How IQ Tests Work: Question Types and Scoring

An IQ test looks like a collection of puzzles, and in a sense it is. But the puzzles are chosen carefully: each type is included because performance on it correlates with performance on the others, and the combination estimates general reasoning ability more reliably than any single type could alone.

The design problem

A test builder faces a specific challenge. They need items that are hard enough to separate people from one another, that do not depend on knowledge only some groups have had the chance to acquire, and that cannot be solved by memorisation. Above all, the items must correlate with each other — that shared variance is what the score is estimating.

This is why tests mix formats. A single question type measures itself plus a lot of noise. Several different types, aggregated, cancel much of the noise and leave the common factor.

The main question types

Matrix reasoning

A grid of figures follows some rule, with one cell left blank. You pick the option that completes the pattern. The rules are things like: a shape rotates a fixed amount each step; elements are added or subtracted across a row; two figures combine to produce a third.

Matrix items are the closest thing testing has to a workhorse. They require no vocabulary, no arithmetic and no cultural knowledge, which makes them unusually portable across languages and backgrounds — the principle behind Raven's Progressive Matrices, one of the most widely used instruments in the field. They load heavily on fluid reasoning: working out an unfamiliar rule on the spot.

Number series

A sequence is given and you supply the next term: 2, 6, 12, 20, 30, … The rule may involve differences, ratios, alternating operations, or a second-order pattern in the differences themselves.

These require basic arithmetic, so they are not culture-free in the way matrix items are. What they probe is the ability to hold several candidate rules in mind, test them against the data and discard the ones that fail — inductive reasoning under working-memory load.

Spatial and mental rotation

You judge whether two shapes are the same object viewed from different angles, or pick which folded net produces a given solid. Solving these means constructing a mental representation and transforming it — visualising, not calculating.

Spatial ability is a distinct broad ability in its own right, and it shows some of the clearest trainability of the abilities tests sample: practice improves spatial task performance substantially, though how far that transfers to unrelated reasoning is another question.

Verbal analogies and vocabulary

"Doctor is to hospital as teacher is to ___". Analogies test whether you can extract a relation and apply it to a new pair. Vocabulary items ask for word meanings directly.

Both lean on crystallised knowledge — what you have accumulated. Vocabulary in particular is among the strongest single predictors of overall test scores in the general population, which surprises people who expect abstract puzzles to dominate. The catch is obvious: these items depend heavily on language exposure and schooling, so they translate badly and disadvantage anyone tested outside their first language. Most international online tests either omit them or handle them with care.

Working memory and processing speed

Full clinical batteries include tasks like repeating digit sequences backwards, or rapidly matching symbols to numbers against the clock. These are separate indices in a Wechsler assessment, contributing to the overall score alongside reasoning.

Timed online tests capture some of this incidentally: when a test has an overall time limit, processing speed is folded into your performance whether or not it is measured separately.

From answers to a score

The conversion happens in stages.

1. Raw score. Correct answers are counted. In many tests items carry different weights, so harder items contribute more than easy ones rather than every item counting the same.
2. Comparison to norms. The raw score is looked up against the distribution produced by the standardisation sample — for age-normed tests, the sample for your age band. This step converts "14 correct" into "at the Nth percentile of the reference group".
3. Scaling. That percentile position is expressed on the IQ scale, where the reference mean is 100 and one standard deviation is 15 points.

Everything hinges on step two. A test with a small, unrepresentative or outdated norm sample produces numbers that look authoritative and mean very little — the arithmetic is fine, but the comparison group is wrong.

Item difficulty and adaptive testing

Not all items are equally informative. An item that everyone answers correctly, or that everyone misses, separates nobody and contributes nothing. The most informative items are those answered correctly by roughly half of the people near the difficulty level being probed.

This is why professional tests are developed using item response theory, which models each item's difficulty and its ability to discriminate between nearby ability levels. Items that perform poorly are cut before the test ships.

It is also why adaptive tests are efficient. When the difficulty of the next item is chosen based on how you answered the last, the test spends its time near your actual ability level instead of on items that are trivial or hopeless for you. An adaptive test can reach the same precision in far fewer items — the principle behind computerised versions of many standardised exams.

What makes a test good

• Reliability — consistency. If the same person takes the test twice under similar conditions, do they get a similar score? Well-constructed instruments report high test–retest reliability over short intervals.
• Validity — whether it measures what it claims. This is established by correlating scores against established instruments and against outcomes the construct should predict.
• Current, representative norms — as above, the comparison group determines what the number means.
• Standardised administration — the same instructions, timing and conditions for everyone. This is the requirement unsupervised online tests structurally cannot meet.

How this test is built

The assessment on this site uses 20 non-verbal items with a 30-minute overall limit, weighted by difficulty, and reports a score on the standard SD-15 scale. It relies on pattern and sequence reasoning rather than vocabulary, so it does not penalise you for taking it outside your first language.

Being unsupervised and self-selected, it is a reasonable indication rather than a formal measurement — a distinction worth keeping in view when you read the result.