This study investigated the dependability of reading comprehension scores across different text genres and response formats for readers with varied language knowledge. Participants included 78 fourth-graders in an urban elementary school. A randomized and counterbalanced 3 × 2 study design investigated three response formats (open-ended, multiple-choice, retell) and two text genres (narrative, expository) from the Qualitative Reading Inventory (QRI-5) reading comprehension test. Standardized language knowledge measures from the Woodcock Johnson III Tests of Achievement (Academic Knowledge, Oral Comprehension, Picture Vocabulary) defined three reader profiles: (a) < 90 as emerging, (b) 90–100 as basic, and (c) > 100 as proficient. Generalizability studies partitioned variance in scores for reader, text genre, and response format for all three groups. Response format accounted for 42.8 to 62.4% of variance in reading comprehension scores across groups, whereas text genre accounted for very little variance (1.2–4.1%). Single scores were well below a 0.80 dependability threshold (absolute phi coefficients = 0.06–0.14). Decision studies projecting dependability achieved with additional scores varied by response format for each language knowledge group, with very low projected dependability on open-ended and multiple-choice scores for readers with basic language knowledge. Multiple-choice scores had similarly low projected dependability levels for readers with emerging language knowledge. Findings evidence interactions between reader language knowledge and response format in reading comprehension assessment practices. Implications underscore the limitations of using a single score to classify readers with and without proficiency in foundational skills.

这项研究调查了具有不同语言知识的读者在不同文本类型和响应格式下阅读理解分数的可靠性。参加者包括城市小学的78名四年级学生。一项随机且平衡的3×2研究设计调查了定性阅读量表（QRI-5）阅读理解测试中的三种回答格式（开放式，多项选择，复述）和两种文本类型（叙述性，说明性）。伍德考克约翰逊三世成绩测验（学术知识，口语理解能力，图片词汇量）中的标准语言知识测评定义了三种读者资料：（a）新兴的<90，（b）基本的90–100和（c）> 100精通。概化性研究针对读者，文本类型，和所有三个组的响应格式。回答格式占各组阅读理解分数差异的42.8％至62.4％，而文本体裁的差异很小（1.2–4.1％）。单项得分远低于0.80的信度阈值（绝对phi系数= 0.06-0.14）。决策研究预测了可靠性，而每个语言知识组的答复得分因不同的分数而有所不同，而对具有基本语言知识的读者的开放性和多项选择分数的预测可靠性却很低。对于具有新兴语言知识的读者，多项选择得分的预测可靠性水平同样较低。研究结果证明了阅读理解评估实践中读者语言知识与回应格式之间的相互作用。