基于项目反应理论的范德堡课堂聆听评估简表在单侧听力损失儿童中的验证

文献信息

中文摘要

目的： 范德堡课堂聆听评估简表（VCLASS）此前被开发为一种监测学龄儿童在课堂中经历的聆听困难的工具。然而，初始验证中使用的先前统计方法并未充分评估问卷的心理测量学特性，未考虑反应的有序性质，也未对比助听与未助听条件。本研究旨在使用项目反应理论（IRT）在助听和未助听两种条件下评估15个条目VCLASS的心理测量学特性。更具体地说，我们旨在（1）评估每种聆听条件（未助听和助听）的潜在结构；（2）确定适当的基于IRT的模型，以估计每种聆听条件的条目区分度和阈值参数；（3）评估基于IRT模型的信度；以及（4）评估每种聆听条件下一致性和非一致性年龄相关差异条目功能（DIF）。
设计： 我们应用基于IRT的分析来分析先前从9至17岁单侧听力损失儿童中收集的反应。所有儿童（n = 1148）在想象自己未使用设备（未助听）的情况下完成了VCLASS；大多数儿童（n = 897）还在想象自己使用个人聆听设备（助听）的情况下完成了VCLASS。对于每种聆听条件（未助听、助听），在考虑反应有序性质的情况下评估VCLASS条目的维度性。接下来，使用最佳拟合模型估计条目区分度（â）和阈值参数（β̂1至β̂4），评估条目和个体拟合，并检查信度。最后，评估每种聆听条件下年龄相关的一致性和非一致性DIF。
结果： 基于多分相关矩阵的平行分析支持未助听和助听条件下均为双因子结构。在两种条件下，反应模式在很大程度上由一个占主导地位的一般学校相关聆听困难因子所捕获；此外，一个条目（未助听时）或四个条目（助听时）加载在一个代表学校相关近距离聆听困难的具体/次要因子上。双因子分级反应模型相对于双因子广义部分信用模型提供了更好的模型拟合。所有条目均显示出良好拟合和高区分度。因子得分估计的基于IRT的边缘信度很强（未助听：因子1和因子2分别为0.88和0.89；助听：因子1和因子2分别为0.86和0.85）。测试信息曲线表明，测量精度最高的地方是样本受访者最多的地方——包括低、中和高但非极端的聆听困难水平——不同聆听条件下的曲线轮廓相当。年龄相关DIF对条目反应的影响可忽略不计，支持年龄不变的测量。
结论： 基于IRT的验证证实，VCLASS为单侧听力损失儿童在未助听和助听聆听条件下提供了可靠且一致的课堂聆听困难测量。通过在建模和缺失数据插补中考虑数据的有序性质，当前结果支持未助听和助听条件下的双因子结构。我们还证明VCLASS具有良好的条目功能和可忽略的年龄相关偏倚。VCLASS在区间水平连续体上提供课堂聆听困难评分，支持其在单侧听力损失儿童的临床和研究环境中使用。

英文摘要

OBJECTIVES: The Vanderbilt Classroom Listening Assessment Short Survey (VCLASS) was previously developed as a tool to monitor listening difficulties experienced by school-aged children in classrooms. However, the previous statistical approach used in the initial validation did not fully evaluate the questionnaire's psychometric properties, did not take into account the ordinal nature of responses, and did not contrast hearing-aided versus hearing-unaided conditions. The present study aimed to evaluate the psychometric properties of the 15‑item VCLASS using item response theory (IRT) in both hearing-aided and unaided conditions. More specifically, we aimed to (1) assess the latent structure for each listening condition (unaided and aided); (2) determine appropriate IRT-based models to estimate item discrimination and threshold parameters for each listening condition; (3) assess reliability from the IRT-based model; and (4) evaluate both uniform and nonuniform age-related differential item functioning (DIF) under each listening condition.
DESIGN: We applied IRT-based analyses to analyze responses previously collected from children aged 9 to 17 years with unilateral hearing loss. All children (n = 1148) completed the VCLASS imagining they were not using a device (unaided); most children (n = 897) also completed the VCLASS imagining they were using their personal listening device (aided). For each listening condition (unaided, aided), dimensionality of VCLASS items was assessed taking into account the ordinal nature of responses. Next, the best-fitting models were used to estimate item discrimination (â) and threshold parameters (β̂1 to β̂4), to evaluate item and person fit, and to examine reliability. Last, age-related uniform and nonuniform DIF were assessed for each listening condition.
RESULTS: Parallel analysis based on polychoric correlations supported a two-factor structure for both the hearing-unaided and hearing-aided conditions. In both conditions, response patterns were largely captured by a dominant, general school-related listening-difficulty factor; additionally, one item (when unaided) or four (when aided) items loaded on a specific/minor factor representing school-related close-proximity listening difficulty. The two-factor graded response model provided better model fit relative to the two-factor generalized partial credit model. All items showed good fit and high discrimination. IRT-based marginal reliability for factor score estimates was strong (unaided: 0.88 and 0.89; aided: 0.86 and 0.85 for factors 1 and 2, respectively). Test information curves indicated greatest measurement precision where there were the most sample respondents-which included low, middle, and high, but not extreme, levels of listening difficulty-with comparable profiles across listening conditions. Age-related DIF effects on item responses were negligible, supporting age-invariant measurement.
CONCLUSIONS: The IRT-based validation confirms that the VCLASS provides a reliable and consistent measure of classroom listening difficulty across unaided and aided listening conditions in children with unilateral hearing loss. By considering the ordinal nature of the data in both modeling and missing data imputation, current results support a two-factor structure across unaided and aided conditions. We also demonstrated that the VCLASS has strong item functioning and trivial age‑related bias. The VCLASS provides listening difficulty scores on an interval‑level continuum for classroom listening difficulties, supporting its use in clinical and research settings for children with unilateral hearing loss.