How the 64 Types IQ test is built and scored
In short: 24 nonverbal puzzles (16 pattern matrices and 8 number series) generated fresh for every person, scored with a Rasch model (item response theory) and put on the familiar IQ scale (mean 100, standard deviation 15). The result is an online estimate for your own interest. It is not a clinical assessment, it is not a diagnosis, and it should not be used for hiring, admission or legal decisions.
The puzzles
- Pattern matrices (in the style of Raven’s Progressive Matrices): a 3×3 grid of figures with the last cell missing and 8 answer options. Each figure has four attributes (shape, number of figures, size, shading). Each attribute either stays fixed, stays the same within a row, changes step by step along the row, or appears as the same three values in a different order in every row. Harder puzzles combine more changing attributes and harder rule types, the main drivers of matrix difficulty found by Carpenter, Just & Shell (1990) and used to build the Sandia matrices (Matzen et al., 2010).
- Number series: find the next number, 6 answer options. Rules go from adding a constant (easiest) to interleaved sequences, Fibonacci-like sums, prime-number steps and alternating operations (hardest), following the rule-complexity account of series completion by Simon & Kotovsky (1963).
- Blueprint: five design difficulty levels, easiest first. Level 1: 2 matrices, 1 series; Level 2: 3 matrices, 2 series; Level 3: 4 matrices, 2 series; Level 4: 4 matrices, 2 series; Level 5: 3 matrices, 1 series.
- Generated, not copied: every puzzle is generated on your device from a random seed, so no two people get the same test and there is no answer sheet to look up. The same seed always rebuilds the same puzzles, which is how a shared result link and the certificate server can re-create your exact test.
- No reading: only shapes and numbers, so the puzzles are the same in all 41 languages of the site. Instructions are translated.
Timing
The test is designed to take about 15 minutes and we suggest working quickly but carefully. There is no enforced time limit and time is not used in the score. You can skip a puzzle and come back; puzzles still unanswered at the end count as wrong. Because time is not controlled, scores from a relaxed session can be higher than they would be under timed conditions.
Scoring
- Model: a Rasch (one-parameter logistic) model. The chance of solving a puzzle depends on the difference between your ability θ and the puzzle’s difficulty b: P = 1 / (1 + e−(θ − b)).
- Difficulties used today (logits, by design level): level 1: -2.2, level 2: -1, level 3: 0.1, level 4: 1.1, level 5: 2.1. They come from the puzzle design, not yet from a calibration on real answers (see “Norms” below).
- Ability estimate: expected a posteriori (EAP) with a standard-normal prior, computed on a grid from −4 to +4 (Bock & Mislevy, 1982). Harder puzzles you solve raise the estimate more than easy ones.
- IQ-scale number: 100 + 15 × θ, rounded and kept between 55 and 145 (a short test can’t measure reliably beyond that).
- 90% range: the estimate ± 1.645 posterior standard deviations, on the same scale. With 24 puzzles this range is typically 15 to 25 points wide, which is why the free result shows the range first.
- Percentile: the share of a normal distribution below your θ (the normal cumulative distribution function). It is a property of the scale, not a count of real test takers, and it is only as good as the norms below.
- Labels (Average, High average, …) use Wechsler-style score bands.
Norms: provisional
A clinical IQ test is normed on a large sample chosen to represent a population, so that 100 means “the population average”. This test is not. Its scale assumes that the design difficulties above are right and that test takers’ ability follows a normal distribution with mean 0 and SD 1 in θ. Neither has been confirmed yet. We store anonymous puzzle-level answers (no name, no email) and will re-estimate the puzzle difficulties with the Rasch model once there are enough answers per puzzle type and level. Until then, treat every score, range and percentile as provisional. When the difficulties change, the test version changes; every Verified IQ Certificate shows the version it was scored with (today: iq-1).
People who choose to take an online IQ test are not a random sample of any population, so even after recalibration the scale describes this test’s takers rather than a whole country.
What the Verified IQ Certificate does and does not show
- Does: the score on the certificate was computed by our server from the answers of that test (the server rebuilds the puzzles from the seed and scores them; it does not take a number from the browser). The record, with a unique random ID, is stored on our server, and anyone can open the verification page to see that the name, score, date and test version match our records.
- Does not: confirm who took the test, that they took it alone, or under what conditions. The name is typed by the buyer. It is not a clinical or professional assessment and is not issued or endorsed by any testing body or society.
Sources
- Rasch, G. (1960). Probabilistic models for some intelligence and attainment tests. Danish Institute for Educational Research.
- Bock, R. D., & Mislevy, R. J. (1982). Adaptive EAP estimation of ability in a microcomputer environment. Applied Psychological Measurement, 6(4), 431–444.
- Embretson, S. E., & Reise, S. P. (2000). Item response theory for psychologists. Lawrence Erlbaum.
- Carpenter, P. A., Just, M. A., & Shell, P. (1990). What one intelligence test measures: A theoretical account of the processing in the Raven Progressive Matrices Test. Psychological Review, 97(3), 404–431.
- Matzen, L. E., Benz, Z. O., Dixon, K. R., Posey, J., Kroger, J. K., & Speed, A. E. (2010). Recreating Raven’s: Software for systematically generating large numbers of Raven-like matrix problems with normed properties. Behavior Research Methods, 42(2), 525–541.
- Simon, H. A., & Kotovsky, K. (1963). Human acquisition of concepts for sequential patterns. Psychological Review, 70(6), 534–546.
- Wechsler, D. (2008). Wechsler Adult Intelligence Scale, Fourth Edition (WAIS-IV). Pearson (score band labels).