Evidence for a unitary structure of spatial cognition beyond general intelligence

Performance in everyday spatial orientation tasks (e.g. map reading and navigation) has been considered functionally separate from performance on more abstract object-based spatial abilities (e.g. mental rotation and visualization). However, evidence remains scarce and unsystematic. With a novel gamified battery, we assessed six tests of spatial orientation in a virtual environment and examined their association with ten object-based spatial tests, as well as their links to general cognitive ability (g). We further estimated the role of genetic and environmental factors in underlying variation and covariation in these spatial tests. Participants (N = 2,660) were part of the Twins Early Development Study, aged 19 to 22. The 6 tests of spatial orientation clustered into a single ‘Navigation’ factor that was 64% heritable. Examining the structure of spatial ability across all 16 tests, three factors emerged: Navigation, Object Manipulation and Visualization. These, in turn, loaded strongly onto a general factor of Spatial Ability, which was highly heritable (84%). A large portion (45%) of this high heritability was independent of g. The results from this most comprehensive investigation of spatial abilities to date point towards the existence of a common genetic network that supports all spatial abilities.

1 Spatial skills are fundamental for everyday life as they make it possible for us to understand and operate on the 2 physical world around us. Studies in primates and other animals have highlighted the importance of spatial 3 ability for evolution and survival. Food-hoarding birds rely on spatial memory to retrieve their caches, which is 4 crucial to their subsistence, and climate harshness has been found to positively drive the evolution of spatial 5 memory skills in black-capped chickadee, another bird species. 1 Spatial skills are also important in modern 6 technologically-oriented societies 2 as individual differences in spatial skills are associated with positive 7 developmental, educational and life outcomes. Spatial ability reliably predicts scholastic and professional 8 success and career choices, particularly in Science, Technology, Engineering and Mathematics (STEM) and 9 related fields, even after controlling for general cognitive ability. [3][4][5] In spite of the increasingly fundamental role 10 that spatial ability has for individuals and contemporary societies, 6 numerous questions remain regarding the 11 nature of spatial ability as well as its origins and structure. 7 12 13 What constitutes good spatial skills? Since its earliest conceptualization, 8 spatial ability has been considered a 14 multifaceted construct comprising several related, yet separable, skills. 9 One of the most widely adopted 15 definitions of spatial ability describes it as the ability to generate, retain, retrieve and transform well-structured 16 visual images. 10 Contrary to this very broad characterization of spatial ability, however, extant research has 17 largely focused on measuring only specific aspects of object-based spatial ability. Among the most widely 18 studied spatial skills are individuals' abilities to mentally rotate shapes, 11 to visualize objects from different 19 perspectives, and to find figures embedded within other shapes. 12 A much smaller body of research has 20 considered larger-scale, practical everyday spatial orientation abilities, such as navigation, map reading and 21 way-finding. 22 23 Until recent years, studies of spatial orientation skills had been hindered by the difficulty in measuring 24 navigation and way-finding abilities in real-life settings utilizing rigorous approaches that are standardized 25 across participants. In addition, assessing navigation in the real environment can be highly costly and time 26 consuming and thus unlikely to be scalable to large samples nation-wide or world-wide. Technological 27 advances in the field of virtual reality (VR) provide a novel powerful tool to study individual differences in 28 spatial orientation skills in realistic settings that can be fully controlled and standardized across participants. 13,14 29 Studies assessing the validity of measuring navigation skills using VR have observed strong correlations (~.60) 30 with performance in real world navigation skills. 13,15 The reliability of assessing spatial abilities in VR is likely 31 to continue increasing as accelerating technological developments provide progressively immersive and realistic 32 tools. 33 34 scale (object-based spatial skills) and the ability to orient in large-scale environments (spatial orientation 48 ability). [19][20][21] This proposition is partly supported by psychological studies suggesting that the two abilities are 49 influenced by separate cognitive processes and brain structures. For example, in a study of the association 50 between performance in object-based psychometric spatial tests and large-scale spatial learning, partial support 51 was found for a differentiation between these skills. Individual differences in measures of spatial learning 52 (measuring skills such as placing landmarks on a map, intra-route distance estimates and route reversal) were 53 unrelated to variation in object-based spatial tests. However, the ability to learn maze and maze reversal, was 54 found to be related to both object-based tests and spatial learning. 22 Other studies in the field of cognitive 55 psychology have found evidence for a partial dissociation between object-based tests and large-scale spatial 56 orientation skills. 23-25 57 58 Neuroimaging studies have also provided preliminary converging evidence for the distinction between object-59 based abilities and spatial orientation skills, suggesting that the two are supported by separate brain networks. 60 Object-based spatial skills, and particularly mental rotation ability, were found to be primarily associated with 61 activation of the parietal lobes. 26 Conversely, variation in learning and remembering the layout of large-scale 62 spaces has been found to be related to processing in the hippocampus and the medial temporal lobes. 27 63

64
Other theoretical accounts and studies, however, have suggested that object-based and spatial orientation skills 65 might be closely related. For example, theories concerning the evolution of sex differences have argued that 66 individual variation in object-based spatial skills, such as mental rotation, are the product of different selection 67 pressures for large-scale spatial orientation abilities between males and females over evolutionary history, 28,29 68 5 therefore suggesting that the two largely reflect common skills. Empirical evidence also supports the idea of a 69 largely unitary set of abilities. A study of the association between object-based spatial abilities, measured with a 70 limited battery of three psychometric tests, and large-scale spatial orientation skills, measured both in realistic 71 settings and a virtual environment, found a substantial correlation between the two. 15 72

73
The proposition of a unitary set of cognitive processes underlying object-based and spatial orientation skills is 74 consistent with the idea that these are aspects of a more general set of cognitive abilities. It is plausible that at 75 the heart of individual differences in all spatial skills is general cognitive ability, or general intelligence (g). G is 76 a psychometric construct that emerged at the beginning of twentieth century from observations that almost all 77 cognitive tests correlate moderately and positively. 30 Individuals performing highly on one cognitive test are 78 also likely to show good performance on other tests of cognitive abilities, and g indexes this covariance 79 observed between cognitive measures. Therefore, g is thought to represent individual differences in the domain-80 general abilities to plan, learn, think abstractly, and solve problems that are necessary for successfully 81 completing cognitive tests. 31 82

83
In our previous work on the factor structure of object-based spatial tests, we have shown that individual 84 differences in spatial abilities cluster into a unitary factor, at both the observed and genetic levels, even after 85 accounting for g. 18 Along the same lines, another study found that the association between object-based and 86 spatial orientation abilities was largely independent of verbal ability. 15 These studies suggest that the coherence 87 of spatial abilities is not simply due to their being part of g, but rather inherent in the spatial domain itself. 88 However, neuropsychological evidence contradicts this view. Case studies of patients with neuropsychological 89 impairments suggest that damage to navigation-related structures in humans typically leads to broad memory 90 deficits that are not limited to the spatial domain. 10 91 92 Extant literature is therefore characterized by contrasting theories and evidence with respect to the factor 93 structure and associations between object-based spatial abilities, assessed mostly through psychometric tests, 94 and large-scale spatial orientation skills, assessed both in real settings and VR. The lack of a cohesive account is 95 likely due to a paucity of studies that have investigated the association between object-based and large-scale 96 spatial orientation skills with a sufficiently diverse battery of tests. In addition, to our knowledge, no study to 97 date has investigated their links within a genetically informative framework, testing the hypothesis that a 98 common genetic network, independent of g, supports performance in all spatial skills. 99

100
The current study addresses these limitations by investigating the structure of spatial ability using two 101 comprehensive online batteries of object-based and spatial orientation skills, administered to a large genetically-102 6 informative sample of twins aged 19 to 22. Importantly, we assessed spatial orientation abilities with an 103 innovative gamified battery of six tests measuring navigation, map reading, wayfinding and large-scale 104 scanning and perspective-taking skills set in a virtual environment. The current work has three main aims: First, 105 we examined, for the first time, the factor structure and origins of spatial orientation skills. Second, we 106 investigated the structure and genetic and environmental origins of spatial ability across sixteen tests of object-107 based and spatial orientation skills. Third, we explored the role that g has in unifying individual differences in 108 performance across tests of spatial abilities. 109 110 Addressing outstanding questions on the factor structure of spatial ability applying a genetically informative skewness and kurtosis (Figure 1; Supplementary Table S1). Therefore, our gamified battery was able to 128 discriminate and reliably capture variation in spatial orientation abilities.  Table S1. 136 137 138 We adopted the twin method (Methods) to calculate heritability estimates for the six measures of spatial 139 orientation; these are presented in presented in Figure 2. Heritability estimates, the extent to which variation in a 140 trait is accounted for by genetic differences, 32 ranged from modest to strong (14-57%). The remaining variance 141 in all tests was accounted for by non-shared environmental factors, environmental factors that do not contribute 142 to similarities between siblings, 32 with the only exception being the test of orientation ability using landmarks, 143 which showed a significant proportion of shared environmental variance (15%). These substantial non-shared 144 environmental estimates might in part reflect measurement error. Because sex differences are often found for spatial abilities (though not always in the same direction), 33,34 we 154 conducted univariate full sex-limitation model (Methods) to examine whether these estimates of heritability 155 differed between males and females. We found no evidence for qualitative genetic sex differences, meaning that 156 the same genetic and environmental factors seemed to influence individual differences in spatial orientation 157 abilities for males and females. No significant quantitative sex differences were found (Supplementary Table  158 S2), that is, differences in the magnitude of genetic and environmental influences. For example, for an overall 159 composite measure of navigation ability, heritability was 52% (95% CI: 0.31; 0.70) for males and 54% for 160 females (95% CI: 0.29; 0.62). These estimates have overlapping confidence intervals, indicating that they are 161 not statistically different from one another. Even with a sample of over 800 complete twin pairs who took part 162 in the spatial orientation battery, the sample size was not sufficient for the sex-limitation model to reliably 163 detect quantitative and qualitative sex differences, if they in fact exist. This is evident from the wide confidence 164 intervals around estimates when calculated for males and females separately. For these reasons, and to increase 165 9 power, the full sample was used in subsequent analyses, combining males and females, and same-and opposite-166 sex twin pairs. 167 168 A single 'navigation' factor captured the variance common across all tests of spatial orientation We applied factor analysis (Methods) to examine the covariance structure across the six tests in the spatial 171 orientation battery. The results showed that the six tests correlated substantially and clustered into one common 172 factor, which we named 'Navigation', as it indexed abilities that are generally described in the literature as 173 spatial navigation skills (see Supplementary Table S3- was shared across all tests; between 66% and 100% of the heritability of each test was captured by the common 183 factor of navigation. Consequently, test-specific genetic effects were found to account for between 0% and 34% 184 of the genetic variance in each test of spatial orientation (Supplementary Table S4). 185 186 Environmental factors were largely specific to each test, as indicated by the considerable size of the specific E 187 paths (bottom of Figure 3), between 64% and 90% of the nonshared environmental variance was found to be 188 specific to each test. The common navigation factor only captured between 10% and 36% of nonshared 189 environmental variance in each test of spatial orientation (Table S4). 190 10 191 Figure 3. Factor structure and genetic and environmental variance common across the six tests of spatial orientation. We applied the 192 common pathway model to parse the genetic (A), shared environmental (C) and nonshared environmental (E) variance that is shared 193 across all the tests (represented by the A, C and E paths leaving from the common Navigation factor) from the genetic and 194 environmental variance that is specific to each test (indexed by the individual A, C, and E latent factors leaving from each rectangle).

195
Each individual test loaded substantially onto a common factor, which we named Navigation factor (loadings ranging from λ = .54 for 196 scanning ability to λ = .75 for navigation based on landmarks). All A, C and E paths are standardized and squared.

198 Substantial associations between measures of spatial orientation and object-based spatial tests
We investigated the structure of spatial ability across a greater diversity of spatial tests. To this end, we 201 extended our analyses beyond the six tests of spatial orientation to incorporate 10 additional tests of object-202 based spatial skills 18 . This additional battery of spatial tasks included measures that very closely align with 203 traditional psychometric tests of spatial ability, including mental rotation, visualization, 2D and 3D drawing 204 ability, and mechanical reasoning. Figure 4 presents phenotypic correlations between the sixteen spatial tests 205 included in the two batteries (spatial orientation and object-based) and their correlations with g. 206 207 Correlations between spatial tests were positive and moderate (.17-.56), with stronger links observed between 208 certain tests within each battery. For example, the four tests assessing navigation and map reading skills in the 209 spatial orientation battery clustered more strongly together (r ranging from . 44   for measures of 2D and 3D drawing, pattern assembly, paper folding, and mental rotation in the object-based 211 battery (r ranging from .34 to .54). 212 213 214 215

223
We conducted a series of confirmatory factor analyses to formally evaluate the covariance structure between the 224 16 spatial tests. We tested different theoretical models about the structure of spatial skills, starting from the 225 simplest model and progressing to increasingly complex representations of the structure of spatial skills. The 226 first model we tested was a one-factor model (Figure 5a), positing that variation in spatial orientation and 227 object-based skills could be largely considered a unitary ability. Although all tests loaded substantially onto a 228 single factor (Figure 5a), model fit indices ( 2 = 692.730 (104), p < .001, CFI = 0.890, TLI = 0.873, RMSEA = 229 0.061, SMRS = 0.059) suggested that this structure did not provide a good fit for the data. 230 231 Secondly, we tested whether including two factors of spatial ability (one for each battery, Figure S1) would 232 provide a more accurate description of the structure of spatial skills. This model provided a good fit ( 2 = 233 316.000 (103), p < .001, CFI = 0.958, TLI = 0.951, RMSEA = 0.037, SMRS = 0.040). However, it also 234 presented one major limitation: due to the substantial difference in test administration and properties of the two 235 batteries, we could not exclude the possibility that the two separate factors emerging from this analysis were a 236 product of differences between the two batteries, rather than underlying a real set of separate, although 237 substantially correlated, abilities. In addition, the two batteries included some cases of parallel measures, so that 238 specific skills were tested in both batteries using different methods (e.g. scanning and perspective taking). 239 240 In order to overcome this limitation, we tested another two-factor model, but this time we constructed the two 241 factors based on theoretically-driven differences between the constructs. The first factor included all those tests 242 that are described in the literature as tapping spatial orientation abilities (navigation, way-finding and map 243 reading) available across the two batteries. This resulted in six tests loading onto a first factor of 'Spatial 244 Orientation': navigation according to directions, navigation according to landmarks, map reading, route 245 memory and two tests originally part of the object-based battery, Elithorne maze and mazes. The second factor 246 of 'Object Manipulation' included the eight remaining tests part of the of object-based battery along with the 247 scanning and perspective-taking measures included the spatial orientation battery ( Figure S2). However, this 248 model did not provide a good fit for the data (supplementary Table S5). 249 250 The last model we examined was based on the structure of the correlations observed between the 16 spatial tests .73 to .95). Based on these strong correlations, we re-specified the model as a hierarchically-structured model of 256 spatial skills: The 16 tests of spatial skills clustered onto three separate abilities (object manipulation, navigation 257 and visualization), which in turn loaded onto a common factor of Spatial Ability ( Figure 6). 258 259 This hierarchical characterization of spatial skills describes the complexity of the structure of individual 260 differences in spatial abilities, while highlighting the strong interconnection between all abilities at a higher 261 level of analysis. The higher order factor of spatial ability accounted for a large portion of individual differences 262 13 in the navigation (R 2 = .791), object manipulation (R 2 = .689) and visualization (R 2 = 1.00) factors. We adopted 263 this hierarchical characterization of individual differences in spatial skills in subsequent analyses.  b. a.
environmental variance, which encompasses those experiences that make children growing up in the same 281 family more similar to one another beyond their genetic similarity, played a meaningful role in accounting for 282 individual differences in spatial skills.

284
This hierarchical AE model ( Figure 6) showed that spatial skills clustered together largely due to shared genetic 285 variance. The common spatial ability factor was in fact highly heritable (84%) and subsumed 67% of the 286 genetic variance in object manipulation. This is calculated, based on path tracing, as the standardized squared 287 genetic variance in the general factor of spatial ability (.84) multiplied by twice the path estimate for object 288 manipulation (.81) divided by the total genetic variance (.84 .81^2+.27) , resulting in 289 (.84.81^2)/(.84.81^2+.27). The common factor of spatial ability accounted for 93% of the genetic variance in 290 the navigation factor and for the entirety of the genetic variance in the visualization factor (see Supplementary  291  Table S6 for the full model including 95% confidence intervals). Nonshared environmental variance accounted 292 for a much smaller proportion of individual differences in the common spatial ability factor (16%). 293 294 Figure 6. Genetic and environmental variance characterizing the hierarchical structure of spatial ability. Within each blue rectangle 295 are the ten tests that were included in the object-based spatial battery, while shaded in red are the six tests part of the spatial orientation  General cognitive ability (g) only partly accounts for the genetic clustering of spatial skills 302 303 It is well established that cognitive skills correlate with each other, and that a substantial portion of variation in 304 different abilities can be accounted for by a general factor of cognitive ability (g), both at the observed and 305 genetic level. 17,35,36 We applied a Cholesky decomposition (Method) to examine to what extent the genetic and 306 environmental variance in spatial ability could be captured by g. The Cholesky approach, similar to hierarchical 307 regression, parses the genetic and environmental variation in each trait into that which is accounted for by traits 308 that have previously been entered into the model and the variance which is unique to a newly entered trait. We 309 applied this method to examine the extent to which the clustering of spatial tests into a common factor of spatial 310 ability could be accounted for by the broader g factor. The results presented in Figure 7 (see Figure S3 for the 311 full model) showed that g accounted for 55% of the genetic variance in the second-order common spatial ability 312 factor. In other words, 45% of the genetic variance in spatial ability was independent of g. 313 314 When we accounted for g at different levels in the models (Supplementary Figures S4 to S9 The current study provides new knowledge on the structure and nature of spatial ability, which addresses three 328 outstanding issues in the field of spatial cognition. First, we examined the structure of spatial orientation 329 abilities, measured with a novel gamified battery set in a virtual environment that included a broad range of 330 measures tapping putatively different aspects of spatial orientation ability. Second, we explored the structure of 331 the associations between spatial orientation skills and object-based spatial tests, a topic that remains mostly 332 unexplored in the cognitive psychology literature and is characterized by strong, contrasting theoretical views. 333 7,15,22 Third, we investigated the extent to which an index of the developmentally stable component of g 334 accounted for the shared variance observed across spatial skills. Across these three broad aims, we leveraged 335 the genetically informative quality of our twin sample to address parallel questions related to the genetic and 336 environmental structure of spatial ability and of its association with g. At every level of analysis our results 337 highlighted communalities rather than differences across tests of spatial ability, largely supporting a unitary 338 structure of spatial cognition. 339 340 Support for the unitary structure of spatial cognition first emerged from phenotypic analyses of our battery of 341 spatial orientation tasks. This finding of a strong general component of variation was remarkable given the 342 breath of spatial orientation skills covered by our newly developed battery. In fact, the development of this 343 innovative, gamified, battery set in a virtual environment was guided by a careful process of literature review 344 aimed at covering all the main domains of spatial orientation described in the existing literature. This resulted in 345 six broad domains that ranged from navigation according to directions and large-scale perspective taking, 346 which, based on Newcombe and Shipley's (2015) taxonomy, could be categorized as extrinsic-dynamic spatial 347 abilities, to route memory and large-scale scanning, which, based on the same taxonomy, could be described as 348 extrinsic-static spatial abilities 7 . Although extrinsic-static and extrinsic-dynamic abilities have been proposed to 349 be separate skills, 7 and a meta-analysis of the effects of training spatial ability partly supported this distinction 350 for a few selected tests, 37 our results contradict this largely theoretical taxonomy. 351 352 We found support for a unitary structure of spatial orientation skills not only at an observed (phenotypic) level, 353 but also in terms of the genetic and environmental factors supporting spatial orientation skills. We found that a 354 common factor of 'navigation ability' that was 64% heritable and captured between 66% and 100% of the 355 heritability of the six individual tests of spatial orientation, and to a lesser extent their nonshared environmental 356 variance (between 10% and 36%). This suggests that, to the extent that measures of spatial orientation covary, 357 they do so largely due to their shared genetic variance. These results push our knowledge of the nature of spatial 358 orientation skills further, providing support for a unitary structure of spatial orientation skills at the genetic 359 level. 360 361 Further support for a unitary structure of spatial cognition emerged when we considered an even greater breadth 362 of spatial tests, including, in addition to our six measures of spatial orientation, ten psychometric tests of object-363 based spatial skills, administered in the same sample as part of another gamified spatial battery. These sixteen 364 tests of spatial skills were specifically selected to cover all the main areas of spatial cognition identified in 365 extant literature, making the current work, to our knowledge, the most comprehensive investigation of spatial 366 abilities to date. We approached the examination of the structure of associations between such a broad umbrella 367 of spatial measures by moving through increasing levels of complexity. 368 369 A simple unitary account of spatial ability, represented by a general factor common to all measures, was found 370 not to provide an accurate description of the foundations of spatial skills. At a first glance, the results could 371 have been interpreted as supporting the existence of three factors of spatial ability. These three factors described 372 individual differences in navigation, object-based abilities and visualization. Existing taxonomies of spatial 373 ability, 7 differentiate not only between static and dynamic spatial skills, but also between intrinsic and extrinsic 374 abilities. Consistent with this account we observed a partial differentiation between object-based spatial tests, 375 such as mental rotation, that are largely concerned with the intrinsic properties of objects, and visualization 376 tests, such as perspective taking and scanning, which are largely concerned with extrinsic relations among 377 objects. 7,38 However, the very strong correlations, from .73 to .95, observed between the object-based, 378 navigation and visualization factors contradicted this putative distinction, and opened the possibility that a 379 coherent, underlying set of abilities held these three factors together. 380 381 Factor analytic evidence supported this hierarchical account of spatial cognition: All sixteen tests were found to 382 load onto three factors (navigation ability, object-based ability and visualization ability), which in turn loaded 383 strongly onto a common factor of spatial ability. A hierarchical structure, which highlights both communalities 384 and differences between cognitive tests, has also been found to provide the most accurate characterization in 385 other domains of cognition, most notably executive functions. [39][40][41][42] Also consistent with what has been observed 386 for individual differences in executive functions, we found that genetic factors were largely shared across all 387 tests of spatial abilities. These results point to the existence of a common genetic network at the basis of 388 individual differences in spatial ability, therefore providing additional support for a unitary account of spatial 389 cognition. 390 391 18 A further line of evidence supporting the existence of a unitary account of spatial cognition was provided by our 392 analyses examining the role of g in the clustering of spatial ability at the genetic and environmental levels. We 393 found that individual differences in g correlated moderately with all individual tests of spatial skills and 394 substantially with the common spatial ability factor. However, nearly half of the substantial genetic variance in 395 spatial ability was found to be independent of the genetic variance in g, measured aggregating multiple 396 cognitive tests over development. Taken together, our results indicate that spatial skills cluster together 397 phenotypically and genetically beyond the simple fact that they are all tests reflecting a general, 398 developmentally stable, capacity for planning, thinking abstractly and solving problems, all skills that are 399 indexed by g. 35 It should be noted that, since the genetic and environmental components of cognitive abilities 400 have differential longitudinal stabilities, aggregating across waves might have resulted in 'cancelling out' 401 environmental variance that is specific to each developmental stage, in favour of aggregating stable genetic 402 variance in g over development. 36

404
In summary, our current work provides a threefold line of support for the unitary nature of spatial cognition, 405 partly independent of other measures of cognitive skills. Interestingly, this unitary account of abilities is at odds 406 with individuals' perceptions of their own ability and feelings towards spatial activities. In our previous work 407 examining the structure of spatial and mathematics anxiety, we found evidence for a separation between the 408 anxiety people feel towards spatial navigation and the anxiety towards object-based skills, such as completing 409 difficult jigsaw puzzles and building flat-pack furniture from instructions. 43 This observed difference in 410 perceptions and feelings towards different spatial activities might contribute to explaining why ideas, theories 411 and taxonomies of spatial cognition have mostly favoured a multifaceted account of spatial skills. 412 413 Although our study provides the most comprehensive investigation of the structure of spatial ability to date in a 414 large sample and addresses several outstanding research questions concerning spatial cognition, it was limited 415 by the technology available to us at the time. Although we developed a new gamified battery set in a virtual 416 environment to reliably examine individual differences in spatial orientation skills, it is possible that assessment 417 in a computer-simulated environment might not be able to capture individual differences in spatial orientation 418 and navigation as well as does assessment in real life settings. It has been proposed that spatial orientation in 419 computer-simulated environments might reflect an allocentric (object-to-object) approximation of the abilities 420 involved in egocentric (self-to-object) real-life spatial orientation. 44 However, studies that have examined the 421 reliability of measuring navigation skills in virtual reality, as compared to real life settings, have found good 422 concordance between the two. 13 While we leveraged the newest technological developments to create a realistic 423 virtual environment to host our gamified test, future studies might explore navigation in virtual reality applying 424 even more immersive tools such as, for example, head-mounted displays (e.g. oculus technology). 425 19 426 Our finding of a unitary structure of spatial cognition across sixteen diverse tests of spatial skills, is likely to 427 inform several disciplines beyond cognitive psychology. Investigations on the nature and structure of spatial 428 ability have concerned researchers in a wide range of scientific disciplines, from evolutionary biology, to 429 neuroscience, ecology and molecular genetics. Our evidence for a largely unitary phenotypic and genetic 430 network supporting individual differences in spatial cognition can serve as a basis for future research on the 431 nature of spatial ability across all these disciplines and will likely provide a shift in our consideration of the 432 architecture of human cognitive abilities. These findings are also likely to inform educational practices and 433 interventions, particularly the development of programs aimed at advancing STEM learning through training 434 spatial skills. 45  Putatively different facets of spatial orientation skills were assessed through a novel gamified battery called 456 'Spatial Spy'. Participants were invited to solve a mystery by collecting clues while orienting and navigating 457 around the streets of a virtual environment (Figure 8). The online battery was developed in Unity 458 (https://unity3d.com) by ETT Solutions. After a comprehensive literature review, we identified four core 459 20 aspects of spatial orientation and navigation skills: 1) navigating when reading a map; 2) navigating based on a 460 previously memorized map or route; 3) navigating following directions (e.g. cardinal points), and 4) navigating 461 using reference landmarks. In addition to these four abilities, the spatial orientation battery included two tests 462 based on paradigms that have been frequently used in the object manipulation spatial literature: perspective-463 taking and scanning. Two research aims motivated the decision to include these two tests in the battery. First, 464 we aimed to explore how perspective taking and scanning measured within a large-scale spatial framework (i.e. 465 within a more naturalistic context approximating a virtual environment) related to the same abilities assessed 466 within a smaller-scale, object manipulation framework (i.e. psychometric tests collected as part of another 467 online battery). Secondly, due to the innovative and experimental nature of the spatial orientation battery, we 468 included measures of scanning and perspective taking, for which we had corresponding data from more 469 traditional psychometric tests, in order to explore the external validity of assessing spatial skills within this new 470 virtual environment. The measures included in this spatial orientation battery are described in detail below. Map Reading (Figure 8a), assessed individual differences in the ability to efficiently read a map to travel from 488 one location to another. Once a map had appeared on the top-right corner of the screen, a flashing yellow dot on 489 the map indicated participants' starting location (A), while a red pointer designated the end-point location on 490 the map (B). Participants were instructed to get from A to B by finding the fastest route and notified that they 491 had 1 minute to complete their mission. If participants could not reach their destination within 60 seconds, they 492 were 'teleported' back to the initial location and allowed a second opportunity to complete the task. The ability 493 21 was assessed though five non-consecutive iterations of increasing difficulty. Each iteration was allocated a 494 score of 2 if participants had successfully travelled from A to B through the quickest (most direct) route, a score 495 of 1 if participants had successful completed the mission but had not selected the fastest route, and a score of 0 496 if participants had failed to complete the mission. This created a final maximum score of 10. The final score 497 was calculated by combining this accuracy score with participants' reaction time (time taken to successfully 498 complete the mission), equally weighted. The test showed good test-retest reliability (r = .69, p< .001) and 499 distribution (Figure 1). 500 501 Memorizing a Route (Figure 8b), assessed individual differences in the ability to travel from one location to 502 another by remembering the content of a map. As for the map reading condition, a map appeared on the top-503 right corner of the screen, with a flashing yellow dot indicating participant's starting location (A), and a red 504 pointer designating the end-point location (B). However, the route memorizing test asked participants to 505 memorise the content of the map before the map disappeared from the screen. Participants were given 20 506 seconds to memorize the map and plan the route before travelling from A to B and were allowed 120 seconds to 507 reach the target location. The number of increasingly difficult iterations, procedure and scoring were the same 508 as those for the previously described map-reading without memory task. Test-retest reliability was good (r = 509 .60, p< .001), and distribution ( Figure 1). 510 511 Navigation according to directions (Figure 8c) assessed participants' skills in navigating around a virtual 512 environment following instructions based on directions. At the start of the task, participants were 'teleported' to 513 one location of the virtual environment and given instructions to navigate around the virtual city in terms of 514 compass points (north, south, east and west). The test included 5 non-consecutive iterations of increasing 515 difficulty and each iteration comprised 4-6 tasks. Each task that was solved correctly was assigned a score of 1. 516 Participants were allowed a maximum of three attempts to respond correctly to each task and consequently 517 proceed to the next set of instructions. After three consecutive failed attempts, the iteration was discontinued 518 and the remaining tasks in that iteration (if any) were assigned a score of 0. Each iteration had a time limit of 519 180 seconds, if the time limit expired before participants had completed all the tasks, the remaining tasks for 520 that iteration were discontinued and assigned a score of 0. There was no progress bar or timer on screen to help 521 participants keep track of time; however, "hurry up" prompts appeared on screen as the time limit approached. 522 At the end of each iteration (either successfully completed or discontinued) participants were teleported to 523 another part of the virtual environment to complete the following iteration. For the first two iterations the image 524 of a compass providing cardinal directions was available on the top-left corner of the screen, but the compass 525 was not available for the last three iterations, making them more difficult to complete. Examples of instructions 526 were: 'Now turn east' and 'You are facing southwest. Go north and immediately turn west'. The final score was 527 22 calculated by combining the accuracy score with participants' reaction time (time taken to successfully 528 complete each iteration), equally weighted. The test showed excellent test-retest reliability (r = .89, p< .001) 529 and distribution of the scores (see Figure 1). 530 531 Navigating based on reference landmarks (Figure 8d) measured the ability to navigate following instructions 532 based on the descriptive features of the destination or other nearby landmarks. The test included 5 non-533 consecutive iterations each comprising 4 or 5 tasks. Each task lasted for a maximum of 60 seconds, so 534 participants had 60 seconds to reach a certain landmark within the virtual environment. If the time limit expired 535 before participants had reached the required landmark, they were discontinued, teleported to the landmark in 536 question, and were able to proceed to the next task. Each task solved correctly, meaning that participants were 537 able to reach the described landmark within the time limit, was assigned a score of 1, while for each trial when 538 participants were not able to reach the location in 60 seconds, they were assigned a score of 0. Neither a map 539 nor a compass was provided to help participants navigate around the environment. Examples of instructions are: 540 'Now reach the tall white pyramid skyscraper', and 'The message instructs you to go to the park near the old 541 clock tower'. The target landmark was visible at the start of the session, but it was not always in plain sight as 542 participants were navigating throughout the city to reach the target landmark. The final score was calculated by 543 combining this accuracy score with participants' reaction time (time taken to successfully complete each 544 iteration), equally weighted. The test showed excellent test-retest reliability (r = .80, p< .001) and distribution of 545 the scores (see Figure 3a). 546 547 Large-scale scanning ability (Figure 8e) measured participants' ability to quickly process visual information 548 and identify a target object, a black briefcase, located somewhere nearby within the virtual city. The target 549 object remained the same across the five non-consecutive iterations of increasing difficulty. When looking for 550 the target, participants' perspective could be rotated freely in any direction, but could not be moved vertically or 551 horizontally around the virtual environment. Participants could identify the target object by clicking on the 552 mouse or trackpad within 60-seconds. Within the time limit, participants were allowed four attempts to 553 correctly spot the target object and, as for all other tasks, they were encouraged to do it as quickly as possible. 554 Feedback was provided after each attempt, and as soon as participants had identified the target object correctly, 555 they were 'teleported' to the next task. It was not possible to pause half-way through the 60-second iteration. 556 The final score was calculated by combining this accuracy score with participants' reaction time (time taken to 557 successfully complete each iteration). The test showed excellent test-retest reliability (r = .80, p< .001) and wide 558 distribution of the scores (see Figure 1). 559 560 Large-scale perspective taking (Figure 8f) measured participants' ability to identify objects from a different 561 perspective in large-scale 'naturalistic' settings. The test comprised five iterations of increasing difficulty that 562 followed the same test rules. Each iteration started with a CCTV-like image showing an aerial shot of a location 563 within the virtual world, and within this location one target object was depicted flashing on screen for ten 564 seconds. During this initial stimulus presentation, participants could not move within the virtual environment, 565 so all participants were exposed to the same image of the flashing target object. After the ten seconds had 566 elapsed, the CCTV image disappeared and participants were teleported back to the target location within the 567 virtual environment, which shifted their perspective shifted back to ground level; they were then instructed to 568 identify the target object as quickly as possible. When looking for the target object (the one that was flashing 569 when presented from the CCTV perspective), participants' perspective could be freely rotated but could not be 570 moved vertically or horizontally around the virtual environment. Participants could identify the target object by 571 clicking on it with their mouse or trackpad within 60-seconds. Within the time limit, participants were allowed 572 four attempts to correctly spot the target object and they were encouraged to do it as quickly as possible. A 573 message would appear on the screen after each attempt (either 'Yes' or 'Try again') to provide participants with 574 feedback on their performance, and each iteration terminated either after a successful attempt, or after 575 participants had used up their four attempts, or if they timed out. A 'Hurry up' message was displayed on the 576 screen a few seconds before the time for each iteration elapsed. The test showed good distribution (Figure 1) 577 and test-retest reliability (r = .67, p< .001). Object manipulation was tested using an online, gamified, battery called 'The King's Challenge' 18 . This test 587 battery measures the major putative dimensions of spatial ability, and is comprised of 10 tests: 1) a mazes task 588 a.

25
(searching for a way through a 2D maze in a speeded task); 2) 2D drawing (sketching a 2D layout of a 3D 589 object from a specified viewpoint); 3) Elithorn mazes (joining together as many dots as possible from an array); 590 4) pattern assembly (visually combining pieces of objects together to make a whole); 5) mechanical reasoning 591 (multiple-choice naïve physics questions); 6) paper folding (visualizing where the holes are situated after a 592 piece of paper is folded and a hole is punched through it); 7) 3D drawing (sketching a 3D drawing from a 2D 593 diagram); 8) mental rotation (mentally rotating objects); 9) perspective-taking (visualizing objects from a 594 different perspective), and 10) cross-sections (visualizing cross-sections of objects). The development of the 595 battery is described in detail elsewhere. 18  CFA is, in most instances, theory-driven and allows for testing hypothesis on the associations between variables 619 and their underlying latent constructs. Alternative theoretical models were compared examining multiple model 620 fit indices. Model fit indices include: a) the Chi-square test, which indicates the correspondence between the 621 expected and the observed covariance matrices, a chi-square value close to zero indicates greater 622