Reasoning
Anchoring Task
Also called: anchoring effect, anchoring and adjustment, anchoring bias, anchoring index, numeric anchoring
Measure how far estimates move toward a number considered first: participants say whether a picture had more or fewer dots than a high or a low number, then estimate the count, scored as the anchoring index.
4-6 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
- 586
- Evidence grade
- C
- Registered clinical studies
- 1
What it measures
Measures how far estimates move toward a number considered just before. Pictures of dots are shown for 2 seconds; the participant says whether there were more or fewer than a number, then estimates how many there were. Every number of dots is shown twice, with a low and a high number. The main score is the anchoring index: the gap between the two estimates over the gap between the two numbers.
Constructs
- Anchoring
- Numerical estimation
- Judgment under uncertainty
- Numerosity perception
Research uses
- Judgment and decision making research
- Cognitive bias research
- Numerical estimation research
- Within-study group comparison
How the paradigm works
The Anchoring Task measures how far numerical estimates move toward a number considered just before. In the paradigm of Tversky and Kahneman (1974), people first judge whether a quantity is higher or lower than a number, the anchor, and then estimate the quantity. Jacowitz and Kahneman (1995) gave each question a high and a low anchor and measured the effect as the difference between the estimates after the two anchors over the difference between the anchors, their anchoring index: 0 means the anchors did not move the estimates, and 1 means the estimates moved as far apart as the anchors.
More about the paradigm
Here the quantity is the number of dots in a picture shown for 2 seconds, too briefly to count, so the true number is known by construction and needs no general knowledge, units or language. Welsh (2021) ran the paradigm on pictures of circles in the same way, with each count shown twice, once with an anchor at half the true number and once at one and a half times it. A session has 10 pairs (5 to 15 as a setting) of about 40 to 120 dots, after 2 practice pictures that show the true number. For each picture the participant chooses "More than" or "Fewer than" the number, then types an estimate. The two pictures of a pair fall in different halves of the session, never one after the other, and the high anchor comes first in half the pairs. No accuracy is shown during the task.
The main score is the anchoring index averaged over pairs: the high-anchor estimate minus the low-anchor estimate, over the difference between the anchors, which with these anchors is the true number. With circles that stayed on screen, Welsh (2021) found a mean of 0.18 in 301 adults. The median signed errors after high and after low anchors show where the estimates sat, and the shares of comparisons on the right side and of estimates on the side the comparison chose show how the task was followed. An estimate under a quarter or over 4 times the number of dots is taken for a slip and left out with its pair. Scores support research and do not diagnose.
Scores
- Anchoring Indexscore
- Mean over pairs of the high-anchor estimate minus the low-anchor estimate, over the high anchor minus the low anchor.
- 0: the numbers did not move the estimates; 1: the estimates moved as far apart as the numbers
- Error After High Numbers%
- Median signed distance of the estimates from the number of dots after a high number, as a percent of it.
- Above 0 is an overestimate
- Error After Low Numbers%
- Median signed distance of the estimates from the number of dots after a low number, as a percent of it.
- Below 0 is an underestimate
- Estimate Error%
- Median distance of the estimates from the number of dots, as a percent of it.
- Lower is closer
- Comparisons Right%
- Share of comparisons on the right side of the number.
- Informational
- Estimates Matching the Comparison%
- Share of estimates on the side of the number that the comparison chose.
- Under half is flagged
- Median Estimate Timems
- Median time from the estimate screen appearing to the estimate.
- Informational
- Anticipationscount
- Pictures with a comparison or an estimate submitted under 100 ms after its screen appeared, left out of the scores.
- 0 expected
- Slipscount
- Pictures whose estimate was under a quarter or over four times the number of dots; they are left out with their pair.
- More than a quarter of the pictures is flagged
- Pairs Scoredcount
- Pairs whose two estimates enter the anchoring index.
- Informational
- Pictures Answeredtrials
- Pictures with an estimate recorded, out of 20 as standard.
- Informational
Use and evidence
586 papers from 2021 to 2025 name this task in their title or abstract, up 41% compared with 2016 to 2020. That ranks it 41 of 495 test families in our census. See the counts for every test.
Show the numbers
| Year | Papers |
|---|---|
| 2025 | 172 |
| 2024 | 93 |
| 2023 | 133 |
| 2022 | 86 |
| 2021 | 102 |
| 2020 | 89 |
| 2019 | 90 |
| 2018 | 78 |
| 2017 | 91 |
| 2016 | 69 |
| 2015 | 66 |
| 2014 | 65 |
| 2013 | 66 |
| 2012 | 48 |
| 2011 | 41 |
| 2010 | 41 |
| 2009 | 39 |
| 2008 | 25 |
| 2007 | 22 |
| 2006 | 33 |
| 2005 | 26 |
| 2004 | 15 |
| 2003 | 11 |
| 2002 | 16 |
| 2001 | 13 |
| 2000 | 22 |
Use grade B
300 to 999 papers from 2021 to 2025.
Evidence grade C
At least 2 test-retest papers, 30 reliability or validity papers or 3 registered clinical studies.
- Papers on test-retest reliability
- 1
- Papers on reliability or validity
- 63
- Registered clinical studies using it as an outcome
- 1
Note on the counts: many papers discuss the bias in medicine, law or finance and run no task.
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.
- Each picture
- Dots drawn by our code show for 2 seconds, too briefly to count, in a circle 300 pixels across. Then the participant chooses "More than" or "Fewer than" a number, the anchor, and types an estimate, a whole number from 1 to 999. There is no time limit for answering.
- Pairs
- 10 pairs (standard; 5 to 15), from about 40 to about 120 dots, evenly spread on a ratio scale. Each number of dots is shown twice, as two pictures with their own dots: once with an anchor at 50% of the true number and once at 150%.
- Order
- One picture of each pair in each half of the session, never one right after the other, in orders drawn from the session's seed. The high anchor comes first in half the pairs.
- Practice
- 2 pictures (32 dots with a high anchor and 128 with a low one), each followed by the true number. No accuracy is shown during the task.
- Scores
- The anchoring index: the high-anchor estimate minus the low-anchor estimate, over the high anchor minus the low anchor, averaged over pairs. Also the median errors after high and after low anchors, the overall error, the comparisons on the right side and the estimates on the side the comparison chose.
- Checks
- An estimate under a quarter or over 4 times the number of dots is a slip and is left out with its pair. Fewer than half the pictures answered is invalid. Slips on more than a quarter of the pictures, no anchoring index and estimates against the comparison on more than half the pictures are flagged and never exclude.
Classic anchoring questions ask about facts, such as the length of a river. Here the quantity is on the screen, so no answer rests on general knowledge, units or culture, as in Welsh (2021), whose pictures of 150 to 350 circles stayed on screen; with fewer dots shown for 2 seconds, the index may differ from his. Person-level anchoring scores have proved unreliable in most tasks (Röseler and colleagues, 2024), so compare groups or conditions. These are research measures of estimates: no result calls anyone biased, and they do not diagnose.
Quality checks
Each session is marked valid, needs review or invalid, with the reason, before you see its scores.
Scored rows: the scored run's pictures, each a comparison with a number and an estimate. Each comparison has a right side, set by the true number of dots, with chance at 50%; comparisons below chance or not above it are flagged for review and never invalidate. The same comparison on at least 90% of pictures invalidates when the comparisons are not above chance. Fewer than half the planned pictures answered is invalid, and any other incomplete run is flagged. An estimate under a quarter or over 4 times the number of dots is a slip, left out with its pair; slips on more than a quarter of the pictures, no anchoring index, and estimates on the other side of the number from the comparison on more than half the pictures are flagged for review and never exclude. Comparisons and estimates under 100 ms are counted as anticipations and left out, and comparisons under 200 ms on at least half the pictures invalidate.
Why these rules
The comparison is how the paradigm makes a person consider the number, and the score rests on the estimates, so wrong comparisons flag and never exclude; a person who gives the same comparison throughout did not compare at all. 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). A first estimate far outside the picture's range is most often a typing slip, as in the advice taking task, so it is left out; many slips can still be a real way of estimating, so they flag and never exclude. No anchoring index is correct, so the index itself is never judged. 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.
No percentile. Scores and quality flags only.
Running this test
Settings you can change
| Setting | Standard | Allowed |
|---|---|---|
| Picture pairs | 10 | 5 to 15 |
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.
- No percentile. Scores and quality flags only. How norms work
- Recruiting online? See the setup guides for Prolific, Sona Systems, CloudResearch Connect and Qualtrics.
Cite and describe
Participants completed the Anchoring Task in a web browser (ConductCognition task version 1.0.0, scoring version 1.0.0; Tversky A, Kahneman D, 1974), with the standard settings (picture pairs: 10). The main score was anchoring index (mean over pairs of the high-anchor estimate minus the low-anchor estimate, over the high anchor minus the low anchor).
ConductCognition versions: task version 1.0.0, scoring version 1.0.0. Every export records the versions each session used, for a methods section.
- Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124-1131. https://doi.org/10.1126/science.185.4157.1124
- Jacowitz, K. E., & Kahneman, D. (1995). Measures of anchoring in estimation tasks. Personality and Social Psychology Bulletin, 21(11), 1161-1166. https://doi.org/10.1177/01461672952111004A high and a low anchor for each quantity, and the anchoring index.
- Welsh, M. B. (2021). Does anything predict anchoring bias? In Proceedings of the 43rd Annual Meeting of the Cognitive Science Society (pp. 3089-3095). https://escholarship.org/uc/item/8c19495c The paradigm on pictures of circles, each count shown with anchors at 50% and 150% of it.
- Lee, C.-Y., & Morewedge, C. K. (2022). Noise increases anchoring effects. Psychological Science, 33(1), 60-75. https://doi.org/10.1177/09567976211024254Anchoring of dot counts, larger for larger arrays.
- Röseler, L., Weber, L., Helgerth, K. A. C., Stich, E., Günther, M., Tegethoff, P., Wagner, F. S., & Schütz, A. (2024). Measurements of susceptibility to anchoring are unreliable. Meta-Psychology, 8. https://doi.org/10.15626/MP.2022.3236Why the index is compared across groups.
Source papers
- Welsh (2021): on pictures of 150 to 350 circles with anchors at 50% and 150% of the true number, the mean anchoring score of 301 adults was 0.18. Röseler and colleagues (2024): across a large set of anchoring studies, the reliability of person-level anchoring scores was very low in most tasks.Paradigm source, Adults
Questions
How long does the Anchoring Task take?
About 4 to 6 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?
Anchoring Index, Error After High Numbers, Error After Low Numbers, Estimate Error, Comparisons Right, Estimates Matching the Comparison, Median Estimate Time, Anticipations, Slips, Pairs Scored and Pictures Answered. Each session is also marked valid, needs review or invalid, with the reason.
How widely is the Anchoring Task used in research?
586 papers from 2021 to 2025 name it in their title or abstract, up 41% compared with 2016 to 2020. That ranks it 41 of 495 test families in the ConductCognition test census (use grade B). Source: ConductCognition test census 0.22.0, counted 3 November 2026 from OpenAlex and ClinicalTrials.gov.
How much reliability evidence does the Anchoring Task have?
In the published research, 1 paper reports its test-retest reliability and 1 registered clinical study uses it as an outcome (evidence grade C in the ConductCognition test census).
Does it report a percentile?
No. Scores and quality flags only.
Which settings can a researcher change?
Picture pairs (5 to 15). 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.