Multilevel Model

80 views
Skip to first unread message

Mia Mclean

unread,
Apr 9, 2021, 6:23:18 PM4/9/21
to GSCA Pro
Hi, 
I am attempting to run a multilevel model with GSCA Pro (MGSCA). The package looks suitable for my data but I am having troubles with running the model. 

My model: 
I have a multilevel model, whereby 38 participants are assessed on 2 components (1 with ~3900 indicators [SCANS], 1 with ~5 indicators [FACTORS]), twice (repeated measures). I have one 2nd level component (1 indicator, measured once, [GA]) for each participant. Participant is the grouping variable.

1. Is this model too big for GSCA to handle? I am able to run a MGSCA on the trial data, grouping by gender. However, when running this model, the program stops running.

2. If not too big, I am wondering if I have to separate level 2 and level 1 observed variables when constructing latent variables. Currently, I have three components with two components, 1 at level 1 [FACTORS], 1 at level 2 [GA], independently predicting an endogenous variable at level 1 [SCANS]. Here [GA] is a participant-level explanatory variable, while [SCANS] and [FACTORS] hold repeated measures for each participant across observed variables. Is this the correct modelling approach? Example papers (Hwang, Takane, Malhorta 2007) did not have level 2 explanatory variables so I was unsure how best to model them. 

Thanks!
Mia

heungsun.hwang

unread,
Apr 10, 2021, 11:32:08 AM4/10/21
to GSCA Pro
Hi Mia,

Unfortunately, the current version of GSCA Pro cannot handle your data, where the number of indicators is far greater than the number of cases. Worse, at the participant level, there are only two cases (T1 and T2). This makes it difficult for GSCA (and any other SEM methods, I think) to analyze the data. Perhaps some regularization has to be adopted deal with the data, which is not implemented into the software. We will consider adding regularization to multilevel GSCA in the future. As you noted, multilevel GSCA does not consider level 2 variables at this point.

Best,
Heungsun  
Reply all
Reply to author
Forward
0 new messages