Evidence for a unitary structure of spatial cognition beyond general intelligence

The study investigates whether spatial orientation abilities (e.g., navigation, map reading, wayfinding) and object-based spatial skills (e.g., mental rotation, visualization) reflect distinct systems or a common underlying structure, and how these relate to general cognitive ability (g). Spatial skills are critical for everyday functioning and STEM success, yet prior research often measured only object-based abilities and rarely included comprehensive spatial orientation measures. Technological advances enable scalable assessment of navigation in virtual reality, which correlates well with real-world performance. The authors aim to: (1) determine the factor structure and genetic/environmental origins of spatial orientation abilities using a novel VR-based gamified battery; (2) examine the broader structure and genetic/environmental origins of spatial ability across 16 spatial tests; and (3) evaluate the extent to which g accounts for the covariance among spatial skills.

Competing theories differentiate small-scale object manipulation from large-scale spatial orientation, supported by distinct cognitive processes and neural substrates (parietal lobes for mental rotation; hippocampal/medial temporal structures for large-scale navigation). Some psychological studies report partial dissociations between object-based tests and spatial learning, while others find substantial correlations between the two domains. Evolutionary accounts posit shared selection pressures linking object-based and large-scale orientation abilities. A robust general cognitive factor (g) underlies covariance among cognitive tests and may contribute to spatial test correlations. Prior work by the authors showed a unitary factor for object-based spatial abilities, largely due to shared genetic variance, but lacked spatial orientation measures. Validation studies show VR navigation correlates strongly (~0.60) with real-world navigation, supporting VR as a proxy for large-scale orientation assessment. Overall, the literature presents mixed evidence regarding unity vs separability of spatial skills and the role of g, motivating a comprehensive, genetically informative evaluation across diverse spatial measures.

Design and sample: Participants were 2,660 twins (868 complete pairs) aged 19–22 (M=21.23) from the UK Twins Early Development Study (TEDS), representative of the British population. Exclusions included major medical, genetic, or neurodevelopmental disorders. Zygosity groups included MZ and DZ same- and opposite-sex pairs. Ethics approval and informed consent were obtained. Measures: Two online gamified batteries assessed spatial abilities with at least five months between administrations (median 265 days). Approximately 74.3% completed both batteries.

1. Spatial orientation battery (Spatial Spy; Unity-based VR-like virtual city): Six tests with training and practice; administered via web browser; completion time ~35–60 min. Tests included:

• Map Reading: navigate from A to B using an on-screen map; five iterations; accuracy (0–2 per iteration) combined with reaction time; test–retest r=0.69.
• Route Memory: memorize a map for 20 s then navigate without the map; five iterations; accuracy+RT; r=0.60.
• Navigation Directions: follow compass-based instructions (N/S/E/W); five iterations; accuracy+RT; r=0.89.
• Navigation Landmarks: navigate to described landmarks without map/compass; five iterations; accuracy+RT; r=0.80.
• Large-scale Scanning: locate a target object (briefcase) within 60 s; five iterations; accuracy+RT; r=0.80.
• Large-scale Perspective-taking: identify a target from an aerial CCTV preview, then from ground-level perspective; five iterations; accuracy+RT; r=0.67. Across the six tests, distributions were approximately normal and average test–retest reliability was r=0.74 (range 0.60–0.89).

2. Object-based spatial battery (The King’s Challenge; 10 tests): Mazes; 2D drawing; Elithorn Mazes; Pattern Assembly; Mechanical Reasoning; Paper Folding; 3D drawing; Mental Rotation; Perspective-taking; Cross-sections.
3. General cognitive ability (g): Longitudinal composite from ages 7, 9, 10, 12, 14, 16 combining multiple verbal and nonverbal tests (e.g., WISC subtests, Raven’s Progressive Matrices, Mill Hill Vocabulary). Spatial tests were not included in g. Analytic strategy: Phenotypic analyses used one randomly selected twin per pair; replication conducted in the cotwins. Descriptive statistics and correlations (R psych); visualization (ggplot2). Structural equation modeling used Mplus v8 and OpenMx v2 with FIML for missing data. Confirmatory factor analysis (CFA) compared one-, two-, three-factor, bifactor, and hierarchical models using χ², CFI, TLI, RMSEA, SRMR, AIC, and Akaike weights. Twin modeling decomposed variance into A (additive genetic), C (shared environment), and E (nonshared environment). Univariate ACE models estimated heritabilities; sex-limitation models tested quantitative/qualitative sex differences. Multivariate genetic analyses used Common Pathway models for the six navigation tests and for the 16-test hierarchical structure. A Cholesky decomposition examined the extent to which g accounts for genetic/environmental variance in the common spatial ability factor.

• Spatial orientation battery validity and reliability: Six VR-based orientation tests showed good distributions and reliability (average test–retest r=0.74). Males outperformed females across all six tests with small-to-moderate effect sizes (largest for map reading, R²=0.17; smallest for scanning, R²=0.03).
• Univariate twin results for orientation tests: Heritabilities ranged from 14% to 57%. Nonshared environmental factors accounted for the remaining variance; a shared environmental component (15%) was observed only for navigation by landmarks. Quantitative sex differences were detected but were small-to-moderate in effect; combined-sex analyses were used subsequently.
• Navigation factor within orientation tests: All six orientation tests loaded on a single Navigation factor explaining 32–57% of variance per test; model fit: χ²(148)=269.94, CFI=0.968, TLI=0.971, RMSEA=0.030, SRMR=0.049. The Navigation factor was 64% heritable (95% CI 0.41–0.91); C=8% (0.00–0.43); E=28% (0.21–0.36). Between 66% and 100% of each test’s heritability was captured by the common Navigation factor; nonshared environmental variance was largely test-specific (64–90%).
• Correlations across 16 spatial tests: Positive phenotypic correlations ranged from r=0.17 to r=0.56; stronger clustering within navigation/map-reading tests (r≈0.44–0.56) and within object-based drawing/assembly/rotation/folding (r≈0.34–0.54).
• Factor structure across 16 tests: A one-factor model fit poorly (χ²(104)=692.73, CFI=0.890, RMSEA=0.061). A two-factor battery-based model fit well but risked method artifacts. A theory-driven two-factor model (Spatial Orientation vs Object Manipulation) did not fit well. A three-factor model (Navigation, Object Manipulation, Visualization) fit well (χ²(101)=351.87, CFI=0.953, RMSEA=0.041), but factors were strongly correlated (r=0.73–0.95). A hierarchical model with these three first-order factors loading on a higher-order Spatial Ability factor fit best by AIC/Akaike weights. The higher-order factor accounted for substantial variance in first-order factors: Navigation R²≈0.79, Object Manipulation R²≈0.69, Visualization R²≈1.00.
• Multivariate twin modeling (hierarchical AE): The higher-order Spatial Ability factor was highly heritable (A=84%, E=16%). The common factor accounted for 93% of the genetic variance in Navigation and 100% in Visualization, and 67% in Object Manipulation. Nonshared environmental influences were primarily test-specific.
• Role of general intelligence (g): Cholesky modeling showed g accounted for 55% of the genetic variance in the common Spatial Ability factor, leaving 45% independent of g. Results were consistent when g was modeled at different levels and when an alternative g measure at age 16 was used.

Findings address whether spatial orientation and object-based spatial skills share a common structure and genetic basis. Despite theoretical claims of separable systems, the six diverse orientation tasks cohered into a single Navigation factor that was substantially heritable, with shared genetics spanning tasks. Expanding to 16 tests revealed three first-order factors—Navigation, Object Manipulation, Visualization—that were strongly intercorrelated and unified by a highly heritable higher-order Spatial Ability factor. Genetic analyses indicated a common genetic network underpinning all spatial skills, with minimal shared environmental influence and largely test-specific nonshared environmental variance. Importantly, nearly half of the genetic variance in Spatial Ability was independent of g, indicating that spatial cognition coherence is not simply a byproduct of general intelligence. This hierarchical, largely unitary structure parallels evidence from executive functions and suggests that visualization abilities bridge across batteries and formats. The results refine theories of spatial cognition by integrating both domain-specific differentiation at a first-order level and substantial higher-order commonality, highlighting a coherent spatial domain with strong genetic foundations beyond g.

The study provides comprehensive phenotypic and genetic evidence for a largely unitary architecture of spatial cognition. Spatial orientation abilities measured in a VR-based environment form a single Navigation factor with substantial heritability. Across 16 diverse spatial tests, a hierarchical model best characterizes spatial ability: three correlated first-order factors (Navigation, Object Manipulation, Visualization) load strongly on a highly heritable general Spatial Ability factor. Approximately 45% of the genetic variance in Spatial Ability is independent of general intelligence, indicating a distinct, cohesive spatial domain. These insights have implications for theories of spatial cognition, neurocognitive models, and educational practice (e.g., spatial training to support STEM). Future research should further validate VR-based assessments against real-world navigation using more immersive technologies (e.g., head-mounted displays), probe neural correlates of the common spatial factor, and investigate molecular genetic underpinnings and developmental dynamics of spatial ability.

Assessment of spatial orientation was conducted in a computer-simulated virtual environment, which may not fully capture real-life navigation (potential allocentric vs egocentric differences). Although prior work shows good VR–real-world concordance, more immersive technologies could improve ecological validity. Differences in administration between the two batteries raise the possibility of method variance, though models accounting for such effects still supported a hierarchical common structure. Nonshared environmental estimates include measurement error. Sex differences were present but small-to-moderate and not the primary focus.