This not only helps practitioners better understand the performance of existing longitudinal LDMs in specific test situations with missing data but also provides a reference to psychometricians for future research on the necessity of imputation methods for missing data in longitudinal learning diagnosis. Therefore, it is necessary to explore the impact of missing data caused by attrition on longitudinal learning diagnosis. Longitudinal studies are particularly susceptible to such bias, as missing data accumulate over time due to attrition. Secondly, it may produce biased results when students with complete data are systematically different from those with missing data. However, this is unfair to those students who were deleted in analysis, because they did not receive any diagnostic feedback. Some studies have previously employed a complete case analysis that deletes any students who dropped out (e.g., Zhan et al., 2019a). For instance, in school-level longitudinal learning diagnosis projects, some students may individually drop out before the end of the study because they move to other schools that are inaccessible to the researchers all students in the class may even drop out altogether because of some unforeseen classroom instructional reasons (see the empirical example in Zhan et al., 2019a).Ī higher percentage of attrition at each point in time means the remaining data at subsequent time points provide less diagnostic information, which may also challenge the robustness of measurement models. Attrition refers to students dropping out prior to the end of the study and do not return. In this current study, we focused on a type of missing data that is common to longitudinal studies, namely, attrition ( Little and Rubin, 2020, p. In practice, missing data are hard to avoid, or even inevitable, in longitudinal learning diagnosis and other longitudinal studies. Although the utility of these models has been evaluated by some simulation studies and a few applications, the harm of ubiquitous missing data in longitudinal designs has not yet been considered and studied.
The diagnostic results of these two model types have a high consistency ( Lee, 2017).
The latter estimates the transition probabilities from one latent class or attribute to another or to the same latent class or attribute. The former estimates the changes in higher-order latent ability over time, and from this, it infers the changes in the lower-order latent attributes.
In recent years, to provide theoretical support for longitudinal learning diagnosis, several longitudinal learning diagnosis models (LDMs) have been proposed, which can be divided into two primary categories: the higher-order latent structure-based models (e.g., Huang, 2017 Lee, 2017 Zhan et al., 2019a) and the latent transition analysis-based models (e.g., Li et al., 2016 Kaya and Leite, 2017 Wang et al., 2018 Madison and Bradshaw, 2018). Longitudinal learning diagnosis not only can be used to diagnose and track students’ growth over time but also can be used to evaluate the effectiveness of diagnostic feedback and corresponding remedial teaching (Tang and Zhan, under review Wang et al., 2020).
Longitudinal learning diagnosis identifies students’ strengths and weaknesses of various attributes throughout a period of time, which also can be seen as an application of learning diagnosis through longitudinal assessments. During the last few decades, to promote student learning, learning diagnosis ( Zhan, 2020) or cognitive diagnosis ( Leighton and Gierl, 2007) through objectively quantifying the learning status of fine-grained attributes (e.g., knowledge, skills, and cognitive processes) and providing diagnostic feedback has been increasingly valued.
0 kommentar(er)
