|Home | About | Journals | Submit | Contact Us | Français|
Hormonal manipulations indicate that early androgens organize sex differences in spatial ability in animal models. In humans, spatial ability is also sexually dimorphic, and information about the effects of prenatal androgens on spatial ability can be obtained from studies of congenital adrenal hyperplasia (CAH) and the ratio of the second and fourth finger lengths (2D:4D). CAH is a condition characterized by prenatal overproduction of adrenal androgens, and several lines of evidence suggest that 2D:4D reflects prenatal androgen exposure. Some studies have found that these proxy measures of prenatal androgenization predict spatial ability, others have found no significant relationship, and yet others have obtained results in opposite directions. In light of these mixed findings, we conducted meta-analyses of published literature and unpublished results to determine if, across studies, either of these indications of prenatal androgens predicts spatial ability. In addition, we applied a trim and fill analysis to the data in search of asymmetry that might be an indication of publication bias. Results indicate that females with CAH have significantly better spatial abilities, and CAH males have worse spatial abilities, than do controls. Little or no relationship exists between 2D:4D and spatial ability. Implications for possible hormonal contributions and the developmental timing of sex differences in spatial cognition are discussed.
The largest known human cognitive sex differences are found in the domain of spatial ability (Maccoby & Jacklin, 1974), with three-dimensional mental rotation tasks showing the largest effect sizes in meta-analytic studies (Voyer, Voyer, & Bryden, 1995). Mental rotation ability is the ability to imagine objects from a perspective other than the one depicted. Sex differences in mental rotations have been observed in African (Mayes & Jahoda, 1988; Owen & Lynn, 1993), East Indian (Owen & Lynn, 1993), and Asian (Mann, Sasanuma, Sakuma, & Masaki, 1990) populations, as well as in Western cultures. Recently, sex differences in spatial ability greater than those observed in mental rotations have been reported for virtual water mazes, computerized versions of mazes used in animal models (Astur, Ortiz, & Sutherland, 1998).
Because virtual water maze and mental rotation performance correlate (Driscoll, Hamilton, Yeo, Brooks, & Sutherland, 2005), and because males outperform females on water mazes in both humans (Astur et al., 1998; Driscoll et al., 2003; Driscoll et al., 2005) and laboratory rats (Jonasson, 2005), rats are likely to represent useful models for possible hormonal contributions to sex differences in human spatial abilities. In rats, spatial ability is masculinized perinatally by testicular hormones. Several studies have shown that neonatal castration impairs maze learning in males (Dawson, Cheung, & Lau, 1975; Isgor & Sengelaub, 2003; Joseph, Hess, & Birecree, 1978; C. L. Williams, Barnett, & Meck, 1990), and neonatal testosterone treatment improves maze performance in females (Dawson et al., 1975; Isgor & Sengelaub, 1998, 2003; Joseph et al., 1978; Roof, 1993; Roof & Havens, 1992; Stewart, Skvarenina, & Pottier, 1975). However, there may be an optimal level of early androgen exposure beyond which spatial ability declines. Early androgen treatment improves spatial ability in female rats, but impairs it in gonadally intact males (Roof, 1993; Roof & Havens, 1992).
Some evidence suggests that early androgens also masculinize human spatial ability (reviewed in Puts, Gaulin, & Breedlove, in press). This evidence includes reported relationships between spatial ability and congenital adrenal hyperplasia (CAH), a condition characterized by elevated prenatal androgen levels. In CAH, an enzyme deficiency causes precursors of cortisol to be shunted down the androgen pathway, leading to an overproduction of adrenal androgens (Pang et al., 1980). Although the hormonal abnormalities of CAH are treated shortly after birth, girls with CAH show signs of elevated prenatal androgen exposure (for example, virilized genitalia) and tend to be masculinized along several behavioral dimensions (Berenbaum, 1999). Some studies have found CAH females to exhibit masculinized spatial abilities (Hampson, Rovet, & Altmann, 1998; Hines et al., 2003; Perlman, 1973; Resnick, Berenbaum, Gottesman, & Bouchard, 1986), although others have not (Baker & Ehrhardt, 1974; Helleday, Bartfai, Ritzen, & Forsman, 1994; Malouf, Migeon, Carson, Petrucci, & Wisniewski, 2006; McGuire, Ryan, & Omenn, 1975; Ripa, Johannsen, Mortensen, & Muller, 2003). Studies of spatial ability in CAH males have obtained equally inconsistent results, with some finding worse spatial ability in CAH males relative to controls (Hampson et al., 1998; Hines et al., 2003) and others finding no significant difference (Baker & Ehrhardt, 1974; McGuire et al., 1975; Resnick et al., 1986).
Possible relationships between early androgens and human spatial ability have also motivated digit ratio studies. The ratio of the lengths of the second and fourth fingers (2D:4D), is a putative marker for early androgens. Males have a lower 2D:4D than do females (Manning, Scutt, Wilson, & Lewis-Jones, 1998), a sex difference present by the end of the first trimester of gestation (Malas, Dogan, Evcil, & Desdicioglu, in press). Because of its early emergence, sexual dimorphism in 2D:4D is thought to be influenced by prenatal sex hormones (Manning et al., 1998). In particular, 2D:4D appears to be influenced by androgens: a more masculine digit ratio has been associated with CAH (Brown, Hines, Fane, & Breedlove, 2002; Okten, Kalyoncu, & Yaris, 2002, but see Buck, Williams, Hughes., & Acerini, 2003), as well as genetic predictor of androgen sensitivity (Manning, Bundred, Newton, & Flanagan, 2003). Because 2D:4D may reflect prenatal androgens, multiple studies have utilized this morphological marker to examine a possible role of early androgens on spatial ability (Alexander, 2005; Austin, Manning, McInroy, & Mathews, 2002; Coolican & Peters, 2003; Csatho, Karadi, & Kallai., 2005; Csatho et al., 2003; Kempel, Burk, & Hennig, 2005; Kempel, Gohlke et al., 2005; Loehlin, Luciano, Medland, & Martin, 2005; Manning & Taylor, 2001; McFadden & Schubel, 2003; Peters & Manning, 2005; Poulin, O’Connell, & Freeman, 2004; Putz, Gaulin, Sporter, & McBurney, 2004; Rahman, Wilson, & Abrahams, 2004; Sanders, Bereckzei, Csatho, & Manning, 2005; Scarbrough & Johnston, 2005; van Anders & Hampson, 2005). The results of these studies have also been quite variable, with some finding positive relationships, others finding negative relationships, and still others finding no significant relationship, even within a single sex (see Putz et al., 2004 for a partial review).
The purpose of the present study is to investigate relationships between CAH and 2D:4D and spatial ability in men and women across studies using meta-analytic methods.
All available published and unpublished studies that examined relationships between spatial ability and either CAH or 2D:4D were obtained. These studies were located via web-based searches using scientific internet search engines (e.g., PubMed, Scirus, Google Scholar), a query regarding such studies posted to an internet listserv (SEXNET) that reaches over 300 researchers of sex differences and sexual behavior, bibliographies of published papers, and personal communication with over 40 researchers in these areas. For studies in which the authors collected 2D:4D and spatial ability data but did not report the correlation between these measures, the correlations were requested from the authors.
Standardized mean differences were the effect size for CAH studies, and correlations were the effect size for 2D:4D studies. All measures of effect size contributing to the meta-analyses are from independent samples. Because it shows a large and reliable sex difference, performance on a three dimensional mental rotations test (3D MRT) was used as the measure of visuospatial ability with two exceptions. First, if a 3D MRT was not administered in a study, then the closest available measure that reliably exhibits a sex difference was used (e.g., 2D MRT). Second, if the seemingly closest available measure to 3D MRT did not exhibit a sex difference among controls (CAH studies), or if another visuospatial test exhibited a larger male advantage in a particular study, then the next closest test to 3D MRT was used. In one study (Hampson et al., 1998), results were presented both with and without an outlier that fell >4 SDs above the mean spatial performance of her group. The decision to use the effect size from the analysis with the outlier removed was made with a coin flip. For 2D:4D studies, correlations between spatial ability and right hand 2D:4D were used, because sex differences in 2D:4D and relationships between 2D:4D and behavioral traits have found to be greater in the right hand than in the left (Manning, 2002; McFadden & Schubel, 2003; T. J. Williams et al., 2000). If right 2D:4D was not reported or available from the authors, then mean 2D:4D was used (one male sample), and finally left 2D:4D was used (one male sample).
All meta-analyses were performed using Comprehensive Meta-analysis Program V.2 (Borenstein, Hedges, Higgins, & Rothstein, 2005). Meta-analyses were conducted on effect sizes using both fixed and random effects models. The fixed effects model assumes that there are no moderators in the relationships between the predictor variable (CAH or 2D:4D) and spatial ability, while the random effects model considers the presence of moderators a possibility. Although we present both the fixed and random effects model results, the random model is the most appropriate for these data, and we limit our discussion to it. In addition to estimating the population mean (CAH studies) or population correlation (2D:4D studies) and confidence interval, we applied a “trim and fill” analysis (Duval & Tweedie, 2000) to the data in search of asymmetry that might be an indication of publication bias. Trim and fill determines where missing studies are likely to fall, adds them to the analysis, and then recomputes the combined effect. An “omit one study” analysis was also performed. This is a type of sensitivity analysis that determines if the results of the meta-analysis would change through the deletion of a study. The effect size to be analyzed for CAH studies was the standardized mean difference (d), which expresses the mean differences between a CAH group and a control group in standard deviation units. Thus, a d of 1 would indicate that the mean of the CAH group was one standard deviation higher than the mean of the control group. The effect size to be analyzed for 2D:4D studies was the correlation coefficient between 2D:4D and spatial ability. Males and females were analyzed both separately and together.
Effect sizes for differences in spatial ability between CAH females and controls were obtained for nine samples from eight studies, involving a total of 128 CAH females and 108 controls (Baker & Ehrhardt, 1974; Hampson et al., 1998; Helleday et al., 1994; Hines et al., 2003; Malouf et al., 2006 [two samples]; McGuire et al., 1975; Perlman, 1973; Resnick et al., 1986, Table 1). CAH Females outperformed controls on spatial tasks. The population standardized mean difference in spatial ability between CAH females and controls was .465 for the random effects model and .336 for the fixed effects model (Figure 1). These results were robust with respect to the deletion of individual studies; “omit one study” analysis produced effect sizes ranging from .303 to .600 under the random effects model (Figure 2). Trim and fill analysis did not reveal any asymmetry in the data and did not change point estimates.
Five studies compared the spatial performance of a total of 61 CAH males to 64 controls (Baker & Ehrhardt, 1974; Hampson et al., 1998; Hines et al., 2003; McGuire et al., 1975; Resnick et al., 1986, Table 1). Overall, CAH males performed worse on spatial tasks than did controls for both fixed and random effects models. For the random effects model, the effect size of the standardized mean difference between CAH males and controls was −.602, and the corresponding value for the fixed effects model was −.583 (Figure 1). The deletion of individual studies in the “omit one study” sensitivity analysis produced effect sizes under the random effects model ranging from −.822 to −.406 (Figure 2), and thus did not change the conclusion that CAH males exhibit poorer spatial performance. Trim and fill analysis suggested the presence of one missing study to the left of the mean effect. With this study imputed, trim and fill shifted the population standardized mean difference from −.602 to −.718 for the random effects model. To the extent that publication bias may be present in this literature, it does not alter the conclusion that CAH males exhibit poorer spatial ability.
Twenty-one correlations between 2D:4D and spatial ability were obtained from 11 published (Austin et al., 2002; Coolican & Peters, 2003; Csatho et al., 2003; Kempel, Gohlke et al., 2005 [3 samples]; McFadden & Schubel, 2003 [2 samples]; Poulin et al., 2004; Putz et al., 2004; Rahman et al., 2004 [2 samples]; Sanders et al., 2005; Scarbrough & Johnston, 2005; van Anders & Hampson, 2005) and five unpublished (Alexander, 2005 [2 samples]; Csatho et al., 2005; Kempel, Burk et al., 2005; Loehlin et al., 2005; Peters & Manning, 2005) studies, involving a total of 101,488 subjects (Table 2). The fixed effects model rendered an effect size estimate of −.028, and the random effects model produced a smaller effect size estimate in the opposite direction: .005 (Figure 3). Sensitivity analysis revealed that these near-zero results were minimally affected by the removal of any given study. When individual studies were deleted from the analysis, the population effect size estimate for the random effects model ranged from −.016 to .008 (Figure 4). Trim and fill analysis suggested that one study was missing to the left of the mean effect. Using the random effects model, the trim and fill point estimate was −.001.
For relationships between 2D:4D and spatial ability in men, 18 effect sizes were obtained from nine published (Austin et al., 2002; Coolican & Peters, 2003; Kempel, Gohlke et al., 2005; Manning & Taylor, 2001; McFadden & Schubel, 2003 [2 samples]; Poulin et al., 2004; Putz et al., 2004; Rahman et al., 2004 [2 samples]; Sanders et al., 2005 [3 samples]) and four unpublished (Alexander, 2005 [2 samples]; Kempel, Burk et al., 2005; Loehlin et al., 2005; Peters & Manning, 2005) studies, involving 117,353 total subjects (Table 2). The effect size estimate was −.030 using the fixed effects model, and −.068 using the random effects model (Figure 3). These near-zero point estimates were also robust with respect to removal of individual studies from the analysis; the “omit one study” analysis produced population correlation estimates ranging from −.073 to −.030 (Figure 4). Trim and fill analysis suggested that three studies were missing to the right of the mean correlation. With these studies imputed, the point estimate under the random effects model was −.015.
Given these results, we offer the tentative conclusions that CAH females have an advantage in spatial ability, CAH males have a disadvantage in spatial ability, and spatial ability is very weakly, if at all, associated with 2D:4D.
We estimate the population standardized mean difference between CAH individuals and controls to be .465 for females and −.602 for males. According to Cohen (1988), these represent small- to medium-sized effects. Results were robust with respect to meta-analytical model (fixed vs. random effects) and sensitivity analyses. “Omit one study” analyses demonstrated minimal effects of deleting individual studies, indicating that these results do not rely on the inclusion of outlier studies or on the application of decision rules to idiosyncratic studies. In addition, trim and fill results suggested no evidence of publication bias in female studies, and, to the extent that publication bias exists in the male studies, it serves to underestimate the strength of the relationship.
One interpretation of these results is that the additional prenatal androgens provided by CAH increased spatial ability in females and decreased it in males. This interpretation accords with research on rats, in which males exhibit superior spatial performance to females (Jonasson, 2005), and testosterone treatment improves spatial ability in females (Dawson et al., 1975; Isgor & Sengelaub, 1998, 2003; Joseph et al., 1978; Roof, 1993; Roof & Havens, 1992; Stewart et al., 1975) but worsens it in males (Roof, 1993; Roof & Havens, 1992).
This interpretation is also consonant both with studies finding behavioral masculinization in CAH females (Berenbaum, 1999) and with studies suggesting that human spatial ability is related to prenatal androgens. For example, second trimester testosterone levels have been found to predict spatial ability positively in girls and negatively (but less clearly) in boys at age seven (Grimshaw, Sitarenios, & Finegan, 1995). Girls with male twins have also been found to exhibit superior spatial ability, possibly because of in utero exposure to androgens produced by their twins (Cole-Harding, Morstad, & Wilson, 1988). Additionally, females with Turner Syndrome, in which androgen and estrogen production are extremely low due to undifferentiated gonads (Hojbjerg Gravholt, Svenstrup, Bennett, & Sandahl Christiansen, 1999; Ross et al., 2002), exhibit specific cognitive deficits in spatial ability (Nijhuis-van der Sanden, Eling, & Otten, 2003). Complete androgen insensitivity syndrome (CAIS) individuals, who have a 46, XY karyotype, develop testes that remain undescended, and produce normal-to-high male levels of testosterone, are nonetheless phenotypically female because they lack functional androgen receptors (Imperato-McGinley et al., 1982). Females with CAIS performed worse on spatial tasks than both their male and non-CAIS female relatives (Imperato-McGinley, Pichardo, Gautier, Voyer, & Bryden, 1991). This finding is consistent with testosterone improving spatial ability in men and in women with functional androgen receptors, although ovarian hormone production in unaffected females may have caused them to differ from CAIS women.
Despite evidence that spatial ability is related to early androgens, and that CAH females are masculinized in other behavioral domains, the interpretation that elevated prenatal androgens produced the observed differences between CAH and unaffected individuals must be made cautiously. First, CAH individuals also differ from unaffected individuals in glucocorticoid levels, which may affect spatial abilities. However, if the relationship between glucocorticoids and spatial ability were simple, one would predict that the direction of the CAH effect on performance would be the same in males and females, since both see a profound lack of glucocorticoids. On the other hand, it is possible that the lack of glucocorticoids affects the two sexes differently. For example, glucocorticoid treatment impaired spatial ability more in female than in male rats (Vicedomini, Nonneman, DeKosky, & Scheff, 1986). On the other hand, stress increases glucocorticoid levels, and both prenatal and postnatal stress have been found to increase spatial ability in female rats and decrease it in males (Bowman et al., 2004; Kitraki, Kremmyda, Youlatos, Alexis, & Kittas, 2004). Second, although CAH is treated soon after birth with a synthetic glucocorticoid, improper management can lead to health complications and impaired cognitive performance (Berenbaum, 2001). This would appear to explain only why CAH males exhibit lower spatial abilities, however, and not why spatial abilities are elevated in CAH females. Of course, it is possible that imperfect glucocorticoid management decreases spatial ability in both males and females with CAH, but that the coincident increase in prenatal androgen masks this effect in females. Third, prenatal androgens may not consistently be elevated in CAH males, although evidence suggests that androgen levels are higher in CAH males during at least some stages of prenatal development. For example, androgen levels assayed via amniocentesis were higher in CAH males than in controls (Dorr & Sippell, 1993; Wudy, Dorr, Solleder, Djalali, & Homoki, 1999), and CAH males were found to have more masculine 2D:4D on both hands compared to their male relatives (Brown et al., 2002; Okten et al., 2002). However, negative feedback on testicular androgen production may normalize or even reduce androgen levels during the critical period for differentiation of spatial ability. Consequently, spatial deficits in CAH males may result from reduced androgens or even be unrelated to prenatal androgens, although only the former possibility explains elevated spatial ability in CAH females. It is also possible that a particular level of prenatal androgen exposure is optimal for adult spatial ability and that either lesser or greater exposure has less effect on this behavior. Finally, our CAH analyses were based on a relatively small amount of data. These analyses should be conducted again as more data cumulate.
In contrast to the small- to medium-sized relationships observed between CAH and spatial ability, correlations between 2D:4D and spatial ability were negligible. Cohen (1988) suggests that a correlation of .1 represents a small effect, and in this meta-analysis, population correlations under the random effects model were .005 for females and −.068 for males. These effect sizes remained very small, regardless of which study was deleted in the “omit one study” sensitivity analysis. Trim and fill analysis suggested some publication bias, and after correcting for this bias by imputing hypothetical missing studies, the point estimates were shifted even closer to zero: −.001 for females, −.015 for males.
Assuming that both 2D:4D and spatial ability sexually differentiate under the influence of prenatal androgens, this essential lack of correlation may appear paradoxical. At least two possible explanations exist. First, the sex difference in 2D:4D accounts for only around 6–9% of the variation in 2D:4D (Coolican & Peters, 2003) and therefore probably only weakly reflects prenatal hormone regimes (van Anders & Hampson, 2005). In fact, this estimate represents the maximum variance in digit ratio that could be attributable to sex differences in prenatal hormones regimes. Because within-sex variation in sex hormones must be lower than between-sex variation, even less variance in digit ratio can be explained by within-sex variation in sex hormones. Males and females also overlap considerably in spatial ability (Maccoby & Jacklin, 1974), so the relationship between these two imperfect hypothetical correlates of prenatal androgens (2D:4D and spatial ability) would also tend to be weak.
2D:4D and spatial ability may also differ in the developmental timing of their putative sensitivity to androgens (Putz et al., 2004; van Anders & Hampson, 2005). 2D:4D should predict sexually dimorphic traits that differentiate under the influence of the same hormones during the same critical period. Sexual orientation appears to be one such trait; homosexual women have more masculine digit ratios on average than do heterosexual women (McFadden et al., 2005; T. J. Williams et al., 2000). However, if androgen levels during the critical periods for 2D:4D and spatial ability sexual differentiation are unrelated, then 2D:4D and spatial ability may also be uncorrelated.
Given that 2D:4D appears to have sexually differentiated by the ninth gestational week (Malas et al., in press), and testicular androgen production cannot begin until the fetal Leydig cell population arises at six weeks postconception (O’Shaughnessy, Baker, & Johnston, 2006), 2D:4D probably sexually differentiates during this interval, and spatial ability probably differentiates sometime thereafter. Hines et al. (2003) suggested that this period may occur as late as the first six months of postnatal life, but the fact that CAH is detected and treated soon after birth, especially in females, would seem to argue against differentiation occurring much after birth. In rats, spatial ability sexually differentiates during the first two of weeks after birth (Dawson et al., 1975; Joseph et al., 1978; Stewart et al., 1975). Because rats are born relatively underdeveloped, this corresponds approximately to the end of the third trimester of gestation in humans (Nunez & McCarthy, 2003), which would appear to be a likely time frame for the sexual differentiation of human spatial ability. If the hormonal abnormalities associated with CAH begin before the critical period for 2D:4D sexual differentiation and persist through the critical period for sexual differentiation of spatial ability, this would explain why both 2D:4D and spatial ability relate to CAH, even though they do not appear to be related to one another.
These results, though tentative, can inform hypotheses regarding several aspects of human sexual differentiation. First, relationships between CAH and spatial ability support the hypothesis that early androgens affect the development of at least one cognitive sex difference, visuospatial cognition. However, decreased prenatal glucorticoid levels in CAH individuals may also affect spatial abilities. Second, these results suggest that the nature of a putative relationship between early androgens and human spatial ability is curvilinear, as in laboratory rats, with very low levels and very high levels of androgens associated with poorer spatial ability. This interpretation relies on higher androgen levels in CAH males than in controls, for which some evidence exists. Finally, moderate associations between CAH and spatial ability, and very small correlations between 2D:4D and spatial ability, provide evidence regarding the timing of sexual differentiation in the neural systems underlying spatial cognition. Specifically, spatial ability probably differentiates after 2D:4D, perhaps in the second or third trimester of gestation, or even in the early postnatal period.
Given the small number of CAH studies analyzed and some evidence for publication bias in studies of 2D:4D, we encourage replication of these results as more data accumulate. We also encourage improved reporting of results. For CAH studies, authors should report means and standard deviations by sex. For both 2D:4D studies and CAH studies, separate correlation matrices by sex should provide intercorrelations of all variables.