Assessing Nonverbal Intelligence: Insights from the JCFS

Most cognitive ability tests in widespread use are either verbal-heavy (vocabulary, comprehension, knowledge) or rely on a fixed sequence of items that takes everyone a similar amount of time. The Jouve Cerebrals Figurative Sequences (JCFS), developed by Xavier Jouve, takes a different approach on both fronts. It is purely nonverbal — every item is a 3×3 grid of figural elements that follow abstract transformation rules — and it administers items adaptively, drawing from a bank of 50 tasks but typically reaching a precise ability estimate after only 6–16 items. The two design choices have specific psychometric consequences worth understanding for anyone choosing among nonverbal cognitive measures.

What the JCFS measures

The JCFS targets fluid-inductive reasoning: the ability to discover, hold in mind, and integrate abstract relations among visual elements. Each item presents a 3×3 grid in which a pattern of figures (typically crosses or similar geometric forms) evolves according to one or more transformation rules — for example, “rotate 90° clockwise across rows,” combined with “increment the number of elements down columns.” The examinee must infer the operating rules and produce the figure that completes the sequence.

This places the JCFS in the same broad family as the Raven Progressive Matrices, the most widely-used nonverbal cognitive measure in the world. Carpenter, Just, and Shell’s classic 1990 cognitive analysis in Psychological Review showed that what makes a Raven-style item difficult is precisely the number and type of abstract rules to be inferred and integrated. Items requiring inference of one rule are easy; items requiring inference of two non-trivial rules and their joint application are substantially harder. The JCFS items are constructed within this framework.

In Carroll’s (1993) factor-analytic taxonomy of human cognitive abilities — the empirical foundation of the modern Cattell-Horn-Carroll (CHC) framework — abilities of this type load primarily on fluid reasoning (Gf), the broad cognitive ability most strongly associated with general intelligence (g). Fluid-reasoning tests are the strongest single contributors to Full-Scale IQ in most major intelligence batteries. The JCFS’s nonverbal, content-light design isolates this construct without contamination by verbal knowledge or culturally specific content.

What makes the JCFS structurally distinctive

Three design features differentiate the JCFS from most existing nonverbal measures:

• Open-ended response format. Where Raven-style tests typically present 6 or 8 candidate completions and ask the examinee to select the correct one, the JCFS requires the examinee to produce the figure that completes the sequence. This eliminates the cueing effects (process of elimination, partial recognition of distractors) that affect multiple-choice formats.
• Adaptive administration. Drawing from a 50-item bank, the JCFS administers items in real time based on previous responses, selecting items whose difficulty is most informative about the examinee’s current ability estimate. This is the same general approach used in modern computerized adaptive testing systems for educational assessment.
• Untimed completion. The JCFS does not impose time limits on items or the test as a whole. This separates the construct of fluid-reasoning capability from fluid-reasoning speed — two related but distinguishable abilities that timed nonverbal tests blend.

The adaptive engine: 2PL IRT with EAP estimation

The JCFS uses a two-parameter logistic (2PL) item response theory model. In this framework, each item has two characteristic parameters:

• Difficulty (b): the latent-trait level at which an examinee has a 50% probability of answering correctly.
• Discrimination (a): how sharply the probability of a correct response increases as ability increases — high-discrimination items distinguish more cleanly between examinees of similar ability.

The 2PL model is more flexible than the simpler one-parameter Rasch model (which constrains all items to the same discrimination) and more parsimonious than the three-parameter model (which adds a guessing parameter, less relevant for open-ended formats).

After each response, the JCFS updates its estimate of the examinee’s ability using Expected A Posteriori (EAP) estimation — a Bayesian approach that combines the response data with a prior distribution to produce a stable posterior estimate. EAP is generally preferred over maximum-likelihood estimation in adaptive testing because it produces stable estimates from the start of the test, before enough responses have been collected to make likelihood-based estimates well-defined.

The combination of an adaptive item-selection algorithm with EAP estimation typically produces precise ability estimates in 6–16 items (average ~10). Wainer, Dorans, Flaugher, Green, and Mislevy’s foundational 2000 text on computerized adaptive testing established the efficiency gains of this approach: an adaptive test typically achieves the same precision as a fixed-form test using roughly half the items.

Psychometric properties

The 2023 psychometric evaluation reports a marginal reliability of approximately ρ = 0.91 across the score range. Marginal reliability is the IRT analog of classical-test-theory reliability, computed as the average conditional precision integrated across the ability distribution. A value of 0.91 is comparable to that of major commercial cognitive batteries.

Reliability is not uniform across the score range. By design, IRT-based adaptive tests produce their highest precision near the center of the distribution where the most items are concentrated, with somewhat lower precision in the tails. The JCFS’s expected conditional precision gradient is reasonable for a test of this length: lower at very high and very low ability levels, but adequate across the range that covers most examinees.

The Visual Sequencing Index (VSI) is reported on a reference scale with mean 100 and SD 15, the same scale used for IQ scores in major commercial batteries. This allows direct comparison between JCFS results and other cognitive ability scores expressed on the same metric.

How the JCFS compares to other nonverbal measures

The choice between the JCFS and other nonverbal cognitive measures depends on what the assessment is being used for:

• vs. Raven’s Progressive Matrices. The JCFS uses the same broad item type (figural pattern completion) but production rather than recognition responses, and adaptive rather than fixed-form administration. The Raven, with longer history and broader normative data, remains the standard for many clinical applications. The JCFS may produce somewhat different ability estimates because of the format change (open-ended vs. multiple-choice).
• vs. WAIS Matrix Reasoning. The Wechsler Adult Intelligence Scale’s Matrix Reasoning subtest is a multiple-choice nonverbal subtest within a broader IQ battery. It serves a different purpose — contributing to a Full-Scale IQ estimate — than a stand-alone nonverbal measure like the JCFS.
• vs. Cattell Culture Fair Intelligence Test. Like the JCFS, the Cattell CFIT was designed to minimize cultural and language influences. It uses fixed-form administration. The JCFS’s adaptive design produces shorter testing times for similar precision.
• vs. other Jouve tests (JCTI). The JCTI is a related Jouve test focused on inductive reasoning. The JCFS is more specifically figural-spatial.

What the JCFS is and is not appropriate for

Several use cases follow from the test’s design:

• Research where nonverbal cognitive ability is a covariate or matching variable. The JCFS produces a score on the standard mean-100/SD-15 scale that can be used as a matching or covariate variable.
• Self-administered assessment in non-English-speaking populations. The purely figural content is not language-bound. The same test can be administered in any language with no translation effects on item content.
• Cognitive assessment when verbal assessment is impractical. Individuals with hearing or speech impairments, or with limited literacy in the test language, can complete the JCFS with minimal accommodation.
• Tracking changes in fluid reasoning over time. The same test bank can be administered repeatedly, with the adaptive engine producing comparable scores from the same item pool.

The test is not appropriate for some applications:

• Stand-alone diagnostic IQ assessment for clinical decisions. A full intelligence battery (WAIS, Stanford-Binet, RIAS) covering verbal and nonverbal abilities is the standard for clinical IQ assessment. The JCFS is a nonverbal-only measure.
• Assessment of crystallized abilities. The JCFS does not measure verbal knowledge, vocabulary, or accumulated factual knowledge. Individuals with high crystallized intelligence and lower fluid reasoning will not show their full cognitive profile on a JCFS-only assessment.
• Severely cognitively impaired examinees. The 2PL adaptive engine is calibrated for ability levels in the typical range. Examinees with very low cognitive functioning may produce unstable estimates.

Limits of the validation evidence

Several limitations are worth holding in mind:

• Single-author validation. The 2023 psychometric evaluation was conducted by the test author. Independent replication of the reliability and validity findings by other research groups is the next step for a test moving toward broader clinical use.
• Sample composition. The validation sample, while sufficient for reliability estimation, has not yet been broken down into demographic-specific subsamples large enough to test for differential item functioning across age, sex, or cultural-background groups.
• Concurrent validity with other batteries. Strong demonstration that JCFS scores correlate with other established nonverbal measures (Raven’s Matrices, WAIS Matrix Reasoning, NNAT) in independent samples is the standard expectation.
• Test-retest stability. Internal consistency (marginal reliability) is well-characterized; test-retest reliability across longer intervals is the next validation priority.
• Age-stratified norms. Cognitive abilities change with age; a single norm reference is less informative than age-stratified norms for individual interpretation.

Frequently Asked Questions

What is the JCFS test?

A nonverbal, computerized adaptive cognitive test consisting of 3×3 grid items that follow abstract transformation rules. Examinees produce the figure that completes the sequence. The test is untimed and uses adaptive item selection from a 50-item bank.

How is the JCFS scored?

The test produces a Visual Sequencing Index (VSI) on a reference scale with mean 100 and standard deviation 15, the same scale as IQ scores in major commercial batteries.

How long does the JCFS take?

Because it is adaptive, administration time varies. Most examinees complete the test in under 30 minutes, with the adaptive engine typically reaching a precise ability estimate after 6–16 items.

Is the JCFS an IQ test?

The JCFS is a measure of nonverbal cognitive ability, specifically fluid-inductive reasoning. It produces a score on the IQ scale but does not assess verbal abilities, working memory, or other components of a full intelligence battery. It is best understood as a nonverbal-only cognitive measure rather than a full IQ test.

What does “marginal reliability” mean?

The IRT analog of classical reliability, computed as the average precision of ability estimates integrated across the ability distribution. The JCFS’s value of approximately 0.91 is comparable to commercial cognitive batteries.

What’s the difference between the JCFS and Raven’s matrices?

Both are nonverbal figural-pattern tests. The Raven uses fixed-form multiple-choice; the JCFS uses adaptive open-ended (production) responses. The Raven has longer history and broader norming; the JCFS is shorter to administer for comparable precision and avoids cueing effects of multiple-choice formats.

Can the JCFS be administered cross-culturally?

The figural-only content avoids most language and culture-specific elements. Like all cognitive tests, however, performance can still be influenced by familiarity with abstract test-taking conventions, which vary across cultural contexts. Cross-cultural use is more defensible than for verbal tests but is not free of all cultural effects.