Executive Function
Fermi Estimates Task
Also called: Fermi problems, Fermi estimation, Fermi questions, order of magnitude estimation, back of the envelope estimates
Measure rough estimation: participants estimate 9 quantities nobody knows offhand, such as the number of trees on Earth, and can type their steps, scored as the median distance from sourced answers in powers of ten.
8-12 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
- 64
- Evidence grade
- D
- Registered clinical studies
- 0
About 10,000 trees in each km²
What it measures
Measures rough estimation of quantities nobody knows offhand and whether people work them out in steps. For each of 9 Fermi problems, written for this test in each of the eight languages, such as how many trees there are on Earth, participants type an estimate and, as an option, the steps they used. The main score is the median distance from the reference answers in powers of ten.
Constructs
- Quantitative estimation
- Problem decomposition
- Planning
- Numeracy
- Mathematical modelling
Research uses
- Judgment and decision-making research
- Numeracy and estimation research
- Mathematics education research
- Within-study comparison of estimates with and without steps
How the paradigm works
The Fermi Estimates Task measures rough estimation of quantities that nobody knows offhand, the Fermi problems of mathematics education and judgment research, and whether people work them out in steps (Albarracín and Gorgorió, 2014). Each of 9 problems, written for this task in each language, asks for a quantity such as how many trees there are on Earth or how many seconds sunlight takes to reach the Earth, with reference answers from 16 to 3.04 trillion, each stated by a named public source. The participant types an estimate as a whole number and, in a second box, the steps they used, one per line; the study can make the steps required or leave the box out. There is no time limit and no feedback, and the order comes from the session's seed. One practice problem comes first and shows its answer.
More about the paradigm
Each estimate's error is the base-10 log of the estimate over the reference answer, so an estimate ten times too high scores 1 and one ten times too low scores -1, whatever the size of the quantity. The main score is the median absolute error, in powers of ten. Beside it come the shares of estimates within a factor of 2 and of 10, the median signed error (whether estimates ran high or low), the share of problems with typed steps and with a stepwise plan of two steps or more, and the median error with and without such a plan. The steps are saved as typed, so a study can code the strategies Albarracín and Gorgorió described, such as a reference point, a grid or a proportion.
Scores
- Median Log Errorscore
- Median of the absolute base-10 log of the estimate over the reference answer.
- 0 is exact and 1 is ten times off in either direction; lower is closer
- Within a Factor of 2%
- Share of estimates from half to twice the reference answer.
- Higher is closer
- Within a Factor of 10%
- Share of estimates from a tenth to ten times the reference answer.
- Higher is closer
- Median Log Ratioscore
- Median of the signed base-10 log of the estimate over the reference answer.
- Above 0, estimates ran high; below 0, low
- Overestimates%
- Share of estimates above the reference answer.
- Informational
- Problems With Steps%
- Share of problems with any typed steps; empty when the study shows no steps box.
- Informational
- Stepwise Plans%
- Share of problems with a stepwise plan: two or more typed steps, one per line.
- Higher means more problems were broken into steps
- Median Stepscount
- Median number of typed steps per problem; empty when the study shows no steps box.
- Informational
- Log Error With a Planscore
- Median absolute log error on problems with a stepwise plan.
- Lower is closer; empty with no such problem
- Log Error Without a Planscore
- Median absolute log error on problems without a stepwise plan.
- Lower is closer; empty with no such problem
- Median Answer Timems
- Median time from a problem appearing to its answer being submitted, for answers of at least 100 ms.
- Informational; the task has no time limit
- Answers Under 100 mscount
- Answers submitted less than 100 ms after their problem appeared.
- 0 expected
- Problems Answeredcount
- Problems with an estimate in the scored run.
- Fewer than half the planned problems invalidates
- Planned Problemscount
- Problems the run planned: 9.
- Informational
Use and evidence
64 papers from 2021 to 2025 name this task in their title or abstract, up 178% compared with 2016 to 2020. That ranks it 399 of 711 test families in our census. See the counts for every test.
Show the numbers
| Year | Papers |
|---|---|
| 2025 | 21 |
| 2024 | 10 |
| 2023 | 17 |
| 2022 | 9 |
| 2021 | 7 |
| 2020 | 7 |
| 2019 | 8 |
| 2018 | 4 |
| 2017 | 4 |
| 2016 | 0 |
| 2015 | 3 |
| 2014 | 5 |
| 2013 | 5 |
| 2012 | 1 |
| 2011 | 4 |
| 2010 | 1 |
| 2009 | 5 |
| 2008 | 2 |
| 2007 | 2 |
| 2006 | 1 |
| 2005 | 1 |
| 2004 | 0 |
| 2003 | 0 |
| 2002 | 1 |
| 2001 | 1 |
| 2000 | 0 |
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
- 6
- Registered clinical studies using it as an outcome
- 0
Measured on ConductCognition
Pooled across the studies that feed our platform norms, by device group. A figure needs 50 or more people.
No pooled figure yet.
Note on the counts: includes classroom studies of mathematical modelling.
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 0.58.0, counted 13 November 2026 from OpenAlex and ClinicalTrials.gov.
Task design
The trial sequence as the test runs in ConductCognition.
- Problems
- 9 Fermi problems written for this library in each language, such as how many trees there are on Earth or how many seconds sunlight takes to reach the Earth, with reference answers from 16 to 3.04 trillion. Each answer is stated by a named public source: NASA, the U.S. Geological Survey, the Smithsonian, a heart charity, an encyclopedia, a university and a peer-reviewed study.
- Answer
- A typed whole number, with the problem's unit beside the box, and a second box for the steps used, one per line. Digits may be grouped, as in 2,500.
- Order
- Drawn from the session's seed. One practice problem comes first, followed by its answer; the 9 problems give no feedback and have no time limit.
- Setting
- The steps box: optional (standard), required on every problem, or not shown.
- Scores
- The median absolute error in powers of ten, the base-10 log of the estimate over the reference answer (the main score); the shares of estimates within a factor of 2 and of 10; the median signed error and the share of overestimates; the shares of problems with typed steps and with a stepwise plan of two steps or more; and the median error with and without a plan.
The problems are this library's own. Errors depend on the problems, so compare conditions or groups within a study. The steps are saved as typed, so a study can code the strategies people use. The scores describe estimates for these problems; they support research and do not judge a person's knowledge.
Quality checks
Each session is marked valid, needs review or invalid, with the reason, before you see its scores.
Scored rows: fermi_estimates_item from the run with the most answered problems, latest on ties. Typed estimates have no chance level, and estimates far from the reference answers are part of what the task measures, so the error never invalidates. Fewer than half the planned problems answered invalidates, and any other incomplete run is flagged. Answers under 100 ms are counted as anticipations, and answers under 200 ms on at least half the problems invalidate. A hidden tab withholds the percentile.
Why these rules
There is no correct option to compare with chance, so accuracy is not judged against chance. Fermi problems ask for quantities nobody knows, so an estimate a hundred times off is a result, not a failed task. 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. A hidden tab could allow a search for the answers, so the percentile is withheld after one.
No percentile. Scores and quality flags only.
Running this test
Settings you can change
| Setting | Standard | Allowed |
|---|---|---|
| Steps box | Optional (standard) | Optional (standard), Required, Not shown |
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 Fermi Estimates Task in a web browser (ConductCognition task version 1.0.0, scoring version 1.0.0; Albarracín L, Gorgorió N, 2014), with the standard settings (steps box: Optional). The main score was median log error (median of the absolute base-10 log of the estimate over the reference answer).
ConductCognition versions: task version 1.0.0, scoring version 1.0.0. Every export records the versions each session used, for a methods section.
- Albarracín, L., & Gorgorió, N. (2014). Devising a plan to solve Fermi problems involving large numbers. Educational Studies in Mathematics, 86(1), 79-96. https://doi.org/10.1007/s10649-013-9528-9Fermi problems as open problems that call for assumptions and estimated quantities before simple calculations, and the plans people devise for them, with strategies such as a reference point, a grid, a concentration, a reduction of the problem and a proportion.
- Burns, B. (2023). Fermi problems as tools for understanding number generation and Benford's law. Proceedings of the Annual Meeting of the Cognitive Science Society, 45. Adults answered the same questions with and without sub-questions; in this first test, sub-questions did not make the estimates more accurate. The steps setting lets a study compare estimates with and without steps.
Source papers
- Albarracín L, Gorgorió N (2014). Devising a plan to solve Fermi problems involving large numbers. Educational Studies in Mathematics, 86(1):79-96.Paradigm source, Students aged 12 to 16
- No normative data for this task.Paradigm source, N/A
Questions
How long does the Fermi Estimates 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?
Median Log Error, Within a Factor of 2, Within a Factor of 10, Median Log Ratio, Overestimates, Problems With Steps, Stepwise Plans, Median Steps, Log Error With a Plan, Log Error Without a Plan, Median Answer Time, Answers Under 100 ms, Problems Answered and Planned Problems. Each session is also marked valid, needs review or invalid, with the reason.
How widely is the Fermi Estimates Task used in research?
64 papers from 2021 to 2025 name it in their title or abstract, up 178% compared with 2016 to 2020. That ranks it 399 of 711 test families in the ConductCognition test census (use grade D). Source: ConductCognition test census 0.58.0, counted 13 November 2026 from OpenAlex and ClinicalTrials.gov.
How much reliability evidence does the Fermi Estimates 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 settings can a researcher change?
Steps box (Optional (standard), Required, Not shown). 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, with its words in the same language. 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.