以羅序模式驗證貝克憂鬱量表第二版之心理計量特性

Abstract

貝克憂鬱量表第二版（Beck Depression Inventory-Second Edition, BDI-II）是研究與臨床上最廣為使用的憂鬱量表之ㄧ，研究者與臨床人員常以BDI-II總分為評量憂鬱嚴重度的指標。雖然已有許多研究文獻應用古典測驗理論的概念與方法驗證BDI-II的心理計量特性，但有四個問題仍未被有效地探討，包括：(1)BDI-II的21個項目是否符合單向度的建構（unidimensionality）？(2)在BDI-II的4分量表設計架構中，同一項目中的各選項是否可以適切地評量受試者的不同嚴重度？(3)BDI-II的項目和選項所反映的憂鬱程度，在憂鬱嚴重度的建構量尺上是否能均勻地分布且完整地涵蓋不同的憂鬱程度，而沒有明顯的間隙、上限或下限效應？以及(4)順序的（ordinal）BDI-II原始總分如何能有效地轉換為等距的（interval）數値，以提升評量結果所賦予的訊息和精準度？ 項目反應理論（item response theory）之羅序模式（Rasch model）以機率的數學模式為基礎，來呈現測量概念中受試者能力、項目難度和評量結果之間的關係。如果評量數據符合模式的假設，則羅序模式可以對項目難度和受試者能力做等距數值的估計，因此能有效地用以驗證量表之心理計量特性。本研究目的為應用羅序模式的概念與方法，驗證BDI-II的四個心理計量相關議題，包括(1)驗證BDI-II項目之單向度建構；(2)檢驗BDI-II各項目之選項是否恰當，(3)檢視BDI-II各項目和選項在憂鬱嚴重度之建構量尺上的分布情形，以及(4)將順序的BDI-II原始總分轉換為等距的羅序分數（logits）。 研究數據取自Steer等人的研究，樣本共包含260位診斷為重鬰症（major depressive disorder）之門診病患。資料分析方法採用羅序模式之部分得分模式（partial credit model）。結果顯示，BDI-II的所有項目均符合單向度建構，除了第21題（缺乏對性的興趣）外。由於該題所評量的症狀也未包含於DSM-IV所定義之憂鬱的診斷準則，因此在實證結果與理論建構的合併考量下，建議後續的使用者將之刪除，以確保BDI-II的評量結果可以精確地反映受試者的憂鬱程度，而未混雜其他概念。這20題符合模式假設的題BDI-II項目，發現其中5題的選項在區辨受試者於該項目上不同嚴重度的精確度不如預期，因此建議未來改版時修改這些選項的評分標準。不過，由這些選項所導致的評量誤差尚在可接受範圍內，因此現階段的研究與臨床使用上，應可暫時保留之。大致而言，這些符合模式假設的20題BDI-II項目與選項所反應的憂鬱程度，在憂鬱嚴重度的建構量尺上能夠均勻地分布且完整地涵蓋不同的憂鬱程度，因此可以有效地區辨受試者之憂鬱程度。最後，本研究依據羅序模式的估算結果，針對符合模式假設之20題BDI-II項目的原始總分，將之轉換為等距的羅序分數，並建立轉換表。 建議未來以BDI-II評量憂鬱症患者之憂鬱嚴重度時，使用20題符合模式假設的項目，並且以轉換後的等距羅序分數取代原始總分，以確保評量結果可以有效並精確地反映受試者的憂鬱程度。

The Beck Depression Inventory-II (BDI-II) total score was one of the most frequently used indicators of depression severity in research and clinics. However, the justification of this use was not well validated. Specifically, there were at least 4 psychometric issues that needed to be investigated empirically, including (1) the unidimensionality which warranted the summation of the BDI-II scores; (2) the appropriateness of category scoring criteria in each item; (3) the distributions of the BDI-II items and their response categories along the continuum of depression severity; and (4) the transformation of ordinal raw BDI-II total scores into interval data, which were more informative and allowed arithmetic computation. The Rasch model, known as a one-parameter logistic model of item response theory, is particularly useful in investigating the aforementioned issues simultaneously. The objectives of this study were to validate the BDI-II with Rasch analyses, including (1) to confirm the unidimensionality by means of examining Rasch model-data fitting, (2) to check the appropriateness of the category scoring criteria, (3) to inspect the distributions of items and their response categories along the continuum of depression severity, and (4) to transform ordinal raw BDI-II total scores into interval Rasch logits. Data were derived from a study with 260 clients diagnosed with major depressive disorders (Steer et al., 2001). The model-data fitting was examined by fit statistics. To fulfill the assumptions of the Rasch model, the remaining analyses were performed with the fitted BDI-II items only. All items fitted the Rasch model’s expectation, and thus the unidimensionality, except the item “Loss of Interest in Sex”. After removal of the misfit item, the remaining 20 BDI-II items all fitted the unidimensionality. Five items in which one or more response categories were less probable to be endorsed across various degrees of depression severity were identified, and thus the category scoring criteria may need to be revised in the future, including the items “Sadness”, “Punishment Feelings”, “Crying”, “Loss of Interest”, and “Changes in Appetite”. However, the resulting errors were within acceptable ranges, so they were kept at this stage. The 20 fitted BDI-II items and their response categories were found to be evenly distributed along the continuum of depression severity, and to cover, with this distribution, the depression severity of the clients with major depressive disorders in this study fairly well. The ordinal raw total scores of the 20 fitted BDI-II items were transformed into interval logits with the Rasch model. Generally, the 20 fitted BDI-II items that constituted a unidimensional construct of depression would be able to discriminate different levels of depression severity and were suitable to measure the depression severity of the clients with major depressive disorders. Moreover, the Rasch transformed interval logits of the total scores of the 20 fitted BDI-II items were suggested for use by future clinicians and researchers because the transformed data can provide more information than ordinal raw BDI-II total scores and allow arithmetic computation and the application of more powerful parametric statistics without the threat of violating statistical assumptions.