项目间多维测验作答时间数据分析:潜在特质速度间效应建模

Modeling of the Effect on Multidimensional Latent Speeds in the Between-Item Multidimensional Response Time

随着计算机测验使用的普及化,被试在心理与教育测验上的作答时间的获取也越发便利。为了充分利用项目作答时间信息,单维与多维的作答时间模型相继被提出。然而,在项目间多维作答时间数据中,潜在特质速度之间可能存在共同关系(比如层阶关系),此时现有的作答时间模型并不能适用。基于此,本研究提出了高阶对数正态作答时间模型与双因子对数正态作答时间模型。在模拟研究中,高阶对数正态作答时间模型与双因子对数正态作答时间模型的各参数都能被准确估计。在瑞文标准推理测验的三组测验项目的作答时间数据中,双因子对数正态作答时间模型表现出更为优秀的拟合效果,同时基于多个统计量说明了一般与局部潜在特质速度同时存在的必要性。因此,在项目间多维测验作答时间数据分析中,非常有必要考虑多维潜在特质速度之间的共同效应。

With the popularity of computer-based testing, it is easier to collect item response times(RTs) in psychological and educational assessments.RTs can provide an important source of information for respondents and tests. Meanwhile, RTs can help in evaluating the speed of respondents,detecting cheating behaviors and designing better tests. RTs can also be used for improving the accuracy of parameter estimation and others.To full use RTs, researchers have invested substantial efforts in developing statistical models of RTs. Most of the proposed models posit a unidimensional latent speed to account for RTs in tests. However, there are many multidimensional tests in psychological and educational assessments.Based on the assumption that each latent speed should be paired with a specific latent ability in multidimensional tests, a multidimensional lognormal response time(MLRT) model was proposed with extending the unidimensional lognormal response time model(ULRTM). In multidimensional tests,there are between-item and within-item multidimensionality. There may be common or hierarchical effects between different latent speeds in the between-item multidimensionality. MLRTM may not be appropriate for this situation.To capture the hierarchical or common effects between different latent speeds, this study proposed a higher-order lognormal response time model(HO-LRTM) and bifactor lognormal response time model(Bi-LRTM) based on the corresponding response models. Model parameters in the HOLRTM and Bi-LRTM can be estimated via maximum likelihood estimation in Mplus. In simulations, the results showed that the parameters of HOLRTM and Bi-LRTM could be accurately estimated. In empirical example, three sets(A, C and D) were chosen from the Raven’s Standard Progressive Matrices. Each set had 12 items. Firstly, the structure of RTs was explored by the empirical kaiser criterion(EKC) and exploratory factor analysis(EFA). The results of the EKC and EFA showed that the latent structure of RTs was a three-dimensional structure. Secondly, according to the different fit indices, the Bi-LRTM fitted better than other models. Furthermore, it was necessary to free estimate speed-slope parameters in the response time models. Finally, this study assessed the unidimensionality of Bi-LRTM based on some statistical indices. These statistical indices showed the general and specific latent speeds of Bi-LRTM impacted RTs in empirical data.Overall, the proposed Bi-LRTM works well in simulation and empirical data. In other words, the common effects on multidimensional latent speeds meet the need for between-item multidimensional response time.