Executive Function
Associative Inference Task
Also called: associative inference paradigm, AIP, paired associate inference task, overlapping pairs task, ABC inference task, relational memory inference task
Measure linking memories that share an item: participants learn overlapping pairs of pictures, then choose pictures linked only through a shared one, scored as inference accuracy.
10-13 minUntimed
Free to start: 10 free participants per account, then $0.99 per participant for academic research, or any yearly plan. Try it here first, with no account.
- Papers, 2021 to 2025
- 52
- Evidence grade
- D
- Registered clinical studies
- 1
What it measures
Measures linking memories that share an item. Participants learn overlapping pairs of pictures, A with B and B with C, then choose which A goes with each C, a pair never shown. The main score is accuracy on those inference choices, with memory for the studied pairs next.
Constructs
- Associative inference
- Relational memory
- Memory integration
- Paired-associate learning
Research uses
- Research on relational memory and the hippocampus
- Schizophrenia and psychosis research at the group level
- Development and aging research on combining memories
How the paradigm works
Associative inference is linking two memories that share an item: after learning that A goes with B and that B goes with C, people can tell that A and C belong together, though they never saw them together. Bunsey and Eichenbaum (1996) showed this in rats, which lost it after hippocampal damage, and Preston, Shrager, Dudukovic and Gabrieli (2004) built the human paradigm from overlapping pairs of faces and houses, and found more hippocampal activity on the inference trials than on the studied pairs. Zeithamova and Preston (2010) and Schlichting and colleagues (2015, 2017) used pairs of objects. The paradigm is used in research on how memories are combined, in development, and in schizophrenia, where Armstrong and colleagues (2012, 2018) found inference impaired at the group level.
More about the paradigm
This test follows the object design of Schlichting and colleagues (2017) with Noto emoji pictures of everyday objects in 12 triads. Each round shows the 24 overlapping pairs, A with B and B with C, side by side for 3.5 s each, in an order in which a triad's two pairs never follow each other, then tests every pair with feedback: a picture appears at the top, and the participant chooses the picture shown with it from three below, the answer and two equally familiar pictures from other triads, by click, tap, or the 1, 2 and 3 keys, with no time limit. After three rounds (two to four in the settings), the instructions explain with drawn shapes that two pictures can go together through a third, and the inference test shows each C above three A pictures, none of them ever shown with it, with no feedback. A final test of the 24 pairs without feedback comes after the inference test, so that it cannot remind people of the links first. Three practice pairs that share no picture come first. Which picture takes which place and every order come from the session's seed. About 10 to 13 minutes.
The main score is inference accuracy: the share of the 12 inference choices that pick the picture linked to the cue; chance is one in three. Inference accuracy over the triads whose two pairs were both answered rightly in the pair test comes next, then accuracy in the pair test, on the A with B and the B with C pairs, and in the rounds. A session that answers fewer than 70% of the pairs rightly in the pair test is flagged for review, because its inferences rest on pairs it did not keep. Scores support research and do not diagnose.
Scores
- Inference Accuracy%
- Share of the 12 inference choices, a C picture above three A pictures never shown with it, that pick the A linked to it through B.
- Higher is better; 0.33 is chance
- Inference Accuracy, Pairs Remembered%
- Inference accuracy over the triads whose A with B and B with C pairs were both answered rightly in the pair test.
- Higher is better; null when no triad qualifies
- Pair Test Accuracy%
- Share of right choices on the 24 studied pairs in the test after the inference test, with no feedback.
- Under 0.7 is flagged for review; 0.33 is chance
- A With B Accuracy%
- Share of right choices on the A with B pairs in the pair test: B above three A pictures.
- Higher is better; 0.33 is chance
- B With C Accuracy%
- Share of right choices on the B with C pairs in the pair test: C above three B pictures.
- Higher is better; 0.33 is chance
- Learning Accuracy%
- Share of right choices in the tests of the rounds, with feedback.
- Higher is better; 0.33 is chance
- Last Round Accuracy%
- Share of right choices in the test of the last round.
- Higher means the pairs were learned
- Median Inference Choice Timems
- Median time from the choices' onset to the choice in the inference test.
- Informational; the task has no time limit
- Median Pair Choice Timems
- Median time from the choices' onset to the choice in the pair test.
- Informational
- Anticipationscount
- Presses within 100 ms of the choices appearing, in the rounds and the tests. They are not taken as choices.
- Quality check
- Test Choices Madecount
- Choices in the inference and pair tests of the scored run.
- Fewer than planned is flagged for review
- Planned Test Trialscount
- Trials in the inference and pair tests: 36.
- Informational
Use and evidence
52 papers from 2021 to 2025 name this task in their title or abstract, up 18% compared with 2016 to 2020. That ranks it 301 of 495 test families in our census. See the counts for every test.
Show the numbers
| Year | Papers |
|---|---|
| 2025 | 12 |
| 2024 | 9 |
| 2023 | 12 |
| 2022 | 11 |
| 2021 | 8 |
| 2020 | 9 |
| 2019 | 7 |
| 2018 | 15 |
| 2017 | 6 |
| 2016 | 7 |
| 2015 | 3 |
| 2014 | 4 |
| 2013 | 3 |
| 2012 | 7 |
| 2011 | 7 |
| 2010 | 7 |
| 2009 | 2 |
| 2008 | 3 |
| 2007 | 5 |
| 2006 | 2 |
| 2005 | 2 |
| 2004 | 0 |
| 2003 | 0 |
| 2002 | 0 |
| 2001 | 1 |
| 2000 | 1 |
Use grade D
30 to 99 papers from 2021 to 2025.
Evidence grade D
Little published measurement evidence found.
- Papers on test-retest reliability
- 0
- Papers on reliability or validity
- 12
- 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 0.22.0, counted 3 November 2026 from OpenAlex and ClinicalTrials.gov.
Task design
The trial sequence as the test runs in ConductCognition.
- Pictures
- 36 Noto emoji pictures of everyday objects in 12 triads, A, B and C, and 6 more in 3 practice pairs that share no picture. Which picture takes which place comes from the session's seed.
- Rounds
- 3 rounds by default, 2 to 4 as a setting. Each shows the 24 overlapping pairs, A with B and B with C, side by side for 3,500 ms each with a 500 ms blank, in an order in which a triad's two pairs never follow each other, then tests each pair once with feedback.
- Each test trial
- A picture at the top and 3 below: the answer and two pictures of the same kind from other triads, all seen equally often. B cues its A, and C cues its B. Choice by click, tap, or the 1, 2 and 3 keys, with no time limit. In practice and in the rounds, Correct or Incorrect shows for 1,000 ms, then a 500 ms blank. Presses under 100 ms are not taken.
- Inference test
- After the rounds, instructions with drawn shapes explain that two pictures can go together through a third. Each C then appears above 3 A pictures, none of them ever shown with it: 12 trials with no feedback.
- Pair test
- The 24 pairs again, after the inference test, with no feedback. In every test the answer takes each of the 3 places equally often.
- Scores
- Inference accuracy, inference accuracy over the triads whose two pairs were answered rightly in the pair test, pair test accuracy, and accuracy in the rounds.
Preston and colleagues (2004) used photographs of faces and houses with two choices, and Schlichting and colleagues (2017) novel objects with three choices and no feedback; this test uses Noto emoji objects, three choices and feedback in the rounds, so its accuracies are not directly comparable with theirs. 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: associative_inference_test from the scored run, a choice among three pictures with no time limit and no feedback. The chance checks use the pair test, whose studied pairs have a learned answer (chance 33%); the inference trials are the measure. Accuracy below chance invalidates and accuracy not above chance is flagged. One place on 90% of choices invalidates when accuracy is not above chance; the answer takes each place equally often. Answers under 200 ms on at least half the choices invalidate. Fewer than 70% of the pairs answered rightly and a run short of its planned choices are flagged for review and withhold 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. An inference through a shared picture means something only once the studied pairs are known: Zeithamova and Preston (2010) left out participants whose direct pairs were not above chance, Schlichting and colleagues (2017) scored inference over the triads whose pairs were remembered, and Armstrong, Williams and Heckers (2012) trained to 70 percent.
Running this test
Settings you can change
| Setting | Standard | Allowed |
|---|---|---|
| Study rounds | 3 | 2 to 4 |
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 Associative Inference Task in a web browser (ConductCognition task version 1.0.0, scoring version 1.0.0; Bunsey & Eichenbaum, 1996; Preston, Shrager, Dudukovic & Gabrieli, 2004; Schlichting et al., 2017), with the standard settings (study rounds: 3). The main score was inference accuracy (share of the 12 inference choices, a C picture above three A pictures never shown with it, that pick the A linked to it through B).
ConductCognition versions: task version 1.0.0, scoring version 1.0.0. Every export records the versions each session used, for a methods section.
- Bunsey, M., & Eichenbaum, H. (1996). Conservation of hippocampal memory function in rats and humans. Nature, 379(6562), 255-257. https://doi.org/10.1038/379255a0Inference across learned pairs that share an item, in rats, lost after hippocampal damage.
- Preston, A. R., Shrager, Y., Dudukovic, N. M., & Gabrieli, J. D. E. (2004). Hippocampal contribution to the novel use of relational information in declarative memory. Hippocampus, 14(2), 148-152. https://doi.org/10.1002/hipo.20009The human paradigm: overlapping face-house pairs, the inference test before the pair test, and choices among familiar pictures.
- Zeithamova, D., & Preston, A. R. (2010). Flexible memories: Differential roles for medial temporal lobe and prefrontal cortex in cross-episode binding. Journal of Neuroscience, 30(44), 14676-14684. https://doi.org/10.1523/JNEUROSCI.3250-10.2010Overlapping pairs of everyday objects, with each inference tested before its pairs.
- Schlichting, M. L., Guarino, K. F., Schapiro, A. C., Turk-Browne, N. B., & Preston, A. R. (2017). Hippocampal structure predicts statistical learning and associative inference abilities during development. Journal of Cognitive Neuroscience, 29(1), 37-51. https://doi.org/10.1162/jocn_a_01028The design followed here: study and test rounds, three choices, the link explained after learning, and inference scored over remembered pairs.
- Armstrong, K., Kose, S., Williams, L., Woolard, A., & Heckers, S. (2012). Impaired associative inference in patients with schizophrenia. Schizophrenia Bulletin, 38(3), 622-629. https://doi.org/10.1093/schbul/sbq145The paradigm in schizophrenia research: patients as a group were not more accurate on inferred than on directly learned face pairs.
- Armstrong, K., Williams, L. E., & Heckers, S. (2012). Revised associative inference paradigm confirms relational memory impairment in schizophrenia. Neuropsychology, 26(4), 451-458. https://doi.org/10.1037/a0028667Smaller training blocks, feedback during training and explicit instructions for the inference trials, which cut the share of patients who could not learn the pairs.
Source papers
- Preston AR, Shrager Y, Dudukovic NM, Gabrieli JDE (2004). Hippocampal contribution to the novel use of relational information in declarative memory. Hippocampus, 14(2):148-152.N 10, 18-23 years
- No normative data for this task.Paradigm source, N/A
Questions
How long does the Associative Inference Task take?
About 10 to 13 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?
Inference Accuracy, Inference Accuracy, Pairs Remembered, Pair Test Accuracy, A With B Accuracy, B With C Accuracy, Learning Accuracy, Last Round Accuracy, Median Inference Choice Time, Median Pair Choice Time, Anticipations, Test Choices Made and Planned Test Trials. Each session is also marked valid, needs review or invalid, with the reason.
How widely is the Associative Inference Task used in research?
52 papers from 2021 to 2025 name it in their title or abstract, up 18% compared with 2016 to 2020. That ranks it 301 of 495 test families in the ConductCognition test census (use grade D). Source: ConductCognition test census 0.22.0, counted 3 November 2026 from OpenAlex and ClinicalTrials.gov.
How much reliability evidence does the Associative Inference 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?
Study rounds (2 to 4). 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.