Internal g-Loading of the International IQ Test

Published by: International IQ Test
Last updated:

For an IQ test modeled on Raven’s Progressive Matrices, the g factor is a useful indicator of internal structural validity. In other words, it helps determine whether the test’s scoring algorithm produces an IQ score that truly reflects a single, general cognitive ability—as expressed in how participants respond to the items.

Data analyzed

We ran the same analysis on two independent samples from the International IQ Test, using the criteria below:

  • Sampling windows
    • 2024: 01/01/2024 to 01/01/2025
    • 2025: 01/01/2025 to 01/01/2026
  • Internal anti-bot filters
  • One unique email address per participant
  • 50,000 participants per year

Method

  1. We converted each participant’s responses into 40 binary variables (0 = incorrect, 1 = correct).
  2. We estimated an individual g score directly from the item responses by extracting the first principal component (PC1) from the 40-item response matrix.
  3. We computed the g-loading of the final IQ score as the correlation between g (from the 40 item responses) and the final score produced by the test’s algorithm.

A higher g-loading indicates that the final score is more closely aligned with the general factor measured by the test.

Results

2024 (N = 50,000)

  • g-loading (corr(g, score)): 0.9437
  • 95% CI: [0.9428, 0.9447]
  • Consistency check: corr(g, total raw score) = 0.9874
  • Variance explained by PC1: ≈ 15.23% (binary items)

2025 (N = 50,000)

  • g-loading (corr(g, score)): 0.9429
  • 95% CI: [0.9419, 0.9439]
  • Consistency check: corr(g, total raw score) = 0.9871
  • Variance explained by PC1: ≈ 15.08% (binary items)

Interpretation

  • The g-loading is very high (~0.943) and is essentially unchanged from 2024 to 2025.
  • The confidence intervals are extremely tight, suggesting a highly precise estimate.
  • Practically speaking, this means the final IQ score behaves like a strongly g-saturated measure: it closely tracks the general cognitive ability captured by the item responses.

Conclusion

Across two independent yearly samples (50,000 participants each), the final score is highly aligned with the g factor extracted from the raw item responses.

Transparency

  • What this shows
    • Internal structural validity: within this test, the final score reflects a broad general cognitive dimension.
    • The result replicates across two independent years.
  • What this does not replace
    • External, clinically supervised validation (e.g., WAIS/WISC) administered under controlled conditions.
Take the IQ Test