學前識字困難高危險群幼兒之鑑別：「學前教師評定搭配認知測驗兩階段篩選機制」區辨效能之檢驗

Identifying Preschool Children at Risk of Learning Difficulties in Chinese Characters Reading: An Evaluation of a Two-Stage Identification Procedure Combining Preschool Teacher Ratings and Cognitive Assessment

閱讀習得與早期語言理解及識字解碼發展有關（Gough & Tunmer, 1986），故語言理解及解碼技能可作為閱讀發展預測指標。本研究主要目的在於早期鑑別學前識字困難高危險群學童，如下逐步檢驗學前教師評定搭配認知測驗兩階段篩選機制，對國小一年級學童識字困難篩選之成效：一、檢驗學前預測變項與識字發展的相關；二、第一階段學前教師評定初篩的分類效果；三、第二階段搭配認知評量的區辨效能；四、分析兩階段篩選機制的整體效能。本研究採縱貫研究，在大班測量早期預測指標分數（教師評定量表、直接測量認知及識字），再追蹤國小一年級學童識字發展。以叢集抽樣篩選171 名大班幼兒，第二年有139 名學童繼續參與。結果顯示，學前識字與國小一年級學童識字呈高度相關，其他與國小一年級學童識字顯著相關者為教師評定之認字解碼與詞素理解兩分項、直接測量之語言理解、注音解碼、快速自動唸名與詞素理解。在階段分析方面，教師評定之識字解碼及詞素理解總分（33% 以下）對國小一年級學童識字困難（25% 以下）篩選的敏感度為53.33%，結果並不理想，故加入學前識字為篩選條件（教師評定及學前識字其中一項低於33%），如此使篩選敏感度提升至93.33%。以第一階段正確預測識字困難及假陽性識字困難為效標，第二階段認知評量為預測變項做區辨分析，結果顯示，語言理解、符號聲韻解碼與快速自動念名符合最佳區辨組合，整體正確性為66.7%，預測敏感度為78.6%。兩階段綜合篩選效能為73.33%，但仍有26.67% 的假陰性錯誤，此與父母親社經地位及幼兒成熟度有關。

Purpose. Reading acquisition is based on early language comprehension and word decoding (Gough & Tunmer, 1986). Therefore, language comprehension and cognitive decoding skills are applied as early predictors of reading development. This study aimed to identify preschool children at risk for learning difficulties in Chinese character reading. The identification procedure was conducted by using language comprehension and decoding skills as early predictors. The efficiency of a two-stage identification procedure that entailed combining preschool teacher ratings and cognitive assessments was evaluated for the following purposes: (1) to investigate the association between early predictors (including preschool teacher-rated scores and assessments of language comprehension, phonological Tzuyin decoding, rapid automatized naming (RAN), morphological awareness, and nonverbal IQ) and character reading ability during preschool and first grade; (2) to verify the classification accuracy of the procedure according to preschool teacher-rated scores and first grade character reading scores (i.e., the first stage of the identification procedure); (3) to examine the significance of classification prediction by applying a set of early predictors at the second stage of the identification procedure; and (4) to evaluate the combined outcome of the two-stage identification procedure. Methods. A total of 171 five-year-old preschoolers were recruited through cluster sampling. The participants´ decoding skills, language comprehension, and morphological awareness were rated by their classroom teachers. Their language comprehension, reading-related cognitive skills (phonological Tzuyin decoding, RAN, and morphological awareness), and character reading skills were assessed individually. One year later, 139 children continued to participate. Their character reading skills were reassessed. Results and Conclusions. Character reading abilities in preschool and the first grade were highly correlated. The individually assessed cognitive skills and abilities that were rated by the preschool teachers, including decoding skills and morphological awareness, were significantly associated with first grade character reading. The classification adequacy recorded at the first stage of the identification procedure, using the sum of preschool teacher-rated scores (which were below 33%) as early predictors, revealed that 53.33% (n = 16) of the children were accurately classified as poor character readers (first grade character reading scores were below 25%). However, the result is not acceptable because of the high proportion of false-negative errors (46.67%; n = 14). To increase the sensitivity of the first identification stage, children´s early character reading scores were used as alternative predictor. Children with either teacher-rated scores or an early character reading ability lower than 33% were identified as having a high risk of becoming poor character readers. This adjustment increased the sensitivity from 53.33% to 93.33% (n = 28). At the second identification stage, a direct discriminant function analysis was performed using three assessment variables as predictors of membership in two groups. The predictors were language comprehension, phonological Tzuyin decoding, and RAN. The groups were distinctly composed of true- and false-positive cases (n = 28 and 20, respectively) selected from the first identification stage. The predictors were significant for discriminating the groups. There was a 66.7% accurate classification rate evidenced through cross-validation of the given classification function. Among the true-positive cases, 78.6% (n = 22) were classified accurately. The combined outcome of the two-stage identification procedure revealed a 73.33% sensitivity for accurate classification of true-positive cases (22 of 30), although there was still a 26.67% false-negative error engendered by the effects of parent social economic status and age.