AIC in Smart PLS 4 and Model Selection Crieria

[MeiPeng]

Dear Respected Scholars

I understand for model selection, we can compare different models based on the AIC and BIC values.

In Smart PLS 3, we could obtain the AIC and BIC values after PLS algo under Model Selection Criteria. But, in Smart PLS 4, only BIC values are available.

In this regard, for model selection, it is sufficient just to compare BIC values without AIC weight?

Looking forward to discussion, explanation, and advice.

Thank you

[jmbecker]

Yes, based on the literature, the BIC and the and GM (Geweke–Meese criterion) are best for model selection and all other criteria are not well suited.

SmartPLS does not include the Geweke–Meese criterion (GM), which is based on the model-complexity adjusted mean square error (MSE) from a saturated (full) model. Such a model is not always well defined and automatic generation could produce strange results (e.g., for moderation or second-order models). You can calculate it by had, but as the BIC is as good as the GM, it is not really necessary.

Alternatively, you could also focus on predictive model comparison using RMSE, which is the best, if your sample sizes are large enough.

Sharma, P. N., Sarstedt, M., Shmueli, G., Kim, K. H., & Thiele, K. O. (2019). PLS-based model selection: The role of alternative explanations in IS research. Journal of the Association for Information Systems, 20(4), 346-397.
Sharma, P. N., Shmueli, G., Sarstedt, M., Danks, N., & Ray, S. (2021). Prediction-oriented model selection in partial least squares path modeling. Decision Sciences, 52(3), 567-607.
Danks, N. P., Sharma, P. N., & Sarstedt, M. (2020). Model Selection Uncertainty and Multimodel Inference in Partial Least Squares Structural Equation Modeling (PLS-SEM). Journal of Business Research, 113, 13-24.