How matrix reasoning works

Matrix reasoning questions present a grid of figures governed by one or more rules across rows, columns or diagonals. To solve them, identify which features change (shape, number, color, position, size, rotation), work out how they change, then pick the option that satisfies every rule at once. Difficulty increases mainly with the number of rules and how much they interact.

Row one: one small black circle, two small black circles, three small black circles. Row two: one medium white square, two medium white squares, three medium white squares. Row three: one large black triangle, two large black triangles, and the missing cell. Two rules run at once: the count increases left to right, and shape, size and shading are constant within each row. The answer is three large black triangles. A classic distractor here is two large black triangles, which satisfies the shape rule but breaks the counting rule, which is why checking every rule before selecting matters.

The taxonomy above is not arbitrary. Carpenter, Just and Shell's 1990 analysis of the Raven Progressive Matrices identified a small set of rules that account for most items, including constant in a row, quantitative pairwise progression, figure addition or subtraction, and distribution of three values. They also found that item difficulty tracked the number of rules that had to be managed at once and the need to hold goals in working memory, which matches what most test takers report: two-rule items feel manageable, four-rule items feel impossible.

Start with the first complete row and list every visible feature. Compare it with the second row: which features follow the same pattern? Form one rule per feature. Predict what the missing cell should look like before looking at the answer options. Then eliminate options that break any rule. Predicting first reduces the pull of attractive but wrong answers that fit only one rule.

Matrix items load working memory and goal management: each candidate rule must be held while the others are tested, which is precisely the process Carpenter and colleagues modeled. Research on speeded testing adds that time limits shift what the test measures by bringing processing speed into the score, which is why untimed practice feels so different from a timed test.