本研究目的在分析魏氏兒童智力量表第四版（WISC-IV）在一般兒童與特殊兒 童組別間之因素恆等性。根據9-16 歲之台灣標準化樣本（N=704）及身心障礙特 殊兒童樣本（N=694），研究者以多樣本高階結構方程模式對「WISC-IV 現行計分 之四因素模式」及「Cattell-Horn-Carroll（CHC）理論依據之五因素模式」兩種不 同解釋取向進行系列性因素恆等性檢驗。兩種架構的主要差異是，四因素內的 「知覺推理」因素在五因素架構內被進一步區分為「流體推理」和「視覺空間」 兩個因素。本研究發現：（1）對一般兒童與特殊兒童而言，四因素與五因素模式 均為合理且具意義之詮釋方式。算術、類同、與符號尋找分測驗具有跨因素負荷 量，表示它們測量到多元認知內涵，分析詮釋時宜納入考量；（2）四因素及五因 素模式在兩組兒童間均具因素恆等性，WISC-IV 分數對兩組兒童具相同的構念意 義、因素與分測驗關連性、及因素與分測驗效度。研究結果支持一般兒童與特殊 兒童的WISC-IV 分數結果可用相同方式詮釋，不論是四或五因素架構均可被用來 對兩組兒童進行有意義之臨床比較。

Purpose: Invariance is a fundamental property of any instrument used to compare individuals from subpopulations. In empirical settings, the Wechsler Intelligence Scale for Children, Fourth Edition (WISC-IV) is frequently employed as a part of psychoeducational assessments. Implicit in this common practice is the assumption that WISC-IV scores have the same meaning for children in various subpopulations. The current WISCIV manual recommends a four-factor scoring structure, whereas the Cattell-Horn-Carroll theory-based five-factor approach recommends that WISC-IV subtests assess five, instead of four, meaningful latent broad factors under g. This study investigated the factorial invariance of both four- and five-factor approaches among large normative and mixed exceptional children samples in Taiwan. Methods: Data from two large and reliable samples were analyzed. The normative sample was part of the Taiwan WISC-IV standardization sample, which consisted of 704 children aged 9 to 16 years. The overall mean fullscale intelligent quotient (FSIQ) was 100.1 (SD = 15.2). The sample consisting of excep-tional children was a heterogeneous sample that included 697 children in the special education system with various diagnoses such as intellectual disability, autism, learning disabilities, attention deficit hyperactivity disorder, and other emotional and behavioral disturbances. The mean FSIQ in this sample was 84.9 (SD = 19.3). Tests for the higher-order confirmatory factor invariance among the normative and exceptional children samples were based on an analysis of the mean and covariance structure models. Both higherorder WISC-IV four- and five-factor approaches were analyzed. Results/Findings: The results based on the multigroup mean and covariance structure analysis revealed the following: (a) Both the four- and five-factor models provided a good data fit for children in both samples, suggesting that both models provide meaningful strategies for interpreting WISC-IV scores. (b) Both models demonstrated full factorial invariance between normative and exceptional samples. (c) Arithmetic, Similarities, and Symbol Search subtests were found with cross-loadings. Conclusions/Implications: For both the four- and fivefactor approaches, the WISC-IV subtests demonstrate the same underlying theoretical latent constructs, the same strength of relationships among factors and subtests, the same validity of each first-order factor, and the same communalities, regardless of clinical status. The results support the same interpretive approach and meaningful comparisons of the WISC-IV between normative and exceptional children in Taiwan. In addition, when performance inconsistencies for subtests within the same latent ability dimension are detected, or when examiners wish to test specific hypotheses, both the main and minor sources of influence for Arithmetic, Similarities, and Symbol Search subtests may warrant consideration.