Keeping It in Three Dimensions: Measuring the Development of Mental Rotation in Children with the Rotated Colour Cube Test (RCCT)

The Mental Rotation Test

In Shepard and Metzler’s (1971) pioneering work, participants were presented with pairs of perspective drawings of 3D cube aggregates and asked to identify whether the second image was either the same or a mirror image. They found a linear relationship between reaction time and the degree of rotation, similar to that found when physically rotating objects – a small rotation of an object takes less time than a large rotation. This suggested that mental image transformations correspond to transformations in the real world (Shepard & Cooper, 1982).

Vandenberg and Kuse (1978) used these black-and-white 3D cube aggregates to produce the Mental Rotation Test (MRT) with a response format that consisted of two targets and two distracters. Participants had to identify two structurally identical, but differently rotated cube aggregates from a multiple-choice selection of four cube aggregates. This test format has become widely applied in research with adults (Geiser, Lehmann, & Eid, 2008; Peters, 2005; Peters et al., 1995; Peters, Manning, & Reimers, 2007; Voyer & Saunders, 2004).

Differences in the characteristics of stimuli, such as using animate objects instead of cubes facilitated mental rotation in women and children (Alexander & Evardone, 2008; Neuburger, Jansen, Heil, & Quaiser-Pohl, 2011; Rosser, Ensing, & Mazzeo, 1985). This indicates that a key difficulty may be related to stimulus identification and encoding (Bialystok, 1989). Mental rotation of an object’s encoded image (Jolicoeur, 1988; Moreau, 2012) may be either matched with a more abstract, structural representation (Hyde, 1981), or directly compared with the nearest and most similar stored view (Hedges & Nowell, 1981). Hence, the initial two sections of our new test assessed perceptual matching of model and target as a baseline ability for mental rotation; only thereafter were the model and target cube differently rotated.

A further facilitating factor in mental rotation performance is colour information (Alington, Leaf, & Monaghan, 1992). Since colour is one of the fundamental properties of an object, it might be perceived pre-attentively like other features such as line orientation (Enns & Rensink, 1991; Treisman, 1986) and may therefore provide less able participants with an additional ‘processing channel’. Children are especially sensitive to colour signals in early stages of retinal perception, whereas size and orientation features are processed in later processing stages (Donnelly et al., 2007). Hence, as in the coloured version of the Raven’s Standard Progressive Matrices Test for children, we used coloured cube images. However, we gradually reduced the relevance of colour differences as a helpful cue in each section. The rationale behind this approach was similar to a visual search task where increased colour similarities resulted in a reduction of feature uniqueness between target and distracters (Gerhardstein & Rovee-Collier, 2002; Treisman, 1988; Treisman & Gelade, 1980).

Developmental Differences in Mental Rotation

Mental rotation is correlated with academic success in topics such as geometry, mathematics, and chemistry (Harris, Hirsh-Pasek, & Newcombe, 2013; Hyde & Linn, 1986). Piaget and Inhelder assumed that children in the preoperational stage until about age seven would only use static imagery. In particular, young children would not understand how changing an object’s direction also changes its features in a coordinated way (dynamic imagery) (Piaget & Inhelder, 1971, p. 120). However, subsequent research demonstrated that this widely accepted assumption was incorrect and underestimated young children’s ability to process rotated objects. The ability to mentally rotate develops in infancy (e.g. Moore & Johnson, 2008, 2011; Quinn & Liben, 2008) and during early childhood, and continues to do so into adolescence (e.g. Kail, 1986; Kosslyn et al., 1990).

Marmor (1975) showed that 5-year-old children were able to rotate the original cube aggregates (Shepard & Metzler, 1971) and found a similar linear increase in reaction time related to angular disparity. Likewise, in a forced choice paradigm study of rotated 2D object shapes, 4- and 5-year-old’s accuracy decreased with the angle of the rotation, but 3-year-old’s accuracy did not (Frick, Hansen, et al., 2013). However, in a different study, 3-year-olds demonstrated the ability to rotate a 2D object although at very slow speeds of 2,500 ms, increasing up to 3,000 ms at larger angles (Krüger et al., 2013; see also Kosslyn et al. 1990). In a Tetris game with dynamic 2D rotated shapes, error rates of 4- and 5-year-olds did not suggest mental rotation ability, but 5-year-olds’ response times increased with greater rotation from 2,200 ms to 3,200 ms (Frick, Ferrara, & Newcombe, 2013). Interestingly, these reaction time studies demonstrated a transition from static to dynamic imagery in very young children already in terms of speed and not just accuracy.

Moreover, recently Schwarzer et al. (2012) used the original Shepard and Metzler cube aggregates in a dynamic video film with 9-months-old infants. Importantly for the assumption that mental rotation mirrors motoric real-life object rotation, the results suggested an active motor component insofar as the more mobile crawlers looked longer at the mirror image of cube aggregates in a habituation task than static infants who could only sit. When using simpler letter stimuli, the motor component was found to be important in a linear fashion in 8- to 10-month-olds, with walkers being more likely to distinguish impossible letter rotation outcomes compared to crawlers, belly crawlers, and sitting infants, respectively (Frick & Möhring, 2013). If manual exploration was permitted, even 6-month-olds showed increased looking time for impossible rotations in a habituation experiment (Möhring & Frick, 2013). This motoric component was also found in adults who were low performers in the Vandenberg and Kuse (1978) test (redrawn version of Peters et al., 1995) as they would gesture more in their explanations regarding differences in the structure of a wooden 3D model of Shepard and Metzler’s cube aggregates (Göksun, Goldin-Meadow, Newcombe, & Shipley, 2013).

Individual Differences in Mental Rotation

Gender differences in mental rotation are widely reported (Linn & Petersen, 1985). In a meta-analysis by Voyer, Voyer, and Bryden (1995), men outperformed women on the MRT by nearly one standard deviation. Other research suggests that gender differences on the MRT tended to disappear with practice in computer games (Haier et al., 1992; Okagaki & Frensch, 1994), through sports activities (Blüchel, Lehmann, Kellner, & Jansen, 2012; Quaiser-Pohl & Lehmann, 2002), by lifting time constraints (Goldstein, Haldane, & Mitchell, 1990; Peters, 2005; Voyer, 2011) and with extensive item-specific practice (Kail, 1986; Kass, Ahlers, & Dugger, 1998), as well as 2D-3D dimensional transformation training (Moreau, 2012; Tzuriel & Egozi, 2007, 2010). Hence many factors influence gender differences in mental rotation performance. We expected that in the current study, gender differences would not show as a main effect because the many trials in the test would give girls sufficient practice with the rotated items and reaction times were not measured.

Socio-economic status (SES) is another variable that influences cognitive development. Children with low SES can fall behind on very early measures of cognitive development such as the Bayley Infant Behaviour Scales (Farah, 2010), school readiness tests (Brooks-Gunn & Duncan, 1997) and executive attention (Mezzacappa, 2004). SES, especially in boys, has been identified as an important factor in influencing spatial cognition and the development of visuo-spatial memory (Levine, Vasilyeva, Lourenco, Newcombe, & Huttenlocher, 2005). Whereas girls and boys with a low SES showed no difference in performance on measures of aerial maps and mental rotation ability, boys from families with middle- and high-SES performed better (Levine et al., 2005). We controlled SES in the current sample of children by considering whether the London council, or their parents, paid for their school meals and included this factor in the statistical analyses. We expected that especially boys from lower SES backgrounds would not perform as well as boys from families that were able to pay for their children’s school meals.

Training Studies and Dimensionality in Mental Rotation

Kail (Kail, 1986; Kail & Park, 1990) indicated that with extensive practice children could reach adult levels of performance on mental rotation tasks, but the training effect was limited to item-specific features of just one object, with no transfer of the practice effect to other objects. This suggested that children stored unique view-specific images of an object without developing an abstract ability to rotate. It is likely that the mental rotation task can be solved with the storage of visual snapshots, similar to visual priming in children (Lange-Küttner, 2010b; Enns & Cameron, 1987). Recent studies on practice and training of mental rotation have focused on dimensionality of the object, in particular on the 2D versus 3D task difficulty (Tzuriel & Egozi, 2007, 2010). The visual information processing system finds 2D stimuli easier to process than 3D (Jansen, Schmelter, Quaiser-Pohl, Neuburger, & Heil, 2013) and this difference appeared to increase with age (Hoyek et al., 2012). This may be the case because degree of rotation is less influential in two dimensions (Bauer & Jolicoeur, 1996; Jolicoeur, Regehr, Smith, & Smith, 1985). Hoyek et al. (2012) found no correlation between 2D stimuli and 3D cubes in 7- to 8-year-olds, which supports the notion of dimension-specific processing.

Two-dimensional stimuli are not really realistic, but flat. However, children favour 3D pictures, become progressively more interested in depth depiction and develop their ability to represent three dimensions in their own graphic constructions (Kosslyn, Heldmeyer, & Locklear, 1980; Lange-Küttner, 1994, 2004, 2009). Girls preferred to unfold cube faces and drew large amounts of surface detail that could distort the overall view of the cube, whereas boys appeared to favour keeping a cube’s visual appearance intact (Lange-Küttner & Ebersbach, 2012). Children of kindergarten age were already able to estimate the volume of 3D cubes (Ebersbach, 2009). Hence, it is both appropriate and beneficial to measure young children’s mental rotation ability in a test with three-dimensional cube images.

What is the particular difficulty when processing three-dimensional stimuli? 2D perception requires the processing of stimuli only within a single plane based on straightforward similarity judgements, whereas three-dimensional perception requires more complex spatial inferences about visually incomplete, hidden-from-view information, where object features have to be interpolated. Superior 2D-3D “dimensionality crossing” (spatial transformations) was identified in males who outperformed women on most mental rotation tasks (Voyer et al., 1995); especially the occluded parts of the cube aggregates were more difficult to process for women (Voyer & Hou, 2006). However, training in 2D-3D spatial transformations successfully improved girls’ performance (Tzuriel & Egozi, 2007, 2010). In adults, 2D training led only to improvements in 2D tasks, whereas 3D training led to improvements in both 2D and 3D tasks (Moreau, 2012). This clearly demonstrates the importance and specificity of dimensionality in the mental rotation task.

The Current Study

A widely used approach for studying children’s mental rotation ability is the reduction of cognitive load through the simplification of stimulus complexity. For instance, outline drawings of human figures provided children with an apparently more simple and socially more suitable test items compared to the classic complex geometric cube aggregates (Estes, 1998). However, when mental rotation performance was measured in terms of increases in reaction time along with angular discrepancy, only 6-year-olds performed similar to adults, while 4-year-olds did not. Moreover, when a hand was used as a mental rotation stimulus this produced an increase in reaction times from about 3,100 ms to nearly 5,000 ms in 5- to 6-year-old children who did not yet attend school, and from about 2,000 ms to nearly 3,500 ms in 7-year-old first graders (Krüger & Krist, 2009).

Nevertheless, bodies instead of cube aggregates support women’s mental rotation ability (Alexander & Evardone, 2008). Interestingly, a hybrid between cube aggregates and human heads, hands and feet also lowered the cognitive load for adults in comparison to the classic cube aggregates, but only when the body parts were orderly attached and not when they were randomly fixed onto the ends of the aggregates (Krüger, Amorim, & Ebersbach, 2014). Quaiser-Pohl, Neuburger, Heil, Jansen, and Schmelter (2014) found that measuring mental rotation ability with cube aggregates and time limits was too difficult for second graders (6- to 9-year-olds) but not for fourth graders (8- to 12-year-olds).

Our aim was to keep the 3D cubes similar to the original stimuli of Shepard and Metzler (1971), but to test children in the multiple choice test format used by Vandenberg and Kuse (1978). We designed the Rotated Colour Cube Test (RCCT) with one single multi-coloured three-dimensional cube and thus simplified the complexity of the geometric cube aggregates (Vandenberg & Kuse, 1978), but not their three-dimensional volume. Similar facilitations were effective in the Three-Mountains-Task that measures the ability to form spatial perspectives when three overlapping mountains (Piaget & Inhelder, 1956) were reduced to a single clearly visible mountain (Liben & Belknap, 1981).

First, we gradually increased task difficulty by varying the rotation of the model cube from an upright canonical position to balancing it on one of its corners without rotating the target and distracters. Thereafter, differences in rotation and colour between the model, target and distracters were gradually introduced. The initial two test Sections (A & B) investigated whether young children could identify test cube items from distracters by simple perceptual matching. Only thereafter were they asked to process rotated cubes in Sections (C &D) (see Fig. 1). We expected children to perform better on simple perceptual matching tasks than in the identification of rotated cubes, and both age and gender differences to appear with increasing task difficulty.

Second, we tested 7- to 10-year-old children as research suggests that differences in mental rotation emerge in this age range (Geiser, Lehmann, Corth, & Eid, 2008; Geiser et al., 2008; Johnson & Meade, 1987; Titze, Jansen, & Heil, 2010; Vederhus & Krekling, 1996). As both time limits (Glück & Fabrizii, 2010; Voyer et al., 1995) and response format (Glück & Fabrizii, 2010) influence mental rotation performance, we removed time restrictions and used the same response configuration as in the Raven Coloured Progressive Matrices Test (RCPM), which requires participants to identify one target from six to eight distracters.

Third, we also tested children with the RCPM as it is a standardized test used to measure non-verbal reasoning. For children, the RCPM is one of the purest measures of fluid intelligence. The RCPM first appeared in 1947 as a specialized form of the Raven’s Standard Progressive Matrices Test, specifically created for testing 5- to 10-year old children. In a meta-analysis by Vijver (1997) of cross-cultural intelligence test scores, the RCPM was the second most used test after the Wechsler Intelligence Scales for children. The aim was to control children’s fluid intelligence, but also to compare the two non-verbal tests with each other for shared variance.

The general assumption is that there are no gender differences in the RCPM scores. This inference was first made by Raven (1939, p. 30) who noted that in the standardization sample, there were no gender difference between boys and girls up to the age of 14 years, both in the mean and the variance of scores. Eysenck and Kamin (1981, p. 41) also noted equal scores between the two genders for children and adults. Jensen (1998, p. 541) concluded that there was no consistent discrepancy between female and male scores in the Raven’s Standard or Coloured Progressive Matrices Tests. Court’s (1983) literature review which summarized 118 studies confirmed that no gender differences in performance were found, and this was confirmed for children by Lynn and Irwing (2004). Hence, in the current study, the RCPM provided an important objective measurement instrument.

Method

Results

Accurate performance was computed in per cent for each section of the Rotated Colour Cube Test (RCCT) and the Raven Coloured Progressive Matrices Test (RCPM). In cases where the Mauchley’s test of Sphericity was violated, we adjusted the degrees of freedom with Greenhouse-Geisser. Statistically significant effects were followed up with post-hoc tests. In the first part of the Results section, we report individual and age differences in the RCCT, in the second part we report individual and age differences on the RCPM, and in the third part we report the comparison of the RCCT and the RCPM overall scores.

Rotated Colour Cube Test (RCCT). A 4 (Sections) by 3 (Age) by 2 (Gender) by 2 (School meal type, FSM) MANOVA was carried out, with repeated measurement for the RCCT Sections. Group means are listed in Table 3 and the statistical effects in Table 4. There was a significant main effect for the RCCT Sections, F(2.23, 195.96) = 330.56, p <  0.001, η² = 0.79, with a very large effect size. Pairwise post-hoc comparisons (p_s <  0.001, two-tailed) confirmed that simple perceptual matching in RCCT A (M = 93.6%) and RCCT B (M = 91.9%) differed significantly from identification of the more difficult rotated targets in RCCT C (M = 83.5%) and RCCT D (M = 48.7%), but not from each other. This seemed to show that the step from a canonical orientation in RCCT A to a more unusual position, balanced on one corner in RCCT B did not increase task difficulty for children. In contrast, the introduction of individually rotated model and distracter cubes in the test panel in Section C led to a decrease in performance of about 10% , and the removal of distinctive and unique cube colours in Section D led to an even more pronounced drop from about 80% to 40% (see Fig. 3).

Furthermore, there were two significant main effects for FSM, F(1, 88) = 8.01, p <  0.01, η ²= 0.08, and age groups, F(2, 88) = 4.73, p <  0.05, η ²= 0.10. Children with free school meals (M = 76.9%) performed overall significantly worse than children with parent financed school meals (M = 81.9%). Multiple comparisons of age differences showed that 7- to 8-year-old children (M = 75.5%), differed from 10-year-olds (M = 81.7%), but no other comparison was significant.

There was also a significant two-way interaction of gender and FSM, F(1,88) = 5.58, p <  0.05, η ²= 0.06, and these factors interacted significantly in a three-way interaction with the RCCT Sections, F(2.23, 195.96) = 6.47, p <  0.01, η² = 0.07. To investigate the three-way interaction between RCCT Sections, gender and FSM, the MANOVA was re-run with a split sample analysis for SES in order to consider how boys and girls from different SES differed in their performance on individual RCCT Sections.

In the state-funded free school meal group, there were neither gender differences, p _s  >  0.61, nor age differences, p_s  >  0.34 in performance. But when parents were able to pay for school meals, boys (M = 85.2%) outperformed girls (M = 78.9 %) and this difference was highly significant, F(1, 47) = 14.27, p <  0.001, η ²= 0.23. Furthermore, only in this high SES group the age difference was significant, F(2, 47) = 6.07, p <  0.01, η ²= 0.21.

Post-hoc t-tests for independent samples were run for boys and girls, per RCCT Section. In each of the two halves of the RCCT, always in the second, more difficult Section (RCCT B and D), a significant gender difference was found: In RCCT A, the mean performance of boys and girls did not differ significantly (boys M = 96.6%; girls M = 95.4%), p >  0.51, but in RCCT B where the cubes had a non-canonical orientation, boys performed significantly better (M = 96.6%) than girls (M = 91.2%), t (51) = – 2.76, p <  0.01. Likewise, in the RCCT C, boys (M = 87.98) and girls (M = 83.80) did not differ significantly, p >  0.16, but in RCCT D where all rotated cubes had the same colour, boys (M = 63.0%) performed significantly better than girls (M = 44.9%), t (51) = – 3.69, p <  0.001 (see Fig. 4, right). Thus, boys from a relatively higher socio-economic background excelled in the more challenging perceptual matching and mental rotation tasks. This confirmed that these two tasks measure related abilities.

Raven Coloured Progressive Matrices Test (RCPM). Accuracy was computed in per cent correct for each type of RCPM Section A, AB and B (see Table 5). As with the RCCT, we conducted a 3 (Section) by 3 (Age) by 2 (Gender) by 2 (School meal type) MANOVA with repeated measurement for each Section.

There was a significant main effect for the factor RCPM Section, F(1.86, 163.93) = 43.11, p <  0.001, η ²= 0.33, with a large effect size (see Table 6). Pairwise post-hoc comparisons (p _s  <  0.001) confirmed significant differences between RCPM A (M = 76.44%), RCPM AB (M = 72.30%), and RCPM B (M = 61%). Sections became increasingly more difficult as the RCPM progressed.

Furthermore, there were two significant between-subjects effects. The effect for FSM, F(1, 88) = 16.51, p <  0.001, η² = 0.16, showed that children with free school meals (M = 63.85%) performed overall significantly worse than children with parent financed school meals (M = 75.76%). The post-hoc tests of the age effect, F(2, 88) = 4.50, p <  0.05, η² = 0.09, showed performance of both the 7-8 and 9-year-old children was worse than in 10-year-olds.

There was a significant two-way interaction of RCPM Sections and FSM, F(1.86, 163.93) = 3.807, p <  0.05, η² = 0.04, that interacted neither with age, p >  0.74, nor with gender, p >  0.23. T-tests for independent samples compared children with free school meals and parent financed school meals per RCPM Section. This revealed a significant difference in each Section (see Fig. 5), which increased the further children progressed in the Raven test, Section A: t(98) = – 2.5, p <  0.05, Section AB: t(98) = – 3.83, p <  0.001, Section B: t(98) = – 4.04, p <  0.001.

Comparison of the RCCT and the RCPM

Because the RCCT was a new test development, we used the established RCPM for cross-validation. The RCCT and RCPM scores were significantly correlated in 7- to 8-year-old children r(28) = 0.52, p = 0.004, in 9-year-old children r(35) = 0.60, p <  0.001, and in 10-year-old children r(32) = 0.72, p <  0.001, and increased with age.

We conducted a more comprehensive MANOVA on the two overall scores of the RCCT and the RCPM, respectively, that allowed for a direct comparison. A 2 (RCCT vs. RCPM) by 3 (Age) by 2 (Gender) by 2 (School meal type) MANOVA revealed no significant gender differences, all p _s >  0.16 (see Table 7). This demonstrated that gender differences were limited to the more difficult RCCT Sections and did not appear when overall scores were used. Furthermore, a significant within-subject main effect, F(1, 88) = 60.71, p <  0.001, indicated that on average the RCPM (M = 69.9%) was more difficult than the RCCT (M = 79.5%). This difference showed a comparably large effect size of η² = 0.41, while all other significant effects sizes were smaller, η²<0.16.

A significant main effect for age groups interacted in a two-way interaction between the two tests, F(2, 88) = 4.13, p <  0.05, η² = 0.86 (see Fig. 6), with an even larger effect size. RCCT scores increased with age by 7.3% (age 7-8 M = 75.2%; age 9 M = 80.2%; age 10 M = 82.5%). Also RCPM scores increased with age, by 11.8% (age 7-8 M = 65.3%; age 9 M = 67.3%; age 10 M = 77.1%). Post-hoc t-tests (paired samples) revealed a significant difference in the two test scores in all three age groups, with the RCCT scores always significantly higher than the RCPM scores, p _s <  0.01. However, Fig. 6 shows that the difference between the test performance reduced with increasing age, and vice versa, the correlations between the two tests increased with age (see the first paragraph of this part of the report).

The significant between-subjects effect for FSM, F(1, 88) = 17,15 p <  0.001, showed that children receiving free school meals (M = 70.4%) scored lower than children with parent financed school meals (M = 78.6%), but importantly, a significant two-way interaction revealed that this varied depending on the test, F(1, 88) = 7.88, p <  0.05.

To test where the difference was located, the MANOVA was run again with a split sample for FSM. It showed that children with free school meals performed much better on the RCCT (M = 76.9%) than on the RCPM (M = 63.9%). However, a two-way interaction showed that this difference became smaller with age, F(1, 47) = 3.56, p <  0.05, 7- to 8-year-olds (RCCT: M = 73.7%; RCPM: M = 61.4%), to 9-year-olds (RCCT: M = 78.6%; RCPM: M = 59.0%), and 10-year-olds (RCCT: M = 78.4%; RCPM: M = 71.1%), as Raven scores were relatively improved in the older children on free school meals. No other statistical effects were significant, p _s >  0.21. Thus the RCCT was especially fair to younger low SES children.

In contrast, the sample of children with parent financed school meals showed the same disparity between tests, RCCT: M = 81.9%; RCPM: M = 75.8% , F(1, 53) = 13.53, p <  0.001, but no reduction of this difference with age, F(1, 53) = 0.78, ns. Instead, there was a main effect of age, F(2, 53) = 3.66, p <  0.05, which interacted with gender, F(2, 53) = 3.93, p <  0.05. Girls’ test scores were similar at 7- to 8-years (M = 77.3%); 9-years (M = 74.8%); and 10-years (M = 77.3%) and showed no improvement, whereas boys’ performance increased with age 7- to 8-years (M = 71.9%); 9-years (M = 83.4%); 10-years (M = 88.5%).

Discussion

This study introduces a new test, the Rotated Colour Cube Test (RCCT) as a measure of perceptual matching and mental rotation in children, using illustrations of a single multi-coloured three-dimensional (3D) cube as a model. We tested 7- to 10-year-old children. The new test is not a mental rotation experiment where angularity of the rotated stimulus and the corresponding reaction times are measured (Shepard & Metzler, 1971). Instead, it is a Mental Rotation test similar to those used with adults by Vandenberg and Kuse (1978) and Peters et al. (1995) that presented rotated models, a rotated target as well as several rotated cubes as distracters. However, while a single cube is a useful reduction of complexity for children in comparison to the cube aggregates of Shephard and Metzler (1971), it does not lend itself to spatial rotation in the same way as a cube aggregate because a single cube lacks a clear one-dimensional extension that points into space, comparable to a vector. Cube aggregates have extensions which have been compared to pictures of gymnasts with outstretched limbs (Alexander & Everdone, 2008). We are currently revising the current version of the test in which we will gradually introduce cube aggregates.

The first part of the test (Sections A, B) measured children’s perceptual matching abilities, and the second part of the test (Sections C, D) more complex perceptual matching and mental rotation abilities. Children performed best in identifying a target cube amongst non-rotated distracters that was identical in colour and orientation to the model cube. Identifying a rotated target cube amongst rotated, similarly coloured distracters was the most difficult task. Additionally, we found gender differences in interaction with children’s low vs. high SES background. Mental rotation performance significantly correlated with children’s performance on the Raven’s Coloured Progressive Matrices Test (RCPM).

We constructed the test format similarly to the RCPM, with a response panel of a target and several distracters, and by increasing task difficulty in each consecutive test section. We assessed a baseline for item identification of a model object in a canonical view (Section A) and in a rotated position (Section B) to make sure that children had an intact object concept. An object concept is not self-understanding because in cube drawings, children often draw just one side, or if they draw more than one side, these multiple cube faces are not integrated (e.g. Lange-Küttner & Ebersbach, 2012). We used colour to distinguish between distracters as it is such an important feature for children that for instance the Raven Test for children only exists in colour, whereas the version for adults is in black-and-white. In our test, children could identify the target by finding the correct spatial configuration of the coloured cube faces. Moreover, colour similarity of the cube distracters was increased during trials in each of the first three sections.

As expected, similar to the RCPM, the new RCCT became more difficult over the four test sections. There was no significant performance decrease in perceptual matching between the first two Sections A and B where the only difference was model cube orientation, except in the sub-sample of high SES boys who performed better in identifying a cube in a non-canonical position standing on a corner. The main difference in test performance arose between simple perceptual identification in RCCT Sections A and B, and the more challenging target cube identification amongst individually rotated distracter cubes in Sections C and D. In Section C, colour similarity was increased between targets and distracters, whereas in Section D colour was completely removed as a distinctive feature. Thus, performance in Sections C and D decreased due to an increase in the number of rotations of the distracters as well as due to a reduction of colour saliency. Indeed, the study showed that it is not just orientation of ONE object as such that is difficult for children, but differences in orientation and similarity between targets and distracters.

The increase in colour similarity between target and distracter cubes to a level where all colours were the same and cubes varied only in orientation occurred gradually over sections C to D (see Appendix, Column ‘Colour Similarity’). But performance deteriorated particularly in section D where no unique object colour were available to distinguish between the distracter cubes. This is especially noteworthy because in both Section C and D, the model and the target could vary (see Appendix, Column ‘Rotation’, Levels III and IV). Hence, distinct unique colours between distracters were particularly helpful as a visual cue in narrowing down attention towards the rotated target. In conclusion, while different orientations of distracters made the RCCT more difficult, different colours of distracters had the opposite effect and made the RCCT easier because colour facilitated clearer discrimination between target and distracters.

Solving a three-dimensional mental rotation task involves the ability to maintain representations of relevant object attributes and their interrelation, while at the same time rotating mental images (Kaufman, 2007). Similar to adults, object colour can be more salient and important than object location (Hyun & Luck, 2007). Integration of the cube faces was easier when objects were different: Differences in the target cube’s orientation and even individually rotated distracters were not especially difficult as long as the distracters’ distinctive object colours were available as a cue. This may be somewhat counterintuitive as mental rotation is a spatial ability. However, colour of cube faces is a feature that defines the cube’s internal structure, but it is not a spatial cue about the location of the cube. Because the cube had only changed orientation and not position, the object-place binding (Lange-Küttner, 2008, 2013) remains intact in mental rotation. In short, rotated objects stay in place. This in turn suggests that feature integration plays an important role in mental rotation.

As in previous studies on visual memory and spatial ability (Lange-Küttner, 2010a; Levine et al., 2005), a pronounced impact of gender and SES was already present in school children: Boys from a higher social-economic background performed better than girls from the same relatively advantaged background in Sections B (non-canonical cube orientation) and D (lack of unique distracter colour), while there was no difference between boys and girls from a low social-economic background. The ‘gearing up’ of the more privileged boys indicated that they were more likely to rise to a challenge (Lange-Küttner, 2012; Lange-Küttner & Green, 2007). This gender by task difficulty effect was particularly apparent when colour cues were not available in Section D, as only the upper middle class boys had a success rate of above 60% while everybody else was below 48% . It could well be that these boys developed more responsiveness and attention towards the less obvious cube features such as contour and line orientation (Enns & Rensink, 1991; Hystegge, Heim, Zettelmeyer, & Lange-Küttner, 2012; Lange-Küttner, Kerzmann, & Heckhausen, 2002; Treisman, 1986).

A limitation of the current test version is that it was given in booklet and not in a computerized form, so precise time measurement was not possible. Lack of a time constraint may have been the reason for the absence of a main effect of gender in mental rotation. Lange-Küttner and Ebersbach (2012) found that boys were comparably more efficient in mental rotation decision making, as they came to more correct conclusions within a given time. The efficient boys with shorter MRT reaction times were also more likely to draw two occluded cubes, whereas this was not the case for girls. Girls worked at their own steady pace independently of the task at hand, but in 6- to 9-year-old boys, mental rotation reaction times were already task-specific. Although we did not measure reaction times in the RCCT, we are currently developing a computerized version that will allow precise time measurements along with the introduction of simple cube aggregates.

Mental Rotation and Fluid Intelligence

Our results showed that in general, the Rotated Colour Cube Test (RCCT) was easier that the Raven’s Coloured Progressive Matrices Test (RCPM), but this difference was more pronounced in 7- to 8-year-olds than in the 10- to 11-year-old children, especially when from families with a lower SES.

We suggest that the RCCT was easier than the Raven test since the spatial reasoning component was less complex. While single three-dimensional cubes were used in the RCCT, in the RCPM both continuous and discrete patterns had to be completed. That is, even if multiple coloured cube faces had to be perceptually integrated and distinguished against competing distracters, the RCCT Sections did not require the formation of a logical sequence of visual pattern fragments which may require executive attention (Jones, Rothbart, & Posner, 2003), or operational intelligence according to Piaget (Inhelder & Piaget, 1958). This might explain why younger children performed comparatively well in the RCCT, except in the last section where colour salience was removed. Therefore, this new test would lend itself to measuring object processing in even younger age groups.

However, besides these differences in the two tests, the significant correlations also suggested strong similarities. The Raven’s Matrices test was identified as measuring both fluid intelligence and spatial ability (Guttman, 1974). Similarly, it was proposed that spatial ability tests load considerably on g (Ullstadius, Carlstedt, & Gustafsson, 2004). We suggest that the similar response format with a perceptual discrimination between target and multiple distracters in both tests may account for the significant correlations in each age group.

In agreement with previous research (Eysenck & Kamin, 1981; Raven, 1939) we found no gender differences on the RCPM, whereas gender differences in the RCCT only emerged in the more difficult Sections B (perceptual matching) and D (mental rotation). This may be due to the difference in dimensionality between the two tests. The RCPM only includes 2D items, whereas the RCCT only consists of 3D items. As the visual information system is sensitive to dimensionality and finds processing of three-dimensional information more difficult (Jansen et al., 2013), this may account for finding gender differences only in the three-dimensional RCCT and not in the RCPM. It is indeed important to have a distinct 3D mental rotation test designed specifically for children, for instance, in order to assess abilities in Math, Science and the Arts from an early age.

The unexpected result of a drop in the levels of performance for children from a low social-economic background on the RCPM supports findings of a study with children from a Pakistani background (Aziz & Farooqi, 1991). Raven’s Progressive Matrices Tests is seen as a culture-fair test of intellectual functioning (Anderson, Kern, & Cook, 1968; Jensen, 1974; Kaplan & Saccuzzo, 1996; Valencia, 1984). However, in the current study this was not the case, which contradicts expectations (American Educational Research Association, American Psychological Association, & National Council on Measurement in Education, 1999, Standard 7.1). The disadvantageous influence of low socio-economic status on both the RCCT and RCPM highlights the need for developing further tests that may help to identify children who can benefit from early intervention strategies (Noble et al., 2015).

While our research sample was relatively large with N = 100, cells were unequal to some degree, though not dramatically so (Howell, 1992, section 13.9, pp. 409). There were 51 boys and 49 girls thus the sample was nearly perfectly gender-balanced. Likewise, for SES there were 47 children on free school meals and 53 children whose parents could afford to pay for school meals. But the smaller cells of the interactions (see Table 1) were unequal, with subsamples between n_s from 5 to 13. So there could have been an element of chance in the obtained significances of some interactions. However, we believe that because the interactions concerned mostly the more difficult sections B and D, that the interaction of task difficulty with individual differences was systematic and genuine. In fact, there was a political initiative insofar as free school meals for all were introduced in London’s entry classes to primary schools (Burns, 2014) at the time when this manuscript was submitted for publication.

The newly developed RCCT distinguishes itself from other available tests for children by preserving and simplifying Shepard and Metzler’s (1971) three dimensional geometric properties and by providing a 3D test specifically adapted for young children. Our research has shown that young children can solve tasks with three dimensional cube illustrations, but increasingly struggle when supportive colour information is reduced. The current study demonstrated that a reduction in task complexity without resorting to 2D images, and varying distracter similarity and colour salience were effective means of adjusting task difficulty. As the MRT is widely used with adults, it was important to create a simplified test for younger children that preserves three-dimensions as well as the use of geometric stimuli, with the aim to bridge the gap between the classic complex three-dimensional cube aggregates used for adults (Shepard & Metzler, 1971) and simplified two-dimensional versions for children.

Bio Sketches

Nikolay Lütke, Bachelor of Science, doctoral student at the London Metropolitan University. Test development of 3D perception and mental rotation in children.

Christiane (Chris) Lange-Küttner, Dr. phil., Associate Professor of Psychology (University of Nicosia), Privatdozentin (Venia Legendi in Psychology, Universität Bremen) and Senior Lecturer (London Metropolitan University). Experimental Psychology and Cognitive Development. Research on factors such as figural versus vector concepts, what-and-where system, spatial fields, memory, speed, repetition, distraction, salience, binding, priming, deterministic and stochastic feedback, boundaries, meaning and rules.

Appendix Table A1