Decision Making
Convex Time Budget Task
Also called: CTB, convex time budgets, intertemporal allocation task, time preference allocation task
Measure time preference from how people divide tokens between a sooner and a later date at six rates of return, scored with the discount factor, present bias and curvature of Andreoni and Sprenger's model.
6-9 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
- 15
- Evidence grade
- D
- Registered clinical studies
- 0
What it measures
Measures time preference from how people divide 100 tokens between a sooner and a later date, a token being worth 20 points sooner and 20 to 36 points later, on 24 screens with the sooner date today or in 4 weeks. The dates are imagined. A model fitted to the splits gives the main score, a weekly discount factor, and an estimate of present bias.
Constructs
- Time preference
- Intertemporal choice
- Present bias
- Discounting
- Utility curvature
Research uses
- Behavioral economics of saving and spending
- Present bias in health behavior research
- Comparison with delay discounting measures
- Within-study group comparison
How the paradigm works
The convex time budget measures time preference from how people divide a budget between a sooner and a later date (Andreoni and Sprenger, 2012). On each of 24 screens the participant divides 100 tokens: a token is worth 20 points on the sooner date and 20, 21, 22, 24, 28 or 36 points on the later date, rates of return from 0% to 80%. The sooner date is today or in 4 weeks, and the later date 4 or 8 weeks after it. The four pairs of dates come in an order drawn for each session, each with its six values in increasing order. The split is set on a bar and confirmed, and any split is allowed. The dates are imagined, and the instructions say so. These values and dates were written for this test.
More about the paradigm
A check question on what a split gives comes first, with feedback.
The main score is the weekly discount factor of the model Andreoni and Sprenger fitted to each person, quasi-hyperbolic discounting with curved utility, estimated from the participant's splits: the value now of a point a week later, where 1 means no discounting. The model also gives present bias (beta, below 1 when a sooner date of today pulls points forward) and the curvature of utility. Fitted to choices made from known values, these 24 screens recover the discount factor closely and present bias only to within a few hundredths, about the size of the bias found for money, so present bias is best compared across groups, as Inukai, Shimodaira and Shiozawa (2024) also found. Simpler summaries need no model: the share of tokens saved for the later date, and how much more is kept for the sooner date when it is today. In the original study, 97 students divided real payments between dates weeks apart; the group's estimates showed an annual discount rate of 30%, curvature of 0.92 and no present bias, and 36 of the 97 never divided a budget (Andreoni and Sprenger, 2012). Across 220 estimates from 28 articles using the method, present bias over money was close to none (Imai, Rutter and Camerer, 2021). Scores support research and do not diagnose.
Scores
- Weekly Discount Factorscore
- Delta of the quasi-hyperbolic model fitted to the splits: the value now of a point one week later. 1 means no discounting. Null when the splits do not pin it down.
- Higher means less discounting
- Present Bias (beta)score
- Beta of the fitted model. Below 1, a sooner date of today pulls points forward; 1 means no present bias. One session estimates it to within a few hundredths, so it is best read across a group.
- Lower means more present bias
- Utility Curvature (alpha)score
- Alpha of the utility c^alpha in the fitted model, 0 to 0.99. Lower means splits that stay even whatever the rate; 0.99 means all-or-nothing splits.
- Informational
- Annual Discount Ratescore
- The weekly discount factor as a yearly rate: delta to the power -365/7, minus 1.
- Higher means steeper discounting
- Model Fit%
- Share of the variation in the splits that the fitted model explains.
- Informational
- Share Saved for Later%
- Share of all tokens put on the later date, a summary that needs no model.
- Higher means more saved for later
- Present Bias in Sharesscore
- Share kept for the sooner date when it is today minus the share when it is in 4 weeks, over screens with the same delay and rate, from -1 to 1.
- Positive means more kept when the sooner date is today
- Saved for Later, Sooner Date Today%
- Share of tokens put on the later date on the screens whose sooner date is today.
- Informational
- Saved for Later, Sooner Date in 4 Weeks%
- Share of tokens put on the later date on the screens whose sooner date is in 4 weeks.
- Informational
- All on One Date%
- Share of screens with every token on one date.
- Informational
- Fewer Later at a Higher Ratecount
- Steps between neighbouring rates of one pair of dates, of 20, at which fewer tokens went to the later date as the rate rose.
- Lower is better
- Check Correctflag
- 1 when the check question on what a split gives was answered correctly, 0 when not.
- 1 supports interpretation
- Decision Timems
- Median time from a screen appearing to the confirmed split.
- Informational
- Very Fast Confirmationscount
- Splits confirmed under 100 ms after the screen appeared.
- Lower is better
- Screenscount
- Scored screens answered.
- Informational
Use and evidence
15 papers from 2021 to 2025 name this task in their title or abstract, down 35% compared with 2016 to 2020. That ranks it 428 of 495 test families in our census. See the counts for every test.
Show the numbers
| Year | Papers |
|---|---|
| 2025 | 3 |
| 2024 | 3 |
| 2023 | 4 |
| 2022 | 2 |
| 2021 | 3 |
| 2020 | 6 |
| 2019 | 6 |
| 2018 | 3 |
| 2017 | 6 |
| 2016 | 2 |
| 2015 | 2 |
| 2014 | 2 |
| 2013 | 4 |
| 2012 | 2 |
| 2011 | 1 |
| 2010 | 0 |
| 2009 | 0 |
| 2008 | 0 |
| 2007 | 0 |
| 2006 | 0 |
| 2005 | 0 |
| 2004 | 0 |
| 2003 | 0 |
| 2002 | 0 |
| 2001 | 0 |
| 2000 | 0 |
Use grade E
Fewer than 30 papers from 2021 to 2025.
Evidence grade D
Little published measurement evidence found.
- Papers on test-retest reliability
- 0
- Papers on reliability or validity
- 3
- Registered clinical studies using it as an outcome
- 0
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 2 November 2026 from OpenAlex and ClinicalTrials.gov.
Task design
The trial sequence as the test runs in ConductCognition.
- Screens
- 100 tokens to divide between a sooner and a later date. A token is worth 20 points on the sooner date and 20, 21, 22, 24, 28 or 36 points on the later date: rates of return of 0%, 5%, 10%, 20%, 40% and 80%.
- Dates
- The sooner date is today or in 4 weeks and the later date 4 or 8 weeks after it: 4 pairs of dates, each with its 6 values in increasing order, the pairs in an order drawn from the session's seed. 24 screens.
- Split
- Set on a bar by drag, tap or the arrow keys, or one token at a time with two buttons, then confirmed. Boxes show the points on each date, and nothing is chosen until the bar is touched. No time limit.
- Imagined dates
- The instructions say the dates are imagined: no points arrive on them.
- Practice
- One check question with feedback on what 50 tokens on each date give.
- Scores
- The quasi-hyperbolic model with curved utility fitted to the splits by least squares, as Andreoni and Sprenger estimated each person: the weekly discount factor first, then present bias and curvature. Also the share of tokens saved for later and how much more is kept for the sooner date when it is today.
- Checks
- Confirmations under 200 ms on half the screens invalidate. The same split on every screen, fewer tokens later at a higher rate on 4 or more steps and a wrong check answer are flagged and never exclude.
Andreoni and Sprenger (2012) paid real money on dates up to 19 weeks away and offered 45 budgets. This test uses its own points, 24 budgets and imagined dates, so its estimates are not on their scale. One session pins down the discount factor well but present bias only to within a few hundredths, so compare groups. 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: convex_time_budget_choice from the scored run, one per screen. A split has no correct answer, so accuracy is not judged against chance, and the same split throughout is accepted. Confirmations under 200 ms on at least half the screens invalidate. The same split on every screen, a run short of its screens, fewer tokens later as the rate rises on at least 4 steps and a wrong check answer are flagged.
Why these rules
There is no correct option to compare with chance, so accuracy is not judged against chance. Putting every token on one date is a common, genuine choice (Andreoni and Sprenger, 2012, found 36 of 97 people who never divided the tokens), so the same split on every screen is flagged and never invalidates; it leaves nothing to fit. Fewer tokens later at a higher rate goes against the law of demand that Andreoni and Sprenger tested, so several such steps suggest the screens were not read, and are flagged for review. A participant who misreads what a split gives may divide the tokens the opposite way, so a wrong check answer is flagged and never invalidates. 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
This test runs with fixed settings.
Need a different version of this test?
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 Convex Time Budget Task in a web browser (ConductCognition task version 1.0.0, scoring version 1.0.0; Andreoni and Sprenger, 2012). The main score was weekly discount factor (delta of the quasi-hyperbolic model fitted to the splits: the value now of a point one week later. 1 means no discounting).
ConductCognition versions: task version 1.0.0, scoring version 1.0.0. Every export records the versions each session used, for a methods section.
- Andreoni, J., & Sprenger, C. (2012). Estimating time preferences from convex budgets. American Economic Review, 102(7), 3333-3356. https://doi.org/10.1257/aer.102.7.3333The method, and the model fitted to each person.
- Imai, T., Rutter, T. A., & Camerer, C. F. (2021). Meta-analysis of present-bias estimation using convex time budgets. The Economic Journal, 131(636), 1788-1814. https://doi.org/10.1093/ej/ueaa115Present bias across 220 estimates from 28 articles using the method.
- Inukai, K., Shimodaira, Y., & Shiozawa, K. (2024). Investigation of the convex time budget experiment by parameter recovery simulation. Journal of Behavioral and Experimental Finance, 43, 100962. https://doi.org/10.1016/j.jbef.2024.100962Simulations finding present bias estimated imprecisely in the range reported.
Source papers
- Andreoni and Sprenger (2012): 97 students and real payments; an annual discount rate of 30%, curvature of 0.92 and no present bias for the group. Imai, Rutter and Camerer (2021): 220 present-bias estimates from 28 articles.Paradigm source, Adults
Questions
How long does the Convex Time Budget Task take?
About 6 to 9 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?
Weekly Discount Factor, Present Bias (beta), Utility Curvature (alpha), Annual Discount Rate, Model Fit, Share Saved for Later, Present Bias in Shares, Saved for Later, Sooner Date Today, Saved for Later, Sooner Date in 4 Weeks, All on One Date, Fewer Later at a Higher Rate, Check Correct, Decision Time, Very Fast Confirmations and Screens. Each session is also marked valid, needs review or invalid, with the reason.
How widely is the Convex Time Budget Task used in research?
15 papers from 2021 to 2025 name it in their title or abstract, down 35% compared with 2016 to 2020. That ranks it 428 of 495 test families in the ConductCognition test census (use grade E). Source: ConductCognition test census 0.22.0, counted 2 November 2026 from OpenAlex and ClinicalTrials.gov.
How much reliability evidence does the Convex Time Budget Task have?
In the published research, 0 papers report its test-retest reliability and 0 registered clinical studies use it as an outcome (evidence grade D in the ConductCognition test census).
Does it report a percentile?
No. Scores and quality flags only.
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.