grad-survey-design
設計問卷的構念操作化、量表開發、信效度和共同方法變異控制
中文翻譯
--- name: "grad-survey-design" description: "套用嚴謹的問卷設計原則，包括構念操作化、Likert 量表開發、信效度評估與共同方法變異控制。當使用者在設計問卷、開發測量題項、需要評估 Cronbach's alpha 或 AVE，或問「這個構念怎麼操作化」「我的量表可靠嗎」「怎麼控制 CMV」時使用。" metadata: category: "WP-31 量化方法" tags: ["survey-design", "questionnaire", "Likert-scale", "reliability", "validity", "CMV", "Cronbachs-alpha"] --- # 問卷設計 (Survey Design) ## 概述 問卷設計是透過系統化的操作化、量表開發與心理計量驗證，把理論構念轉換成可測量的題項。嚴謹的問卷能確保觀察到的分數，可靠而有效地代表想測的構念，同時控制共同方法變異這類方法造成的假象。 ## 何時使用 - 測量感知、態度、信念或行為意圖 - 把理論架構中的潛在構念操作化 - 開發或改編多題項的 Likert 量表 - 規劃依賴自陳式資料的量化研究 ## 何時不要使用 - 有客觀行為資料或檔案資料可用，而且更合適 - 這個構念用實驗或觀察來測量比較好 - 母體無法透過問卷觸及（識字率極低、沒有抽樣架構） - 研究問題是探索性的，構念還沒有明確定義 ## 假設 ``` 鐵律：問卷測量的是「感知」，不是客觀現實，而且當預測變數和效標 來自同一個來源時，共同方法變異會膨脹相關係數。 ``` 關鍵假設： 1. 受訪者對題項的理解與原意一致（語意等值） 2. 回答是誠實的，不會被社會期許系統性地扭曲 3. 這些題項對構念領域有足夠的取樣 4. 同一量表內的題項，是同一個潛在構念的反映性指標 ## 方法 ### 步驟 1：構念操作化 從理論界定每個構念的概念領域。指明構面與子構面。從文獻、專家判斷與質性資料產生題項池（每個構面至少 3-5 題）。 ### 步驟 2：量表設計與預試 選擇回應格式（5 點或 7 點 Likert）。避免雙重題幹、誘導式或含糊的題項。進行認知訪談或專家小組審查。以 N ≥ 30 進行前測。 ### 步驟 3：評估信度與效度 信度：Cronbach's alpha ≥ 0.70、組合信度（CR）≥ 0.70。收斂效度：AVE ≥ 0.50、因素負荷量 ≥ 0.60。區別效度：Fornell-Larcker 準則或 HTMT < 0.90。公式見 `references/`。 ### 步驟 4：控制共同方法變異 程序上的補救：在時間上把預測變數和效標分開、使用不同的量表格式、保證匿名。統計上的補救：Harman 單因子檢定（必要但不充分）、標記變數技術、含共同方法因子的 CFA。 ## 輸出格式 ```markdown ## 問卷設計：[研究標題] ### 構念操作化 | 構念 | 構面 | 題項 | 來源 | |-----------|-----------|-------|--------| | [名稱] | [構面] | x 題 | [改編自] | ### 信度評估 | 構念 | 題項 | Cronbach's α | CR | AVE | |-----------|-------|-------------|-----|-----| | [名稱] | x | x.xx | x.xx | x.xx | ### 效度評估 | 檢定 | 結果 | 門檻 | 評估 | |------|--------|-----------|------------| | 因素負荷量（最小值） | x.xx | ≥ 0.60 | [通過/未通過] | | AVE | x.xx | ≥ 0.50 | [通過/未通過] | | HTMT（最大值） | x.xx | < 0.90 | [通過/未通過] | ### CMV 控制 | 補救措施 | 類型 | 結果 | |--------|------|--------| | [補救措施] | [程序/統計] | [發現] | ### 限制 - [記下任何違反假設之處] ``` ## 常見陷阱 - Cronbach's alpha 是信度的下限，而且假設 tau 等價，建議用 CR - 信度高但效度低，代表你很精準地在測錯東西 - 反向計分的題項能降低默從偏誤，但在 CFA 中常常形成人為的方法因子 - Harman 單因子檢定被廣泛使用，但偵測 CMV 的檢定力非常低 - 翻譯與回譯不保證跨文化的測量不變性 - 回收率低於 30%，即使樣本數夠大，也會引發無回應偏誤的疑慮 ## 參考資料 - DeVellis, R. F. (2017). *Scale Development: Theory and Applications* (4th ed.). Sage. - Podsakoff, P. M., MacKenzie, S. B., Lee, J.-Y., & Podsakoff, N. P. (2003). Common method biases in behavioral research. *Journal of Applied Psychology*, 88(5), 879-903. - Hair, J. F., Black, W. C., Babin, B. J., & Anderson, R. E. (2019). *Multivariate Data Analysis* (8th ed.). Cengage.
英文原文
--- name: "grad-survey-design" description: "Apply rigorous survey design principles including construct operationalization, Likert scale development, reliability and validity assessment, and common method variance control. Use this skill when the user designs questionnaires, develops measurement items, needs to evaluate Cronbach's alpha or AVE, or when they ask 'how do I operationalize this construct', 'is my scale reliable', or 'how do I control for CMV'." metadata: category: "WP-31 量化方法" tags: ["survey-design", "questionnaire", "Likert-scale", "reliability", "validity", "CMV", "Cronbachs-alpha"] --- # 問卷設計 (Survey Design) ## Overview Survey design translates theoretical constructs into measurable items through systematic operationalization, scale development, and psychometric validation. Rigorous surveys ensure that observed scores reliably and validly represent the intended constructs while controlling for method artifacts such as common method variance. ## When to Use - Measuring perceptions, attitudes, beliefs, or behavioral intentions - Operationalizing latent constructs from a theoretical framework - Developing or adapting multi-item Likert scales - Planning a quantitative study that relies on self-report data ## When NOT to Use - Objective behavioral data or archival data are available and more appropriate - The construct is better measured through experiments or observations - Population is unreachable via survey (extremely low literacy, no sampling frame) - Research question is exploratory and constructs are not yet well-defined ## Assumptions ``` IRON LAW: A survey measures PERCEPTIONS, not objective reality — and common method variance inflates correlations when predictor and criterion come from the same source. ``` Key assumptions: 1. Respondents understand items as intended (semantic equivalence) 2. Responses are honest and not systematically biased by social desirability 3. The construct domain is adequately sampled by the items 4. Items within a scale are reflective indicators of the same underlying construct ## Methodology ### Step 1 — Construct Operationalization Define each construct's conceptual domain from theory. Specify dimensions and sub-dimensions. Generate item pool from literature, expert judgment, and qualitative input (3-5 items per dimension minimum). ### Step 2 — Scale Design and Pretesting Choose response format (5-point or 7-point Likert). Avoid double-barreled, leading, or ambiguous items. Conduct cognitive interviews or expert panel review. Pilot test with N ≥ 30. ### Step 3 — Assess Reliability and Validity Reliability: Cronbach's alpha ≥ 0.70, composite reliability (CR) ≥ 0.70. Convergent validity: AVE ≥ 0.50, factor loadings ≥ 0.60. Discriminant validity: Fornell-Larcker criterion or HTMT < 0.90. See `references/` for formulas. ### Step 4 — Control for Common Method Variance Procedural remedies: separate predictor and criterion temporally, use different scale formats, guarantee anonymity. Statistical remedies: Harman's single-factor test (necessary but not sufficient), marker variable technique, CFA with common method factor. ## Output Format ```markdown ## Survey Design: [Study Title] ### Construct Operationalization | Construct | Dimensions | Items | Source | |-----------|-----------|-------|--------| | [name] | [dim] | x items | [adapted from] | ### Reliability Assessment | Construct | Items | Cronbach's α | CR | AVE | |-----------|-------|-------------|-----|-----| | [name] | x | x.xx | x.xx | x.xx | ### Validity Assessment | Test | Result | Threshold | Assessment | |------|--------|-----------|------------| | Factor loadings (min) | x.xx | ≥ 0.60 | [pass/fail] | | AVE | x.xx | ≥ 0.50 | [pass/fail] | | HTMT (max) | x.xx | < 0.90 | [pass/fail] | ### CMV Controls | Remedy | Type | Result | |--------|------|--------| | [remedy] | [procedural/statistical] | [finding] | ### Limitations - [Note any assumption violations] ``` ## Gotchas - Cronbach's alpha is a lower bound of reliability and assumes tau-equivalence; CR is preferred - High reliability with low validity means you are precisely measuring the wrong thing - Reverse-coded items reduce acquiescence bias but often form artifactual method factors in CFA - Harman's single-factor test is widely used but has very low power to detect CMV - Translation and back-translation do not guarantee measurement invariance across cultures - Response rate below 30% raises non-response bias concerns even with adequate sample size ## References - DeVellis, R. F. (2017). *Scale Development: Theory and Applications* (4th ed.). Sage. - Podsakoff, P. M., MacKenzie, S. B., Lee, J.-Y., & Podsakoff, N. P. (2003). Common method biases in behavioral research. *Journal of Applied Psychology*, 88(5), 879-903. - Hair, J. F., Black, W. C., Babin, B. J., & Anderson, R. E. (2019). *Multivariate Data Analysis* (8th ed.). Cengage.