Decision Making
Two-Step Task
Also called: two-stage task, two-stage decision task, Daw two-step task, two-stage Markov decision task, model-based learning task
Measure model-based and model-free learning: participants choose a boat that usually reaches one of two islands, then an animal there that may find a star.
9-12 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
- 70
- Evidence grade
- C
- Registered clinical studies
- 4
A star!What it measures
Measures how much choices rely on a model of the task (model-based) or on past rewards alone (model-free). Participants choose a boat that usually sails to one of two islands, then an animal there that may find a star; the animals' chances drift slowly. The main score is the model-based index: how a star after a rare trip to the other island changes the next boat choice, over 3 blocks of 50.
Constructs
- Model-based and model-free learning
- Goal-directed and habitual control
- Reinforcement learning
- Sequential decision-making
Research uses
- Research on compulsivity, addiction and other habit-related conditions
- Development and aging of goal-directed choice
- Within-study comparison of decision strategies
How the paradigm works
The two-step task separates two ways of learning from reward (Daw, Gershman, Seymour, Dayan and Dolan, 2011). A model-free learner repeats what was rewarded. A model-based learner also uses a map of the task: it knows where each first choice usually leads and plans with it. The two make different predictions when a first choice leads somewhere unusual and a reward follows there: a model-free learner repeats that first choice, while a model-based learner switches to the other one, which usually leads to the rewarded place. Most people show some of both (Daw and colleagues, 2011), and studies of thousands of people online link less model-based choice to compulsive symptoms (Gillan, Kosinski, Whelan, Phelps and Daw, 2016).
More about the paradigm
Each trial starts with a choice between two boats. Each boat sails to its usual island on 70% of trials and to the other island on 30%, and which island each boat usually reaches never changes. On the island the participant chooses one of two animals, and the animal finds a star or not. Each animal's chance of a star drifts slowly between 25% and 75%, by steps with a standard deviation of 2.5 percentage points per trial, the values of Daw and colleagues, so learning never stops. The participant clicks or taps a picture, or presses F or J, within 3 seconds for each choice (2 to 4 seconds in the settings); a choice not made in time ends the trial. A session has 3 blocks of 50 trials (2 to 4 in the settings) with exactly 15 rare trips in each block, after 10 practice trials in which a line says whether the boat reached its usual island. The pictures are emoji from the open Noto set. The chance series come from this task's code with one fixed seed, so every participant meets the same series; the session's seed sets which island each boat usually reaches, which animal carries which series, the rare trips and the sides. Each star earns 1 point (0 to 10 in the settings; 0 counts stars without points), and the study decides what points are worth.
The main score is the model-based index, the reward-by-transition interaction on staying with the same boat: the stay probability after a star and a common trip minus that after a star and a rare trip, minus the same difference after no star. A model-based learner scores above 0 and a purely model-free learner near 0 (the range is -2 to 2). The model-free index is the effect of a star on staying, whatever the trip. The task also reports the four stay probabilities, how much slower the animal choice is after a rare trip (a sign that the boats' islands are known), how often the better animal was chosen, the share of trials with a star and the time of each choice. Scores support research and do not diagnose.
Scores
- Model-Based Indexscore
- (Stay after a star and a common trip minus after a star and a rare trip) minus (stay after no star and a common trip minus after no star and a rare trip). From -2 to 2.
- Above 0 shows model-based choice; near 0 shows model-free choice alone
- Model-Free Indexscore
- Stay after a star (common plus rare trips) minus stay after no star (common plus rare trips). From -2 to 2.
- Above 0 means a star makes repeating the boat more likely
- Stay After a Star, Common Trip%
- Share of trials after a star on a common trip on which the same boat was chosen again. Blank with fewer than 5 such trials.
- Informational
- Stay After a Star, Rare Trip%
- Share of trials after a star on a rare trip on which the same boat was chosen again.
- Lower than after a star on a common trip in model-based choice
- Stay After No Star, Common Trip%
- Share of trials after no star on a common trip on which the same boat was chosen again.
- Informational
- Stay After No Star, Rare Trip%
- Share of trials after no star on a rare trip on which the same boat was chosen again.
- Higher than after no star on a common trip in model-based choice
- Stay Probability%
- Share of trials on which the boat of the trial before was chosen again.
- Informational; high values can also mean repeating out of habit
- Slowing After a Rare Tripms
- Median time of the animal choice after a rare trip minus after a common trip.
- Above 0 suggests the participant knows where each boat usually goes
- Better Animal Chosen%
- Share of animal choices that picked the animal with the higher chance of a star at that moment.
- Above 0.5 means the animals' chances were tracked
- Stars per Trial%
- Share of completed trials that ended with a star.
- Informational; model-based and model-free choice earn about the same in this design
- Points Wonpoints
- Points for the stars found.
- Informational
- Median Boat Choice Timems
- Median time from the boats' onset to the boat choice.
- Informational
- Median Animal Choice Timems
- Median time from the island's onset to the animal choice.
- Informational
- Missed Trialscount
- Trials with a choice not made in time. They are left out of the scores.
- More than 10% is flagged; more than 20% makes the session invalid
- Anticipationscount
- Presses within 100 ms of the choices' onset. They are not taken as choices.
- Quality check
- Trials Completedcount
- Trials with both choices made.
- Informational
- Planned Trialscount
- Trials in the session: 50 per block.
- Informational
Use and evidence
70 papers from 2021 to 2025 name this task in their title or abstract, up 37% compared with 2016 to 2020. That ranks it 225 of 395 test families in our census. See the counts for every test.
Show the numbers
| Year | Papers |
|---|---|
| 2025 | 23 |
| 2024 | 22 |
| 2023 | 8 |
| 2022 | 7 |
| 2021 | 10 |
| 2020 | 17 |
| 2019 | 11 |
| 2018 | 9 |
| 2017 | 9 |
| 2016 | 5 |
| 2015 | 5 |
| 2014 | 2 |
| 2013 | 2 |
| 2012 | 3 |
| 2011 | 1 |
| 2010 | 1 |
| 2009 | 0 |
| 2008 | 0 |
| 2007 | 0 |
| 2006 | 0 |
| 2005 | 0 |
| 2004 | 0 |
| 2003 | 1 |
| 2002 | 0 |
| 2001 | 0 |
| 2000 | 0 |
Use grade D
30 to 99 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
- 0
- Papers on reliability or validity
- 17
- Registered clinical studies using it as an outcome
- 4
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.
- Pictures
- Emoji from the open Noto set: two boats (a sailboat and a ship), two islands (a beach and a volcano) and two animals on each island (a crab and a flamingo, a goat and a lizard).
- Trips
- Each boat reaches its usual island on 70% of trials and the other island on 30%: exactly 15 rare trips in each block of 50, in a seeded order. Which island each boat usually reaches is drawn from the session's seed and never changes.
- Stars
- Each animal's chance of a star drifts by Gaussian steps with a standard deviation of 0.025 per trial between reflecting bounds at 0.25 and 0.75, from a uniform start. The four series come from this task's code with one fixed seed, the same in every session; which animal carries which series is drawn from the session's seed, and each star is drawn in advance from its chance.
- Trial
- The boats until a choice by click, tap, or the F and J keys; the chosen boat for 500 ms; the island with its two animals until a choice; then the star or no star for 1,000 ms and a 500 ms blank. Each boat, and each animal on its island, is on the left on half the trials of a block.
- Time to choose
- 3 seconds (standard; 2, 2.5, 3 or 4) for each choice. A choice not made in time shows a line for 1,000 ms, ends the trial with no star and leaves it out of the scores.
- Points
- 1 point (standard; 0 to 10) for each star, with the total shown. 0 counts stars without points. The study decides what points are worth.
- Blocks
- 3 blocks (standard; 2 to 4) of 50 trials, with a rest between them.
- Practice
- 10 trials with every animal at a 50% chance; on each island a line says whether the boat reached its usual island. The first two practice trips are common and 3 of the rest are rare.
Daw and colleagues used Tibetan characters in colored boxes, a 2 second limit and 201 trials, after 50 practice trials on other symbols. This test uses its own story and chance series, a 3 second standard limit, 150 trials and a practice that names the rare trips, so its values compare within a study rather than with theirs. Model-based choice earns no more stars in this design, so the index has no better direction and the test gives no percentiles. 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: two_step_test from the scored run, two choices per trial within the time limit. No choice is correct in advance, so accuracy is not judged against chance. The same side on at least 95% of first choices invalidates. More than 20% of trials missed invalidates and more than 10% is flagged. Presses under 100 ms are not taken as choices. A run short of its planned trials, or with too few trial pairs for the indices, is flagged for review.
Why these rules
There is no correct option to compare with chance, so accuracy is not judged against chance. The boats change sides at random, so choosing one side on nearly every trial means the boats were not looked at; online studies of this task excluded the same key on more than 95% of trials and more than 10% of trials missed (Gillan and colleagues, 2016), or more than 20% missed (Donegan and colleagues, 2023). Daw and colleagues (2011) report 2 missed trials of 201 at a 2 second limit, so missing more than a fifth means the participant was not doing the task.
No percentile. Scores and quality flags only.
Running this test
Settings you can change
| Setting | Standard | Allowed |
|---|---|---|
| Blocks | 3 | 2 to 4 |
| Time to choose | 3 seconds (standard) | 2 seconds, 2.5 seconds, 3 seconds (standard), 4 seconds |
| Points per star | 1 | 0 to 10 |
| Payment line | No mention of money (standard) | No mention of money (standard), The study pays for points |
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 Two-Step Task in a web browser (ConductCognition task version 1.0.0, scoring version 1.0.0; Daw, Gershman, Seymour, Dayan & Dolan, 2011), with the standard settings (blocks: 3; time to choose: 3 seconds; points per star: 1; payment line: No mention of money). The main score was model-based index ((Stay after a star and a common trip minus after a star and a rare trip) minus (stay after no star and a common trip minus after no star and a rare trip)).
ConductCognition versions: task version 1.0.0, scoring version 1.0.0. Every export records the versions each session used, for a methods section.
- Daw, N. D., Gershman, S. J., Seymour, B., Dayan, P., & Dolan, R. J. (2011). Model-based influences on humans' choices and striatal prediction errors. Neuron, 69(6), 1204-1215. https://doi.org/10.1016/j.neuron.2011.02.027The two-stage task: 70% and 30% transitions, reward chances drifting between 0.25 and 0.75, and staying analyzed by reward and transition.
- Gillan, C. M., Kosinski, M., Whelan, R., Phelps, E. A., & Daw, N. D. (2016). Characterizing a psychiatric symptom dimension related to deficits in goal-directed control. eLife, 5, e11305. https://doi.org/10.7554/eLife.11305The task online: 200 trials, a 2.5 second limit and the exclusion rules.
- Kool, W., Cushman, F. A., & Gershman, S. J. (2016). When does model-based control pay off? PLOS Computational Biology, 12(8), e1005090. https://doi.org/10.1371/journal.pcbi.1005090In this design, model-based choice earns no more reward than model-free choice.
- Donegan, K. R., Brown, V. M., Price, R. B., Gallagher, E., Pringle, A., Hanlon, A. K., & Gillan, C. M. (2023). Using smartphones to optimise and scale-up the assessment of model-based planning. Communications Psychology, 1, 31. https://doi.org/10.1038/s44271-023-00031-yReliability of the model-based index and shorter versions of the task.
Source papers
- Daw ND, Gershman SJ, Seymour B, Dayan P, Dolan RJ (2011). Model-based influences on humans' choices and striatal prediction errors. Neuron, 69(6):1204-1215.N 17, Adults (mean age 25.8)
- Gillan CM, Kosinski M, Whelan R, Phelps EA, Daw ND (2016). Characterizing a psychiatric symptom dimension related to deficits in goal-directed control. eLife, 5:e11305.N 1413, Adults online
Questions
How long does the Two-Step Task take?
About 9 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?
Model-Based Index, Model-Free Index, Stay After a Star, Common Trip, Stay After a Star, Rare Trip, Stay After No Star, Common Trip, Stay After No Star, Rare Trip, Stay Probability, Slowing After a Rare Trip, Better Animal Chosen, Stars per Trial, Points Won, Median Boat Choice Time, Median Animal Choice Time, Missed Trials, Anticipations, Trials Completed and Planned Trials. Each session is also marked valid, needs review or invalid, with the reason.
How widely is the Two-Step Task used in research?
70 papers from 2021 to 2025 name it in their title or abstract, up 37% compared with 2016 to 2020. That ranks it 225 of 395 test families in the ConductCognition test census (use grade D). Source: ConductCognition test census 0.14.0, counted 31 October 2026 from OpenAlex and ClinicalTrials.gov.
How much reliability evidence does the Two-Step Task have?
In the published research, 0 papers report its test-retest reliability and 4 registered clinical studies use 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?
Blocks (2 to 4), time to choose (2 seconds, 2.5 seconds, 3 seconds (standard), 4 seconds), points per star (0 to 10) and payment line (No mention of money (standard), The study pays for points). 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.