科研 Skill

实验经济学设计

experimental-economics-design

Designs incentivized lab, online, or lab-in-field experiments: treatment and payoff design, deception/IRB norms, and oTree/z-Tree setup. Use when designing a behavioral-economics study.

发布者 Codar安全审核通过v0.1.0更新于 2026-11-18

SKILL.md

Experimental Economics Design

Purpose

Covers the design of incentivized economics experiments — laboratory, online (Prolific/MTurk/CloudResearch), or lab-in-field — distinct from the field-RCT skills elsewhere in this marketplace (rct-core-design and siblings in the social-science plugin), which focus on randomized policy or program evaluation in the field. Experimental economics instead tests a specific behavioral or strategic hypothesis (risk preferences, bargaining, public-goods contribution, market behavior, discrimination) under controlled, usually monetarily incentivized conditions, often grounded in a game-theoretic or decision-theoretic model whose predictions the experiment is built to distinguish.

Use this skill for designing the experiment itself: treatments, incentive structure, session protocol, and software setup. For sample-size and power calculations once the design is fixed, pair with rct-power-calculations; for randomization mechanics, pair with rct-randomization; for analyzing the resulting data, pair with rct-data-analysis or ols-regression.

When to Use

Use when the request mentions:

  • lab experiment, economics experiment, behavioral experiment design
  • oTree, z-Tree, LIONESS Lab, or similar experiment software
  • incentive-compatible design, induced-value theory, show-up fee
  • within-subject vs. between-subject design for an economic game
  • deception in experimental economics, IRB for incentivized studies
  • online experiment on Prolific/MTurk for an economic task

Do not use for:

  • field-RCT program evaluation design (rct-core-design)
  • survey experiments embedded in opinion polling without monetary incentives (conjoint-experiment, survey-research-architect)
  • game-theoretic model derivation itself without an empirical test plan

Core Workflow

Step 1 — State the Theoretical Prediction(s) Being Tested

Write down the competing point predictions (e.g., "the standard self-interest model predicts X; a fairness/reciprocity model predicts Y") the experiment is built to distinguish. An experiment without a model to test against is a demonstration, not a test — design the treatments so that the competing theories make different, observable predictions.

Step 2 — Choose the Game/Task and Treatment Structure

  • Select the canonical paradigm that operationalizes the construct (ultimatum/dictator game for fairness, public-goods game for cooperation, trust game for trust/reciprocity, first-price/second-price auction for bidding behavior, BDM mechanism for willingness-to-pay elicitation).
  • Decide between-subject (cleaner causal identification, no spillover across conditions, needs more subjects) vs. within-subject (more statistical power per subject, but risks order effects and demand effects — randomize order and test for it).
  • If multiple treatments, map each to the specific model comparison from Step 1; do not add treatments that do not sharpen a theoretical contrast.

Step 3 — Design the Incentive Structure

  • Apply induced-value theory: subjects' payoffs must be monotonically tied to the decisions being studied, salient (paid in real money, not hypothetical "points" unless explicitly studying hypothetical-bias), and dominant (large enough relative to subjects' alternative use of time to matter).
  • Decide the payment rule: pay every decision, or randomly select one decision per subject to pay (random incentive/lottery system) — the latter preserves incentive compatibility across multiple within-subject tasks at lower cost; document why it does when used.
  • Set a show-up fee that does not swamp the treatment-relevant payoff variation, and state currency/PPP-adjusted amounts for cross-country comparisons.

Step 4 — Decide on Deception and Clear It Ethically

Economics experiments conventionally prohibit deception (unlike some psychology traditions) because a participant pool that learns it has been deceived distrusts instructions in all future sessions, contaminating the subject pool for every other researcher using it. Default to no deception; if a design element could be perceived as deceptive (e.g., a non-strategic confederate/robot player), disclose this to subjects before or immediately after the session per field norms, and document the justification for IRB. Pair with rct-ethics-registration for the registration and informed-consent documentation mechanics.

Step 5 — Build the Session Protocol

  • Write the instructions subjects actually read: neutral framing (avoid loaded terms like "partner" vs. "opponent" unless framing is the treatment itself), comprehension-check questions before play, and a fixed script for the experimenter to avoid uncontrolled experimenter-demand variation across sessions.
  • Decide group/matching protocol: fixed partners, random re-matching each round (stranger matching), or perfect strangers (no repeat matching) — this affects whether reputation/repeated-game effects can contaminate a one-shot-game inference.
  • Pre-register the number of rounds, matching protocol, and primary outcome before data collection.

Step 6 — Choose and Configure the Platform

  • oTree (Python/Django-based): best for multiplayer interactive games with real-time matching; open source, runs in-lab or online.
  • z-Tree: the long-standing lab standard, Pascal-based, strong at complex market/auction experiments; in-lab only (no native online mode).
  • Qualtrics/LIONESS Lab: adequate for single-player decision tasks (risk elicitation, individual choice) without real-time multiplayer interaction; not suitable for synchronous bargaining games.
  • Prolific/CloudResearch/MTurk for online subject recruitment: specify screening criteria, attention checks, and bot/inattentive-respondent exclusion rules set before data collection.

Step 7 — Plan Attrition, Comprehension Failures, and Exclusions

State in advance how subjects who fail comprehension checks, drop out mid-session, or show evidence of inattention (online) will be handled — excluded, re-shown instructions, or retained with a robustness check both ways. Deciding this after seeing the data invites p-hacking.

Prerequisites

  • A specific behavioral or game-theoretic hypothesis with competing point predictions, not just "see if people cooperate."
  • A funded or budgeted subject-payment plan (induced-value incentives, show-up fees); experiments without real monetary stakes should explicitly flag the hypothetical-bias risk.
  • IRB/ethics approval appropriate to the institution before running human subjects, and a lab or online-panel account (oTree/z-Tree server, or a Prolific/MTurk account) to field the study.

Limitations

  • Lab behavior may not generalize to naturalistic field settings (external validity); lab-in-field designs mitigate but do not eliminate this gap — state the intended scope of generalization explicitly.
  • Online samples (Prolific/MTurk) differ demographically from traditional undergraduate lab pools and from the general population; do not claim population-representative estimates without a matched or weighted sample.
  • This skill does not implement the statistical analysis of resulting data; pair with rct-data-analysis or ols-regression for that, and with rct-power-calculations for a priori sample-size planning.
  • Platform-specific quirks (oTree version compatibility, z-Tree Windows-only lab requirement) change over time; verify current software documentation before finalizing the technical build.

Quality Checklist

  • Competing theoretical predictions stated before the design is built
  • Incentive structure satisfies monotonicity, salience, dominance
  • Deception avoided, or explicitly justified and disclosed per field norms
  • Matching protocol (fixed/random/stranger) stated and justified
  • Comprehension checks and exclusion rules pre-registered, not decided post hoc
  • Platform choice matches the interaction structure (multiplayer real-time vs. individual choice)

示例提示词

我想设计一个激励相容的最后通牒博弈实验,用oTree在线收集数据。

所属插件

经济金融与管理研究面向经济、金融与管理学研究:DID/IV/RDD/合成控制、面板与时间序列计量、事件研究与因子模型、期权与组合、管理学论文写作、NSFC 申请与质性/混合方法。

安装该插件会一并获得同领域的 40 个 Skill / MCP。

同插件的其他科研 Skill

常见问题

实验经济学设计能做什么?

Designs incentivized lab, online, or lab-in-field experiments: treatment and payoff design, deception/IRB norms, and oTree/z-Tree setup. Use when designing a behavioral-economics study。

如何安装实验经济学设计?

在科搭 Codar 左侧点击「插件」,在「组件」页签搜索「实验经济学设计」并安装;在 Claude Code 或 Codex 中可以发送本页的安装提示词,或下载 Skill 包后解压到 ~/.claude/skills 或 ~/.codex/skills。

实验经济学设计需要哪些权限?

该科研 Skill没有声明额外权限;执行具体操作时仍由 Agent 按自身规则请求确认。

实验经济学设计经过官方审核吗?

是。SpeedSkills 收录的每个版本都会记录内容摘要与权限声明,当前版本已通过安全审核。