PLSpredict for assessment of the model's predictive power

Research topics can be discussed in this area.
Post Reply
katarina86
PLS Junior User
Posts: 5
Joined: Thu Feb 21, 2019 9:44 am
Real name and title: PhD Katarina Njegic Assistant Professor

PLSpredict for assessment of the model's predictive power

Post by katarina86 » Wed Oct 09, 2019 3:14 pm

Dear all,

I want to assess predictive power of my model. It has 5 latent variables, and the maximum number of arrows pointing at one construct is two. I have around 120 observations in my sample. I'm not sure how many folds should I choose. It is suggested in the literature (Hair j., Risher J., Sarstedt M., Ringle C. (2019). When to use and how to report the results of PLS-SEM. European Business Review, 31(1), 2-24.) that every fold(subset) has to meet minimum sample size guidelines. Since my sample is modest, is it enough to use, for example, only 3 folds? Or should I use more, 4 or 5, or even more... Also, the results differ depending on the number of folds I set. So I don't know which number of folds to chose.

Thank you in advance for your time and answers.

Kind regards,
Katarina

jmbecker
SmartPLS Developer
Posts: 1045
Joined: Tue Mar 28, 2006 11:09 am
Real name and title: Dr. Jan-Michael Becker

Re: PLSpredict for assessment of the model's predictive power

Post by jmbecker » Thu Oct 10, 2019 8:49 am

Usually it is recommended to use 10 folds. In this case your training data is 90% of your total sample size. This training data should meet the requirements for minimum sample size.

A good new source for applying PLSPredict is also the following:
Shmueli, G., Sarstedt, M., Hair, J.F., Cheah, J.-H., Ting, H., Vaithilingam, S., & Ringle, C.M. (2019). Predictive model assessment in PLS-SEM: Guidelines of using PLSpredict, European Journal of Marketing, 53(11), 2322-2347.
When choosing a value for k, researchers have to ensure that the training sample in a single fold still meets the model’s minimum sample size requirements (Kock and Hadaya, 2018). For example, with 200 observations and k=5, each fold’s training sample has 160 observations. Hence, the minimum sample to estimate the underlying model must be 160 (or higher). Predictive studies typically set k to 10 (Shmueli et al., 2016). We recommend following this convention as long as the minimum sample size requirements are met. If a k of 10 does not result in the minimum required training sample size, researchers should choose a higher k value to increase the size of the training sample in each cross-validation run.
Dr. Jan-Michael Becker, University of Cologne, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Jan_Michael_Becker
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de

katarina86
PLS Junior User
Posts: 5
Joined: Thu Feb 21, 2019 9:44 am
Real name and title: PhD Katarina Njegic Assistant Professor

Re: PLSpredict for assessment of the model's predictive power

Post by katarina86 » Thu Oct 10, 2019 10:11 am

Thank you very much for the answer, professor! It is very helpful! I understand it better now.

Best wishes,
Katarina

Post Reply