Hi Mohd,

Since you are using SmartPLS, assuming you are doing variance-based SEM, sample size is determined by:

1) 10 times the largest number of formative indicators used to measure a single construct, or

2) 10 times the largest number of structural path directed at a particular construct in the structural model

The above are extracted from:

Hair, J. F., Hult, G. T. M., Ringle, C. & Sarstedt, M. (2013). A Primer on Partial Least Squares Structural Equation Modeling (PLS-SE). Sage Publications, Inc.

Regards,

Fung

Hi Mohd,

Since you are using SmartPLS, assuming you are doing variance-based SEM, sample size is determined by:

1) 10 times the largest number of formative indicators used to measure a single construct, or

2) 10 times the largest number of structural path directed at a particular construct in the structural model

The above are extracted from:

Hair, J. F., Hult, G. T. M., Ringle, C. & Sarstedt, M. (2013). A Primer on Partial Least Squares Structural Equation Modeling (PLS-SE). Sage Publications, Inc.

Regards,

Fung

The selection of the sample size should take into consideration the desired power levels,   effect sizes and significance (Cohen 1992).

So the target sample size should be computed in that context .

In addition to the above considerations,  the sample size will also be dependent on the number of LV's and the number of indicators per LV.   The guidance varies and include;

a. 20 subjects per variable - (Costello and Osborne 2005)

b. 5 subjects per variable (Bentler and Chou 1987)

c. 10 cases per indicator ( a widely used but incorrectly understood rule of thumb)  (Chin 2011)

d. 10 times per largest number of paths from IV's going into a DV in a PLS SEM model (Chin)

Generally however, larger sample sizes will tend to produce more reliable results.

A critique on PLS is given in the 2010 Antonakis paper in the Leadership Quarterly "On making causal claims: A review and recommendations" (sec 4.2.1.2 on page 1103)

You might therefore want to use the opportunity to examine some alternatives to PLS.(eg Two-stage Least Squares 2SLS)

Thank you all for your quick guidelines and reference. It really helps me in determine my sample size.

However, a strong word by Antonakis (2010) did worry me in using PLS.  Any advice?

Please, beware that the statistic for the estimator plays an important role depending on the sample. It is possible to work with small samples as well. I'm attaching here a reference that you might find interesting. 

http://www3.nd.edu/~kyuan/courses/sem/readpapers/Bentler-Yuan-MBR1999.pdf

For some guidance on  how to ensure rigour when using PLS SEM in your data analysis I would recommended the following Editor's advice from MIS Quarterly:

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

Very much thank you Silburn Clarke and Pedro Pires. It helps me out indeed.

A common mistake in determining sample size is to use the often-cited 10 times rule and use it citing Hair et al. (2013). It is a mistake! Hair in his book "a primer on partial least squares structural equation modeling states that: "UNFORTUNATELY(!!) some researchers believe that sample size consideration do not play a role in the application of PLS-SEM. This idea is fostered by the often cited 10 times rule which ..." (page 20)

He says that it is not the proper way and he says it is better to use Cohen (1992)or use programs such as G*Power. 

I suggest you to use G*Power using Dattalo (2008) settings (alpha= 0.05 and beta=0.80).  use F test (linear multiple regression: fixed model. R2 deviation from zero) and for the type of power set " a priori" compute required sample size= given alpha, power and effect size"

In the use of PLS-SEM, there is no problem and you can back it up with a set of journal articles. PLS is useful specially when your distribution is not normal, and your sample size is not that big. it is also useful for formative measurement 

Cheers 

I have a very long experience in practicing PLS-SEM on real world studies. My conclusions are the followings:

1) When the effects are strong you do not need much individuals.  For example only one spoon of soup is enough to find it too salted. In the paper  "Tenenhaus, Pages,  Ambroisine, Guinot : PLS methodology to study relationships between hedonic

judgements and product characteristics, Food Quality and Preference 16 (2005) 315–325", only 6 orange juices were enough for the PLS-SEM model.

2) Computing confidence intervals by bootstrapping gives you the right information on what you can do with your data.  Considering the significant parameters, you extract exactly from the data what they can give.

3) I do not dare to work with less than 6 individuals.

4) I do not believe in any rules about sample sizes: the proof of the pudding is in the eating.

Dear Michel,

what do you mean by enough? Find significant effects? I'm not sure, but is it really a good idea to use bootstrap on 6 observations. Shouldn't the standard errors be deflated with 6 observations and e.g. 999 bootstrap draws? And couldn't that be the reason why you get significant results? 

Kind regards,

Florian

I was surprised by the results of my paper on orange juices.  But the significant results found in that paper were validated by sensory data experts. Also, using on the same data PLS regression models yields the same significant results. Again "the proof of the pudding in the eating". 

PLS-SEM is exploratory.  It is a way to communicate some results on your data.  Your findings have to be validated by some experts of the field.  I will never use PLS-SEM as a tool to validate some theory.

Thank you!

The minimum sample size in PLS-SEM is obtained from the following formula:

10 multiplied by the maximum number of indicators of a latent variable

Srs

Meu modelo tem duas variáveis latentes, cada uma com dois indicadores.

Minha amostra é de 57 questionários válidos.

Vou utiliizar PLS-SME e estou inseguro com minha amostra.

Como justificar?

Obrigado.

the power analysis suggest on the basis of your predictors and the effect size through G*Power.

Check this out:

https://www.smartpls.com/resources/ebook_on_pls-sem.pdf

I am also stuck in determining the sample size. i am getting the following error when calculation the PLS algorithm using SmartPLs Ver 3 " Sample size too small : There must be at least 16 cases/observations.? Any who can help?