視覺類比量尺的診斷分類模型

Abstract

視覺類比量尺（visual analogue scale, VAS）使受試者根據題目的敘述，在連續的視覺化量尺上進行標記，來反應受試者於試題欲測量潛在特質的傾向。由於VAS具有等距的特性，因此相較於間斷量尺（如李克特量尺），VAS在個體層面上得以提供更細緻的區辨度。鑒於目前所知的文獻中並未有針對VAS資料的診斷分類模型（diagnostic classification model, DCM），因此本研究旨在發展針對VAS資料的DCM。由於VAS資料為連續且具有雙邊界（doubly bounded）特性，本研究透過結合beta response model （BRM）以及log-linear cognitive diagnosis model（LCDM）組成針對連續雙邊界資料的beta diagnostic classification model （BDCM），並以馬可夫鏈蒙地卡羅（Markov chain Monte Carlo, MCMC）作為模型參數的估計方法。模擬研究中透過操弄特質數以及樣本數比較兩種模型：（1）應用BDCM於VAS資料以及（2）使用LCDM於二分資料，比較兩者之間試題參數回復以及分類準確率的差異。研究結果顯示，在試題參數回復上，BDCM所需的樣本小於LCDM，且在分類準確率上BDCM也優於LCDM。實徵研究針對Holland職業代碼（Holland code）發展的VAS職業興趣量表進行分析，並針對受試者的特質分類進行探討。

Visual analogue scale (VAS) enables participants to mark their responses on a continuous visual scale based on the item descriptions that reflect their tendencies toward the measured latent traits in the given items. VAS appears to provide more fine-grained discrimination at the individual level compared to categorical scale (i.e., Likert’s scale), given its interval properties. To the author’s best knowledge, there is currently no diagnostic classification model (DCM) designed for VAS data. Therefore, this study aims to develop a novel DCM model specifically tailored for VAS data. As VAS data are continuous and exhibit doubly bounded characteristics, this study integrates the beta response model (BRM) and the log-linear cognitive diagnosis model (LCDM) into beta diagnostic classification model (BDCM), suitable for continuous bounded data. Markov Chain Monte Carlo (MCMC) was employed as the estimation method for model parameters. The simulation study manipulated the number of attributes and sample sizes and compared two models: (1) applying BDCM to VAS data and (2) utilizing LCDM with dichotomous data. Specifically, the simulation study compared item parameter recovery and classification accuracy between the two models. The results suggest that, in terms of item parameter recovery, the sample size required for BDCM is smaller than that for LCDM. Additionally, in terms of classification accuracy, BDCM outperforms LCDM. An empirical study was conducted to examine the VAS Career Interest Scale based on the Holland Code, and investigated the trait classification of participants.