econ-behavioral
用行為經濟學的偏誤和推力設計來影響決策
中文翻譯
---
name: "econ-behavioral"
description: "運用行為經濟學的概念（有限理性、展望理論、心理帳戶、推力理論）分析決策偏誤。當使用者需要理解人們為什麼做出不理性的經濟決策、設計選擇架構，或運用推力（nudge）影響行為時使用。就算使用者只說「為什麼客戶老是做出不好的選擇」、「怎麼鼓勵大家多存錢」或「設計一個更好的預設選項」也要用。"
metadata:
category: "WP-17 經濟學院"
tags: ["economics", "behavioral-economics", "decision-making"]
---
# 行為經濟學
## 概述
行為經濟學研究心理因素如何讓人偏離理性的經濟預測。古典經濟學假設行為者是理性的，行為經濟學則記錄系統性的偏誤，並設計介入措施（推力）來改善決策。
## 框架
```
IRON LAW: Biases Are Systematic, Not Random
Behavioral biases are PREDICTABLE patterns, not noise. Loss aversion
doesn't sometimes make people risk-seeking and sometimes not — it
consistently makes people overweight losses relative to equivalent gains
(roughly 2:1 ratio). Use specific bias names and their documented effects,
not vague "people are irrational."
```
### 核心概念
**有限理性（Bounded Rationality）**（Simon）：因為認知資源有限，人們會「滿意即可」（找「夠好」的），而不是最佳化。
**展望理論（Prospect Theory）**（Kahneman & Tversky）：
- **損失趨避**：損失帶來的痛，大約是等量獲得帶來的快樂的 2 倍
- **參考點依賴**：人們是相對於參考點來評價結果，不是用絕對值
- **邊際敏感度遞減**：從 $0 到 $100 的差別，感覺比 $1000 到 $1100 的差別大
**心理帳戶（Mental Accounting）**（Thaler）：人們會把錢分成心裡的「桶子」（房租、娛樂、儲蓄），並區別對待，違反了金錢的可替代性。
**框架效應（Framing Effect）**：同樣的資訊用不同方式呈現，會導致不同的決定。「90% 存活率」vs「10% 死亡率」，同一個事實，不同的選擇。
### 商業應用的關鍵偏誤
| 偏誤 | 定義 | 商業應用 |
|------|-----------|---------------------|
| **錨定效應（Anchoring）** | 先看到的數字會影響後續的估計 | 在折扣前先秀出較高的「原價」 |
| **預設效應（Default effect）** | 人們會堅持預先選好的選項 | 訂閱、器官捐贈用 opt-out 比 opt-in 好 |
| **社會認同（Social proof）** | 人們會跟著別人做 | 「1,000 位以上客戶選了這個方案」 |
| **稀缺性（Scarcity）** | 供應有限會提高感知價值 | 「庫存只剩 3 件」 |
| **稟賦效應（Endowment effect）** | 人們會高估自己已擁有的東西 | 免費試用讓取消感覺像損失 |
| **現時偏誤（Present bias）** | 人們會過度重視眼前的獎勵，輕忽未來的 | 「今天免費開始」> 「12 個月內幫你省錢」 |
| **沉沒成本謬誤（Sunk cost fallacy）** | 過去的投入影響未來的決定（本來不該） | 「我已經看了 2 小時，應該把電影看完」 |
| **現狀偏誤（Status quo bias）** | 偏好現狀、不喜歡改變 | 既有客戶就算有更好的選擇也很少轉換 |
### 推力設計框架（Thaler & Sunstein）
有效推力的 **EAST 框架**：
- **Easy（容易）**：降低阻力。簡化表單、預先填入資料、減少步驟。
- **Attractive（吸引人）**：讓期望的行動在視覺上醒目、有吸引力。
- **Social（社會性）**：顯示別人在做什麼。同儕比較、使用者見證。
- **Timely（及時）**：在做決定的當下送出推力，不要太早或太晚。
### 分析步驟
1. **找出決策情境**：使用者/客戶正在做什麼選擇？
2. **對應相關偏誤**：哪些系統性偏誤可能在起作用？
3. **評估目前的選擇架構**：這個決策現在是怎麼呈現的？
4. **設計介入措施**：用 EAST 框架運用推力
5. **測試**：拿介入措施跟目前的設計做 A/B 測試
## 輸出格式
```markdown
# Behavioral Analysis: {Decision Context}
## Decision Context
- Decision-maker: {who}
- Choice: {what they're deciding}
- Current behavior: {what they typically do}
- Desired behavior: {what we want them to do}
## Biases Identified
| Bias | How It Manifests | Impact |
|------|-----------------|--------|
| {bias} | {specific manifestation} | H/M/L |
## Current Choice Architecture
{How the decision is currently structured and why it triggers biases}
## Proposed Nudges
| Nudge | EAST Principle | Expected Effect |
|-------|---------------|----------------|
| {intervention} | Easy/Attractive/Social/Timely | {predicted change} |
## Testing Plan
- Control: {current design}
- Treatment: {nudged design}
- Metric: {conversion rate / opt-in rate / etc.}
- Sample size: {N}
```
## 範例
### 正確的做法
**情境：** 提高一家台灣公司的退休儲蓄參與率
**起作用的偏誤：**
- **現狀偏誤**：員工不加入，因為要主動 opt in
- **現時偏誤**：退休是幾十年後的事，現在花錢感覺比較急
- **損失趨避**：每月從薪水扣款感覺像損失
**推力設計：**
| 推力 | 原則 | 介入措施 |
|-------|-----------|-------------|
| 自動加入 | **Easy**（預設） | 從 opt-in 改成 opt-out（預設提撥 3%） |
| 逐步提高 | **Timely** | 「每次年度調薪時，提撥比例增加 1%」，時間點配合加薪，讓扣款不會感覺像損失 |
| 社會認同 | **Social** | 「你 78% 的同事都有參加退休計畫」 |
**預測效果**：光是自動加入，通常就能讓參與率從約 30% 提高到約 90%（文獻上有充分記錄）✓
### 錯誤的做法
- 「人是不理性的，所以我們應該操縱他們」→ 行為經濟學找出的是系統性模式，不是隨機的不理性。推力應該幫助人們做出符合「他們自己」所述目標的決定，不是違背他們的利益去操縱。違反 Iron Law 和倫理原則。
## 常見陷阱
- **推力是自由意志家長主義**：它保留選擇權，同時引導人走向更好的結果。如果推力把選擇拿掉了，那就不是推力，是強制命令。
- **偏誤會互相作用**：損失趨避 + 錨定 + 框架可以疊加。「省下 NT$300」（獲得框架）vs「別再每個月損失 NT$300」（損失框架 + 錨定），後者因為偏誤複合而更強。
- **文化差異**：有些偏誤因文化而異。社會認同在集體主義文化（台灣、日本）比個人主義文化強。要依情境校準。
- **推力疲勞**：同時用太多推力會降低效果。優先選影響最大的那個。
- **倫理界線**：利用偏誤賣人們不需要的東西（dark pattern）是剝削，不是推力。檢驗標準：如果對方知道這個推力，會感謝你嗎？
## 參考資料
- 展望理論的數學，見 `references/prospect-theory.md`
- dark pattern 與合乎倫理的推力，見 `references/ethics-of-nudging.md`英文原文
---
name: "econ-behavioral"
description: "Apply behavioral economics concepts including bounded rationality, prospect theory, mental accounting, and nudge theory to analyze decision-making biases. Use this skill when the user needs to understand why people make irrational economic decisions, design choice architectures, or apply nudges to influence behavior — even if they say 'why do customers make bad choices', 'how do we encourage people to save more', or 'design a better default option'."
metadata:
category: "WP-17 經濟學院"
tags: ["economics", "behavioral-economics", "decision-making"]
---
# Behavioral Economics
## Overview
Behavioral economics studies how psychological factors cause people to deviate from rational economic predictions. Where classical economics assumes rational actors, behavioral economics documents systematic biases and designs interventions (nudges) to improve decisions.
## Framework
```
IRON LAW: Biases Are Systematic, Not Random
Behavioral biases are PREDICTABLE patterns, not noise. Loss aversion
doesn't sometimes make people risk-seeking and sometimes not — it
consistently makes people overweight losses relative to equivalent gains
(roughly 2:1 ratio). Use specific bias names and their documented effects,
not vague "people are irrational."
```
### Core Concepts
**Bounded Rationality** (Simon): People satisfice (find "good enough") rather than optimize because cognitive resources are limited.
**Prospect Theory** (Kahneman & Tversky):
- **Loss aversion**: Losses hurt ~2x more than equivalent gains feel good
- **Reference dependence**: People evaluate outcomes relative to a reference point, not in absolute terms
- **Diminishing sensitivity**: The difference between $0 and $100 feels larger than between $1000 and $1100
**Mental Accounting** (Thaler): People categorize money into mental "buckets" (rent, fun, savings) and treat them differently, violating fungibility.
**Framing Effect**: Same information presented differently leads to different decisions. "90% survival rate" vs "10% mortality rate" — same fact, different choices.
### Key Biases for Business Application
| Bias | Definition | Business Application |
|------|-----------|---------------------|
| **Anchoring** | First number seen influences subsequent estimates | Show high "original price" before discount |
| **Default effect** | People stick with the pre-selected option | Opt-out > opt-in for subscriptions, organ donation |
| **Social proof** | People follow what others do | "1,000+ customers chose this plan" |
| **Scarcity** | Limited availability increases perceived value | "Only 3 left in stock" |
| **Endowment effect** | People overvalue what they already own | Free trials make cancellation feel like a loss |
| **Present bias** | People overweight immediate rewards vs future | "Start free today" > "Save money over 12 months" |
| **Sunk cost fallacy** | Past investments influence future decisions (shouldn't) | "I've already watched 2 hours, I should finish the movie" |
| **Status quo bias** | Preference for current state over change | Existing customers rarely switch, even when better options exist |
### Nudge Design Framework (Thaler & Sunstein)
**EAST Framework** for effective nudges:
- **Easy**: Reduce friction. Simplify forms, pre-fill data, reduce steps.
- **Attractive**: Make the desired action visually prominent and appealing.
- **Social**: Show what others are doing. Peer comparisons, testimonials.
- **Timely**: Deliver the nudge at the moment of decision, not before or after.
### Analysis Steps
1. **Identify the decision context**: What choice is the user/customer making?
2. **Map relevant biases**: Which systematic biases are likely at play?
3. **Evaluate current choice architecture**: How is the decision currently presented?
4. **Design interventions**: Apply nudges using EAST framework
5. **Test**: A/B test the intervention against the current design
## Output Format
```markdown
# Behavioral Analysis: {Decision Context}
## Decision Context
- Decision-maker: {who}
- Choice: {what they're deciding}
- Current behavior: {what they typically do}
- Desired behavior: {what we want them to do}
## Biases Identified
| Bias | How It Manifests | Impact |
|------|-----------------|--------|
| {bias} | {specific manifestation} | H/M/L |
## Current Choice Architecture
{How the decision is currently structured and why it triggers biases}
## Proposed Nudges
| Nudge | EAST Principle | Expected Effect |
|-------|---------------|----------------|
| {intervention} | Easy/Attractive/Social/Timely | {predicted change} |
## Testing Plan
- Control: {current design}
- Treatment: {nudged design}
- Metric: {conversion rate / opt-in rate / etc.}
- Sample size: {N}
```
## Examples
### Correct Application
**Scenario:** Increasing retirement savings enrollment in a Taiwanese company
**Biases at play:**
- **Status quo bias**: Employees don't enroll because they'd have to actively opt in
- **Present bias**: Retirement is decades away; spending now feels more urgent
- **Loss aversion**: Monthly salary deduction feels like a loss
**Nudge design:**
| Nudge | Principle | Intervention |
|-------|-----------|-------------|
| Auto-enrollment | **Easy** (default) | Change from opt-in to opt-out (3% default contribution) |
| Escalation | **Timely** | "Increase contribution by 1% at each annual raise" — timed to coincide with salary increase so deduction doesn't feel like a loss |
| Social proof | **Social** | "78% of your colleagues contribute to the retirement plan" |
**Predicted effect**: Auto-enrollment alone typically increases participation from ~30% to ~90% (well-documented in literature) ✓
### Incorrect Application
- "People are irrational, so we should manipulate them" → Behavioral economics identifies systematic patterns, not random irrationality. Nudges should help people make decisions aligned with their OWN stated goals, not manipulate against their interests. Violates Iron Law and ethical principles.
## Gotchas
- **Nudges are libertarian paternalism**: They preserve choice while steering toward better outcomes. If the nudge removes choice, it's not a nudge — it's a mandate.
- **Biases interact**: Loss aversion + anchoring + framing can combine. "Save NT$300" (gain frame) vs "Stop losing NT$300/month" (loss frame + anchoring) — the latter is stronger due to compounding biases.
- **Cultural variation**: Some biases vary across cultures. Social proof is stronger in collectivist cultures (Taiwan, Japan) than individualist ones. Calibrate for context.
- **Nudge fatigue**: Too many nudges simultaneously reduce effectiveness. Prioritize the highest-impact one.
- **Ethical boundary**: Using biases to sell products people don't need (dark patterns) is exploitation, not nudging. The test: would the person thank you for the nudge if they knew about it?
## References
- For prospect theory mathematics, see `references/prospect-theory.md`
- For dark patterns vs ethical nudges, see `references/ethics-of-nudging.md`