soc-user-research

asgard-ai-platform ｜ 規劃與釐清 ｜ 原文 ｜ MIT

依問題選訪談、焦點團體、問卷或現場觀察來做使用者研究

---
name: "soc-user-research"
description: "用訪談、焦點團體、問卷和現場觀察來設計並執行使用者研究。使用者需要了解客戶需求、驗證產品假設、蒐集質性洞察，或設計研究時使用，就算他們說的是「我們需要跟使用者聊聊」、「我們怎麼驗證這個想法」或「我們的客戶實際上怎麼想」也一樣。"
metadata:
  category: "WP-18 社會科學院"
  tags: ["social-science", "user-research", "qualitative-methods"]
---

# 使用者研究方法

## 概述

使用者研究是透過直接的觀察和對話，而不是假設，來找出人們需要什麼、想要什麼、做什麼。這個 skill 涵蓋四種核心方法：深度訪談、焦點團體、問卷和現場觀察，以及各自該在什麼時候用。

## 框架

```
IRON LAW: Method Follows Question

Choose the method based on what you need to learn, not what's convenient.
"Why" questions → Interviews/Observation (qualitative)
"How many" questions → Surveys (quantitative)
"What do they really do" → Observation (behavioral)

Running a survey to answer "why" produces misleading data.
```

### 方法選擇

| 方法 | 最適合 | 樣本 | 深度 | 成本 |
|--------|---------|--------|-------|------|
| **深度訪談** | 了解動機、痛點、心智模型 | 8-15 人 | 非常高 | 中 |
| **焦點團體** | 探索反應、產生點子、社會動態 | 每組 6-10 人，2-3 組 | 中 | 中 |
| **問卷** | 大規模衡量普及程度、偏好、人口統計 | 100 份以上的回覆 | 低 | 低-中 |
| **現場觀察** | 了解情境中的實際行為（不是自己回報的） | 5-10 場 | 非常高 | 高 |

### 深度訪談指南

1. **暖身**：建立融洽關係（2 分鐘）：「跟我聊聊你的角色／你的一天」
2. **情境**：了解他們的世界（5 分鐘）：「帶我走一遍你上次…的經過」
3. **核心問題**：探索主題（20 分鐘）：開放式、不用引導式問題
4. **追問**：對有趣的線索深入挖掘：「多說一點」、「為什麼？」
5. **收尾**：總結並確認（3 分鐘）：「我理解得對嗎，你的意思是…？」

**規則**：
- 問過去的行為，不要問假設的未來（問「你當時做了什麼？」，不要問「你會怎麼做？」）
- 絕對不要問「你會用這個嗎？」，人在預測自己的行為上糟透了
- 沉默是一種工具，讓他們自己把空白填滿

### 問卷設計

1. 從篩選問題開始（確認受訪者資格）
2. 從一般到具體
3. 敏感的／人口統計的問題放最後
4. 限制在 15-20 題（5-7 分鐘填完）
5. 盡可能使用已驗證的量表（Likert、NPS、SUS）

**要避免的題型**：
- 雙重問題：「這個產品快速而且可靠嗎？」（是指哪一個？）
- 引導式：「你不覺得我們的 app 很容易用嗎？」
- 假設式：「你願意為這個功能每月付 $10 嗎？」

### 分析

**質性**（訪談、觀察）：
- 親和圖：把觀察結果分組成主題
- 在 5 位以上的受訪者之間找模式
- 逐字引用，不要轉述

**量化**（問卷）：
- 先看描述統計（平均數、分布）
- 依區隔做交叉分析
- 比較時看統計顯著性（p < 0.05）

## 輸出格式

```markdown
# User Research Plan: {Project}

## Research Questions
1. {what we need to learn}

## Method
- Type: {interview / focus group / survey / observation}
- Rationale: {why this method for this question}
- Sample: {who, how many, recruitment criteria}
- Timeline: {duration}

## Discussion Guide / Survey Instrument
{Key questions or survey structure}

## Analysis Plan
{How findings will be synthesized}
```

## 範例

### 正確的應用
**情境：** 了解使用者為什麼在結帳時放棄一個外送 app
- **方法**：深度訪談（需要了解「為什麼」，不是「有多少」）
- **樣本**：過去 30 天內放棄結帳的 10 位使用者（透過 app 內資料招募）
- **關鍵問題**：「帶我走一遍你上一筆沒有完成的訂單。發生了什麼事？」（行為性、過去式、開放式 ✓）

### 錯誤的應用
- 問卷問「如果我們取消外送費，你會完成訂單嗎？」→ 假設式。使用者會說會，但行為不一定會改變。應該觀察實際行為，或用真正的實驗來測試。

## 注意事項

- **5 位使用者能找出 85% 的可用性問題**（Nielsen）：對可用性測試來說，超過 5 位之後邊際效益遞減。要了解動機的話，需要 8-15 位。
- **自己回報的行為 ≠ 實際行為**：人們會高估自己吃得多健康、多常運動，以及願意付多少錢。觀察和行為資料 > 自己回報。
- **招募偏誤**：如果你招募的是「我們 app 的使用者」，就會漏掉非使用者和已流失的使用者。要小心定義母體。
- **訪談者偏誤**：訪談者的反應（點頭、「太棒了！」）會影響回答。要保持中立。
- **問卷衡量的是你問的東西，不是重要的東西**：如果你沒想到要問某個痛點，問卷就不會揭露它。要先用質性研究找出對的問題。

## 參考資料

- 訪談腳本範本請看 `references/interview-templates.md`
- 問卷設計最佳實務請看 `references/survey-design.md`

---
name: "soc-user-research"
description: "Design and conduct user research using interviews, focus groups, surveys, and field observation. Use this skill when the user needs to understand customer needs, validate product assumptions, gather qualitative insights, or design a research study — even if they say 'we need to talk to users', 'how do we validate this idea', or 'what do our customers actually think'."
metadata:
  category: "WP-18 社會科學院"
  tags: ["social-science", "user-research", "qualitative-methods"]
---

# User Research Methods

## Overview

User research uncovers what people need, want, and do — through direct observation and conversation, not assumptions. This skill covers four core methods: depth interviews, focus groups, surveys, and field observation, plus when to use each.

## Framework

```
IRON LAW: Method Follows Question

Choose the method based on what you need to learn, not what's convenient.
"Why" questions → Interviews/Observation (qualitative)
"How many" questions → Surveys (quantitative)
"What do they really do" → Observation (behavioral)

Running a survey to answer "why" produces misleading data.
```

### Method Selection

| Method | Best For | Sample | Depth | Cost |
|--------|---------|--------|-------|------|
| **Depth Interview** | Understanding motivations, pain points, mental models | 8-15 people | Very high | Medium |
| **Focus Group** | Exploring reactions, generating ideas, social dynamics | 6-10 per group, 2-3 groups | Medium | Medium |
| **Survey** | Measuring prevalence, preferences, demographics at scale | 100+ responses | Low | Low-Med |
| **Field Observation** | Understanding actual behavior in context (not self-reported) | 5-10 sessions | Very high | High |

### Depth Interview Guide

1. **Warm-up**: Build rapport (2 min) — "Tell me about your role/day"
2. **Context**: Understand their world (5 min) — "Walk me through the last time you..."
3. **Core questions**: Explore the topic (20 min) — Open-ended, no leading questions
4. **Probing**: Go deeper on interesting threads — "Tell me more about that", "Why?"
5. **Wrap-up**: Summarize and confirm (3 min) — "Did I understand correctly that...?"

**Rules**:
- Ask about past behavior, not hypothetical future ("What did you do?" not "What would you do?")
- Never ask "Would you use this?" — people are terrible at predicting their own behavior
- Silence is a tool — let them fill the gap

### Survey Design

1. Start with screening questions (qualify respondents)
2. Move from general to specific
3. Put sensitive/demographic questions last
4. Limit to 15-20 questions (5-7 min completion)
5. Use validated scales where possible (Likert, NPS, SUS)

**Question types to avoid**:
- Double-barreled: "Is the product fast and reliable?" (which one?)
- Leading: "Don't you think our app is easy to use?"
- Hypothetical: "Would you pay $10/month for this feature?"

### Analysis

**Qualitative** (interviews, observation):
- Affinity mapping: Group observations into themes
- Look for patterns across 5+ participants
- Quote verbatim — don't paraphrase

**Quantitative** (surveys):
- Descriptive stats first (means, distributions)
- Cross-tabulate by segments
- Statistical significance for comparisons (p < 0.05)

## Output Format

```markdown
# User Research Plan: {Project}

## Research Questions
1. {what we need to learn}

## Method
- Type: {interview / focus group / survey / observation}
- Rationale: {why this method for this question}
- Sample: {who, how many, recruitment criteria}
- Timeline: {duration}

## Discussion Guide / Survey Instrument
{Key questions or survey structure}

## Analysis Plan
{How findings will be synthesized}
```

## Examples

### Correct Application
**Scenario:** Understanding why users abandon a food delivery app at checkout
- **Method**: Depth interviews (need to understand "why", not "how many")
- **Sample**: 10 users who abandoned in the last 30 days (recruit via in-app data)
- **Key question**: "Walk me through your last order that you didn't complete. What happened?" (behavioral, past-tense, open-ended ✓)

### Incorrect Application
- Survey asking "Would you complete your order if we removed the delivery fee?" → Hypothetical. Users will say yes but behavior may not change. Should observe actual behavior or test with a real experiment.

## Gotchas

- **5 users find 85% of usability problems** (Nielsen): For usability testing, diminishing returns after 5. For understanding motivations, need 8-15.
- **Self-reported behavior ≠ actual behavior**: People overestimate how healthy they eat, how often they exercise, and how much they'd pay. Observation and behavioral data > self-report.
- **Recruitment bias**: If you recruit "users of our app", you miss non-users and churned users. Define the population carefully.
- **Interviewer bias**: The interviewer's reactions (nodding, "great!") influence responses. Stay neutral.
- **Surveys measure what you ask, not what matters**: If you didn't think to ask about a pain point, the survey won't reveal it. Use qualitative research first to discover the right questions.

## References

- For interview script templates, see `references/interview-templates.md`
- For survey design best practices, see `references/survey-design.md`