Dear Users,
I would like to compare two models and show that model A is better than model B.
Model B can be interppreted as a part of model A - so the models can be seen as nested models.
The only measure of model fit I can find in PLS is R-square:
- thus, unfortunately I could only do a descriptive comparison of the R-square of both models
- but, R-square does not even take into account that model A is more complex and has more parameters and should therefore show a better fit. Something like the AIC (which considers the models' complexity) would help me more.
What can I do to compare the fit of two models in PLS?
Is there also a possibility for a test of this comparison (like the modification indices MI in Lisrel)??
Thank you very much for your help
sandra
comparison of the fit of two models
Thank you,
however, if I use the global fit index proposed by Tenenhaus et al. it is still only a descriptive comparison, isn't it?
And, I am not sure, if this global fit index does consider model complexity (like for instance AIC does)?
So wouldn't I actually end up with a similar comparison like the comparison of R^2 for my dependent LV in the two models?
Thank you for your help!
however, if I use the global fit index proposed by Tenenhaus et al. it is still only a descriptive comparison, isn't it?
And, I am not sure, if this global fit index does consider model complexity (like for instance AIC does)?
So wouldn't I actually end up with a similar comparison like the comparison of R^2 for my dependent LV in the two models?
Thank you for your help!
-
- PLS Expert User
- Posts: 54
- Joined: Wed Oct 19, 2005 5:53 pm
- Real name and title:
Dear Sandra,
if I remember correctly, the goodness-of-fit index proposed by Tenenhaus goes something like this:
GOF = sqrt(average(AVE)*average(Rsq))
where average(AVE) is the mean of all AVE values of all LVs and average(Rsq) is the mean of all Rsquare values of all endogeneous LVs.
Thus, the GOF suffers from the same issues as the direct comparison of Rsq-values across models of different complexity/ nested models.
However, you could test the Rsq-difference of both models for significance just like in a regular regression model.
Cheers,
Stefan
if I remember correctly, the goodness-of-fit index proposed by Tenenhaus goes something like this:
GOF = sqrt(average(AVE)*average(Rsq))
where average(AVE) is the mean of all AVE values of all LVs and average(Rsq) is the mean of all Rsquare values of all endogeneous LVs.
Thus, the GOF suffers from the same issues as the direct comparison of Rsq-values across models of different complexity/ nested models.
However, you could test the Rsq-difference of both models for significance just like in a regular regression model.
Cheers,
Stefan