Dear community,

I have a question regarding SEM. I work with pupils' test-data and would like to model the growth over time. I have three time points from three consecutive school years. This is a panel study so the same individuals are tested multiple times. I have this model in mind, see the figure attached.


My tests have shown that the RMSEA decreases drastically as soon as I insert a pathwath from t1 to t3, so the model fit improves. I wonder what this means from a theoretical perspective. How can the ability of a child of t1 affect t3 without being mediated through the ability in 2? How would you interpret such a model (all coefs are highly significant)? What does this mean in your view? I simply struggle a bit with this direct effect of t1 to t3 that is apparently independet of t2.

Thanks a lot.

------------------------------
Hans Bauer
------------------------------

#SPSSStatistics

Dear Hans, 

Logically, it does make a sense that knowledge acquired in t1 can be used directly at t2 stage, as well as at t3 stage. 

What you can do to make the study meaningful is to test two different models:

Model 1: Make a new construct Ability t12 = Ability t1 + Ability t2. Then, make a model in the SEM like this: Ability t2 --> Ability t3; and Ability t12 --> Ability t3. . 

Model 2: What you have already mentioned. 

The new construct, Ability t12 is the mediating effect. Comparing the two models, you will be able to make a clearer decision on whether t1 is directly affecting t3 or it is affecting t3 better with a mediating effect of t2. 

Hope this helps.

------------------------------
Vandit Desai
------------------------------