Is Rasch model analysis applicable in small sample size pilot studies for assessing item characteristics? An example using PROMIS pain behavior item bank data
Is Rasch model analysis applicable in small sample size pilot studies for assessing item characteristics? An example using PROMIS pain behavior item bank data
复制标题
DOI：
10.1007/s11136-013-0487-5
复制
发表时间：
2014-03-01
影响因子：
3.5
通讯作者：
Revicki, Dennis A.
中科院分区：
文献类型：
--
作者：
Chen, Wen-Hung;Lenderking, William;Revicki, Dennis A.
通常认为 Rasch 模型需要大样本才能获得稳健的项目参数估计。最近，小样本拉希分析被建议作为项目心理测量特性的初步评估。本研究旨在使用小样本量评估 Rasch 分析结果。使用了 10 个 PROMIS 疼痛行为项目。从总共 800 名受试者中随机抽取 30、50、100 和 250 个样本，以及 30 个目标样本，各抽取 10 次。对每个样本和完整样本进行了 Rasch 分析。在完整样本中，有 104 例极端得分、没有空类别、2 个错误排序的项目和 4 个不适合的项目。对于250、100、50、30和目标30的样本，极端分数的平均数分别为42.2、17.1、9.6、6.1和1.2；无效类别的平均数量为 1.0、3.2、8.7、13.4 和 8.3；项目参数排序错误的项目平均数量为 0.1、0.8、2.9、4.7 和 3.7；拟合残差超过 +/- 2.5 的项目的平均数量分别为 0.8、0.3、0.1、0.2 和 0.3。基于小样本（千分之一货币符号 50）的 Rasch 分析发现，与大样本（千分之一日元 100）相比，参数排序错误的项目数量较多。然而，被确定为不合适的物品较少。小样本的结果得出了与大样本相反的结论。基于小样本的拉希分析在用于探索性目的时应极其谨慎。
Large samples are generally considered necessary for Rasch model to obtain robust item parameter estimates. Recently, small sample Rasch analysis was suggested as preliminary assessment of items' psychometric properties. This study is to evaluate the Rasch analysis results using small sample sizes.Ten PROMIS pain behavior items were used. Random samples of 30, 50, 100, and 250, and a targeted sample of 30 were drawn 10 times each from a total of 800 subjects. Rasch analysis was conducted for each of these samples and the full sample.In the full sample, there were 104 cases of extreme scores, no null categories, two incorrectly ordered items, and four misfit items. For samples of 250, 100, 50, 30, and targeted 30, the average numbers of extreme scores were 42.2, 17.1, 9.6, 6.1, and 1.2; the average numbers of null categories were 1.0, 3.2, 8.7, 13.4, and 8.3; the average numbers of items with incorrectly ordered item parameters were 0.1, 0.8, 2.9, 4.7, and 3.7; and the average numbers of items with fit residuals exceeding +/- 2.5 were 0.8, 0.3, 0.1, 0.2, and 0.3, respectively.Rasch analysis based on small samples (a parts per thousand currency sign50) identified a greater number of items with incorrectly ordered parameters than larger samples (a parts per thousand yen100). However, fewer items were identified as misfitting. Results from small samples led to opposite conclusions from those based on larger samples. Rasch analysis based on small samples should be used for exploratory purposes with extreme caution.