Probing Interaction Effects

This forum is closed, and read-only.
Locked
Alexander Kapeller
PLS Junior User
Posts: 3
Joined: Thu Sep 06, 2007 9:00 am
Real name and title:

Probing Interaction Effects

Post by Alexander Kapeller »

Hi All

I am interested in the construction of interaction effects.
The calculation in smart pls seems to be pretty convenient.

I read the book of Aiken and West as well as some related articles.
e.g. Preacher, K. J., Curran, P. J., & Bauer, D. J. (2006). Computational tools for probing interaction effects in multiple linear regression, multilevel modeling, and latent curve analysis. Journal of Educational and Behavioral Statistics, 31, 437-448.

I am now especially interested in probing the interaction at several levels of z (=moderator)

in Y = a + b1x + b2z + b3xz + e

this i can do with the covariance variance matrix. smart pls delivers in the output the correlation matrix of the latent variables.
so my request is how to derive the covariance matrix of the LV?

the computational way is to multiply the correlation with the variance, but where do i get the variance for the "latent variables" in smart pls or via output calculation procedures in spss etc.

Or is the varince of the latent variables set to "1" in the pls computation and the correlation is the covariance the same time?

Thanks for any suggestions in advance.

Alex
User avatar
ghozali
PLS Expert User
Posts: 39
Joined: Sat Oct 15, 2005 1:18 am
Real name and title: Prof. Imam Ghozali, Ph.D
Location: Indonesia

Post by ghozali »

To get covariance matrix from correlation data, you have also need mean, and its standard deviation. Output of PLS bootstraping will give you mean as well as std dev
Faculty of Economics, Diponegoro University
Jl. Erlangga Tengah 17 Semarang, Indonesia
ghozali_imam@yahoo.com
User avatar
Diogenes
PLS Super-Expert
Posts: 899
Joined: Sat Oct 15, 2005 5:13 pm
Real name and title:
Location: São Paulo - BRAZIL
Contact:

Post by Diogenes »

Hi,
in bootstrap we have the standard error of the paths coefficients, not the variance of the LV.

One way to have the covariance matrix computed is to copy/paste the Unstandardized LV scores to SPSS and use them to do this.

In this version of the SmartPLS (2.0 M3) there is a bug in the calculation of the unstandardized LV scores (viewtopic.php?t=812), but this has a little effect in the final results, for example, if you compute the correlation matrix using this scores, probably they will be a little different from the SmartPLS results. In the link above I had explained how to compute by hand to have the correct LV unstandardized scores.

Best regards.
Bido
Alexander Kapeller
PLS Junior User
Posts: 3
Joined: Thu Sep 06, 2007 9:00 am
Real name and title:

Post by Alexander Kapeller »

Thanks for the advice.

Bido, one more question, how do I calculate the covariance matrix from the unstandardized LV's

Best
Alex
User avatar
Diogenes
PLS Super-Expert
Posts: 899
Joined: Sat Oct 15, 2005 5:13 pm
Real name and title:
Location: São Paulo - BRAZIL
Contact:

Post by Diogenes »

Hi Alex,
just copy the unstandardized scores from SmartPLS

Report / Index Values / Results / Latent Variable Scores (unstandardised)

and past them to SPSS or Excel (or other statistical software) and compute the covariances between these scores.
Best regards.
Bido
Locked