Beta is Minus one

3 replies [Last post]
khan's picture
Offline
Joined: 06/05/2010

Hi everyone,

I have one problem and this from will be helpful.

I am using PLS for analyzing my data, i have created product terms for testing moderating effect. But one of the interaction term (moderating effect) gives me -1.25 path coefficient. Can anybody help where is the problem? I double checked the data all the interaction terms are fine.

Thank you.

Ryne's picture
Offline
Joined: 07/31/2009
I'm going to assume that PLS

I'm going to assume that PLS is partial least-squares, which is a technique I'm not terribly familiar with. I'll also assume that your beta coefficient is supposed to be standardized, which is why a value in excess of one is a problem.

The more general question is whether standardized coefficients in any model can exceed one, and how to interpret them. Standardized coefficients can exceed one whenever a coefficient represents only part of a total effect. The most common example in the structural modeling approach is a two factor model with a negative factor correlation and cross-loadings.

If you reached your beta coefficient by standardizing the input variables, you could have a second issue. The product of two predictor variables with unit variance is not guaranteed to have unit variance itself. Your interaction term could not only have covariances with lots of other terms in the model, but could also have means and variances very far from zero and one, respectively.

The short version of this can be stated that it's OK if 1 standard deviation change in X leads to >1 SD change in Y, provided that change in X also leads to change in predictor Z can can adjust the total change in Y down under 1.

Hope this helps,
ryne

khan's picture
Offline
Joined: 06/05/2010
Hi Ryne Thanks for your

Hi Ryne

Thanks for your comments.

I am not sure weather the beta coefficient is standardized or not, this is the one PLS (partial least square) gives you by default.
I didn't standardized the input data, all the constructs are continues and measured on five Likert scale.

I am really poor in statistics and in PLS as well and really don't understand mush of your explanation. Could you provide me some step by step instruction how to overcome this problem? if you need more information about the data or the constructs please let me know.

Regards,

khan's picture
Offline
Joined: 06/05/2010
Hi, I solved the problem by

Hi,

I solved the problem by standardizing the data.

Thank you for your time.