SEM. Possiible predict output using SEM coefficients?
I have the following model in SEM in JASP:
L ~ D
D ~ Q
Q ~ M
M ~ P
P ~ I
I ~ A
A ~ M
All these variables linearly interrelate. They're logged variables to make this so.
I didn't standardize the data before estimation, and I'm using the unstandardized regression coefficients output of this model to predict output from input, in R, using the following R code:
------------------------------
x=USER SPECIFIED
x=log10(x) # log the input
x=x*-0.375
x=x*-0.875
x=x*-0.617
x=x*1.174
x=x*0.661
x=x*0.971
x=x*-0.947
x=10^(x) # de-log the output
x
---------------------------------
The problem is I'm not getting output within the expected range (and I do de-log the output. And the inputs tested are within the range of data of the model). Are the regression coefficients useable in this way? Or is what I'm trying to do just non-sensical?
The baseline test of the model has a small p-value (9e-29), and the p-values of the regression coefficients are all < 0.05, indeed many are so small that they're reported in JASP as 0 (and the 95% confidence intervals are good, with no zero within any). All the +- signs of the regression coefficients are as expected. The GFI is 0.969. So I presume the model is good? All be it the CFI is only 0.828, which is a worry?
P.S. The data set is quite sparse and I've used FIML setting within JASP to populate where there is missing data values.
Comments
I am quite confused. Can you share your .jasp file? Have tried running all this in R? Are the results then different from JASP?