effect sizes the median split measure of gender categorization is employed. This method should be considered only an approximation of the gender measures employed in the regression analyses. To facilitate comparison of performance in the two tasks, the effect sizes and regression analyses nominally employ the 127 subjects exposed to both spatial tasks. However, for within-sex regression analyses, in order to maximize sample size, all appropriate subjects were employed.
An analysis of internal reliability for the BSRI measure revealed a Cronbach’s alpha of .76 (N = 154) for the 20 feminine items and .72 (N = 154) for the 20 masculine items.
With N = 176, the mean masculine score was 4.76 (SD = 0.77) and the mean feminine score was 4.7 (SD = 0.56).
Tables 1a, 1b and 1c show the cross-tabulations of self-reported gender possession (measured by the BSRI and median split technique) with sex, institution and A-level course. An initial log-linear exploration indicated that an interaction existed between gender trait possession, institution and A-level course. Fitz-Gibbon (1985) found a relationship between institution type, A-level course and the Ravens Advance Progressive Matrices. Phillips (1979) noted a relationship between S & M performance and measures of intelligence. Thereform, in the present study, the variables, institution type and A-level course, were controlled for in each of the regression analyses by being included in the pool of predictors for all analyses.
Phillips S & M Mental Rotation Task
Descriptive statistics. With N = 127, the overall mean = 8.19 and standard deviation = 4.53. The breakdown of S & M performance with sex and gender categories is shown below in Table 2.
It can be seen from the table that the mean difference associated with sex is large; however, a large difference in mean performance between the gender categories also exists. The variability in mean mental rotation performance associated with the gender categories is graphically represented in Fig. 2.
Effect sizes. All subjects. The female vs. male effect size was d = 0.78. This value is consistent with the recent 3-D mental rotation studies reviewed by Linn & Petersen (1985) and Halpern (1992) and similar to Phillips’s (1979) original observations. The effect associated with perceived gender traits (median split technique) is shown in Table 3.
It should be noted that in this sample the effect sizes found with the androgynous subjects vs. feminine subjects and androgynous vs. undifferentiated were both d = 0.80, a comparable effect size with that of the sex variable. Thus, in this spatial task, self-perceived gender trait possession is associated with a relatively large effect size.
Female subjects. In the comparison of the gender categories, the maximum effect size obtained was d = 0.46 with androgynous vs. feminine subjects. Therefore, with female subjects only, the effect size is reduced.
Male subjects. In the comparison of the gender categories with male subjects, the maximum effect size was d = 1.08 with androgynous rs. undifferentiated subjects. With male subjects, therefore, the possession of distinct gender traits is associated with a relatively large effect size. This is shown in Fig. 3.
S & M task regression analysis. The following variables were employed as predictors in the (forward) stepwise regression analyses: Institute (FE or other), class (maths A-level or other A-level course), M (masculinity), F (femininity), M x F, M F, and sex (female or male). Two further variables which were included, M[sup2] and F[sup2], were employed as controls against ‘moderator effects’ (Lubinsky & Humphreys, 1990). The S & M final score was employed as the criterion.
Regression analysis. All subjects. The results of this analysis showed F(4,122) = 8.89, with p .84.
It can be noted from this analysis that the variable of sex is highly significant. However, the gender variable, M x F, adds to the model and is also significant in this analysis.
Female subjects. The obtained model showed F(2,119) = 6.08 with p .36.
Male subjects. The model obtained with this analysis showed F(3,50) = 5.46, with p .49.
Female subjects. With female subjects the model produced an F(2,92) = 6.78, with p .52.
Male subjects. The model achieved (Table 11) showed F(1,29) = 5.26, with p .17.
Therefore, while self-perceived gender traits were important for the GEFT regression analyses, it should be noted that relatively high performance was associated with either high masculine trait possession alone or a high masculine score in combination with a relatively low feminine score.
Combined spatial score
A final analysis was made with the S & M and GEFT scores combined. Each of the predictor variables identified above were employed in the stepwise regression and the criterion was the combined standardized scores from the S & M and GEFT performances.
The model produced an F(4,122) = 6.52, with p