A multilevel latent transition analysis (LTA) with a mixture IRT measurement model (MixIRTM) is used for investigating the effectiveness of an intervention. Comparing to LTA–MixIRTM, the proposed model includes a multilevel structure which has different results in group membership assignment in empirical results.
The computer program Mplus was used to estimate parameters
of all models in this study
First, I have to say, I feel shocked to find two repeated sentences in such top journal.（2.4,paragraph 2, the last sentence“There was one missing item response from a single student in the final sample of 310 students.”）
And one sentence on the page after table 4，P2，“The most common error made by students in both student-level latent classes on most items was not finding a common denominator, which reveals students’ lack of basic misunderstanding of fractions”, this sentence does not make sense.
Since in the pre-test most students had minimum scores on the pre-test, they all incline to answer the easy time correctly, the pattern are similar then which is difficult to classify while on the post-test, for inventions, they answer more items correctly, the patterns are more different, which may cause different classification.
Since the author said, the control and treatment group is not modeled in the proposed model, in the post-test, all the subjects are used for classification, the effect can’t be divided for the control or treatment, then how can they be sure that “The increase in variance on the post-test occurred for teachers in both latent classes mainly from the EAI intervention effect and to a lesser extent from instruction in the BAU condition”，may they modeled separately to see the effect.
The IRT model in Mplus is GRM, may the model be another one for example, PCM.
The multimodality can be shown in the paper.
The model may also be used in policy implementation, before implementation and after implementation，the real effect in related filed.