Page 1 of 1

Moderating Effects and two concept questions

Posted: Wed Dec 14, 2005 7:56 pm
by kehanxu
Hi, there

In terms of the Moderating Hypothesis, We can have X as independent construct and Y as dependent construct. Z is the Moderating construct. X has reflective indicators X1-X2-X3, Y has Y1-Y2-Y3. Z has Z1, Z2, Z3. And we have Z1*X1, Z1*X2, and 7 others for the Interaction Effects’ construct.

But when I draw the three construct together, how I can know the Moderating effect is strong not weak? For example, X-> Y is 0.25; Z-> Y is -0.04; X*Z-> Y is 0.46. So can we say the negative relationship between Z and Y means Moderating Effect is strong (or weak)? And how to explain the positive 0.46 between the Interaction Effects (X*Z) and the dependent construct (Y)?

Thank you very much for your advice and time.

Also, what is the concept of ‘loadings’ and ‘weights’? Both SmartPLS 2.0 and AMOS 5 manual do not contain the two terms. Does Alpha means loadings and a small '1' in AMOS model means weights?

Posted: Thu Jan 05, 2006 10:24 am
by stefanbehrens

provided your path coefficients (X->Y and Z*X->Y) are both significant (Z->Y will likely not be), your results indicate a strong moderating effect of Z on the X->Y relationship. Chin (1998) provides the following interpretation of the Z*X->Y path coefficient:

0.25 is the X->Y path coefficient at Z=0
0.25+0.46=0.71 is the X->Y path coefficient at Z=1

Hope this helps.


'Goodness of Fit'

Posted: Fri Jan 06, 2006 8:42 am
by kehanxu
Dear Stefan,

Thank you very much for your kind help and advice.

Do we have measurement like 'Goodness of Fit' from SmartPLS output? I may need to compare the output of SmartPLS with AMOS for the same model.

Posted: Sun Jan 08, 2006 1:33 pm
by stefanbehrens
There really are no "real" goodness-of-fit statistics in PLS.
However, squared multiple correlations (R2) of your dependent variables allow you check overall model fit. For example, Chin(1998) calls an R2 of ~.66 substantial, ~.35 moderate and ~.17 weak model fit.

Beyond that you can check the predictive validity of your model using the Stone-Geisser Q2>0 criterium. Unfortunately, SmartPLS in its current release does not provide for a blindfolding procedure and thus cannot calculate the Q2 statistic. I guess, however, that the next release is supposed to address this shortcoming.