Hi Louise,

In SmartPLS we can use Composite Reliability to measure reliability.  In order to measure Cronbach Alpha for reliability, you can go back to mainstream statistical software like SPSS etc.  Some researchers prefer to use Composite Reliability (CR) rather than Cronbach Alpha because Cronbach Alpha is being criticized for its lower bound value which underestimates the true reliability (Peterson & Kim, 2013). However, CR can be used as an alternative as its CR value is slightly higher than Cronbach Alpha whereby the difference is relatively inconsequential (Peterson & Kim, 2013) - see below bottom for the complete reference.

In order to adequately measure the moderating effect, you need to collect larger sample size because when many moderators are being introduced in a study, the interaction effect between IVs and moderators will create more new predictors (IV1 x Moderator1, IV-n x Moderator-n etc.) that can influence DV.  Some statistical analyses method used dependent on no. of predictors x certain constant to determine the no. of sample required. With small sample size, most of the moderating effect will not be significant.

Peterson, R. A., & Kim, Y. (2013). On the Relationship between Coefficient Alpha and Composite Reliability. Journal of Applied Psychology,98(1), 194-8. DOI:10.1037/a0030767

Hi Louise,

In SmartPLS we can use Composite Reliability to measure reliability.  In order to measure Cronbach Alpha for reliability, you can go back to mainstream statistical software like SPSS etc.  Some researchers prefer to use Composite Reliability (CR) rather than Cronbach Alpha because Cronbach Alpha is being criticized for its lower bound value which underestimates the true reliability (Peterson & Kim, 2013). However, CR can be used as an alternative as its CR value is slightly higher than Cronbach Alpha whereby the difference is relatively inconsequential (Peterson & Kim, 2013) - see below bottom for the complete reference.

In order to adequately measure the moderating effect, you need to collect larger sample size because when many moderators are being introduced in a study, the interaction effect between IVs and moderators will create more new predictors (IV1 x Moderator1, IV-n x Moderator-n etc.) that can influence DV.  Some statistical analyses method used dependent on no. of predictors x certain constant to determine the no. of sample required. With small sample size, most of the moderating effect will not be significant.

Peterson, R. A., & Kim, Y. (2013). On the Relationship between Coefficient Alpha and Composite Reliability. Journal of Applied Psychology,98(1), 194-8. DOI:10.1037/a0030767

Refer to

https://www.researchgate.net/publication/286480610_PLS_Guidelines

Method PLS Guidelines

Hey Louise, 

I first assume you are dealing with reflectively measured constructs, right? To assess their internal consistency reliability, you have two choices. Chronbach's alpha (the traditional criterion, which you can assess with SPSS and the like) considers the intercorrelations of your construct items and thereupon estimates the reliability. Here, it is assumed, that all indicators have the same loading on the your construct. Hence, the Cronbach's Alpha is also sensitive to the number of items in your scale and tends to underestimate your internal consistency reliability.

In PLS, however, some items will be more important for your construct than others, which implies different outer loadings on the construct. These different outer loadings are taken into account when you determine the Composite Reliability (CR) of your construct. The CR ranges from 0 to 1. Generally, a value between 0.60 and 0.70 is considered acceptable for exploratory research. However, you don't want to have values above 0.90, as this indicates that all your construct items are very much the same (e.g., only slightly rephrased) and you are unlikely to measure all necessary fascets of your construct. 

I hope that helps you to understand the difference between both criterions :)

If you want to read more on Smart PLS, I can really recommend you the book "A primer on partial least squares strucural equation modeling (PLS-SEM)" by Hair et al. (2013)

Hi Dr. Kelly

I think your using reflective construct PLS approach, as formative PLS approach does not require to have checked outer loadings for reliability and validity. In this model, we assume the indicators are not interchangeable and have also insignificant correlation.

In reflective constructed model, indicators are assumed to be interchangeable and thereby highly correlated, and that's why we need to thoroughly check their outer loadings, internal consistency and reliability, and their discriminant validity as well.

In my views, I strongly support to use composite reliability test than the cronbach''s alpha test, as the latter further undermines or exaggerates the consistency of the outer loadings.

In regard to the number of sample size, I think the main purpose of using smart PLS than the traditional SEM model is that Sem-pls is considered to be robust in handling small sample size. It's also good at addressing reliability issues as compare to that of the traditional sem model (involves certain complication). So a sample size of 65 is good enough to capture the relationship between independent and dependent latent variables.

Thanks !

Check this out:

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

I have just learnt a lot from your comprehensive answers. Thanks for sharing

I agree with the guidance given by previous scholars. Their is no contradiction in the information given. synthesize the information and take a position.