心理計量假設對政治學實證研究之影響性評估

2005-08-012024-05-17https://scholars.lib.ntu.edu.tw/handle/123456789/694201摘要：政治學者以往多忽略測量理論背後的假設，導致在進行測量工作時沒有考慮到使用 不同量表對實證結果產生的差異。根據最近的一份研究顯示(黃旻華，2004)，「單面向性的假設」、「對原始資料量表的尺度假設」、「是否採信單項強度測量」、「是否主張相等的鑑別度」、「是否主張相等的難度」等五項心理計量學中的假設，都會對於迴歸分析結果有不可忽視的影響力，而即便兩量表間的相關係數高達0.995，這樣的結論仍然都成 立，這說明了理解這五項假設的重要性。然而黃文並沒有針對原始資料的性質來進行討論，使得上述發現可能只是因為黃文所用資料特性所致，而無法將這樣的結論套用在分析其它原始資料上。基於此點，本計劃有系統的操縱資料特性，包括「偏離常態分配程度」、「任意合併答項相鄰類別」、「原始資料的遺失值多寡」，來對於黃文的研究反覆進行模擬分析，觀察這五項假設的影響 力是否會隨資料性質而變化，並一般化原始資料特性如何改變這五項假設的影響力，提供未來政治學者在選擇測量量表時的參考。<br> Abstract: Political scientists usually pay little attention to the assumptions behind the measurement theory. This phenomenon leads them to conduct a measurement job without a due consideration to the possibility that applying different scales may result in different conclusions in a quantitative empirical analysis. According to a recent study reported by Huang (2004), the following five psychometric assumptions are all very important in deciding the result of regression analysis: “the assumption of unidimensionality”, “the metric assumption of the raw data’s scale”, “whether we count the information of attitudinal intensity from a single item”, “the assumption of the equal difficulty parameter”, and “the assumption of the equal discrimination parameter”. This conclusion stands even if two scales are highly correlated as high as 0.995. This finding highlights the importance of clear understanding on the five psychometric assumptions. Nevertheless, Huang (2004) has no discussion on the possibility that his finding is driven by certain statistical properties of the raw data. Therefore, we cannot rule out that Huang’s finding is only a special case under a particular set of data but not others and that not all of the five psychometric assumptions are influential all the time. In view of this possibility, this project will apply different methods of manipulating data properties of Huang’s study, including the degree of violating normal distribution, arbitrarily combining of adjacent response categories, and changing the proportion of missing data, to replicate his analysis through an intensive simulation study. Two research agenda are expected to be fulfilled in this project. First, a comprehensive understanding should be reached as to the relationship between different data properties and the relevance of the five psychometric assumptions. Second, a generalization of how data properties can affect the importance of the five assumptions should be concluded to provide a guideline for political scientists regarding how to decide what scales should be used.量表負載心理計量學迴歸分析模擬實驗Scale-LadenessPsychometricsRegression AnalysisSimulation Study心理計量假設對政治學實證研究之影響性評估