Datasets:
Tasks:
Other
Modalities:
Text
Formats:
json
Languages:
English
Size:
< 1K
Tags:
computational-social-science
llm-agents
ai-evaluation
benchmark
behavioural-economics
psychology
License:
SiliconSubjects: 17 protocols with human anchors for auditing simulated subject populations
6dd335a verified | """ | |
| Silicon Subjects — protocol suite builder. | |
| Writes data/protocols.jsonl: classic experimental protocols from behavioural | |
| economics and psychology, each paired with *human anchor* statistics taken from | |
| the published literature, split along the three axes a simulated population has | |
| to reproduce: | |
| 1. location — the level of the response (mean) | |
| 2. dispersion — the spread of the response (sd) | |
| 3. effect — the direction and magnitude of the treatment contrast | |
| Anchors are literature aggregates, not raw subject-level data. They are meant as | |
| reference points for auditing a simulated population, NOT as exact replications; | |
| always re-check against the cited source before drawing conclusions. | |
| Usage: | |
| python scripts/build_protocols.py [--out-dir data] | |
| """ | |
| from __future__ import annotations | |
| import argparse | |
| import json | |
| from pathlib import Path | |
| from typing import Any, Dict, List, Optional | |
| def P( # noqa: N802 - short builder alias | |
| protocol_id: str, | |
| name: str, | |
| domain: str, | |
| paradigm: str, | |
| source: str, | |
| response: Dict[str, Any], | |
| conditions: List[Dict[str, str]], | |
| anchors: Dict[str, Dict[str, Any]], | |
| effect: Optional[Dict[str, Any]] = None, | |
| shape: Optional[str] = None, | |
| notes: str = "", | |
| ) -> Dict[str, Any]: | |
| return { | |
| "protocol_id": protocol_id, | |
| "name": name, | |
| "domain": domain, | |
| "paradigm": paradigm, | |
| "source": source, | |
| "response": response, | |
| "conditions": conditions, | |
| "anchors": anchors, | |
| "effect": effect, | |
| "shape": shape, | |
| "notes": notes, | |
| } | |
| NUM = lambda lo, hi, unit: {"type": "numeric", "min": lo, "max": hi, "unit": unit} # noqa: E731 | |
| BIN = lambda lo_label, hi_label: { # noqa: E731 | |
| "type": "binary", | |
| "min": 0, | |
| "max": 1, | |
| "unit": f"1 = {hi_label}, 0 = {lo_label}", | |
| } | |
| # --------------------------------------------------------------------------- | |
| # 1-8. Social preferences | |
| # --------------------------------------------------------------------------- | |
| DICTATOR = P( | |
| "dictator_game", | |
| "Dictator game", | |
| "social_preferences", | |
| "one-shot unilateral allocation", | |
| "Engel (2011) 'Dictator games: a meta study', Experimental Economics 14:583-610", | |
| NUM(0, 10, "dollars transferred out of a $10 endowment"), | |
| [ | |
| { | |
| "condition_id": "single", | |
| "label": "single decision", | |
| "prompt": ( | |
| "You are taking part in an economics experiment. You have been given $10. " | |
| "You may transfer any amount from $0 to $10 to another participant, who " | |
| "cannot respond and cannot affect your payoff. You keep whatever you do " | |
| "not transfer. How many dollars do you transfer? Answer with a single number." | |
| ), | |
| } | |
| ], | |
| { | |
| "single": { | |
| "mean": 2.83, | |
| "sd": 2.60, | |
| "confidence": "medium", | |
| "note": "Meta-analytic mean share is ~28% of the stake; individual-level dispersion " | |
| "varies widely across subject pools, so treat sd as approximate.", | |
| } | |
| }, | |
| shape="bimodal — mass at $0 and at half the stake", | |
| notes="The canonical other-regarding-preference measure. A simulated population that gives " | |
| "the 'fair' 50% is reproducing a norm, not the human distribution.", | |
| ) | |
| ULTIMATUM_OFFER = P( | |
| "ultimatum_offer", | |
| "Ultimatum game — proposer offer", | |
| "social_preferences", | |
| "one-shot bargaining, proposer role", | |
| "Oosterbeek, Sloof & van de Kuilen (2004) 'Cultural differences in ultimatum game " | |
| "experiments', Experimental Economics 7:171-188; Camerer (2003)", | |
| NUM(0, 10, "dollars offered out of a $10 pie"), | |
| [ | |
| { | |
| "condition_id": "proposer", | |
| "label": "proposer", | |
| "prompt": ( | |
| "You are the proposer. You and one other participant will split $10. You " | |
| "propose how much of the $10 the other person receives. If they accept, you " | |
| "both get the proposed split; if they reject, neither of you gets anything. " | |
| "How many dollars do you offer? Answer with a single number." | |
| ), | |
| } | |
| ], | |
| { | |
| "proposer": { | |
| "mean": 4.00, | |
| "sd": 1.30, | |
| "confidence": "medium-high", | |
| "note": "Meta-analytic mean offer is ~40% of the pie, with a strong mode at 50%.", | |
| } | |
| }, | |
| shape="unimodal with a spike at 50%", | |
| notes="Offers cluster at the equal split; very low offers are rare because proposers " | |
| "anticipate rejection.", | |
| ) | |
| ULTIMATUM_REJECT = P( | |
| "ultimatum_reject_low", | |
| "Ultimatum game — rejection of a low offer", | |
| "social_preferences", | |
| "one-shot bargaining, responder role", | |
| "Camerer (2003) Behavioral Game Theory; Oosterbeek et al. (2004)", | |
| BIN("accept", "reject"), | |
| [ | |
| { | |
| "condition_id": "low_offer", | |
| "label": "offer of $2 out of $10", | |
| "prompt": ( | |
| "You are the responder. The proposer offers you $2 out of $10 and keeps $8. " | |
| "If you accept, you get $2 and they get $8. If you reject, neither of you " | |
| "gets anything. Answer 1 if you reject the offer, 0 if you accept it." | |
| ), | |
| } | |
| ], | |
| { | |
| "low_offer": { | |
| "mean": 0.50, | |
| "sd": None, | |
| "confidence": "medium", | |
| "note": "Rejection probability for offers at or below 20% of the pie is roughly " | |
| "40-60%; it rises steeply as the offer falls.", | |
| } | |
| }, | |
| notes="Human responders punish unfairness at a material cost to themselves. Populations " | |
| "that never reject are behaving like pure payoff maximisers.", | |
| ) | |
| PUBLIC_GOODS = P( | |
| "public_goods_oneshot", | |
| "Public goods game — one shot", | |
| "social_preferences", | |
| "linear voluntary contribution mechanism, one-shot", | |
| "Ledyard (1995) in Kagel & Roth (eds.), Handbook of Experimental Economics; " | |
| "Chaudhuri (2011) 'Sustaining cooperation in laboratory public goods experiments', " | |
| "Experimental Economics 14:47-83", | |
| NUM(0, 10, "tokens contributed out of 10"), | |
| [ | |
| { | |
| "condition_id": "oneshot", | |
| "label": "one-shot, 4 players, MPCR 0.4", | |
| "prompt": ( | |
| "You are in a group of 4 people. Each person receives 10 tokens. Any token " | |
| "you put into a common pool is multiplied by 1.6 and then split equally " | |
| "among the 4 members. Tokens you keep are worth 1 point each to you. This " | |
| "is a single, one-shot decision with the same 4 people. How many of your 10 " | |
| "tokens do you contribute to the pool? Answer with a single number." | |
| ), | |
| } | |
| ], | |
| { | |
| "oneshot": { | |
| "mean": 4.50, | |
| "sd": 2.60, | |
| "confidence": "medium", | |
| "note": "One-shot contributions typically run 40-60% of the endowment; " | |
| "individual-level spread is large.", | |
| } | |
| }, | |
| notes="Contributions start well above the Nash prediction of zero. Free-riding is a " | |
| "minority behaviour, not the modal one.", | |
| ) | |
| PUBLIC_GOODS_DECAY = P( | |
| "public_goods_decay", | |
| "Public goods game — decay across rounds", | |
| "social_preferences", | |
| "repeated voluntary contribution, partners condition", | |
| "Ledyard (1995); Chaudhuri (2011); Isaac & Walker (1988)", | |
| NUM(0, 10, "tokens contributed out of 10"), | |
| [ | |
| { | |
| "condition_id": "round_1", | |
| "label": "round 1 of 10", | |
| "prompt": ( | |
| "You are in a group of 4 people and will play 10 rounds with the same " | |
| "people. Each round every person receives 10 tokens; tokens placed in a " | |
| "common pool are multiplied by 1.6 and split equally among the 4 members. " | |
| "This is round 1. How many of your 10 tokens do you contribute? Answer with " | |
| "a single number." | |
| ), | |
| }, | |
| { | |
| "condition_id": "round_10", | |
| "label": "round 10 of 10 (final round)", | |
| "prompt": ( | |
| "You are in a group of 4 people and have played 9 rounds with the same " | |
| "people. Each round every person receives 10 tokens; tokens placed in a " | |
| "common pool are multiplied by 1.6 and split equally among the 4 members. " | |
| "The average contribution started around 4 tokens and has fallen steadily. " | |
| "This is round 10, the final round. How many of your 10 tokens do you " | |
| "contribute? Answer with a single number." | |
| ), | |
| }, | |
| ], | |
| { | |
| "round_1": {"mean": 4.50, "sd": 2.60, "confidence": "medium", "note": ""}, | |
| "round_10": {"mean": 1.80, "sd": 2.00, "confidence": "medium", "note": ""}, | |
| }, | |
| effect={ | |
| "a": "round_1", | |
| "b": "round_10", | |
| "mean_diff": 2.70, | |
| "cohen_d": 1.10, | |
| "confidence": "medium-high", | |
| "source": "Ledyard (1995); Chaudhuri (2011)", | |
| "note": "The decay of cooperation without punishment or communication is one of the " | |
| "most robust regularities in experimental economics.", | |
| }, | |
| notes="A population that holds contributions flat across rounds is missing the entire " | |
| "dynamics of conditional cooperation.", | |
| ) | |
| TRUST_SEND = P( | |
| "trust_game_send", | |
| "Trust game — amount sent", | |
| "social_preferences", | |
| "one-shot investment game, trustor role", | |
| "Johnson & Mislin (2011) 'Trust games: a meta-analysis', Journal of Economic " | |
| "Psychology 32:865-889", | |
| NUM(0, 10, "dollars sent out of $10 (tripled before transfer)"), | |
| [ | |
| { | |
| "condition_id": "trustor", | |
| "label": "trustor", | |
| "prompt": ( | |
| "You are given $10 and paired with a stranger. You may send any amount from " | |
| "$0 to $10. Any amount you send is tripled by the experimenter and given to " | |
| "the stranger, who then decides how much, if anything, to send back to you. " | |
| "How many dollars do you send? Answer with a single number." | |
| ), | |
| } | |
| ], | |
| { | |
| "trustor": { | |
| "mean": 5.00, | |
| "sd": 2.70, | |
| "confidence": "medium", | |
| "note": "Meta-analytic mean send is ~50% of the endowment.", | |
| } | |
| }, | |
| notes="Sending is positive in nearly every subject pool, but far from the payoff-maximising " | |
| "100%.", | |
| ) | |
| TRUST_RETURN = P( | |
| "trust_game_return", | |
| "Trust game — amount returned", | |
| "social_preferences", | |
| "one-shot investment game, trustee role", | |
| "Johnson & Mislin (2011)", | |
| NUM(0, 18, "dollars returned out of $18 available"), | |
| [ | |
| { | |
| "condition_id": "trustee", | |
| "label": "trustee", | |
| "prompt": ( | |
| "A stranger was given $10 and sent you $6. The experimenter tripled it, so " | |
| "you have $18. You may send back any amount from $0 to $18 to the stranger. " | |
| "How many dollars do you send back? Answer with a single number." | |
| ), | |
| } | |
| ], | |
| { | |
| "trustee": { | |
| "mean": 6.50, | |
| "sd": 4.00, | |
| "confidence": "approximate", | |
| "note": "Meta-analytic 'trustworthiness' is ~0.37 of the amount available; values " | |
| "here are approximate and sensitive to the amount sent.", | |
| } | |
| }, | |
| notes="Reciprocity, not pure self-interest: trustees typically return enough to make " | |
| "sending worthwhile.", | |
| ) | |
| PRISONERS = P( | |
| "prisoners_dilemma", | |
| "Prisoner's dilemma — one shot", | |
| "social_preferences", | |
| "one-shot 2x2 symmetric game", | |
| "Sally (1995) 'Conversation and cooperation in social dilemmas', Rationality and " | |
| "Society 7:58-92", | |
| BIN("defect", "cooperate"), | |
| [ | |
| { | |
| "condition_id": "oneshot", | |
| "label": "one-shot, anonymous, no communication", | |
| "prompt": ( | |
| "You and one other person each choose to cooperate or defect, simultaneously " | |
| "and anonymously. You will never meet again. If both cooperate, each gets $6. " | |
| "If both defect, each gets $2. If you cooperate and the other defects, you get " | |
| "$0 and they get $9. If you defect and the other cooperates, you get $9 and " | |
| "they get $0. Answer 1 to cooperate, 0 to defect." | |
| ), | |
| } | |
| ], | |
| { | |
| "oneshot": { | |
| "mean": 0.47, | |
| "sd": None, | |
| "confidence": "medium", | |
| "note": "Meta-analytic cooperation in the one-shot prisoner's dilemma is ~45-50%; " | |
| "it collapses toward 0 with experience and rises with communication.", | |
| } | |
| }, | |
| notes="Cooperation in the one-shot game is the benchmark case where a purely rational " | |
| "population and a human population diverge most sharply.", | |
| ) | |
| # --------------------------------------------------------------------------- | |
| # 9-14. Heuristics and biases | |
| # --------------------------------------------------------------------------- | |
| ASCH = P( | |
| "asch_conformity", | |
| "Asch line-judgement conformity", | |
| "social_influence", | |
| "unambiguous perceptual judgement under group pressure", | |
| "Asch (1951, 1955); Bond & Smith (1996) 'Culture and conformity', Psychological " | |
| "Bulletin 119:111-137", | |
| BIN("give the correct answer", "conform to the group's wrong answer"), | |
| [ | |
| { | |
| "condition_id": "critical", | |
| "label": "critical trial, unanimous wrong majority", | |
| "prompt": ( | |
| "You are judging which of three comparison lines matches a standard line in " | |
| "length. The correct match is obvious and you are confident about it. Six " | |
| "other participants answer before you and all six give the same clearly " | |
| "incorrect answer. It is now your turn, aloud. Answer 1 if you give the " | |
| "group's incorrect answer, 0 if you give the obviously correct one." | |
| ), | |
| } | |
| ], | |
| { | |
| "critical": { | |
| "mean": 0.32, | |
| "sd": None, | |
| "confidence": "medium", | |
| "note": "~32% of critical trials draw a conforming response in the classic " | |
| "setting; roughly 75% of participants conform at least once and ~25% never do.", | |
| } | |
| }, | |
| notes="The cleanest available test of whether a simulated subject is willing to contradict " | |
| "an unambiguous fact under social pressure.", | |
| ) | |
| ANCHORING = P( | |
| "anchoring", | |
| "Anchoring — comparative anchor then estimate", | |
| "heuristics_biases", | |
| "anchoring-and-adjustment", | |
| "Jacowitz & Kahneman (1995) 'Measures of anchoring in estimation tasks', Personality " | |
| "and Social Psychology Bulletin 21:1161-1166; Tversky & Kahneman (1974)", | |
| NUM(0, 100, "estimated percentage of African countries in the UN"), | |
| [ | |
| { | |
| "condition_id": "high", | |
| "label": "high anchor (65%)", | |
| "prompt": ( | |
| "Is the percentage of African countries in the United Nations more or less " | |
| "than 65%? Now estimate, as accurately as you can, the actual percentage of " | |
| "African countries in the United Nations. Answer with a single number " | |
| "between 0 and 100." | |
| ), | |
| }, | |
| { | |
| "condition_id": "low", | |
| "label": "low anchor (10%)", | |
| "prompt": ( | |
| "Is the percentage of African countries in the United Nations more or less " | |
| "than 10%? Now estimate, as accurately as you can, the actual percentage of " | |
| "African countries in the United Nations. Answer with a single number " | |
| "between 0 and 100." | |
| ), | |
| }, | |
| ], | |
| {}, | |
| effect={ | |
| "a": "high", | |
| "b": "low", | |
| "mean_diff": 20.0, | |
| "cohen_d": 0.90, | |
| "confidence": "approximate", | |
| "source": "Jacowitz & Kahneman (1995)", | |
| "note": "Absolute estimates are item-specific, so only the contrast is anchored here; " | |
| "the high-low gap is typically 15-25 percentage points.", | |
| }, | |
| notes="Deliberately has no level anchor: the point of this protocol is the contrast, not " | |
| "the absolute estimate. This is the pattern to use whenever only an effect size is " | |
| "trustworthy.", | |
| ) | |
| ASIAN_DISEASE = P( | |
| "asian_disease", | |
| "Asian disease problem — risky-choice framing", | |
| "heuristics_biases", | |
| "gain vs. loss frame, logically equivalent options", | |
| "Tversky & Kahneman (1981) 'The framing of decisions and the psychology of choice', " | |
| "Science 211:453-458", | |
| BIN("choose the risky option", "choose the sure option"), | |
| [ | |
| { | |
| "condition_id": "gain", | |
| "label": "gain frame (lives saved)", | |
| "prompt": ( | |
| "Imagine the US is preparing for an unusual disease expected to kill 600 " | |
| "people. Two programmes are proposed. If programme A is adopted, 200 people " | |
| "will be saved. If programme B is adopted, there is a one-third probability " | |
| "that 600 people will be saved and a two-thirds probability that no one will " | |
| "be saved. Answer 1 for programme A (the sure option), 0 for programme B." | |
| ), | |
| }, | |
| { | |
| "condition_id": "loss", | |
| "label": "loss frame (lives lost)", | |
| "prompt": ( | |
| "Imagine the US is preparing for an unusual disease expected to kill 600 " | |
| "people. Two programmes are proposed. If programme C is adopted, 400 people " | |
| "will die. If programme D is adopted, there is a one-third probability that " | |
| "nobody will die and a two-thirds probability that 600 people will die. " | |
| "Answer 1 for programme C (the sure option), 0 for programme D." | |
| ), | |
| }, | |
| ], | |
| { | |
| "gain": {"mean": 0.72, "sd": None, "confidence": "high", "note": ""}, | |
| "loss": {"mean": 0.22, "sd": None, "confidence": "high", "note": ""}, | |
| }, | |
| effect={ | |
| "a": "gain", | |
| "b": "loss", | |
| "mean_diff": 0.50, | |
| "cohen_d": 1.00, | |
| "confidence": "high", | |
| "source": "Tversky & Kahneman (1981)", | |
| "note": "Options A/C and B/D are formally identical; only the description differs.", | |
| }, | |
| notes="One of the most replicated effects in the literature. A population that is " | |
| "frame-invariant here is not behaving like a human sample.", | |
| ) | |
| ENDOWMENT = P( | |
| "endowment_effect", | |
| "Endowment effect — WTA vs. WTP", | |
| "heuristics_biases", | |
| "buyer vs. seller valuation of the same object", | |
| "Kahneman, Knetsch & Thaler (1990) 'Experimental tests of the endowment effect and " | |
| "the Coase theorem', Journal of Political Economy 98:1325-1348", | |
| NUM(0, 10, "dollars"), | |
| [ | |
| { | |
| "condition_id": "wtp", | |
| "label": "buyer — maximum willingness to pay", | |
| "prompt": ( | |
| "A coffee mug is for sale in the experiment. You do not own one. What is the " | |
| "largest amount, from $0 to $10, that you would be willing to pay to buy this " | |
| "mug? Answer with a single number." | |
| ), | |
| }, | |
| { | |
| "condition_id": "wta", | |
| "label": "seller — minimum willingness to accept", | |
| "prompt": ( | |
| "You have just been given a coffee mug and now own it. What is the smallest " | |
| "amount, from $0 to $10, that you would be willing to accept to sell this " | |
| "mug? Answer with a single number." | |
| ), | |
| }, | |
| ], | |
| { | |
| "wtp": {"mean": 3.00, "sd": 1.50, "confidence": "approximate", "note": ""}, | |
| "wta": {"mean": 6.50, "sd": 2.00, "confidence": "approximate", "note": ""}, | |
| }, | |
| effect={ | |
| "a": "wta", | |
| "b": "wtp", | |
| "mean_diff": 3.50, | |
| "cohen_d": 0.80, | |
| "confidence": "approximate", | |
| "source": "Kahneman, Knetsch & Thaler (1990); Horowitz & McConnell (2002) meta-analysis", | |
| "note": "The WTA/WTP ratio is typically 2-3; absolute dollar values depend on the " | |
| "object and the subject pool.", | |
| }, | |
| notes="Ownership alone shifts valuation. A population that quotes the same number in both " | |
| "roles is treating this as a pure valuation problem.", | |
| ) | |
| CRT = P( | |
| "crt_batball", | |
| "Cognitive reflection — bat and ball", | |
| "heuristics_biases", | |
| "intuitive-but-wrong vs. reflective answer", | |
| "Frederick (2005) 'Cognitive reflection and decision making', Journal of Economic " | |
| "Perspectives 19:25-42", | |
| BIN("intuitive wrong answer (10 cents)", "reflective correct answer (5 cents)"), | |
| [ | |
| { | |
| "condition_id": "batball", | |
| "label": "bat and ball", | |
| "prompt": ( | |
| "A bat and a ball cost $1.10 in total. The bat costs $1.00 more than the " | |
| "ball. How much does the ball cost? Answer 1 if your answer is 5 cents " | |
| "(correct), 0 if your answer is 10 cents." | |
| ), | |
| } | |
| ], | |
| { | |
| "batball": { | |
| "mean": 0.40, | |
| "sd": None, | |
| "confidence": "approximate", | |
| "note": "Correct rates on the bat-and-ball item run roughly 30-50% in student " | |
| "samples and lower in general-population samples.", | |
| } | |
| }, | |
| notes="Scored as a binary here so that no free-text grading is needed. If you want the " | |
| "full 3-item CRT, run it as three binary protocols.", | |
| ) | |
| CONJUNCTION = P( | |
| "conjunction_fallacy", | |
| "Conjunction fallacy — Linda problem", | |
| "heuristics_biases", | |
| "representativeness vs. the conjunction rule", | |
| "Tversky & Kahneman (1983) 'Extensional versus intuitive reasoning', Psychological " | |
| "Review 90:293-315", | |
| BIN("choose the single (logically correct) option", "choose the conjunction"), | |
| [ | |
| { | |
| "condition_id": "between", | |
| "label": "between-subjects, ranking removed", | |
| "prompt": ( | |
| "Linda is 31 years old, single, outspoken and very bright. She majored in " | |
| "philosophy. As a student she was deeply concerned with issues of " | |
| "discrimination and social justice, and she participated in anti-nuclear " | |
| "demonstrations. Which is more probable? Option 1: Linda is a bank teller. " | |
| "Option 2: Linda is a bank teller and is active in the feminist movement. " | |
| "Answer 1 if you choose option 2 (the conjunction), 0 if you choose option 1." | |
| ), | |
| } | |
| ], | |
| { | |
| "between": { | |
| "mean": 0.80, | |
| "sd": None, | |
| "confidence": "medium", | |
| "note": "Roughly 80-90% choose the conjunction in the standard version; rates fall " | |
| "with transparent wording or within-subjects designs.", | |
| } | |
| }, | |
| notes="Choosing the conjunction violates the conjunction rule. This is the classic case " | |
| "where a well-aligned model looks *worse* than a human sample.", | |
| ) | |
| # --------------------------------------------------------------------------- | |
| # 15. Reasoning | |
| # --------------------------------------------------------------------------- | |
| WASON = P( | |
| "wason_selection", | |
| "Wason selection task — abstract vs. deontic", | |
| "reasoning", | |
| "conditional-reasoning / cheater-detection", | |
| "Wason (1968); Cosmides & Tooby (1992) 'Cognitive adaptations for social exchange', " | |
| "Cognition 31:187-276", | |
| BIN("any other selection", "exactly the two cards that can falsify the rule"), | |
| [ | |
| { | |
| "condition_id": "abstract", | |
| "label": "abstract rule (D / K / 3 / 7)", | |
| "prompt": ( | |
| "Four cards are on a table, showing D, K, 3 and 7. Every card has a letter on " | |
| "one side and a number on the other. The rule is: if a card has a D on one " | |
| "side, then it has a 3 on the other side. You may turn over only the cards " | |
| "needed to test whether the rule is false. The correct answer is the D card " | |
| "and the 7 card. Answer 1 if you would turn over exactly those two cards, " | |
| "0 otherwise." | |
| ), | |
| }, | |
| { | |
| "condition_id": "deontic", | |
| "label": "social-contract rule (under-age drinking)", | |
| "prompt": ( | |
| "You are checking whether people are violating the rule 'if a person is " | |
| "drinking beer, then they must be over 21'. Four people are in front of you: " | |
| "one drinking beer, one drinking coke, one aged 25, one aged 16. You may check " | |
| "only the ones needed to find a violation. The correct answer is the beer " | |
| "drinker and the 16-year-old. Answer 1 if you would check exactly those two " | |
| "people, 0 otherwise." | |
| ), | |
| }, | |
| ], | |
| { | |
| "abstract": {"mean": 0.15, "sd": None, "confidence": "medium-high", "note": ""}, | |
| "deontic": {"mean": 0.70, "sd": None, "confidence": "medium", "note": ""}, | |
| }, | |
| effect={ | |
| "a": "deontic", | |
| "b": "abstract", | |
| "mean_diff": 0.55, | |
| "cohen_d": 1.20, | |
| "confidence": "medium-high", | |
| "source": "Cosmides & Tooby (1992); Wason (1968)", | |
| "note": "The content effect is large: the same logical form is easy when framed as " | |
| "cheater detection and hard when framed abstractly.", | |
| }, | |
| notes="The correct answers are stated in the prompts, so this measures whether the " | |
| "population's *selection* matches human selection, not whether it can solve the logic.", | |
| ) | |
| # --------------------------------------------------------------------------- | |
| # 16-17. Prosocial behaviour and moral judgement | |
| # --------------------------------------------------------------------------- | |
| BYSTANDER = P( | |
| "bystander_effect", | |
| "Bystander effect — alone vs. in a group", | |
| "social_influence", | |
| "diffusion of responsibility", | |
| "Darley & Latané (1968) 'Bystander intervention in emergencies', Journal of Personality " | |
| "and Social Psychology 8:377-383; Latané & Nida (1981)", | |
| BIN("do not help", "help"), | |
| [ | |
| { | |
| "condition_id": "alone", | |
| "label": "participant alone", | |
| "prompt": ( | |
| "You are alone in a room filling out a questionnaire. You hear someone in the " | |
| "next room fall, cry out in pain and groan. No one else is present. Answer 1 " | |
| "if you would go to help, 0 if you would not." | |
| ), | |
| }, | |
| { | |
| "condition_id": "group", | |
| "label": "two passive others present", | |
| "prompt": ( | |
| "You are in a room with two other people, filling out a questionnaire. You " | |
| "hear someone in the next room fall, cry out in pain and groan. The other two " | |
| "people look up, then go back to their questionnaires without reacting. Answer " | |
| "1 if you would go to help, 0 if you would not." | |
| ), | |
| }, | |
| ], | |
| { | |
| "alone": {"mean": 0.75, "sd": None, "confidence": "medium", "note": ""}, | |
| "group": {"mean": 0.35, "sd": None, "confidence": "medium", "note": ""}, | |
| }, | |
| effect={ | |
| "a": "alone", | |
| "b": "group", | |
| "mean_diff": 0.40, | |
| "cohen_d": 0.85, | |
| "confidence": "medium", | |
| "source": "Darley & Latané (1968); Latané & Nida (1981)", | |
| "note": "Meta-analytic effect is moderate and smaller than textbook accounts suggest, " | |
| "but consistently present.", | |
| }, | |
| notes="A simulated population that helps at the same rate in both conditions shows no " | |
| "diffusion of responsibility at all.", | |
| ) | |
| TROLLEY = P( | |
| "trolley_dilemma", | |
| "Trolley problem — switch vs. footbridge", | |
| "moral_judgment", | |
| "sacrificial moral dilemmas, means vs. side-effect", | |
| "Foot (1967); Thomson (1985) 'The trolley problem', Yale Law Journal 94:1395-1415; " | |
| "large-N replications (e.g. Hauser et al. 2007)", | |
| BIN("do nothing", "act to save the five"), | |
| [ | |
| { | |
| "condition_id": "switch", | |
| "label": "switch (impersonal)", | |
| "prompt": ( | |
| "A runaway trolley is heading for five people who will be killed if it " | |
| "continues. You are standing next to a switch. If you pull it, the trolley " | |
| "diverts to a side track where it will kill one person instead. Answer 1 if " | |
| "you pull the switch, 0 if you do not." | |
| ), | |
| }, | |
| { | |
| "condition_id": "footbridge", | |
| "label": "footbridge (personal)", | |
| "prompt": ( | |
| "A runaway trolley is heading for five people who will be killed if it " | |
| "continues. You are on a footbridge above the track next to a large stranger. " | |
| "The only way to stop the trolley is to push the stranger off the bridge onto " | |
| "the track; his body will stop the trolley, killing him but saving the five. " | |
| "Answer 1 if you push the stranger, 0 if you do not." | |
| ), | |
| }, | |
| ], | |
| { | |
| "switch": {"mean": 0.85, "sd": None, "confidence": "medium-high", "note": ""}, | |
| "footbridge": {"mean": 0.25, "sd": None, "confidence": "medium-high", "note": ""}, | |
| }, | |
| effect={ | |
| "a": "switch", | |
| "b": "footbridge", | |
| "mean_diff": 0.60, | |
| "cohen_d": 1.30, | |
| "confidence": "medium-high", | |
| "source": "Thomson (1985); Hauser et al. (2007)", | |
| "note": "Both dilemmas sacrifice one to save five; the personal/impersonal contrast " | |
| "drives a very large difference in responding.", | |
| }, | |
| notes="The two dilemmas are consequentially identical. A population that gives the same " | |
| "answer to both is doing utilitarian arithmetic, not reproducing human moral judgement.", | |
| ) | |
| PROTOCOLS: List[Dict[str, Any]] = [ | |
| DICTATOR, | |
| ULTIMATUM_OFFER, | |
| ULTIMATUM_REJECT, | |
| PUBLIC_GOODS, | |
| PUBLIC_GOODS_DECAY, | |
| TRUST_SEND, | |
| TRUST_RETURN, | |
| PRISONERS, | |
| ASCH, | |
| ANCHORING, | |
| ASIAN_DISEASE, | |
| ENDOWMENT, | |
| CRT, | |
| CONJUNCTION, | |
| WASON, | |
| BYSTANDER, | |
| TROLLEY, | |
| ] | |
| def _validate(p: Dict[str, Any]) -> None: | |
| ids = {c["condition_id"] for c in p["conditions"]} | |
| for cid in p["anchors"]: | |
| assert cid in ids, f"{p['protocol_id']}: anchor for unknown condition {cid}" | |
| e = p.get("effect") | |
| if e: | |
| assert e["a"] in ids and e["b"] in ids, f"{p['protocol_id']}: bad effect condition" | |
| assert e["mean_diff"] != 0, f"{p['protocol_id']}: zero effect anchor" | |
| assert p["response"]["min"] < p["response"]["max"] | |
| def main() -> None: | |
| ap = argparse.ArgumentParser() | |
| ap.add_argument( | |
| "--out-dir", default=str(Path(__file__).resolve().parents[1] / "data") | |
| ) | |
| args = ap.parse_args() | |
| for p in PROTOCOLS: | |
| _validate(p) | |
| out = Path(args.out_dir) | |
| out.mkdir(parents=True, exist_ok=True) | |
| with (out / "protocols.jsonl").open("w", encoding="utf-8", newline="\n") as f: | |
| for p in PROTOCOLS: | |
| f.write(json.dumps(p, ensure_ascii=False) + "\n") | |
| n_eff = sum(1 for p in PROTOCOLS if p["effect"]) | |
| n_loc = sum(1 for p in PROTOCOLS if p["anchors"]) | |
| print(f"wrote {len(PROTOCOLS)} protocols -> {out / 'protocols.jsonl'}") | |
| print(f" with location/dispersion anchors: {n_loc}") | |
| print(f" with a treatment-effect anchor: {n_eff}") | |
| if __name__ == "__main__": | |
| main() | |