JCWS: A Verbal Abilities Test

The Jouve-Cerebrals Word Similarities (JCWS) is a 150-item open-response verbal-reasoning test built around a distinctive item format that combines partial-letter cues with semantic and analogical reasoning. Each item presents a target word’s letters scrambled or partially given, plus a verbal clue: for example, “(A _ I K L N T) means nearly the same as SPEAKING” — the examinee infers that the answer is TALKING from both the letter constraints and the meaning relation. The format unifies vocabulary depth, orthographic processing, and inductive reasoning in a way that more conventional verbal tests do not, and it produces a Verbal Reasoning Index (VRI) on the standard IQ metric (M = 100, SD = 15) with reliability ξ = .981 in a 407-examinee development sample (Cogn-IQ, 2025). The JCWS is positioned by its manual as an active research instrument rather than a fully-validated clinical battery.

Three subtests, one composite

The JCWS contains 150 items distributed equally across three subtests. Each subtest uses the same letter-cue format but emphasizes a different cognitive operation:

• Nearly The Same (NTS, 50 items, ξ = .952). The examinee produces a word that shares meaning with a verbal clue, given partial letters of the target. Example format: “(A _ I K L N T) means nearly the same as SPEAKING.” This is the closest to a traditional vocabulary task, but the letter constraint adds an orthographic-retrieval component that pure synonym tests do not require.
• Is To As (ITA, 50 items, ξ = .958). The examinee completes a four-term verbal analogy with the same letter-cue scaffold. Example format: “(A _ I K L N T) is to SPEAKING as WALKING is to MOVING.” The format taps verbal concept formation (Gc/VL) plus inductive reasoning (Gf/I) — both required to identify the relevant relation and to retrieve the target word.
• Which Relates To (WRT, 50 items, ξ = .945). The examinee identifies a word that fits a logical progression. Example format: “(E N O) relates to TWO, which relates to THREE, which relates to (F _ _ U).” This subtest emphasizes general sequential reasoning (Gf/RG) on top of the lexical and orthographic demands shared with NTS and ITA.

The three subtest scaled scores (M = 10, SD = 3) combine into the Verbal Reasoning Index (VRI) through a Modified Tellegen-Briggs Formula 4 with cubic correction, which addresses tail-compression bias and stabilizes scores at the distribution extremes. The VRI is reported with a 95% confidence interval of approximately ±4 points (SEM ≈ 2.07). Reliable change between two administrations requires roughly ≥ 6 VRI points (Cogn-IQ, 2025).

Reliability via JRRE (Jouve’s Xi)

Reliability for the JCWS is computed using Jouve’s Randomized Reliability Estimation (JRRE; Jouve, 2025), also referred to as Jouve’s Xi (ξ). Where Cronbach’s alpha computes a single internal-consistency value from the observed item covariances and assumes essentially tau-equivalent items, JRRE estimates reliability through repeated randomized split-half resampling: items are randomly partitioned into halves, the half-scores are correlated, the correlation is Spearman-Brown adjusted, and the procedure is averaged across many permutations. The output is a reliability distribution with confidence intervals rather than a single point estimate, robust to any particular split or item ordering. Jouve (2025) reports a comprehensive simulation study comparing JRRE against Cronbach’s α, KR-20, McDonald’s ω_t, eigenvalue-based lower bounds, and Guttman’s λ coefficients across Rasch, 2PL, 3PL, 4PL, GRM, GPCM, and NRM models, documenting the conditions under which classical coefficients underestimate or distort reliability and JRRE provides more accurate estimates.

For the JCWS development sample (N = 407, 150 items), JRRE yields:

• Total scale (VRI): ξ = .981 (95% CI [.972, .986])
• NTS: ξ = .952 (1,774 iterations)
• ITA: ξ = .958 (2,607 iterations)
• WRT: ξ = .945 (5,552 iterations)

These values are high in absolute terms and consistent with what would be expected for a long, content-homogeneous verbal-reasoning instrument. The negative skew in the JRRE distribution (typical at high reliability) reflects the ceiling compression that occurs as reliability approaches 1.0; the manual emphasizes median and percentile bands of the distribution rather than mean-only point estimates.

Subtest intercorrelations and what they imply

Within-test subtest correlations in the development sample (N = 407) range from .746 to .868, with an average intercorrelation of .728. This is high — substantially higher than typical between-subtest correlations in multi-construct cognitive batteries — and it reflects the subtests sharing the same item-format scaffold and the same underlying verbal-reasoning factor. The intercorrelation pattern supports computing a single VRI composite while retaining subtest-level scores for profile description.

Factor structure: a strongly unidimensional general factor

A factor analysis (N = 28) examined the JCWS subtests jointly with the JCCES subtests (Verbal Analogies, Math Problems, General Knowledge) to characterize the JCWS’s structure within a broader Gc nomological network. Factorability indices supported the analysis: Kaiser-Meyer-Olkin sampling adequacy = .822 (“very good”), Bartlett’s test of sphericity χ² = 170.75 (p < .001), matrix determinant 0.0009 (no multicollinearity), average intercorrelation .728 (Cogn-IQ, 2025).

The general-factor evidence is striking:

• Explained Common Variance (ECV) = 92.3% — of the variance shared across all six indicators (three JCWS subtests, three JCCES subtests), 92.3% loads on the general factor.
• Omega Hierarchical (ω_h) = 1.000 — indicating a unidimensional structure where the general factor accounts for essentially all reliable variance.
• First eigenvalue = 4.65 (77.5% of total variance), with an eigenvalue ratio of 7.05:1 between the first and second components, well above conventional thresholds for unidimensionality.
• g-loadings for JCWS subtests: NTS .939, WRT .923, ITA .899; for JCCES indicators: VA .920, GK .856, MP .813. All six load strongly on the general factor.

The CHC mapping in the manual labels the JCWS subtests as primarily Gc/VL (lexical knowledge) for NTS, with VL/LD plus secondary I (induction) or RG (general sequential reasoning) for WRT and ITA. These narrow-ability labels describe content emphasis rather than separate factors in the analysis — the empirical structure is dominated by g.

Convergent validity

The JCWS technical manual reports convergent-validity correlations against four criterion measures (Cogn-IQ, 2025). Both observed and disattenuated coefficients are reported, where disattenuation uses the Spearman correction with reliability inputs of ξ = .981 for the JCWS VRI and the published reliability estimates for each criterion.

VRI convergent validity:

• WAIS-IV Verbal Comprehension Index: r = .86 (N = 15); disattenuated r’ = .89
• IAW Vocabulary Proficiency Index: r = .80 (N = 57); disattenuated r’ = .83
• JCCES Cognitive Acumen Index (CAI): r = .66 (N = 28); disattenuated r’ = .68
• JCCES Verbal Acumen Index (VAI): r = .71 (N = 28); disattenuated r’ = .74
• JCCES Verbal Analogies (VA) subtest: r = .83 (N = 28); disattenuated r’ = .91
• JCCES Math Problems (MP) subtest: r = .51 (N = 28); disattenuated r’ = .54
• JCCES General Knowledge (GK) subtest: r = .66 (N = 28); disattenuated r’ = .69

The pattern is informative for what it tells us about the JCWS’s construct. The strongest correlation is with the JCCES Verbal Analogies subtest (.83 observed, .91 disattenuated) — a subtest that uses an analogous A:B::C:D format without the letter-cue scaffold. This indicates that the JCWS’s letter-cue items, despite their distinctive surface format, are tapping the same verbal-analogical-reasoning construct as more conventional analogy tests. The .80 correlation with the IAW VPI (a pure vocabulary measure) places the JCWS in the same Gc neighborhood as established vocabulary measures. The lower correlation with JCCES MP (.51) — the math-problems subtest — is the appropriate differential-validity pattern: the JCWS does not measure quantitative ability.

The .86 correlation with WAIS-IV VCI is high but based on a small sample (N = 15). The manual flags this coefficient as provisional, noting that “replication with larger Ns is in progress; treat those specific coefficients as provisional.” This is appropriate transparency about a research-instrument validity coefficient that has not yet been independently replicated.

The letter-cue item format and what it adds

The JCWS’s signature design feature — partial letters of the target word presented alongside the semantic clue — is unusual among standardized verbal tests. Most vocabulary and verbal-analogy measures present either pure semantic content (definitions, analogy stems with no orthographic information) or pure orthographic content (letter puzzles with no semantic anchor). The JCWS combines both, which has measurement implications worth understanding.

The orthographic constraint serves two functions. First, it sharply narrows the response space. Without the letter cue, a synonym task admits hundreds of plausible responses to a typical clue; with it, the response space typically narrows to a small handful of words that satisfy both the meaning and the letter pattern. This makes automated scoring tractable while preserving the production-rather-than-recognition format. Second, the orthographic constraint adds a retrieval-from-partial-cue component that tests have shown loads on Glr (long-term retrieval) over and above the pure Gc loading of definitional vocabulary tasks.

The trade-off is in construct purity. A pure Gc/VL measure isolates lexical knowledge; the JCWS combines lexical knowledge with orthographic-retrieval ability and (in WRT) sequential reasoning. The factor analysis indicates these are well-integrated within a single general factor in this sample, but users should understand that the JCWS is not equivalent to a pure vocabulary measure such as the IAW VPI — the formats overlap substantially (r = .80 disattenuated to .83) but are not interchangeable.

Practical implications

For researchers and clinicians considering the JCWS:

• Use the JCWS when extended-ceiling verbal reasoning is the construct of interest. The 150-item bank with three subtests provides measurement at higher ability levels than typical clinical batteries reach, useful for research with high-ability samples and for situations where standard tests reach their ceiling.
• Recognize the research-instrument framing. The technical manual explicitly positions the JCWS as developing psychometric evidence, particularly for criterion correlations with the WAIS where current N = 15. High-stakes clinical applications should rely on instruments with more accumulated independent-replication evidence.
• Account for the linguistic and orthographic dependencies. Cormier et al. (2022) showed that examinee characteristics — particularly linguistic background and exposure to the test language — affect cognitive test performance more strongly than test characteristics in many contexts. The JCWS’s letter-cue format adds an orthographic-language-specific component that compounds this consideration. Cross-linguistic and cross-cultural use requires either norm reference within the target population or careful interpretive caveats.
• Pair with measures of other broad abilities. The factor analysis indicates the JCWS is essentially unidimensional within its content space. A complete cognitive profile requires complementary measures of fluid reasoning, processing speed, working memory, and other broad CHC abilities.
• Plan for the 150-item length. The instrument is thorough but long; the manual recommends scheduling breaks and monitoring effort, and noting that the 95% confidence interval of approximately ±4 VRI points reflects the high reliability of the long form.

Open research directions

Independent replication of the convergent-validity correlations — particularly the WAIS VCI association where the current N = 15 is small — is the most important next step in the JCWS evidence base. Differential item functioning analyses across linguistic, educational, and demographic subgroups would clarify the bounds of fair use. Test-retest reliability across longer intervals would complement the strong internal-consistency evidence already documented. The manual is transparent about these directions and frames the JCWS appropriately as an instrument under continued psychometric development rather than a finalized clinical battery.

The takeaway

The JCWS is a 150-item open-response verbal-reasoning test with a distinctive letter-cue item format, three subtests (Nearly The Same, Is To As, Which Relates To), and a composite Verbal Reasoning Index on the IQ metric. JRRE reliability of ξ = .981 across N = 407 and a unidimensional factor structure (ECV = 92.3%, ω_h = 1.000, g-loadings .899–.939) support a strong general-verbal-reasoning interpretation. Convergent-validity correlations span r = .80–.86 with established verbal measures (IAW VPI, WAIS VCI), with the strongest disattenuated correlation (.91) against the JCCES Verbal Analogies subtest. The instrument is appropriately positioned by its technical manual as an active research instrument with continuing psychometric development.