Reasoning5–10 minBurden: MediumEMA: Low

Figure Matrices

A figural reasoning task in the format of Raven's matrices: a 3 × 3 matrix whose rows follow one to three rules, the bottom-right panel missing, and eight options. Every item is generated from a seed.

Inductive reasoningFluid reasoningFigural reasoning
Category
Reasoning
Typical duration
5–10 min
Participant burden
Medium
EMA suitability
Low

Try this task

Configure parameters and run an interactive preview exactly as participants will experience it. No data is recorded.

Configure preview

Adjust parameters below, then start the preview on the right.

Include practice trials

Shown with feedback before the main task

Task parameters

Shared evenly across the levels, easiest first. Default 12, the length of the short forms of Raven’s Advanced Progressive Matrices (Arthur & Day, 1994; Bors & Stokes, 1998). Harris et al. (2020) gave 5 items per Sandia Matrices item type.

With feedback; one per level in turn. Default 3 (our choice; no published value).

Levels

The number of rules, as in the one-, two- and three-relation items of the Sandia Matrices (Matzen et al., 2010), which get harder in that order (Harris et al., 2020). Every rule runs along the rows, the same in each row, with the rule types of Carpenter, Just & Shell (1990): shapes turning by 45° or 90°, a dot moving one or two places around the edge, or one shape more or fewer per step (pairwise progressions), and each row holding the three fills once (distribution of three values). Anything without a rule is the same in all nine panels.

Default none: time pressure lowers accuracy in matrix tests (Gonthier, 2023). Krautter et al. (2021) capped figural matrices at 90 s per item. An item not answered in time counts as an omission.

Default none. Published short forms time the whole test: 15 min for the 12 items of the APM short form (Arthur & Day, 1994, as given by Frischkorn & von Bastian, 2021), 25 min for the 20 items of the HeiQ short forms (Pallentin et al., 2024), about 75 s per item. The full APM (36 items) is also given with 30 or 40 min (Hamel & Schmittmann, 2006). With a limit, items come in order with no going back and a countdown is shown; the item on screen at time-up is kept but not scored, and the items never shown count as not reached. Practice is not timed.

Empty: each participant gets a new form, and the seed it was drawn from is stored with the result. The same seed gives every participant the same items.

Eight options per item in two rows of four, as in Raven-type matrices: every combination of the right and one wrong value of three of the item’s dimensions (a 2 × 2 × 2 facet design; Pallentin, Danner & Rummel, 2023; Hu et al., 2021), so every value appears in four options and the answer cannot be found from the options alone. Level 3 tests its three rules; levels 1 and 2 test their rules and one or two constant dimensions. About 7 min with these settings. Outputs: correct, answered, not reached (with a time limit for all items), accuracy per level and per rule, median RT of correct answers; each trial stores the matrix, the options and which rules the chosen option breaks.

5–10 minBurden: MediumEMA: Low

This is a researcher preview. No participant data is recorded.

Simulated participant view

9:41

Figure Matrices

A figural reasoning task in the format of Raven's matrices: a 3 × 3 matrix whose rows follow one to three rules, the bottom-right panel missing, and eight options. Every item is generated from a seed.

No data is recorded

Participant experience on smartphone

A 3 × 3 grid of panels with a dashed panel and a question mark at the bottom right; below it eight option panels in two rows of four, tapped to answer. On a 360-pt-wide phone the matrix panels are about 86–104 pt and the options about 78 pt, depending on the screen height. Figures are drawn in one dark ink on white, as in Figure Series: shapes that point one way, a dot on one of eight places around the edge, one to five shapes, and empty, striped or filled shapes, so no rule depends on colour. Optionally a countdown for the whole block.

When to use

For a matrices measure with many parallel forms: items are generated, so sessions can use new items (no seed) or one fixed form (a set seed). Suited to baseline and occasional sessions, and to studies of the figural matrices used in aptitude and selection tests, timed as a whole block if needed.

When not to use

Not a normed intelligence test and not a substitute for Raven's matrices: the items follow published rule types and distractor principles but have no norms or IRT parameters yet, and their rules are simpler than the hardest APM items (no figure addition or subtraction, no logic rules). Not a quick EMA probe. For a fixed set of 15 items, use Matrix Reasoning; for rows of figures, Figure Series.

How to use in a study

The defaults give 12 items, 4 per level, levels 1 to 3 (one, two and three rules), easiest first, without a time limit. Leave the seed empty to give each participant and each session new items; set a seed to give everyone the same form. To time the block as published short forms do, set a time limit for all items (e.g. 15 min for 12 items); the item on screen at time-up is kept but not scored, and the rest count as not reached. Per-level accuracy shows where performance breaks down; the trial data say which rules a chosen wrong option breaks.

Researcher-configurable parameters

  • Number of items: default 12, the length of the APM short forms (Arthur & Day, 1994; Bors & Stokes, 1998); Harris et al. (2020) gave 5 per Sandia Matrices item type
  • Levels: 1, 2 and 3 rules (the one-, two- and three-relation items of Matzen et al., 2010; Harris et al., 2020)
  • Time limit per item: default none (Gonthier, 2023); Krautter et al. (2021) capped figural matrices at 90 s per item
  • Time limit for all items: default none; the 12-item APM short form is given with 15 min (Frischkorn & von Bastian, 2021), the 20-item HeiQ short forms with 25 min (Pallentin et al., 2024)
  • Seed: empty (new items per run; the seed is stored with the result) or a number (the same items for everyone)
  • Practice items with feedback: default 3, one per level (our choice; no published value)

Outputs collected

  • Correct answers and items answered
  • Items not reached, when the block has a time limit
  • Accuracy per level and per rule (turning shapes, moving dot, number of shapes, three fills per row)
  • Median RT of correct answers
  • Every trial: level, rules, the matrix, the eight options, the correct option, the response, the rules the chosen option breaks, touch-down RT, time-outs, and whether the item was cut off by the block time
  • The seed, from which every item of the run can be rebuilt

Interpretation notes

Rules run along the rows, the same in every row, each row from its own starting value (rule types of Carpenter, Just & Shell, 1990): the shapes turn by 45° or 90° per step, a dot moves one or two places around the edge, or there is one shape more or fewer per step (pairwise progressions); or each row holds the three fills once, as a Latin square (distribution of three values). Anything without a rule is the same in all nine panels, so the level is the number of rules, as in the Sandia Matrices; unlike the Sandia generator, which also runs relations down columns and along diagonals, all rules here run along the rows, so the levels are not comparable with Sandia norms. Options: the eight options are a 2 × 2 × 2 facet design, the right and one wrong value of three dimensions (Pallentin, Danner & Rummel, 2023; the balanced answer sets of I-RAVEN, Hu et al., 2021), so every value appears in four options: choosing by most shared features, by majority value or as the odd one out is right at chance (1 in 8) on generated items, and at level 3 rule-free looks at the matrix (what is new in the last row, what looks like the panel above) stay near chance too, apart from the fill rule, where the answer is by definition the fill missing from its row. Level 3 tests its three rules; levels 1 and 2 also test one or two constant dimensions, so they are easier. Without a time limit the task measures power; RTs are timed from the drawn item to the touch-down, but this timing has not been validated on real devices. The items have no norms; compare groups or sessions within a study.

Scientific evidence

  • In the Raven matrices, individuals differ mainly in inducing abstract relations and in managing many problem-solving goals in working memory (Carpenter, Just & Shell, 1990).
  • Software that combines the relation types of the Standard Progressive Matrices produced generated items that covered and extended the difficulty range of the originals in a norming study (Matzen et al., 2010).
  • In an IRT review of these generated items, each with eight response options, item types grew harder from one-relation to logic problems; the authors caution that generated distractors need checking (Harris et al., 2020).
  • Distractors from a facet design, with every value shown equally often, prevent counting strategies (Pallentin, Danner & Rummel, 2023); short forms of that test take about 75 s per item (Pallentin et al., 2024).
  • 12-item short forms of the Advanced Progressive Matrices showed psychometric properties similar to the 36-item test (Arthur & Day, 1994) or acceptable ones (Bors & Stokes, 1998), in a much shorter testing time.
  • Even mild time pressure lowers accuracy and constructive matching in Raven's matrices (Gonthier, 2023).