Two- vs. three-dimensional presentation of mental rotation tasks: Sex differences and effects of training on performance and brain activation.

The well-documented sex difference in mental rotation favoring males has been shown to emerge only for 2-dimensional presentations of 3-dimensional objects, but not with actual 3-dimensional objects or with virtual reality presentations of 3-dimensional objects. Training studies using computer games with mental rotation-related content have demonstrated training effects on mental rotation performance. Here, we studied the combined effect of a two-week mental rotation (MR) training on 2-dimensional vs. 3-dimensional presentations of a classic Shepard-Metzler task (presented in a pretest-training-posttest design) and their accompanying cortical activation patterns assessed via EEG in a sample of 38 male and 39 female adolescents of about 15 years of age. Analysis of one performance parameter (reaction times) displayed only main effects of dimensionality (with shorter RTs on the 3D vs. 2D version of the MR task) and of training (significant shortening of RTs), but no significant sex difference. Analysis of the other performance parameter (scores) in the MR task revealed a sex difference favoring males that first, appeared only in the 2D version, but not in the 3D version of the MR task and, secondly, diminished after training. Neurophysiologically we observed a complex sex × dimensionality × training × hemisphere interaction showing that the hypothesized decrease of brain activation (increase in neural efficiency) with training emerged for males in both 2D and 3D conditions, whereas for females this decrease was found only in the 3D but not with the 2D version of the MR task.

In differential psychology a revived interest in sex differences in personality traits and especially in cognitive ability variables can be observed recently. In the latter domain some of the largest sex differences can be found in visuo-spatial abilities. Especially the ability of mental rotation, as measured, e.g., by the classic mental rotation task by Shepard and Metzler (1971) produces the largest and most consistent gender differences in the spatial ability domain, with meta-analyses by Linn and Petersen (1985) and Voyer, Voyer, and Bryden (1995) showing effect sizes around 0.95 favoring males. More recently, however, the generality of this phenomenon has been challenged. In a study by McWilliams, Hamilton, and Muncer (1997) it has been shown that the male advantage disappears completely when the rotation task was presented in the form of true three-dimensional models, as compared to presenting 2-dimensional prints of 3-dimensional objects. Two studies (Larson et al., 1999; Parsons et al., 2004) have employed virtual reality spatial rotation tasks as compared to a classic paper-and-pencil (PP) version and, while replicating the sex difference in the PP version, they found no sex effects in the virtual environment. This suggests that the female disadvantage might not lie in the process of mental rotation per se but in the derivation of a 3-dimensional representation from a 2-dimensional image.

Furthermore, the stability of the sex difference is called into question by studies showing that mental rotation performance can be enhanced through practice: Several studies demonstrated a considerable performance improvement through practicing computer games such as Tetris, Blockout, a 3D version of Tetris, or action video games in general (De Lisi & Cammarano, 1996; Feng, Spence, & Pratt, 2007; Haier et al., 1992; Okagaki & Frensch, 1994). With respect to sex differences Kass, Ahlers, and, Dugger (1998) showed that a brief mental rotation training caused an increase in mental rotation performance in women to the level of men, an effect that was still evident when re-testing the participants three weeks later.

Finally, the malleability of mental rotation performance has been demonstrated in studies relating everyday spatial activities and spatial abilities: Newcombe, Bandura, and Taylor (1983) found a substantial positive relationship between the Spatial Relations test of the Differential Aptitude Test (DAT) and the frequency with which a multitude of daily spatial activities was performed only in females (r = .40). Similar findings were reported by Quaiser-Pohl and Lehmann (2002) who found significant relationships of spatial technical activities, sport activities as well as computer activities with mental rotation test performance only in females, not in males. However, in a more thorough analysis of computer game preference Quaiser-Pohl, Geiser, and Lehmann (2006) reported a relationship of action-and-simulation-playing (involving spatial demands) with MRT performance only in boys, not in girls (which might be due to the sample being secondary school children here as opposed to undergraduates in the 2002 study of this group of researchers).

From this collection of evidence it can be concluded that (1) spatial abilities seem rather malleable, i.e., they can be changed through (everyday or experimentally induced) practice and that these effects are rather gender-specific, and (2) that the emergence of the sex difference in mental rotation seems to depend strongly on the mode of task presentation (especially 2D vs. 3D presentation).

However, to our knowledge, the combined effect of these two aspects has not been studied yet, which is a first major aim of our study. Even more important for this paper is that we followed a neuroscientific approach in this study: We wanted to study the effects of a mental rotation training on neurophysiological measurements of brain activation and compare, first, this between sexes and, second, between 2- and 3-dimensional presentations of the very same mental rotation task. To provide a basis for the derivation of our research questions and hypotheses, we will outline the current state of knowledge regarding neuroscience correlates of human cognitive ability in general and spatial ability in particular in the following paragraphs. Subsequently we will elaborate on recent research on practice-related changes of brain activation patterns.

Neurophysiological correlates of general cognitive ability or intelligence have been studied intensely throughout the past 20 years since the first report on a negative relationship between brain activation and intelligence (Haier et al., 1988), i.e., brighter (as compared to less intelligent) individuals display a lower and more focused brain activation during cognitive task performance. This observation led to the postulation of the neural efficiency hypothesis of human intelligence, according to which individual differences in intelligence are associated with differences in the general efficiency of brain functioning. Even though the neural efficiency hypothesis could be corroborated by more than 30 studies using different neurophysiological methods (such as EEG, PET or fMRI) and employing a wide range of cognitive tasks (from elementary cognitive to complex reasoning demands, see Neubauer & Fink, 2009, for a review), other studies called the generality of this phenomenon into question. In their review of the pertinent literature Neubauer and Fink (2009) suggested several moderating variables, with sex being among the most important ones. In two studies the research group around Neubauer (Neubauer, Fink, & Schrausser, 2002; Neubauer, Grabner, Fink, & Neuper, 2005) has shown that the neural efficiency hypothesis received support in the verbal condition only for females and in the figural condition only for males.

Directly related to this assumption and perhaps the most critical moderating variable appears to be the level of prior knowledge or expertise in a certain domain. In the study of Grabner, Stern, and Neubauer (2003), lower and higher IQ taxi drivers were tested with a novel (intelligence-related) and familiar (expertise-related) task. Neural efficiency could only be observed in the novel but not in the familiar task. Referring to the finding, that “once an elaborate domain-specific knowledge base has been constructed, intelligence loses its impact” (p. 95), we concluded that the acquisition of expertise (through learning) in a certain field seems to increase the efficiency of the brain (cf. also Grabner, Neubauer, & Stern, 2006).

Despite the already uncovered variables moderating the neural efficiency phenomenon, there still exist some contradictory findings of positive or null intelligence-activation correlations that are in need of explanation (e.g., Geake & Hansen, 2005; Gray, Chabris, & Braver, 2003; Klimesch, Doppelmayr, Pachinger, & Russegger, 1997; cf. Neubauer & Fink, 2009, for a review). In addition, from a theoretical viewpoint, it is largely unknown where the more efficient brain functioning in brighter individuals derives from. Hypotheses that have already been put forward in this context, for instance, trace neural efficiency back to a higher degree of myelination of the brain (Miller, 1994), to more strongly pruned neural networks (Haier, 1993), to a higher level of dendritic and axonal arborization (Garlick, 2002) or to the availability of more gray matter resulting in less energy use (Haier, Jung, Yeo, Head, & Alkire, 2004).

When it comes to more specialized cognitive abilities we find much less research at hand. A first group of studies that should be considered here show that cortical activation patterns correlated with figural–spatial cognitive processes can quite well be discriminated from those reflecting rather verbal information processing (Burbaud et al., 2000; Reichle, Carpenter, & Just, 2000; Sohn et al., 2004). Most of the research on neurophysiological correlates of visuo-spatial abilities comes from neuroscience approaches to mathematical performance, with a focus on parietal areas of both hemispheres, which are involved in visuo-spatial processes and skills (Cabeza & Nyberg, 2000; Trojano et al., 2004). As a review by Houde and Tzourio-Mazoyer (2003) suggests, visuo-spatial processes are also involved in mathematical performance, especially in exact computations that require more complex operations than merely retrieving arithmetic facts from long-term memory. Additional evidence for the link between superior mathematical performance and figural–spatial representations comes from neuropsychological and neurophysiological studies on mathematical giftedness (O'Boyle & Benbow, 1990; O'Boyle, Benbow, & Alexander, 1995; Singh & O'Boyle, 2004). O'Boyle and co-workers have described two central characteristics of mathematically gifted male adolescents: (a) enhanced development and subsequent processing reliance on the capacities of the right hemisphere, which is hypothesized to be strongly involved in spatial abilities (cf. Vogel, Bowers, & Vogel, 2003), and (b) enhanced interhemispheric interactions, suggesting a more fine-tuned and coordinated exchange of information between the hemispheres (see also O'Boyle et al., 2005). Interestingly, this pattern of results was not found in female adolescents, who displayed rather diffuse bilateral activation, which was speculated to reflect a verbal processing advantage (O'Boyle & Gill, 1998). In sum, superior performance in complex mathematical tasks appears to involve visuo-spatial processes and neural networks. At present, though, it is a matter of debate, whether a differential reliance on verbal and figural–spatial representations or strategies might be jointly responsible for individual differences in general, and gender differences in particular.

Neuroimaging studies of mental rotation in particular are scant and can be divided into two groups: Those which have a special focus on sex differences versus others seeking for general brain correlates not exploring sex differences. For the latter Zacks (2008) has recently provided a review and meta-analysis, however, considering only fMRI and PET studies. The meta-analysis identifies mainly areas of the posterior parietal cortex (extending down into the superior posterior occipital cortex) which are engaged by mental rotation tasks. In many mental rotation tasks, motor areas of the posterior frontal cortex are engaged, too. Most recent studies have focused on relationships between brain activation and angular disparity (Weiss et al., 2009), or have examined the neural correlates of 2- versus 3-dimensional mental rotation of three-dimensional objects (Kawamichi, Kikuchi, Noriuchi, Senoo, & Ueno, 2007; Kawamichi, Kikuchi, & Ueno, 2007). At this instance, however, it should be stressed that—in contrast to our present study—in all neuroscientific studies on mental rotation even the 3D MRTs were presented in the traditional way, i.e., on a monitor with 2D-presentation. We have not yet located any single study that compared brain activation patterns of 2-dimensional vs. real or virtual 3-dimensional presentations of 3D-objects.

A second larger group of brain activation studies focused on sex differences during mental rotation. On the basis of the neural efficiency hypothesis outlined above Jaušovec and Jaušovec (2008) reported the hypothesized inverse brain activation–ability relationships in the visuo-spatial intelligence domain for men and in the emotional intelligence domain for women (when performing a task requiring recognizing emotions), which seems similar to the findings from Neubauer et al. (2002, 2005) when comparing verbal vs. visuo-spatial (rotation) tasks (see above). Roberts and Bell (2003) compared brain activation of males and females for 2D vs. 3D rotation tasks (again involving 3-dimensional rotation but not stimulus presentation) and found mostly hemispheric differences in the parietal cortex. In the 2D task males showed more right hemisphere, females more left hemisphere activation, while in the 3D task both sexes showed more right hemispheric involvement. Interestingly, however, behavioral performance sex differences resulted only in the 3D task, not the 2D task (and were favoring males, as hypothesized). The same authors (2000) reported that the generally stronger parietal activation in males vs. females during mental rotation emerges only in adulthood, but could not be observed in 8-year-olds. Other studies have focused not on brain activation per se but on the communication of brain areas that can be measured via EEG coherence measures. Rescher and Rappelsberger (1999) observed sex differences in local coherence measures as well as in interhemispheric coherence pointing towards a generally more symmetrical coherence in females (cf. Gootjes, Bruggeling, Magnee, & Van Strien, 2008, for a very recent study on this issue). Finally, one study has looked into sex differences in brain morphology in relation to MRT performance (Koscik, O'Leary, Moser, Andreasen, & Nopoulos, 2009): Women had more gray matter in the parietal lobe than men and this was disadvantageous for women, whereas men had greater parietal surface area which constituted a performance advantage for men on MRT. From this the authors conclude that the structural sex difference could be a neurobiological substrate for the sex difference in mental rotation performance.

As we combined the neuroscientific approach to the gender difference in mental rotation (as well as the comparison of the classic versus virtual reality presentation) with the study of training effects of a mental rotation training here, the effects of training on brain activation patterns shall finally be dealt with. Research on the neurophysiological correlates of practice or training has been conducted for a wide variety of demands, inter alia comprising motor learning, passive visual perception, mirror reading, artificial grammar learning, verb generation, working memory, and reasoning (cf. Kelly & Garavan, 2005 for a review). Three types of changes were found to accompany skill acquisition or improvement in all these domains: (a) the brain region engaged by a task remains constant but the activation in this region increases (probably due to its stronger involvement or the recruitment of additional brain circuits; e.g. Olesen, Westerberg, & Klingberg, 2004), (b) the brain region remains constant but the activation decreases (probably due to higher neural efficiency; e.g., Haier, Siegel, MacLachlan et al., 1992), or (c) the engaged brain regions change and/or some regions are more and others less activated afterwards (probably due to a functional reorganization of the underlying neural networks; e.g., Gevins & Smith, 2000).

However, only a few studies have dealt with the role of individual differences in this context, which have turned out to play a crucial role in activation changes through learning or practice. For instance, Haier, Siegel, MacLachlan et al. (1992) trained participants on the computer game Tetris over 4 to 8 weeks which resulted in a sevenfold performance improvement. PET scans before and after the practice period revealed decreases of activation from pre- to posttest in a number of brain regions. Most interestingly, these practice-related activation decreases were significantly associated with participants' performance change (the higher the performance improvement, the stronger the activation decrease) and their general intelligence level (the brighter the individual, the more the activation was reduced after training; Haier, Siegel, Tang, Abel, & Buchsbaum, 1992). Similarly Neubauer, Grabner, Freudenthaler, Beckmann, and Guthke (2004) employing a pretest–training–posttest learning found that the degree of efficiency increases from pre- to posttest was correlated substantially (up to .54) with participants' intelligence: Likewise, brighter individuals more strongly decreased their cortical activation from pre- to posttest, particularly in one of those brain regions most strongly associated with reasoning processes (i.e., the prefrontal cortex, that is part of the fronto-parietal network proposed by Jung & Haier, 2007).

Finally, it should be mentioned that our study focused on the age group of adolescents. Repeatedly, the considerable importance of spatial ability for educational outcomes in the so-called STEM domains has been emphasized (science, technology, engineering, maths; cf. Wai, Lubinski, & Benbow, 2009). From this practical–educational viewpoint supporting adolescent girls in the spatial ability domain can be considered a very important goal for societies that need more young people educated in technical domains and that cannot fulfill this demand with men alone who self-select such professions. It is generally acknowledged that today's knowledge-based societies especially need to recruit more women for STEM educations and professions.

To sum up, we want to answer the following questions with this study; the questions are always formulated with respect to (a) behavioral effects and (b) neurophysiological effects:

Is the male advantage in mental rotation tasks reduced or even annihilated when 3-dimensional figures are presented in a ‘real’ (virtual) 3D mode as compared to the ‘standard’ 2-dimensional presentation? On the basis of the reported evidence we presume a reduced sex difference with 3D (vs. 2D).

Do males show a higher neural efficiency as compared to females in the 2D version? Is the sex difference in neural efficiency reduced with the 3D presentation?

Does training of mental rotation reduce the female disadvantage? We predict a smaller sex difference in mental rotation performance after training.

Does training increase neural efficiency? Is there a gender difference in the increase in neural efficiency? We predict higher neural efficiency after training as well as a reduced sex difference in neural efficiency after the training.

What are the joint effects of training and presentation mode (2D vs. 3D) with respect to sex differences in mental rotation performance? As these two experimental factors have, at least to our knowledge, not been studied conjointly before, we cannot make any predictions regarding this question.

What are the joint effects of training and presentation mode (2D vs. 3D) with respect to sex differences in neural efficiency? Again, we cannot make any predictions regarding this question.

Method

Participants

77 participants (38 males, 39 females) were selected from a large pre-test pool of participants (N = 929) on the basis of their visuo-spatial intelligence, i.e., aiming at a large variability. Furthermore, participants were also selected to present a normal distribution (Skewness = − 0.17, SDSkewness = 0.28; Kurtosis = − 0.87, SDKurtosis = 0.56). Males and females were matched regarding their IQ scores in order to avoid a confounding effect of sex difference on spatial ability. On average study participants were 15.02 years old (SD = 0.55). Furthermore, all participants were right-handed and had normal or corrected-to-normal vision. None of them reported any medical or psychological disorders and all subjects knew that they were going to be tested twice with a training in-between. Participants provided written informed consent of their parents before the experiment started. The participants were given vouchers for completing the training sessions and participating in the EEG posttest session.

There was no control group since the introduction of a control group would not have been easily justified as we were primarily interested in the interactions of sex with training and sex with 2- vs. 3-dimensional mental rotation tasks.

Psychometric tests

Prior to the EEG sessions, participants were screened with respect to their visuo-spatial intelligence by using a figural subtest (cube task [“Würfel erkennen”]) of the well-established German/Austrian intelligence test “Intelligenz-Struktur-Analyse” (ISA; ITB & Gittler, 1998). This intelligence test was constructed according to the tradition of tests derived from Thurstone's model of intelligence. Psychometric properties reported in the manual show satisfactory reliability coefficients, and conclusive information on the validity of the test battery is given. Moreover, measures of the participants' temporary mood during the EEG session were included as control variables and were assessed with the “State-Trait-Anxiety Inventory” (STAI) by Spielberger, Gorsuch, and Lushene (1970).

Experimental tasks

Overall, two experimental and two tasks, for other purposes not dealt with in this paper, with 45 trials each were presented while EEG was recorded. For the two experimental tasks pairs of Shepard–Metzler (SM) Figures were built out of eleven three-dimensional cubes (see Fig. 1). In one experimental condition SM-Figures were shown on a screen in a conventional 2D presentation mode. Participants were instructed to judge whether the presented figures were identical or different ones. In the second experimental task all SM-Figures were presented in a 3D presentation mode. For that purpose, a 3D projector was used that projects two images involving slightly different presentations (i.e., different angles) of one object. The 3D effect results from presenting the object from one angle to one eye and from the other angle to the other eye, an effect that is created through active 3D glasses. To allow for a better comparison also the 2D presentation was performed with participants wearing these glasses, but for 2D they were switched-off.

Fig. 1

Schematic time course and EEG measurement intervals for the Shepard—Metzler-Figures. The intervals relevant for the ERD computation are depicted as R (reference interval) and A (activation interval).

EEG recording

The EEG was measured by means of gold electrodes (9 mm diameter) located in an electrode cap in 33 positions (according to the international 10–20 system); a ground electrode was located on the forehead, the reference electrode was placed on the nose. To register eye movements, an electrooculogram (EOG) was recorded bipolarly between two gold electrodes diagonally placed above and below the inner, respectively, the outer canthus of the right eye. This electrode placement allows for detecting both vertical and horizontal eye movements using only one EOG channel. The EEG signals were filtered between 0.1 Hz and 100 Hz; an additional 50 Hz notch filter was applied to avoid power line contamination. Electrode impedances were kept below 5 kΩ for the EEG and below 10 kΩ for the EOG. Trigger signals for the stimulus presentation and the responses were also recorded. All signals were sampled at a frequency of 256 Hz.

Based on visual inspection of the topographical distribution of the event-related desynchronization (ERD), for further analyses, the ERD data was aggregated for different electrode locations, distinguishing the hemispheres as well as anteriofrontal (AF), frontal (F), frontocentral (FC), centrotemporal (CT), centroparietal (CP), parietotemporal (PT) and parietooccipital (PO) brain areas. The electrode positions were aggregated as follows: anteriofrontal left (FP1 and AF3), anteriofrontal right (FP2 and AF4), frontal left (F3 and F7), frontal right (F4 and F8), frontocentral left (FC1 and FC5), frontocentral right (FC2 and FC6), centrotemporal left (C3 and T3), centrotemporal right (C4 and T4), centroparietal left (CP1 and CP5), centroparietal right (CP2 and CP6), parietotemporal left (P3 and T5), parietotemporal right (P4, T6), parietooccipital left (PO3, PO5, and O1) and parietooccipital right (PO4, PO6, and O2). The midline electrodes (FZ, CZ, and PZ) were not included in the analyses (as we were also interested in hemispheric differences).

For the presentation of the experimental tasks, a PC (g.STIMunit, Guger Technologies, Austria) with an external response-console consisting of two horizontally arranged buttons for the YES-responses on the top of the response-console and two horizontally arranged buttons for the NO-responses on the bottom of the console was used. In order to avoid a confounding effect with hemispheric differences, participants were instructed to respond simultaneously with their index fingers in case that a YES-response was required and with their thumbs in case that a NO answer was required.

Training

In between the two experimental test sessions a training phase at participants' home took place. All pupils underwent a 14-day long computer-based mental rotation training intervention with seven different training modules. Every second day participants had to complete one out of the seven modules. We considered seven training modules to be sufficient for a long-term increase of the individual mental rotation performance. Furthermore, a training intervention consisting of seven modules was meant to be short enough to keep the teenagers' commitment and attention until the whole training was completed. However, each module combined the following tasks:

Cube tasks from the Intelligence-Structure-Test (“Intelligenz-Struktur-Test 2000R”) by Liepmann, Beauducel, Brocke, and Amthauer (2007): Participants are presented five reference cubes with different signs on their sides. In addition, in each of the 10 trials one critical cube is shown, and participants have to indicate which of the reference cubes is shown. The critical cubes are rotated and the appropriate cube has to be indicated.

Brick tasks from the Bricks-Test (“Bausteine-Test BST”) by Birkel, Schein, and Schumann (2002): Participants are presented four different reference bricks, each of them composed of four three-dimensional cubes. Additionally, in each of the 70 trials one critical figure is presented, and participants have to judge which two of the reference bricks were used to create the critical figure. Participants are not allowed to use one reference brick twice.

Various trials of the computer game Tetris (programmed by members of the research team responsible for this study): Initially, participants are presented with an empty rectangular playing field. On each trial of the game one of seven randomly created shapes appear at the top of the screen and descend towards the bottom. Each of the Tetris shapes is composed of four squares and can be used to create lines at the bottom of the screen. The player is allowed to press keys which will either rotate the shape counterclockwise in increments of 90° or slide it horizontally on the playing field. The aim is to pile the shapes such that their components create continuous lines across the playing field. As soon as a line is created, it will disappear and any block above it will drop down to that level. The game continues with shapes descending at increasingly faster rates. The game is over when the bricks pile up to the top of the playing field. The number of removed lines is the participants' score.

The seven training modules differed with regard to their difficulty. The first module combined the easiest cube and brick tasks whereas the last module combined the most difficult and most complex ones. The difficulty of the computer game Tetris increased as soon as a line disappeared. After a line disappeared the shapes descended at faster rates.

Participants' training data was registered and analyzed. Only participants who completed the whole training were invited for retest. However, 12 study participants (8 females, 4 males) did not complete the training and were therefore not invited for the posttest session. Those 12 participants did not differ from the ones who completed the training with regard to their visuo-spatial intelligence or their mental rotation performance in the pretest as was revealed by t-tests.

Procedure

The EEG session started with mounting the electrodes and checking the impedances. Subsequently, the participant was seated comfortably in the darkened sound-attenuating EEG recording room, and two 2-min EEG sequences under resting conditions were recorded, the first one with eyes closed, the second one with eyes open. Then, the participants started to work on the experimental tasks described above. Another two 2-min resting EEG sequences (with eyes closed and eyes open, respectively) were recorded. In order to reduce any possibility of task influence on outcome, the presentation order of the tasks was counterbalanced. During the EEG sessions, short breaks of 5 min each were allowed. In total, the EEG session lasted about 2 h.

Cortical activation was quantified by means of the event-related desynchronization approach (ERD; Pfurtscheller & Aranibar, 1977; see also Pfurtscheller & Lopes da Silva, 1999), which is based on the phenomenon that the amount of alpha power decreases during cognitive task performance (activation interval) compared to a resting state (reference interval).

In order to measure participants' learning progress we administered pre- and posttests and, furthermore, calculated residual gain scores. For computing residual gain scores the raw pre- and posttest scores were first transformed into T-values with a mean of 50 and a standard deviation of 10. Then residual gain scores were computed separately for each sex and each presentation mode (2D and 3D) following Formula 1 from Guthke, Jäger, and Schmidt (1983), since these scores are uncorrelated with the initial performance status. However, residual gain scores were exclusively used to measure participants' learning progress (i.e., change from pre to posttest). Residual gain scores were not used for indicating sex differences in mental rotation performance.

EEG analyses

For reasons given in Klimesch (1999) the frequency borders of the analyzed alpha bands were determined individually for each participant by using the dominant EEG frequency indicated by the highest amplitude (peak) in the alpha band (the so-called Individual Alpha Frequency, IAF) as an anchor point. First, power spectra for all recording positions were calculated from the resting EEG (2 min with eyes open). In a next step, the center of gravity in the frequency range between 7 and 13 Hz was calculated for each electrode position. In determining the IAF, we aggregated the gravity frequencies over both resting conditions with eyes open and over all leads. Three different frequency windows with a bandwidth of 2 Hz each were defined: lower1 alpha band (L1 = [IAF—4 Hz] to [IAF—2 Hz]), lower2 alpha band (L2 = [IAF—2 Hz] to IAF) and upper alpha band (U = IAF to [IAF + 2 Hz]).

As depicted in Fig. 1, each EEG trial started with the presentation of a fixation cross for 3 s. After the 3 s, the test stimulus was presented and the participant had to respond as fast and accurately as possible to the stimulus by pressing either the YES-buttons or the NO-buttons, upon which the stimulus was deleted from the screen. Each response was followed by an inter-trial interval of 4 s.

The next step in calculating the ERD was to check all trials individually for artifacts (eye movements, blinks, muscle tension, etc.) by visual inspection. Trials containing artifacts were completely eliminated from the ERD analyses since artifacts are a major source of contamination of the EEG. Artifacts cause a change in the electrical activity over the scalp (e.g., eye-movements influence the electric fields of scalp areas adjacent to the eyes) and since the EEG records the electrical activity over the scalp, the EEG is often significantly distorted by such artifacts. The EEG recording is a combination of neural potentials and interfering potentials caused by artifacts and, thus, it is essential to adjust neural potentials in order to make inferences about event-related processes in the human brain (Croft & Barry, 2000).

The power of background activity in the individually defined alpha bands was computed for each entire trial. Afterwards, the band power (μV2) in the reference and activation intervals was averaged over all remaining trials, separately.

The percentage decrease (or increase) in alpha power from the reference interval to the activation intervals was defined as: %ERD = ([R − A]/R) × 100. Positive %ERD values indicate decreases in alpha power (cortical activation or desynchronization) and negative %ERD values indicate increases in alpha power (= event-related synchronization, ERS, usually interpreted as cortical deactivation).

Results

Psychometric data

Descriptive statistics of the visuo-spatial test scores as well as performance data (i.e., number of correct responses and reaction time) in the Shepard–Metzler tasks are given in Table 1. As expected, due to the process of pre-selection no significant sex difference emerged in the visuo-spatial scores (F1, 75 = 0.12, p > .05). In order to analyze sex differences and training effects in mental rotation (MR) performance separate analyses were conducted for the number of correctly answered Shepard–Metzler tasks as well as for mean reaction time (RT).

Table 1

Descriptive statistics (M, SD) of the visuo-spatial IQ and performance data (number of correct responses and RT) in the 2D and 3D MR task.

	Male		Female		Total
	M	SD	M	SD	M	SD
Visuo-spatial score	50.53	9.88	49.80	8.60	50.16	9.20
Pretest
Correct MR tasks 2D	38.58	4.29	35.13	5.59	36.83	5.25
Correct MR tasks 3D	38.79	4.69	37.05	3.88	37.91	4.36
RT 2D MR tasks	4.36	0.75	4.16	0.75	4.26	0.75
RT 3D MR tasks	4.07	0.76	4.17	0.88	4.12	0.82
Posttest
Correct MR tasks 2D	40.68	4.36	39.17	4.50	39.92	4.47
Correct MR tasks 3D	40.63	4.19	41.63	3.31	41.14	3.78
RT 2D MR tasks	3.69	0.69	3.85	0.66	3.77	0.67
RT 3D MR tasks	3.68	0.57	3.76	0.68	3.72	0.62

Note. Visuo-spatial scores are depicted as T-values.

Scores

A three-way repeated measures ANOVA with DIMENSIONALITY (2D versus 3D presentation mode) and TIME (pre- versus posttest) as within-subjects variables and SEX as between-subjects variable was performed for the number of correctly solved Shepard–Metzler tasks. It is important to note that the repeated measures ANOVA was only performed for the solution rate but not for residual gain scores.

As expected, the between-subjects variable SEX had a significant effect on MR performance (F1, 75 = 4.02, p < .05, partial η = .06) suggesting that male participants receive higher scores than female participants (M♀ = 38.24, SE♀ = 0.50 versus M♂ = 39.67, SE♂ = 0.51). Furthermore, the analysis revealed a significant effect for the within-subjects variable DIMENSIONALITY (F1, 75 = 6.26, p < .05, partial η = .08). Overall, higher scores have been achieved when the MR tasks were presented in a 3D mode (M2D = 38.39, SE2D = 0.47 versus M3D = 39.56, SE3D = 0.37). In addition, a significant and stable interaction between DIMENSIONALITY and SEX can be reported (F1, 75 = 5.42, p < .05, partial η = .07, statistical power 1–β = .98). Fig. 2 as well as post-hoc t-tests indicate that sex differences appeared only in the 2D presentation mode (t75 = 5.38, p < .01), but diminished in the 3D presentation mode (t75 = 0.90, p > .05). However, it is important to note that the interaction described in Fig. 2 does not refer to residual gain scores but indicates mental rotation performance (i.e., performance scores) only.

Fig. 2

MR Performance (score). Interaction between SEX and DIMENSIONALITY.

Moreover, the repeated measures ANOVA yielded a highly significant main effect TIME (F1, 75 = 68.02, p < .01, partial η = .48) depicting that MR performance increased from pre- to posttest across sex and dimensionality. Cohen's ds1 indicate that on average, participants improved their MR performance score by 0.52 standard deviations from pre- to posttest in the 2D presentation mode and by 0.53 standard deviations in the 3D presentation mode. The correlations between pre- and posttest scores were r = .37 (p < .05) for tasks in the 3D mode and r = .53 (p < .01) for tasks in the 2D mode. The less-than-perfect correlations suggested interindividual differences in learning progress. Finally, a significant interaction between SEX and TIME can be reported (F1, 75 = 9.40, p < .01, partial η = .11). The interaction indicated that girls could profit more from the training intervention. However, as described in the method section residual gain scores were derived to investigate participants' learning progress in more detail. Analyses of the residual gain scores led to the result that residual gains did not differ significantly between girls and boys (F1, 75 = 1.31, p > .05). Nevertheless, the statistical power 1 − β = .86 indicated a stable SEX × TIME interaction.

Reaction times

In accordance with the analysis described above a repeated measures ANOVA with DIMENSIONALITY (2D versus 3D presentation mode) and TIME (pre- and posttest) as within-subjects variables and SEX as between-subjects variable was computed to analyze sex differences in and training effects on RT. Again, the analyses yielded a significant effect of DIMENSIONALITY (F1, 75 = 6.47, p < .01, partial η = .08) suggesting that participants displayed lower RT on MR tasks in the 3D mode than in the 2D mode (M2D = 4.01, SE2D = 0.07 versus M3D = 3.92, SE3D = 0.07). Moreover, reaction time decreased significantly from pre- to posttest, irrespective of gender or dimensionality (TIME: F1, 75 = 67.31, p < .01, partial η = .47). There was, however, neither a significant effect for SEX nor any significant interaction. Correlation coefficients for RT in the pre- and posttest were comparable to those reported for MR performance scores and were r = .38 (p < .01) for the 3D presentation mode and r = .55 (p < .01) for the 2D presentation mode.

Physiological (ERD) data

In order to investigate sex differences as well as training-related changes at the neurophysiological level a five-way multiple measures ANOVA was computed for the %ERD in the upper alpha band. DIMENSIONALITY (2D versus 3D presentation mode), TIME (pre- and posttest), HEMISPHERE (left versus right) and AREA (AF, F, FC, CT, CP, PT and PO) were treated as within-subjects variables whereas SEX was handled as a between-subjects variable. A summary of all significant effects is presented in Table 2.

Table 2

Effects of the ANOVA for upper alpha %ERD.

Effect	df, dferror	F	η2
HEMISPHERE	1, 73	3.00†	.04
AREA	6, 68	47.98**	.81
AREA × HEMISPHERE	6, 68	4.85**	.30
AREA × TIME	6, 68	3.93**	.26
SEX × DIMENSIONALITY × TIME × HEMISPHERE	1, 73	8.56**	.11

Note. For the sake of clarity, only significant effects are presented.

* p < .05. ** p < .01. † p < .10.

The analysis revealed a main effect of HEMISPHERE suggesting that the right hemisphere was associated with less %ERD than the left one (Mright = − 6.61, SEright = 3.46 versus Mleft = − 3.36, SEleft = 3.43). As expected, the second main effect, AREA, indicated the highest amount of desynchronization over parietooccipital positions whereas the highest amount of synchronization could be found for frontal positions (see Fig. 3). The effect of AREA, moreover, interacted with HEMISPHERE as well as with TIME. Regarding the AREA × HEMISPHERE interaction (statistical power 1 − β = .98) less %ERD was found in the right hemisphere over the FC, CT, CP and PT positions. As shown in Fig. 3 the AREA × TIME interaction (statistical power 1 − β = .96) obviously emerged due to the larger pretest–posttest differences in the %ERD regarding the AF, F, FC and CT positions.

Fig. 3

%ERD. Interaction between AREA and TIME. Error bars indicate ± 1 SE of the mean.

The analysis furthermore yielded a SEX × DIMENSIONALITY × TIME × HEMISPHERE interaction. The statistical power 1 − β = .82 indicated a stable interaction (see Fig. 4) that can mainly be traced back to dimensionality specific sex differences in the right hemisphere at time 2 (posttest). As expected boys' cortical activation generally decreased from pre- to posttest. For girls such an expected decrease emerged only when they were presented with the 3D version of the MR task. If girls were presented with the 2D version of the MR task, there was no reduction in brain activation at all.

Fig. 4

%ERD. Interaction between DIMENSIONALITY × HEMISPHERE × TIME × SEX.

Discussion

As this was—to our knowledge—the first attempt to assess neurophysiological correlates of 2D vs. 3D presentations of objects that have to be mentally rotated (i.e., Shepard–Metzler figures) we had—on the basis of the neural efficiency hypothesis—been hoping to elucidate a source of behavioral findings reported already before, namely that (1) real / virtual reality presentations lead to smaller or even eliminated sex differences in mental rotation performance and (2) that practice reduces (or even eliminates) the sex difference in mental rotation.

Effects of 3D vs. 2D presentation

Basically, the first expectation was confirmed behaviorally, but, we could not find a directly corresponding neurophysiological effect. We observed a general facilitating effect of 3-dimensional presentation (compared to 2-dimensional), both with respect to scores as well as for reaction times. However, only for scores sex interacted as expected with the mode of presentation, which qualified the main effect in sex from a general male advantage to one that emerges only in the 2D version. The 3D presentation of the Shepard–Metzler figures seemed to ‘release’ the information processing load for all individuals, but more strongly for females. In studies reporting a similar disappearance of the sex difference when using either real objects (McWilliams et al., 1997) or employing virtual reality presentation (Larson et al., 1999; Parsons et al., 2004) several explanations have been proposed.

A first explanation could be the following: Keeping in mind that 2D tasks were clearly harder for girls and boys than 3D tasks and that boys outperformed girls in the 2D tasks of the pretest, it might be argued that boys have already had better rudimentary skills for solving 2D tasks before the training. Consequently, the training would foster different processes for boys and girls. Whereas the training could serve to automatize boys' skills for completing 2D tasks, the training for girls could only serve to develop rudimentary mental rotation skills which boys had already developed before the training without automatizing these skills. The advantage of 3D tasks may, therefore, lie in the fact that girls and boys do not differ with regard to their initial skills necessary for solving 3-dimensional MR tasks (as indicated by the diminishing sex differences in the 3D tasks).2

Another relevant explanation that has also been advanced by Voyer et al. (1995) in their meta-analysis on sex differences in spatial abilities essentially goes back to Horan and Rosser (1984). They hypothesized that some spatial tasks require transforming a spatial problem presented in two dimensions to a solution in three dimensions, a phenomenon they called ‘dimensionality crossing’, which the authors claimed to be responsible for sex differences favoring males (although Voyer et al., 1995 point out that in their meta-analysis other tasks requiring dimensionality crossing do not show significant sex differences). Nevertheless, McWilliams et al. (1997) argue that their usage of real 3-dimensional objects reduced task complexity for females because their problems with cross-dimensionality may be resolved. Parsons et al. (2004), when explaining the disappearance of a sex difference with virtual reality mental rotation tasks, argued in the same vein by stating that virtual reality objects do not require the creation of 3D cognitive representations from 2D drawings, which should be the process that inflates sex differences in paper–pencil tests of mental rotation. In addition they put forward a second hypothesis: As their subjects had to ‘superimpose’ the rotated stimulus on a target stimulus by rotating the stimulus manually by grasping and moving a sphere shaped ‘cyberprop’ this task involves motor aspects too and provides immediate feedback about ‘success’ of the rotation. According to the authors this liberates individuals from the need to double-check their answer, an inefficient strategy that according to Linn and Petersen (1985) should be a reason for the slower performance of women on mental rotation tasks.

Based on our findings we argue that the female disadvantage in solving mental rotation tasks might not lie in the process of mental rotation per se but in the derivation of a 3-dimensional representation from a 2-dimensional image. This argument is based on the following observations: First, there were neither significant sex differences nor any interactions with sex in the analysis of reaction times. Sex differences solely occurred for scores. This result is consistent with recent findings reported by Moé, Meneghetti, and Cadinu (2009) as well as Clements-Stephens, Rimrodt, and Cutting (2009). Second, we used only ‘augmented’ and not ‘virtual’ reality presentations here (i.e., without any motor component and involving no immediate feedback as in Parson et al.'s study) and therefore, the observed dimensionality main effect as well as its interaction with sex cannot be due to any motor and/or immediate feedback effects. The task as presented here would still involve the need of double-checking the answer, which is why we conclude that the reduced sex difference in the 3D presentation mode should rather be due to the reduced need of dimensionality-crossing, which should make the task less complex, especially for females. It is puzzling, however, that this reduced complexity is not accompanied by any physiological effect; at least based on the assumption that dimensionality-crossing would require working memory, it should have led to a stronger activation of the prefrontal cortex with 2D tasks, especially in women. A physiological effect has been found only in combination with the effects of training, i.e., when comparing the first vs. second mental rotation task presentation, which shall be dealt with in the following.

Effects of training

It should be noted that regarding the effect sizes the effects of 2D vs. 3D presentation mode were rather low, especially when we compared them to the quite large effect of the factor TIME, i.e., the pretest vs. posttest distinction. The two-week training led to a strongly increased performance in the Shepard–Metzler tasks, both for the analysis of scores and for reaction times. It should, however, be mentioned that the improvement cannot be attributed directly to the training; it could also be a simple repetition effect, as we had no control group. The reason for not including a control group is that with such laborious pretest–training–posttest studies in a neurophysiological laboratory the introduction of a control group would not have been easily justified as we were primarily interested in the interactions of sex with training and sex with dimensionality.

With respect to the sex by training interaction we could confirm the hypothesis that the sex difference diminishes after training, but this finding was obtained only for scores not for RTs and it could be argued that for this measure the interaction could also result from some kind of ceiling effect (with girls scoring lower in the pretest having more ‘space’ to improve). In fact, an additional analysis of residual gains that took the difference in starting values into account showed no significant sex difference; therefore the interaction should probably not be overestimated.

Regarding neurophysiological effects of training we observed a significant area by time interaction showing that a pre- to posttest decrease of activation (i.e., increase in neural efficiency) was observable only in frontal areas (including anteriofrontal and frontocentral areas). This finding is in nice correspondence to a recently published study testing adolescent girls (Haier, Karama, Leyba, & Jung, 2009) that found a decrease of the BOLD response in fMRI in frontal areas after three months of Tetris practice. In that study, structural MRI changes were assessed, too, but these occurred in other brain areas than those displaying functional (BOLD) changes with practice.

Joint effects of dimensionality and training

The only neurophysiological effect of the grouping variable SEX was found involving the repeated-measure variable TIME together with DIMENSIONALITY and HEMISPHERE. This four-way interaction (depicted in Fig. 4) showed that the expected general increase in neural efficiency from pre- to posttest holds generally for males whereas for females it can only be observed in the 3D version, but not in the 2D presentation of the MR task. As mentioned in the introduction, training should—in the cognitive domain—generally lead to an increase in neural efficiency (i.e., a decrease in cortical activation from pre- to posttest; Kelly & Garavan, 2005; cf. also Neubauer & Fink, 2009). Especially in view of the fact that the training performed by our participants was quite comprehensive and even ‘exhaustive’ (performed on 7 days within two weeks) it is nevertheless surprising that the females did not show any decrease in their brain activation from pretest to posttest. And maybe from this viewpoint the TIME × DIMENSIONALITY interaction for scores should not be underestimated even if it does not hold when analyzed with residual gain scores. The finding of female performance improvement from pre- to posttest only in the 3D condition accompanied by a neural efficiency increase for that same condition in our view showed that the mode of presentation of spatial tasks (mental rotation) plays an important role for females although the effect size was rather low for the dimensionality effect alone. Perhaps, only in combination with the training effect the dimensionality effect was strong enough to emerge behaviorally as well as physiologically.

Some important limitations of the study must also be mentioned: First, as compared to ‘full-blown’ virtual reality simulations the ‘augmented reality’ employed here only involves 3D vs. 2D presentation of stimuli (like in a 3D cinema), but no possibility to move around in space or to ‘grasp’ objects by means of some kind of device. In fact ‘real’ virtual reality simulations are extremely difficult to combine with neurophysiological measurements because these usually strongly restrict the possibility to move around, either completely (like in an fMRT scanner) or mainly (like with EEG, where movement artifacts pose a serious problem to any measurement). It could be argued that virtual reality or even ‘real world’ presentations of objects might reduce the sex difference more strongly. Insofar it is nevertheless surprising that we could find behavioral as well as physiological effects at all. But taking EEG measurements during real handling of three-dimensional objects might be an interesting endeavor for future studies.

Second, since we did not control for reasons given for prematurely ending training sessions, it might be argued that only highly motivated participants completed the rather intense and tedious mental rotation training. Thus, the strong training effect reported in the current study might be confounded with participants' motivation for training. Future research should control for (achievement) motivation when investigating training in the field of mental rotation.

Third, most studies showing sex differences in mental rotation tested adults. With the adolescents tested here, it might be argued that especially with respect to puberty and hormonal processes going on in this phase effects might be ‘blurred’ (Hampson & Kimura, 1992; Silverman & Phillips, 1993). However, the reason for testing in this age range was first that sex difference in mental rotation appear already under 13 years but become more prominent in 13- to 18-year-olds (Voyer et al., 1995). Secondly, as has been argued in the introduction adolescents have been selected here because of the considerable importance of spatial ability for educational outcomes especially with respect to the STEM domains (science, technology, engineering, maths; cf. Wai et al., 2009). The results presented here offer another argument that the female disadvantage in mental rotation is not ‘carved in stone’ but rather seems amenable to interventions.