Spatial vs. Abstract Reasoning: One Factor or Two?

1,679 words · 7 min read · 3 references cited

Spatial reasoning and abstract reasoning are routinely treated as two distinct cognitive abilities in lay accounts and in some applied testing contexts. The first is associated with rotating mental images, navigating environments, and visualising three-dimensional structure; the second with detecting patterns, reasoning about analogies, and manipulating symbols disconnected from physical referents. The intuition that these are different mental skills is strong. The empirical question is whether the data agree.

What the two constructs are supposed to measure

Spatial reasoning, in the psychometric tradition, refers to the ability to mentally represent and transform objects and their geometric relations — rotating a shape in the mind, predicting what a folded figure will look like unfolded, identifying the same object viewed from different angles. Spatial-visualisation tests like the Vandenberg-Kuse Mental Rotations Test, the Paper Folding Test, and the Spatial Construction subtest of standard batteries all draw on this capacity.

Abstract reasoning typically denotes pattern detection and rule-induction in symbolic content that does not require imagining physical space. Tasks include matrix-completion tests like Raven’s Progressive Matrices, analogy problems (“hand is to glove as foot is to ___”), and sequence-extrapolation problems (“2, 4, 8, 16, ?”). The content varies; the demand is to extract a generative rule from the structure presented.

The standard psychometric question is whether these tasks index two separable latent abilities or the same underlying capacity expressed through different surface features. Answers depend on the breadth of the test battery, the size of the sample, and the criterion against which abilities are measured.

The factor-analytic case for one factor

The clearest small-sample test of the question in a single battery comes from an analysis of the Jouve-Cerebrals Test of Induction (JCTI) alongside four General Ability Measure for Adults (GAMA) subtests — Matching, Analogies, Sequences, and Construction (Jouve, 2018). The analysis pooled 118 adult test-takers and asked whether their performance on the five tasks was best described by one factor or two.

General reasoningfactorConstruction(spatial)JCTI (matrixinduction)Analogies & SequencesMatching
Figure 1. In a five-test reasoning battery a single general factor (roughly 42% of variance) loaded all tasks, absorbing the spatial-versus-abstract split (Jouve, 2018).

The result was unambiguous: principal-axis factoring with parallel analysis retained a single factor that explained roughly 42% of observed variance. Loadings were substantial across tasks, with the largest on Construction (which has heavy spatial-visualisation demands), then JCTI (matrix-style induction), then Analogies and Sequences (more abstract content), and Matching last. Forcing a two-factor solution produced near-zero correlation between the putative factors, prominent cross-loadings (Analogies loading on both factors in particular), and worse fit indices than the one-factor solution. Within this five-test battery, the apparent spatial-versus-abstract distinction was not stable enough to justify treating them as separable constructs.

The general pattern is consistent with what factor analyses of nonverbal reasoning batteries typically find: a strong general reasoning factor dominates, and putative subdimensions account for relatively little additional variance once g is extracted. The post’s finding is one specific instance of a broader literature on general intelligence saturation in nonverbal reasoning tasks.

The case for treating spatial as distinct

The single-factor finding from a five-test battery is not the whole story. Wai, Lubinski, and Benbow (2009) drew on more than 50 years of cumulative data from the Study of Mathematically Precocious Youth (SMPY) and other longitudinal samples to argue that spatial ability has incremental predictive validity for STEM educational and occupational outcomes above and beyond general reasoning and verbal ability. Across multiple cohorts and follow-up windows extending into mid-career, individuals high on spatial ability were over-represented in engineering, the physical sciences, and technology fields — even after controlling for general intelligence and verbal ability.

General reasoning (g)Spatial-specificabilitySTEM outcomes(engineering, sciences)incremental
Figure 2. In richer batteries and against STEM criteria, spatial ability adds predictive validity above general reasoning and verbal ability (Wai, Lubinski & Benbow, 2009).

This is consistent with a layered picture rather than a contradiction of the single-factor finding. In a small battery with only one or two indicators per facet, the residual variance attributable to a “spatial-specific” factor is modest and noisy — a single-factor solution wins on parsimony and fit. In larger and richer batteries with multiple indicators per cognitive facet, and against differentiated criterion outcomes (engineering performance, technical drawing, mechanical reasoning), spatial ability emerges as a reliable specific factor that adds predictive value beyond general g.

Both can be true: the variance is dominated by g in any battery, and there is enough specific spatial variance to matter for the small set of outcomes where spatial demand is heavy. The right inference depends on what is being predicted and how many indicators are available.

The trainability of spatial skills

One area where the spatial-distinct angle has practical bite is training. Uttal, Meadow, Tipton, Hand, Alden, Warren, and Newcombe (2013) conducted a meta-analysis of 217 spatial-training studies and reported a robust positive average training effect, with Hedges’ g ≈ 0.47 across studies. Critically, the gains transferred to non-trained spatial tasks (rather than just improving performance on the trained task), persisted in delayed post-tests, and applied across age groups, training types, and gender. The training effect was largest for novices and for paper-and-pencil spatial-skill courses, but the broader pattern was consistent.

This cuts against any strong “spatial ability is fixed and reflects only g” reading. The variance underlying spatial-task performance contains a substantial trainable component that does not look like generic reasoning ability — people get better at spatial tasks specifically when given practice on spatial tasks specifically, and the gains do not collapse onto general intelligence improvements. From a training-design perspective, treating spatial as a target-able skill is empirically defensible regardless of what the latent-variable structure looks like in any given factor analysis.

Resolution: depends on the question being asked

The right framing is not whether spatial and abstract reasoning are “really one thing or two” in the abstract, but what the answer is being used for. For a brief screening battery aimed at producing a single index of nonverbal reasoning, the factor-analytic evidence supports a total score interpretation: most of the systematic variance lives in a general factor, and parsing performance into spatial and abstract subscales overstates what the battery can measure. For a STEM-aptitude assessment or a domain-specific training programme, treating spatial as a focus area is empirically warranted: it has incremental validity for relevant outcomes and it is trainable.

The same logic applies more broadly to the literature on fluid versus crystallized intelligence and on cognitive abilities in general. Latent-variable structures are tools for organising data; the right level of resolution depends on what kind of decision the data are supporting. A two-factor split that does not survive in a five-test battery may be perfectly justified in a fifteen-test battery; a single-factor solution that wins on parsimony at one sample size may lose at a larger one.

Practical implications

Three implications follow. First, in clinical and educational screening with brief reasoning batteries, the safer interpretation of subtest variation is task-specific noise rather than meaningfully separable abilities — a total score is more defensible than a “spatial-vs-abstract profile” carved from few indicators. Second, in domains where spatial demand is heavy (engineering, surgery, architecture, navigation), assessing spatial ability specifically is empirically supported and predictive over and above general reasoning. Third, spatial skills are trainable with measurable transfer to non-trained tasks; if a training context calls for spatial competence, focused practice produces real gains.

Frequently asked questions

Are spatial reasoning and abstract reasoning the same thing?

Mostly. Within a small reasoning battery, performance on spatial-visualisation tasks and abstract-pattern tasks loads heavily on a single general-reasoning factor (Jouve, 2018), and the apparent spatial-versus-abstract distinction does not stably split into two factors. In larger batteries with multiple indicators per facet, spatial ability emerges as a specific factor with incremental validity (Wai, Lubinski, & Benbow, 2009) but the bulk of shared variance still sits in g.

Can spatial ability be trained?

Yes. Uttal et al.’s (2013) meta-analysis of 217 training studies reported an average training effect of approximately 0.47 standard deviations, with transfer to non-trained spatial tasks and persistence in delayed post-tests. Effects were broadly consistent across training types, age groups, and sex.

Why do some IQ tests separate spatial and abstract subscales if they’re correlated?

Because the goal varies. For brief screening, a single nonverbal-reasoning total score has the most defensible psychometric backing. For domain-specific assessment (STEM aptitude, occupational selection), separating spatial gives clinicians and educators a more interpretable profile, and the evidence for spatial-specific predictive validity supports doing so. Both interpretations are valid at their own resolution.

Does spatial ability predict success in STEM fields?

Yes, with incremental validity above general intelligence and verbal ability. The Wai-Lubinski-Benbow (2009) synthesis of more than 50 years of longitudinal data showed that high-spatial individuals are substantially over-represented in engineering, the physical sciences, and technology, and that this pattern persists into mid-career outcomes after controlling for other cognitive factors.

Is the JCTI a spatial test or an abstract-reasoning test?

It is both, in the sense that any non-trivial reasoning task draws on overlapping cognitive demands. The JCTI is a matrix-style inductive-reasoning test in the Raven’s family; its content is largely abstract-pattern and spatial-arrangement combined. Empirically, JCTI performance loads strongly on the same general reasoning factor that loads spatial-visualisation tasks like Construction (Jouve, 2018), reinforcing the broader point that the spatial/abstract distinction is more about surface content than separable latent ability.

References

  • Jouve, X. (2018). How many dimensions underlie the JCTI and the GAMA subtests? Evidence for a dominant general factor. Cogn-IQ Research Papers. https://pubscience.org/ps-1mFWV-3f180b-jGlP
  • Uttal, D. H., Meadow, N. G., Tipton, E., Hand, L. L., Alden, A. R., Warren, C., & Newcombe, N. S. (2013). The malleability of spatial skills: A meta-analysis of training studies. Psychological Bulletin, 139(2), 352–402. https://doi.org/10.1037/a0028446
  • Wai, J., Lubinski, D., & Benbow, C. P. (2009). Spatial ability for STEM domains: Aligning over 50 years of cumulative psychological knowledge solidifies its importance. Journal of Educational Psychology, 101(4), 817–835. https://doi.org/10.1037/a0016127

Related questions

What is the g factor?

The g factor — Charles Spearman's name for the common variance that runs through all cognitive tests — is the most replicated and the most contested construct in the science of human intelligence. Whenever a sufficiently varied battery of mental tests is administered to a sufficiently varied sample of people, the same statistical regularity emerges: scores on every test correlate positively with scores on every other test, and a single general factor explains a substantial share of the differences between people. g has survived 120 years of methodological scrutiny because the pattern it describes is genuinely there in the data. What it is at the level of brains and minds, and what it does and does not justify in policy, education, and selection, is a separate set of questions that the data do not settle on their own. Read more →

IQ vs. EQ: which matters more?

Few claims in pop psychology are as widely repeated as "EQ matters more than IQ." It originated with Daniel Goleman's 1995 trade book, was adopted enthusiastically by corporate training programs, and now anchors a global industry of leadership coaching. But the underlying claim — that emotional intelligence outpredicts cognitive intelligence for real-world outcomes — does not survive contact with the meta-analytic literature. The actual research tells a more interesting story: IQ is by far the strongest predictor of job and academic performance, emotional intelligence adds modest incremental value mostly in interpersonal domains, and most "EQ" measures used in popular accounts are largely re-branded personality traits. Read more →

What does an IQ of 130, 140, or 150 mean?

If you've received a score of 130, 140, or 150 on an IQ test — or if you're simply curious about what these numbers represent — you've likely found that the internet offers more mythology than explanation. These scores place individuals well above average, but what that means practically, statistically, and psychologically requires more than a percentile table. Each of the three numbers sits in a different statistical neighborhood, and each has different implications for what an IQ test can and cannot say about the person who scored it. Read more →

How do gender and education relate to cognitive outcomes?

A 2010 study by Jouve, drawing on 251 examinees of the Jouve Cerebrals Test of Induction (JCTI), found that males scored higher than females on inductive reasoning at both middle/high-school and college levels, with no reliable interaction between gender and education stage. The mean gender gap was approximately 4.7 raw-score points (Hedges' g ≈ 0.40) at the school level, where the comparison was statistically marginal (p ≈ .06), and approximately 6.3 points (g ≈ 0.61) at the post-secondary level (p < .01). Education stage produced a substantially larger effect (η²p = 0.114) than gender (η²p = 0.066), and the gender × education interaction was not reliable (η²p = 0.001) — meaning the gender effect was roughly constant across stages, with the apparent "divergence" reflecting differential statistical power across cell sizes rather than a substantive interaction (Cogn-IQ, 2025). Read more →

What is the factor-analytic case for a single reasoning factor?

The clearest small-sample test of the question in a single battery comes from an analysis of the Jouve-Cerebrals Test of Induction (JCTI) alongside four General Ability Measure for Adults (GAMA) subtests — Matching, Analogies, Sequences, and Construction (Jouve, 2018). The analysis pooled 118 adult test-takers and asked whether their performance on the five tasks was best described by one factor or two.

Why treat spatial ability as distinct anyway?

The single-factor finding from a five-test battery is not the whole story. Wai, Lubinski, and Benbow (2009) drew on more than 50 years of cumulative data from the Study of Mathematically Precocious Youth (SMPY) and other longitudinal samples to argue that spatial ability has incremental predictive validity for STEM educational and occupational outcomes above and beyond general reasoning and verbal ability. Across multiple cohorts and follow-up windows extending into mid-career, individuals high on spatial ability were over-represented in engineering, the physical sciences, and technology fields — even after controlling for general intelligence and verbal ability.