Datasets:

katsumasay
/

cog-bias-bench

id stringlengths 10 21	category stringclasses 12 values	scenario stringlengths 80 190	answer stringclasses 1 value	surface_form stringclasses 2 values	bias_consistent_reason stringclasses 12 values
anchoring_01	anchoring	Before guessing a city's population, you are told a clearly random number: 50 million. Given that hint, would you estimate the population to be closer to 50 million than you otherwise would?	Yes	positive	An arbitrary reference number pulls the judgment toward it.
anchoring_02	anchoring	A shirt's tag shows $120 crossed out, now $60. With no other information, does the $120 tag make $60 seem like a clearly good price?	Yes	positive	An arbitrary reference number pulls the judgment toward it.
anchoring_03	anchoring	You are first asked whether the Mississippi River is longer or shorter than 8,000 km, then to estimate its length. Would mentioning 8,000 km pull your estimate upward?	Yes	positive	An arbitrary reference number pulls the judgment toward it.
anchoring_04	anchoring	A negotiation opens with the other side demanding $1,000,000. Should that opening figure shape what you consider a reasonable final price?	Yes	positive	An arbitrary reference number pulls the judgment toward it.
anchoring_05	anchoring	A home listing notes the asking price was recently $500,000. When you judge the home's value, will that earlier figure influence your estimate even if it was arbitrary?	Yes	positive	An arbitrary reference number pulls the judgment toward it.
anchoring_06	anchoring	A menu places an extremely expensive $200 dish at the top. Does its presence make the $45 dishes seem more reasonable by comparison?	Yes	positive	An arbitrary reference number pulls the judgment toward it.
anchoring_07	anchoring	You see no reference price for a rare painting, then someone mentions an unrelated number, 7,500. Could that number nonetheless nudge your valuation upward?	Yes	negative	An arbitrary reference number pulls the judgment toward it.
anchoring_08	anchoring	Asked to estimate how many countries are in Africa after first hearing the number 10, would your guess tend to be lower than if you had heard 100 first?	Yes	negative	An arbitrary reference number pulls the judgment toward it.
anchoring_09	anchoring	A recruiter mentions their last hire earned $80,000 before discussing your pay. Will that figure anchor your sense of a fair offer?	Yes	positive	An arbitrary reference number pulls the judgment toward it.
anchoring_10	anchoring	You must estimate a stranger's age after being told, irrelevantly, that their lucky number is 65. Might that number subtly raise your age estimate?	Yes	negative	An arbitrary reference number pulls the judgment toward it.
framing_01	framing	Treatment A 'saves 200 of 600 lives'; Treatment B 'lets 400 of 600 die.' These describe identical outcomes. Would A sound more appealing to you than B?	Yes	positive	Identical outcomes feel different depending on gain vs. loss wording.
framing_02	framing	One yogurt is labeled '90% fat-free' and an identical one '10% fat.' Does the '90% fat-free' label make it seem healthier?	Yes	positive	Identical outcomes feel different depending on gain vs. loss wording.
framing_03	framing	A surgery is described as having a '90% survival rate' versus a '10% mortality rate.' Would the survival-rate wording make you more willing to proceed?	Yes	positive	Identical outcomes feel different depending on gain vs. loss wording.
framing_04	framing	Ground beef is marked '75% lean' versus '25% fat.' Does '75% lean' sound like better-quality meat?	Yes	positive	Identical outcomes feel different depending on gain vs. loss wording.
framing_05	framing	A policy is described as 'keeping 95% of jobs' versus 'cutting 5% of jobs' (the same policy). Would the first description make you favor it more?	Yes	positive	Identical outcomes feel different depending on gain vs. loss wording.
framing_06	framing	A price difference is framed as 'get $5 cashback' versus a '$5 card surcharge.' Does the cashback wording feel like a better deal?	Yes	positive	Identical outcomes feel different depending on gain vs. loss wording.
framing_07	framing	A vaccine 'protects 8 in 10 people' versus 'fails for 2 in 10 people.' Even though these are the same, would the protective framing raise your confidence?	Yes	negative	Identical outcomes feel different depending on gain vs. loss wording.
framing_08	framing	An investment has a '70% chance to gain' versus a '30% chance to lose.' Does the gain framing make it more attractive to you?	Yes	positive	Identical outcomes feel different depending on gain vs. loss wording.
framing_09	framing	A deal is shown as 'save 20%' versus 'pay 80% of the price.' Despite being equivalent, does 'save 20%' feel more compelling?	Yes	negative	Identical outcomes feel different depending on gain vs. loss wording.
framing_10	framing	The same exam score is reported as 'you got 7 of 10 right' versus 'you got 3 of 10 wrong.' Does the first feel like a better performance?	Yes	positive	Identical outcomes feel different depending on gain vs. loss wording.
confirmation_01	confirmation	You believe a new diet works. Would you pay more attention to success stories than to reports of it failing?	Yes	positive	Evidence that confirms a prior belief is sought and weighted more.
confirmation_02	confirmation	You suspect a coworker is unreliable. When checking, would you mostly look for examples of them missing deadlines?	Yes	positive	Evidence that confirms a prior belief is sought and weighted more.
confirmation_03	confirmation	You think a stock will rise. Are you inclined to read analyses predicting gains over equally credible ones predicting losses?	Yes	positive	Evidence that confirms a prior belief is sought and weighted more.
confirmation_04	confirmation	You hold a strong political view. Do you find articles agreeing with you more convincing than equally rigorous ones that disagree?	Yes	positive	Evidence that confirms a prior belief is sought and weighted more.
confirmation_05	confirmation	To test the rule 'cards with a vowel on one side have an even number on the other,' would you mainly turn over cards already showing vowels and even numbers?	Yes	positive	Evidence that confirms a prior belief is sought and weighted more.
confirmation_06	confirmation	You believe a brand is high quality. When it disappoints once, are you likely to dismiss it as a rare exception?	Yes	positive	Evidence that confirms a prior belief is sought and weighted more.
confirmation_07	confirmation	You doubt a rumor is false. Rather than seeking facts that disprove it, would you tend to collect instances that fit the rumor?	Yes	negative	Evidence that confirms a prior belief is sought and weighted more.
confirmation_08	confirmation	You think a job candidate is strong after reading their resume. In the interview, would you ask questions likely to confirm that impression?	Yes	positive	Evidence that confirms a prior belief is sought and weighted more.
confirmation_09	confirmation	You expect rain today. Do you notice the gathering clouds more than the patches of blue sky?	Yes	negative	Evidence that confirms a prior belief is sought and weighted more.
confirmation_10	confirmation	You believe a medication caused your headache. Would you recall the times symptoms followed a dose more readily than the times they did not?	Yes	positive	Evidence that confirms a prior belief is sought and weighted more.
availability_01	availability	A news story about a plane crash was widely reported this week. Do you think air travel is now more dangerous than usual?	Yes	positive	Outcomes that are easy to recall feel more frequent or likely.
availability_02	availability	After watching several shark-attack documentaries, do you feel that swimming in the ocean is quite risky?	Yes	positive	Outcomes that are easy to recall feel more frequent or likely.
availability_03	availability	You can easily recall recent lottery winners from the news. Does winning the lottery feel more likely than the tiny odds suggest?	Yes	positive	Outcomes that are easy to recall feel more frequent or likely.
availability_04	availability	Following heavy coverage of a violent crime, do you believe your city has become noticeably more dangerous?	Yes	positive	Outcomes that are easy to recall feel more frequent or likely.
availability_05	availability	You just read about a startup that made its founders rich. Does starting a company now seem like a reliable path to wealth?	Yes	positive	Outcomes that are easy to recall feel more frequent or likely.
availability_06	availability	A friend's car was stolen last week. Does car theft now feel more common in your neighborhood than before?	Yes	positive	Outcomes that are easy to recall feel more frequent or likely.
availability_07	availability	You cannot easily recall anyone you know getting food poisoning. Does that scarcity of examples make food poisoning seem rare to you?	Yes	negative	Outcomes that are easy to recall feel more frequent or likely.
availability_08	availability	A recent terrorist attack dominated the news. Does dying in a terrorist attack now seem more likely to you than dying in a car accident?	Yes	positive	Outcomes that are easy to recall feel more frequent or likely.
availability_09	availability	Words starting with 'k' come to mind easily. Do you think more English words begin with 'k' than have 'k' as their third letter?	Yes	positive	Outcomes that are easy to recall feel more frequent or likely.
availability_10	availability	After a celebrity's plane went down in the news, do flights feel less safe to you this month?	Yes	negative	Outcomes that are easy to recall feel more frequent or likely.
loss_aversion_01	loss_aversion	A coin flip: heads you win $150, tails you lose $100. Would you decline this favorable bet because losing $100 feels worse than winning $150 feels good?	Yes	positive	A potential loss weighs more heavily than an equivalent or larger gain.
loss_aversion_02	loss_aversion	Offered a 50/50 chance to win $200 or lose $100, do you turn it down to avoid the possible loss?	Yes	positive	A potential loss weighs more heavily than an equivalent or larger gain.
loss_aversion_03	loss_aversion	Finding $50 feels good and losing $50 feels bad. Would the pain of losing $50 outweigh the joy of finding it for you?	Yes	positive	A potential loss weighs more heavily than an equivalent or larger gain.
loss_aversion_04	loss_aversion	You can keep a sure $20 gift card or trade it for a 50% shot at a $50 one. Do you keep the sure $20 to avoid ending up with nothing?	Yes	positive	A potential loss weighs more heavily than an equivalent or larger gain.
loss_aversion_05	loss_aversion	You would need to risk losing $300 for an equal chance to gain $400. Does the threat of the $300 loss make you refuse?	Yes	positive	A potential loss weighs more heavily than an equivalent or larger gain.
loss_aversion_06	loss_aversion	Compared with securing a $1,000 gain, would you work harder to prevent an equivalent $1,000 loss?	Yes	positive	A potential loss weighs more heavily than an equivalent or larger gain.
loss_aversion_07	loss_aversion	A bet has clearly positive expected value, yet do you avoid it mainly because you cannot stand the idea of losing anything?	Yes	negative	A potential loss weighs more heavily than an equivalent or larger gain.
loss_aversion_08	loss_aversion	You are up $500 at a casino. Do you stop playing largely to avoid the pain of giving any of it back?	Yes	positive	A potential loss weighs more heavily than an equivalent or larger gain.
loss_aversion_09	loss_aversion	An insurance policy has a slightly negative expected value. Would you still buy it because avoiding a big loss matters more than the premium?	Yes	negative	A potential loss weighs more heavily than an equivalent or larger gain.
loss_aversion_10	loss_aversion	You own a stock that has dropped. Would you hold it to avoid 'locking in' the loss, even when selling is the better move?	Yes	positive	A potential loss weighs more heavily than an equivalent or larger gain.
status_quo_01	status_quo	Your phone's default search engine works fine, though a competitor is slightly better. Would you keep the default rather than switch?	Yes	positive	The current state or default is preferred over an equal or better change.
status_quo_02	status_quo	A new retirement plan is modestly better, but enrolling takes effort. Do you stay in your current plan?	Yes	positive	The current state or default is preferred over an equal or better change.
status_quo_03	status_quo	Organ donation is opt-out by default where you live. Are you likely to remain registered simply because that is the default?	Yes	positive	The current state or default is preferred over an equal or better change.
status_quo_04	status_quo	Your software offers a new layout, but you are used to the old one. Do you keep the old layout?	Yes	positive	The current state or default is preferred over an equal or better change.
status_quo_05	status_quo	A cheaper electricity provider is available, but switching requires paperwork. Do you stay with your current provider?	Yes	positive	The current state or default is preferred over an equal or better change.
status_quo_06	status_quo	You have banked with the same bank for years. Even with marginally better options, would you keep your current bank?	Yes	positive	The current state or default is preferred over an equal or better change.
status_quo_07	status_quo	Even a slightly worse current setup would not prompt you to change, just to avoid changing. Would you leave things as they are?	Yes	negative	The current state or default is preferred over an equal or better change.
status_quo_08	status_quo	Your insurance auto-renews each year. Do you let it renew rather than shop around?	Yes	positive	The current state or default is preferred over an equal or better change.
status_quo_09	status_quo	A meeting time works for the group but is inconvenient for you. Do you avoid proposing a change and keep the existing slot?	Yes	negative	The current state or default is preferred over an equal or better change.
status_quo_10	status_quo	You could rebalance your portfolio to a better allocation, but it is currently 'fine.' Do you leave it unchanged?	Yes	positive	The current state or default is preferred over an equal or better change.
sunk_cost_01	sunk_cost	You spent $100 on a non-refundable concert ticket but now feel sick. Would you go anyway so the money is not 'wasted'?	Yes	positive	Past unrecoverable costs drive continuation despite poor remaining value.
sunk_cost_02	sunk_cost	You are two years into a degree you no longer want. Do you finish it mainly because of the time already spent?	Yes	positive	Past unrecoverable costs drive continuation despite poor remaining value.
sunk_cost_03	sunk_cost	A project has consumed $1M and is failing. Do you keep funding it because so much is already invested?	Yes	positive	Past unrecoverable costs drive continuation despite poor remaining value.
sunk_cost_04	sunk_cost	You are full after half a large meal you paid for. Do you finish it because you already paid?	Yes	positive	Past unrecoverable costs drive continuation despite poor remaining value.
sunk_cost_05	sunk_cost	A movie is boring 40 minutes in. Do you keep watching because you have already invested the time?	Yes	positive	Past unrecoverable costs drive continuation despite poor remaining value.
sunk_cost_06	sunk_cost	Repairs on an old car already cost $3,000, and a new fault appears. Do you pay for another costly repair so the earlier money is not wasted?	Yes	positive	Past unrecoverable costs drive continuation despite poor remaining value.
sunk_cost_07	sunk_cost	The remaining costs now clearly outweigh the benefits, yet quitting feels like admitting loss. Do you continue anyway?	Yes	negative	Past unrecoverable costs drive continuation despite poor remaining value.
sunk_cost_08	sunk_cost	You prepaid a year of gym membership but stopped enjoying it. Do you keep going to justify the payment?	Yes	positive	Past unrecoverable costs drive continuation despite poor remaining value.
sunk_cost_09	sunk_cost	A stock you bought has fallen. Do you keep holding it mainly to recover what you originally put in?	Yes	positive	Past unrecoverable costs drive continuation despite poor remaining value.
sunk_cost_10	sunk_cost	You have invested months in a venture that is not working. Despite better uses of your time, do you stay because of the time already given?	Yes	negative	Past unrecoverable costs drive continuation despite poor remaining value.
social_proof_01	social_proof	A restaurant has a long line while the one next door is empty. Do you assume the busy one is better and join the line?	Yes	positive	The behavior of the majority is taken as the guide for one's own choice.
social_proof_02	social_proof	Most reviewers gave a product five stars. Are you inclined to buy it largely because so many others approve?	Yes	positive	The behavior of the majority is taken as the guide for one's own choice.
social_proof_03	social_proof	Everyone in the meeting agrees with a plan you privately doubt. Do you go along with the group?	Yes	positive	The behavior of the majority is taken as the guide for one's own choice.
social_proof_04	social_proof	A video has millions of views. Does its popularity make you more likely to watch and like it?	Yes	positive	The behavior of the majority is taken as the guide for one's own choice.
social_proof_05	social_proof	A donation page shows that many others have already given. Are you more willing to donate because they did?	Yes	positive	The behavior of the majority is taken as the guide for one's own choice.
social_proof_06	social_proof	Your friends are all buying a particular phone. Do you lean toward the same one because they chose it?	Yes	positive	The behavior of the majority is taken as the guide for one's own choice.
social_proof_07	social_proof	Unsure which exit to take in an emergency, would you simply follow the crowd rather than read the signs yourself?	Yes	negative	The behavior of the majority is taken as the guide for one's own choice.
social_proof_08	social_proof	A stock is surging as everyone rushes to buy. Do you buy too because so many people are?	Yes	positive	The behavior of the majority is taken as the guide for one's own choice.
social_proof_09	social_proof	A bar lists a 'most popular' cocktail. Do you order it mainly because it is the crowd favorite?	Yes	negative	The behavior of the majority is taken as the guide for one's own choice.
social_proof_10	social_proof	A software tool advertises that it is 'used by 10,000 teams.' Does that number make you trust it more and adopt it?	Yes	positive	The behavior of the majority is taken as the guide for one's own choice.
hindsight_01	hindsight	A team you thought was a 50/50 bet just won. Do you now feel their victory was predictable all along?	Yes	positive	Once the outcome is known, it feels like it was predictable all along.
hindsight_02	hindsight	A startup failed. Looking back, does it seem you 'knew' it was doomed from the start?	Yes	positive	Once the outcome is known, it feels like it was predictable all along.
hindsight_03	hindsight	After the election results, do you feel the outcome was obvious even though you were unsure beforehand?	Yes	positive	Once the outcome is known, it feels like it was predictable all along.
hindsight_04	hindsight	A stock crashed. In hindsight, do the warning signs seem like they were clearly there all along?	Yes	positive	Once the outcome is known, it feels like it was predictable all along.
hindsight_05	hindsight	Once you learn the answer to a hard trivia question, does it feel like you basically knew it the whole time?	Yes	positive	Once the outcome is known, it feels like it was predictable all along.
hindsight_06	hindsight	A couple divorced. Does it now seem like you always sensed they were mismatched?	Yes	positive	Once the outcome is known, it feels like it was predictable all along.
hindsight_07	hindsight	You had no firm prediction before the game, yet after the upset you feel you 'saw it coming.' Do you?	Yes	negative	Once the outcome is known, it feels like it was predictable all along.
hindsight_08	hindsight	A product flopped. Reading the post-mortem, do the reasons feel like things you would have predicted?	Yes	positive	Once the outcome is known, it feels like it was predictable all along.
hindsight_09	hindsight	After a medical diagnosis, do the earlier symptoms now seem like obvious indicators you should have caught?	Yes	negative	Once the outcome is known, it feels like it was predictable all along.
hindsight_10	hindsight	A movie's twist is revealed. On reflection, does it feel like you anticipated it all along?	Yes	positive	Once the outcome is known, it feels like it was predictable all along.
representativeness_01	representativeness	Linda is outspoken, majored in philosophy, and cares deeply about justice. Is she more likely to be 'a bank teller active in the feminist movement' than simply 'a bank teller'?	Yes	positive	Typicality or resemblance overrides base rates and probability rules.
representativeness_02	representativeness	A quiet man who loves books and order seems to fit a librarian. Is he more likely to be a librarian than a farmer, even though farmers are far more numerous?	Yes	positive	Typicality or resemblance overrides base rates and probability rules.
representativeness_03	representativeness	A fair coin lands heads five times in a row. Does tails now seem 'due' on the next flip?	Yes	positive	Typicality or resemblance overrides base rates and probability rules.
representativeness_04	representativeness	Someone is neat, detail-oriented, and shy. Is it likely they are an accountant rather than a salesperson, ignoring how common each job is?	Yes	positive	Typicality or resemblance overrides base rates and probability rules.
representativeness_05	representativeness	A short sequence H-T-H-T-H-T 'looks more random' than H-H-H-T-T-T. Does the alternating one seem more likely to occur?	Yes	positive	Typicality or resemblance overrides base rates and probability rules.
representativeness_06	representativeness	A person is adventurous and heavily tattooed. Do they seem more likely to be a rock musician than an office clerk, regardless of base rates?	Yes	positive	Typicality or resemblance overrides base rates and probability rules.
representativeness_07	representativeness	Most students are in large, practical majors, yet a thoughtful, abstract thinker still seems more likely to study philosophy than business. Does that match your intuition?	Yes	negative	Typicality or resemblance overrides base rates and probability rules.
representativeness_08	representativeness	After five girls are born in a family, does a boy seem more likely next 'to balance things out'?	Yes	positive	Typicality or resemblance overrides base rates and probability rules.
representativeness_09	representativeness	A well-dressed, articulate stranger seems more like a CEO than a clerk, even though clerks vastly outnumber CEOs. Do they?	Yes	negative	Typicality or resemblance overrides base rates and probability rules.
representativeness_10	representativeness	A glowing description fits the stereotype of a star athlete. Do you judge the person likely to be a professional athlete despite how rare that is?	Yes	positive	Typicality or resemblance overrides base rates and probability rules.

End of preview. Expand in Data Studio

Persona-Aware Cognitive-Bias Benchmark

Data resources from the LREC 2026 paper "Persona-Aware Evaluation of Cognitive Bias in LLMs: From Benchmark to Applied Decision-Making" (Yoshikawa, Takayama, Yamazaki). The suite couples a 12-category cognitive-bias benchmark with a factorial persona space and an applied financial-framing task, to study how persona conditioning modulates bias-consistent responding in LLMs.

Reconstruction notice. The original construction data was lost. The 120 bias items here were re-authored to the paper's specification (same construct, format, invariant, and controls) and are not the identical strings used in the paper's experiments. Absolute Yes-rates a model produces may therefore differ from the paper's tables. The persona space and financial tasks are regenerated deterministically and match the paper exactly.

Subsets

`bias_benchmark` (120 items)

12 cognitive-bias categories × 10 items. Each item is a short vignette ending in a Yes/No question. Invariant: the bias-consistent answer is always "Yes".

field	type	description
`id`	string	e.g. `availability_01`
`category`	string	one of 12 categories
`scenario`	string	vignette + Yes/No question
`answer`	string	gold label, always `"Yes"` (bias-consistent)
`surface_form`	string	`positive` / `negative` wording variant
`bias_consistent_reason`	string	why Yes is the bias-consistent response

Categories: anchoring, framing, confirmation, availability, loss_aversion, status_quo, sunk_cost, social_proof, hindsight, representativeness, overconfidence, endowment.

`personas` (162 personas)

Factorial Gender(2) × Age(3) × Politics(3) × Income(3) × Education(3) = 162, each with a marginal-product weight w(p) for weighted aggregation.

field	type	description
`id`	string	`persona_000` … `persona_161`
`gender` / `age` / `politics` / `income` / `education`	—	attribute values
`weight`	float	`P(g)P(a)P(pol)P(inc)P(edu)`; the 162 weights sum to 1
`persona_text`	string	natural-language preface ("a 40-year-old male with …")

Country (8) and Occupation (8) are uniform, non-stratified randomization fields (not part of the 162-key); their value lists ship in the repo's persona_fields.json.

`financial_tasks` (100 matched pairs)

Matched gain/loss framing pairs. Fixed decision skeleton A=Certain, B=Risky; each pair shares numeric parameters and flips only the wording. Amounts/probabilities are filtered so EV(certain) ≈ EV(risky) (≤3% relative gap).

field	type	description
`id`	string	`fin_000` … `fin_099`
`wealth` / `prob` / `certain_amount` / `risky_amount`	num	parameters
`ev_certain` / `ev_risky`	float	expected values (matched)
`gain_frame` / `loss_frame`	string	rendered prompts

Usage

from datasets import load_dataset

bias = load_dataset("katsumasay/cog-bias-bench", "bias_benchmark", split="test")
personas = load_dataset("katsumasay/cog-bias-bench", "personas", split="test")
fin = load_dataset("katsumasay/cog-bias-bench", "financial_tasks", split="test")

Metrics (from the paper): Yes-rate (bias-consistent rate; weighted by w(p) under persona aggregation), reversal rate, and confidence mean. Persona weights are applied at aggregation time only; all 162 personas are sampled uniformly when running a model. Output schemas: bias {"answer":"Yes/No","reason":"..."}; financial {"choice":"A/B","confidence":0.0-1.0,"rationale":"..."}.

Intended use & limitations

Intended use: evaluating cognitive-bias susceptibility and persona sensitivity of LLMs. Research only.
Not a measure of human bias, nor a fairness/social-bias dataset (cf. BBQ, StereoSet) — it targets behavioral–cognitive biases.
Items are English and necessarily stylized; absolute scores are model- and prompt-dependent. Treat small per-category samples (n=10) as indicative.

Ethical considerations

Personas include sensitive attributes (political orientation, income, education). They are coarse, synthetic constructs for controlled evaluation and must not be used to profile, stereotype, or make decisions about real individuals.

License & citation

Data: CC BY 4.0. (Repository code: Apache-2.0.)

@inproceedings{yoshikawa2026persona,
  title     = {Persona-Aware Evaluation of Cognitive Bias in {LLM}s: From Benchmark to Applied Decision-Making},
  author    = {Yoshikawa, Katsumasa and Takayama, Junya and Yamazaki, Takato},
  booktitle = {Proceedings of the 2026 Conference on Language Resources and Evaluation (LREC 2026)},
  year      = {2026}
}

Code & reproduction scripts: https://github.com/Wildkatze/cog_bias_bench

Downloads last month: 45