Online "IQ tests" are everywhere, and almost none of them can tell you what they claim to. A number that looks like an IQ needs a large, representative sample behind it, and a short quiz on a website rarely has one.
This guide explains what reasoning problems actually measure, how our reasoning test scores them without inventing a norm, and what you can do with the result.
A cognitive reasoning test measures how well a person solves unfamiliar problems by finding rules and following logic. This test has 24 original problems written for this site, six each of number and letter series, verbal reasoning, figural matrices and cube folding, modelled on the item types of the International Cognitive Ability Resource. It reports problems solved, never an IQ.
Four kinds of problem
The test follows the four item types of the International Cognitive Ability Resource (ICAR), a research project that showed short, unproctored sets of such problems track general reasoning ability (Condon & Revelle, 2014).
- ✦Number and letter series: find the rule behind a sequence and extend it.
- ✦Verbal reasoning: decide what must follow from a few statements, and what merely could.
- ✦Figural matrices: several rules act at once on shapes, numbers, shading or lines in a three-by-three grid.
- ✦Cube folding: picture how a flat net folds into a cube, and which view of it is possible.
Why the problems are our own
ICAR's item bank is licensed for academic research only, and its terms do not allow the items on a website. So every problem in our test was written for the site from the general item type, then checked by computation to have exactly one answer: each matrix against every rule that fits its cells, and each cube by folding its net geometrically.
Original items have a cost: they have not yet been calibrated on any sample. That is why the result is a count of problems solved, not an IQ.
What the score is, and what it is not
Your score is how many of the 24 problems you solved. The likely range around it comes from the binomial error model (Lord, 1957), which needs only the number of items and correct answers, not a comparison sample.
The other reference point is guessing. Each problem has three to eight options, so pure guessing averages about 4.7 correct and often reaches 7. A total at or below that cannot be read as reasoning, and the result says so plainly.
There is no IQ, percentile or "smarter than N% of people", because none of those can be computed without a norm sample. Six problems per kind also makes each kind's score coarse, so the result names a stronger or weaker kind only when the gap is larger than measurement error.
Getting better at reasoning problems
Practice makes people better at the kinds of problem they practise. Spatial skills respond especially well: a meta-analysis of 217 training studies found that gains lasted and carried over to untrained spatial tasks (Uttal et al., 2013). Whether practice raises reasoning in general is far less certain; working-memory training, for example, has not shown convincing transfer to nonverbal reasoning (Melby-Lervåg & Hulme, 2013).
A few habits help on every kind: work one rule or attribute at a time, predict the answer before reading the options, and for cubes, find the opposite faces first — they are never visible together.
Using the result well
Do not retake the same problems to measure progress; you will remember them. Judge practice by how new problems feel. And never use a score like this for hiring, admission or a diagnosis: it was not built or validated for any of them.
If you want to see how you think about work rather than how you solve puzzles, the career interests guide is a different and complementary angle.