A small sample, let's say 15 cases would be enough.

But you can search on bibliography by Prof. Tenenhaus or Prof. Zinzi

Thank you, Sergio.

I'm dumbfounded by the small number of case possible for PLS.

How much credibility would I put in a study with a sample size of 15, if you hadn't give me this info...Can you refer any published article with sample size this small?

How about other PLS user?

Samples of 100 and more are more reliable.

Hello doctor!

sample size depends on many things: 1- complexity of model 2- statistical power 3- type I error 4- amount of R^2 that you expect to discover

Hair et al, use Cohen directions in determining the required sample size in OLS regression.

There are other ways too, for example, you can use 10 times of the number of most arrows headed toward a construct, or you can use software like Gpower to determine your sample size.

But I must insist that all if these methods are not completely accurate.

Key factors in determining sample size are the desired power of the model and its effect.

A great paper on this area is by Jacob Cohen in 1992 titled "A power primer" published in Quantitative methods in Psychology

See the Journal of Academy of Marketing Science article by Marko Sarstedt et al.

I am always reluctant to have small sample size as a reason for choosing any data analysis technique. Should it not be a combination of factors.

Brad

I have used PLS path modeling on a sample of size n= 6 with success.

I attach my publication where I worked with 6 subjetcs.

I have a lot of experience in using PLS-PM  in the industry with a number of subjects between 6 and 12 and I obtain very useful results. I don't dare to go below 6.

In fact the problem is the meaning of a correlation computed on 6 subjects.  It means something on highly stuctured data as the orange juice example.

convergence is practically ever a problem for pls. See. Henseler article for exceptions.

the power and Cohen advice is sound but as Michel states it really depends on the data. What happens if you are studying a population with a rare disease and only 11 people have been diagnosed.... Do you not try and analyse that!

regards brad

Yes PLS can provide you a valid results in case of small sample, due to the non parametric characteristic.

PLS doesn't lock to the normality of data that why no issue if the sample is small.

there are generally wrong assumption that PLS works well with little sample size (even 15 or less). however, such assumption is more discussed and criticized in recent years. you can see following paper for instance:

Marcoulides, G. A., & Saunders, C. (2006). PLS: A silver bullet?. Management Information Systems Quarterly, 30(2), 1.

The MISQ Editor's has provided some  guidance on how to ensure rigour when using PLS SEM.  This may prove useful to you

http://www.misq.org/skin/frontend/default/misq/pdf/appendices/2011/EdCommentsV35N2Appendices.pdf

scholars need to appreciate what Michel T originally said. Their work on sensory matters, wine etc is very diagnostic. Small n large no of variables. Rgs brad

Thanks guys. I learn a lot.

Small sample size in PLS - any thoughts?

You can refer to my answer shared in this RG link below:

https://www.researchgate.net/post/What_is_the_minimum_sample_acceptable_for_structural_equation_modelling_using_smartPLS

The minimum sample size is obtained from the following formula: 10 multiplied by the maximum number of indicators of a latent variable.

you may also refer to the following reference:

F. Hair Jr, Joe, et al. "Partial least squares structural equation modeling (PLS-SEM) An emerging tool in business research." European Business Review 26.2 (2014): 106-121.

In this article, the author has given certain guidelines related to the small Sample size. Further, I have attached a workshop slide related to Smart PLS.

Regards

Amir Ali