Skewness/Kurtosis

Frequently asked questions about PLS path modeling.
Post Reply
brill5
PLS Junior User
Posts: 1
Joined: Sun Sep 24, 2017 7:25 pm
Real name and title: Tom Brill

Skewness/Kurtosis

Post by brill5 » Sun Sep 24, 2017 7:30 pm

I have a PLS analysis involving two items (from 7 point Likert-type scales) with negative skewness (-1.321, -1.029) and the same two items have positive kurtosis (1.490, 1.201).

In https://www.smartpls.com/documentation/ ... s-sem-book, mention is made that the limits are +/- 1. I have heard that PLS works with values up to +/- 2.2 but can't find any formal cites.

Given that SmartPLS touts working with nonnormally distributed data, what is the limit? ... and the formal cites for that limit? If SmartPLS doesn't work with the items mentioned above, how do I transform the values?
* Log10 transformation won't work for Likert-type scores

User avatar
cringle
SmartPLS Developer
Posts: 797
Joined: Tue Sep 20, 2005 9:13 am
Real name and title: Prof. Dr. Christian M. Ringle
Location: Hamburg (Germany)
Contact:

Re: Skewness/Kurtosis

Post by cringle » Fri Sep 29, 2017 4:23 pm

These descriptive statistics in SmartPLS give you the info if your data is normal or not. PLS-SEM is well-know to provide robust results when data is non-normal (as in your case). However, you may be cautious to some extent as explained in the section on distributional assumptions in this article on PLS-SEM: https://www.researchgate.net/publicatio ... n_Modeling

Best
Christian

lpfeil
PLS Junior User
Posts: 2
Joined: Tue Jul 03, 2018 9:06 am
Real name and title: Lisa Pfeil

Re: Skewness/Kurtosis

Post by lpfeil » Tue Jul 03, 2018 9:13 am

I'm interested in this topic, too. I have a PLS SEM model with Likert-type scales and many of my variables are negatively skewed (majority of variables have skewness and kurtosis >|1|, one has skewness>|2|). I chose PLS because of the explorative nature of my study and my main objective which is predicting and finding influences on one target variable.

Nevertheless, I wonder if the skewness is a problem in this case, as the above cited chapter and the Hair et al. (2017) book still keep saying that skewness and kurtosis below |1| are not critical - what if my data are more skewed?

Thank you in advance,

Lisa

Post Reply