Executive Function
Category Learning Task
Also called: perceptual category learning, information-integration category learning, rule-based category learning, categorization task
Measure how people learn two groups of lines from feedback, under a rule-based or an information-integration structure, scored as accuracy in the last block.
8-12 minUntimed
Free to start: 10 free participants per account, then $0.99 per participant for academic research, or any yearly plan. Preview any test free after you sign up.
- Papers, 2021 to 2025
- 123
- Evidence grade
- D
- Registered clinical studies
- 1
What it measures
Measures how well people learn two groups of lines from feedback. Lines differ in length and angle; participants sort each into group A or B and see if they were right. In the standard information-integration structure, length and angle must be combined; in the rule-based structure, length alone decides. The main score is accuracy in the last block of 50.
Constructs
- Category learning
- Learning from feedback
- Rule-based and information-integration learning
- Strategy use
Research uses
- Research on explicit and procedural learning systems
- Aging and Parkinson's disease research
- Within-study comparison of rule-based and information-integration learning
How the paradigm works
Category learning tasks measure how people learn to sort things into groups from feedback. Ashby and Maddox contrasted two kinds of category structure (Ashby, Alfonso-Reese, Turken and Waldron, 1998; Ashby and Maddox, 2005). In a rule-based structure, a rule that is easy to put into words sorts the items, such as "long lines are in one group". In an information-integration structure, the best boundary combines two features in a way that is hard to put into words, and the evidence suggests that learning it relies on gradual, feedback-driven learning in the striatum. Studies of patients, aging and brain imaging have used the pair to separate these two kinds of learning (Ashby and Maddox, 2011).
More about the paradigm
Each trial shows one line in a pale square. Lines differ in length, from 60 to 260 px, and in angle, from 10 to 82 degrees above horizontal. The participant sorts each line into group A or group B by clicking or tapping, or with the F and J keys, with no time limit, then sees Correct or Wrong and the line's group for 1 second. The two groups are overlapping normal distributions of lines, so an ideal observer is right on 96% of the lines in a block. In the standard information-integration structure the boundary is diagonal, so length and angle must be combined; in the rule-based structure the same groups are turned so that length alone decides, and the two structures are equally hard for an ideal observer. A session has 4 blocks of 50 lines (3 to 6 in the settings) with a rest and the block's score between blocks, after 6 practice trials with a circle and a square. The lines and their order, and which side of the boundary is group A, come from the session's seed. The lines and the group values were made for this task.
The main score is accuracy in the last block: the share of lines sorted into their group. The task also reports accuracy in each block and overall, the gain from the first block to the last, and which strategy model fits the last block best: a one-dimensional rule on length or on angle, a boundary that combines both, or guessing, compared by the Bayesian information criterion as in Maddox and colleagues (2010). Rule strategies are common early in training on the information-integration structure (Maddox and colleagues, 2010), and in some studies most participants never leave them (Roark and Chandrasekaran, 2023), so the share of participants best fit by a rule model is a useful group result. Scores support research and do not diagnose.
Scores
- Last Block Accuracy%
- Share of lines in the last block sorted into their group.
- Higher is better; 0.5 is chance and 0.96 the ideal observer
- Overall Accuracy%
- Share of all test lines sorted into their group.
- Higher is better; early answers are guesses by design
- Block 1 Accuracy%
- Share of lines in block 1 sorted into their group.
- Higher is better
- Block 2 Accuracy%
- Share of lines in block 2 sorted into their group.
- Higher is better
- Block 3 Accuracy%
- Share of lines in block 3 sorted into their group.
- Higher is better
- Block 4 Accuracy%
- Share of lines in block 4 sorted into their group. Blank when the session had 3 blocks.
- Higher is better
- Block 5 Accuracy%
- Share of lines in block 5 sorted into their group. Blank when the session had fewer blocks.
- Higher is better
- Block 6 Accuracy%
- Share of lines in block 6 sorted into their group. Blank when the session had fewer blocks.
- Higher is better
- Learning Gain%
- Last block accuracy minus block 1 accuracy.
- Positive values mean accuracy rose during the session
- Rule Model Fits Best0/1
- 1 when a one-dimensional rule, on length or on angle, fits the answers of the last block best by the Bayesian information criterion. Blank with fewer than 20 answers in that block.
- The share of 1s is the share of participants best fit by a rule
- Integration Model Fits Best0/1
- 1 when a boundary that combines length and angle fits the answers of the last block best.
- Expected in the information-integration structure once it is learned
- Guessing Fits Best0/1
- 1 when guessing, with or without a bias toward one group, fits the answers of the last block best.
- 1 suggests the groups were not learned
- Information-Integration Structure0/1
- 1 when the session ran the information-integration structure, 0 for the rule-based structure.
- Informational
- Median Answer Timems
- Median time from a line's onset to the answer.
- Informational; the task has no time limit
- Anticipationscount
- Presses within 100 ms of a line's onset. They are not taken as answers.
- Quality check
- Lines Answeredcount
- Test lines answered in the scored run.
- Fewer than planned is flagged for review
- Planned Linescount
- Test lines in the session: 50 per block.
- Informational
Use and evidence
123 papers from 2021 to 2025 name this task in their title or abstract, up 7% compared with 2016 to 2020. That ranks it 167 of 395 test families in our census. See the counts for every test.
Show the numbers
| Year | Papers |
|---|---|
| 2025 | 32 |
| 2024 | 37 |
| 2023 | 18 |
| 2022 | 17 |
| 2021 | 19 |
| 2020 | 19 |
| 2019 | 18 |
| 2018 | 24 |
| 2017 | 27 |
| 2016 | 27 |
| 2015 | 22 |
| 2014 | 34 |
| 2013 | 27 |
| 2012 | 19 |
| 2011 | 23 |
| 2010 | 30 |
| 2009 | 23 |
| 2008 | 17 |
| 2007 | 18 |
| 2006 | 16 |
| 2005 | 17 |
| 2004 | 10 |
| 2003 | 8 |
| 2002 | 2 |
| 2001 | 6 |
| 2000 | 2 |
Use grade C
100 to 299 papers from 2021 to 2025.
Evidence grade D
Little published measurement evidence found.
- Papers on test-retest reliability
- 0
- Papers on reliability or validity
- 6
- Registered clinical studies using it as an outcome
- 1
How the grades work
Use: A is 1,000 or more papers from 2021 to 2025, B 300 to 999, C 100 to 299, D 30 to 99 and E fewer than 30. Evidence: A means at least 25 papers report its test-retest reliability and at least 25 registered clinical studies use it as an outcome; B, at least 8 test-retest papers, and either 5 registered clinical studies or 150 reliability or validity papers; C, at least 2 test-retest papers, 30 reliability or validity papers or 3 registered clinical studies; D, little published measurement evidence found. The counts measure use. They include every paper that names the task in its title or abstract. Grades describe the published research on this task, not ConductCognition's version of it.
Source: ConductCognition test census 0.14.0, counted 31 October 2026 from OpenAlex and ClinicalTrials.gov.
Task design
The trial sequence as the test runs in ConductCognition.
- Lines
- One dark line in a pale 300 px square, 60 to 260 px long and 10 to 82 degrees above horizontal, drawn by code for this task. On a small screen the square scales down as a whole.
- Groups
- Two overlapping normal distributions of lines on a 0 to 100 scale, where 1 unit is 2 px of length or 0.72 degrees of angle: each group's mean lies 13 units from the boundary, with a standard deviation of 8 across it and 16 along it. An ideal observer is right on 96% of the lines in a block.
- Structures
- Information-integration (standard): a diagonal boundary, so length and angle must be combined. Rule-based: the same groups turned 45 degrees, so length alone decides.
- Trial
- The line stays until the participant chooses A or B by click, tap, or the F and J keys, with no time limit. Then Correct or Wrong and the line's group for 1,000 ms, with the line still shown, and a 500 ms blank.
- Blocks
- 4 (standard; 3 to 6) of 50 lines, 25 from each group, in a seeded order, with a rest and the block's score between blocks.
- Practice
- 6 trials sorting a circle (group A) and a square (group B), with the same feedback.
- Counterbalancing
- Which side of the boundary is group A is drawn from the session's seed and saved on every trial.
- Strategy models
- On the last block: guessing, a one-dimensional rule on length or on angle, and a boundary that combines both, fitted by maximum likelihood and compared by the Bayesian information criterion.
The sources ran several hundred trials on lines of their own values, often with four groups. This test uses its own lines and two groups in 200 trials, so its accuracies are not comparable with theirs. Rule strategies are common early in training on the information-integration structure, and the strategy models show who uses them. The scores support research and do not diagnose.
Quality checks
Each session is marked valid, needs review or invalid, with the reason, before you see its scores.
Scored rows: category_learning_test from the scored run, a choice of group A or B with no time limit (chance 50%). Accuracy below chance invalidates and accuracy not above chance is flagged. One group on 90% of answers invalidates when accuracy is not above chance; each block holds 25 lines of each group. Answers under 200 ms on at least half the lines invalidate. A run short of its planned lines is flagged for review and withholds the percentile.
Why these rules
Accuracy significantly below chance (one-sided exact test, p < .05) cannot come from guessing; it means the response mapping was reversed or the task was not followed (Slick, Sherman and Iverson 1999 treat below-chance forced-choice performance as definite response bias). Chance-level accuracy can be a real result in impaired participants, so it is only flagged. One option on at least 90% of answered trials, when the correct options are balanced and accuracy is not above chance, means the answers did not depend on the items (the task counterpart of long-string responding in surveys; Curran 2016). An answer under 200 ms is faster than a simple visual reaction, so it cannot reflect the item. Following response-time effort screening for untimed tests (Wise and Kong 2005), the session is invalid only when at least half the answers are that fast. Early answers are guesses by design, so a learner can stay near chance for a while; only accuracy over the whole run is judged.
Running this test
Settings you can change
| Setting | Standard | Allowed |
|---|---|---|
| Category structure | Information-integration (standard) | Information-integration (standard), Rule-based |
| Blocks | 4 | 3 to 6 |
Yearly plans can change settings within these limits. Free and pay-as-you-go studies use the standard settings.
Need a setting this test does not have?
Devices and languages
- Untimed test: runs on desktops, laptops, tablets and phones. Participant devices
- Languages: English, Spanish, French, German, Italian, Portuguese (Brazil), Chinese (Simplified) and Japanese.
- Available on: Every plan, including the free plan.
- Recruiting online? See the setup guides for Prolific, Sona Systems, CloudResearch Connect and Qualtrics.
Cite and describe
Participants completed the Category Learning Task in a web browser (ConductCognition task version 1.0.0, scoring version 1.0.0; Ashby, Alfonso-Reese, Turken & Waldron, 1998; Ashby & Maddox, 2005), with the standard settings (category structure: Information-integration; blocks: 4). The main score was last block accuracy (share of lines in the last block sorted into their group).
ConductCognition versions: task version 1.0.0, scoring version 1.0.0. Every export records the versions each session used, for a methods section.
- Ashby, F. G., Alfonso-Reese, L. A., Turken, A. U., & Waldron, E. M. (1998). A neuropsychological theory of multiple systems in category learning. Psychological Review, 105(3), 442-481. https://doi.org/10.1037/0033-295X.105.3.442
- Ashby, F. G., & Maddox, W. T. (2005). Human category learning. Annual Review of Psychology, 56, 149-178. https://doi.org/10.1146/annurev.psych.56.091103.070217
- Ashby, F. G., & Maddox, W. T. (2011). Human category learning 2.0. Annals of the New York Academy of Sciences, 1224, 147-161. https://doi.org/10.1111/j.1749-6632.2010.05874.xThe definitions of rule-based and information-integration tasks.
- Maddox, W. T., Pacheco, J., Reeves, M., Zhu, B., & Schnyer, D. M. (2010). Rule-based and information-integration category learning in normal aging. Neuropsychologia, 48(10), 2998-3008. https://doi.org/10.1016/j.neuropsychologia.2010.06.008Lines that vary in length and orientation, structures related by a 45 degree turn, feedback for 1 second, and strategy models fitted to the last block by the Bayesian information criterion.
- Edmunds, C. E. R., Carpenter, K., Wills, A. J., & Milton, F. (2025). The effect of deferring feedback on rule-based and information-integration category learning. PLOS ONE, 20(4), e0313726. https://doi.org/10.1371/journal.pone.0313726Two groups of lines sampled from normal distributions, answered with two keys.
- Roark, C. L., & Chandrasekaran, B. (2023). Stable, flexible, common, and distinct behaviors support rule-based and information-integration category learning. npj Science of Learning, 8, 14. https://doi.org/10.1038/s41539-023-00163-0In learning categories of sounds, most participants used rules even in information-integration tasks.
Source papers
- Ashby FG, Maddox WT (2005). Human category learning. Annual Review of Psychology, 56:149-178.Paradigm source, Adults
- Maddox WT, Pacheco J, Reeves M, Zhu B, Schnyer DM (2010). Rule-based and information-integration category learning in normal aging. Neuropsychologia, 48(10):2998-3008.N 59, Younger and older adults
- No normative data for this task.Paradigm source, N/A
Questions
How long does the Category Learning Task take?
About 8 to 12 minutes, as listed in the test library.
What device does a participant need?
Untimed test: runs on desktops, laptops, tablets and phones. The browser must be Chrome, Edge, Firefox or Safari.
What scores does it produce?
Last Block Accuracy, Overall Accuracy, Block 1 Accuracy, Block 2 Accuracy, Block 3 Accuracy, Block 4 Accuracy, Block 5 Accuracy, Block 6 Accuracy, Learning Gain, Rule Model Fits Best, Integration Model Fits Best, Guessing Fits Best, Information-Integration Structure, Median Answer Time, Anticipations, Lines Answered and Planned Lines. Each session is also marked valid, needs review or invalid, with the reason.
How widely is the Category Learning Task used in research?
123 papers from 2021 to 2025 name it in their title or abstract, up 7% compared with 2016 to 2020. That ranks it 167 of 395 test families in the ConductCognition test census (use grade C). Source: ConductCognition test census 0.14.0, counted 31 October 2026 from OpenAlex and ClinicalTrials.gov.
How much reliability evidence does the Category Learning Task have?
In the published research, 0 papers report its test-retest reliability and 1 registered clinical study uses it as an outcome (evidence grade D in the ConductCognition test census).
Which settings can a researcher change?
Category structure (Information-integration (standard), Rule-based) and blocks (3 to 6). Settings can be changed on yearly plans; free and pay-as-you-go studies use the standard settings.
Which languages can participants use?
Participants can take this test in English, Spanish, French, German, Italian, Portuguese (Brazil), Chinese (Simplified) and Japanese. English is on every plan, and the other languages come with yearly plans. Have a native speaker read the machine-translated text before launch.
Is this a clinical test?
No. ConductCognition is for research use only. It runs, scores and exports the test; clinical interpretation is outside the platform.
Every test: the full test list.
Run this test in your next study
Start free with 10 free participants on any test, then use participant credits or a yearly plan.