引用
Three Common Myths in Quantitative Social Research
作者：譚康榮（Tony Tam） | 首次發表於 2020-07-06 | 第 26 期 December 2001
DOI：https://dx.doi.org/10.6786/TJS.200112.0251
研究紀要（Research Notes）
DOI：https://dx.doi.org/10.6786/TJS.200112.0251
研究紀要（Research Notes）
論文資訊 | Article information
摘要 Abstract
統計顯著檢定和多變項模型對量化的社會研究具有革命性的貢獻。然而，即使是偉大的工具也有可能被誤用。本篇研究紀要討論量化研究中對於這兩個基本工具，最爲常見的三類誤用：（1）對統計顯著檢定的誤解；（2）錯誤地依賴淨效果做爲推論的基礎；以及（3）對於控制變項的過度信任。對於每一種誤用，我指出其觀念上的迷思和誤解的來源，並提出補救之道。我會以大學畢業生的虛擬數據資料和近年來有關美國勞力市場性別不平等的辯論爲例，閻明其間的問題以及補救方式。本文所指出的問題與補救方；對直用於任何以統計模型爲基礎的多變項分析（例如OLS迴歸、對數迴歸、tobit、其他受限依變項迴歸分析、以及相聯方程式模型等） 。
關鍵字：統計顯著檢定、多變項模型、統計控制、測量誤差
關鍵字：統計顯著檢定、多變項模型、統計控制、測量誤差
Significance testing and multivariate models are revolutionary tools for quantitative social research. However, even great tools have the potential for misuse. This paper discusses three common ways in which these fundamental tools have been misused: (1) misinterpreting significance tests, (2) a misplaced focus on net effects, and (3) overconfidence in the control variables. For each misuse, I identify a conceptual myth, specify the source of the misconception, and suggest a remedy. In addition, numerical data on college graduates and a recent debate over gender inequality in the American labor market illustrate these problems and remedies. The problems and remedies are applicable to all kinds of multivariate analysis based on statistical modeling (eg, OLS regression, logit, tobit, other limited dependent variable regressions, simultaneous equations models, and so forth).
Keywords: Significance Test, Multivariate Model, Statistical Control, Measurement Error
Keywords: Significance Test, Multivariate Model, Statistical Control, Measurement Error
- 關鍵詞:
