I used PROCESS model 1 to calculate a moderation effect.
Y = general health questionnair
X = effort reward imbalance
W= overcommitment
The results show :
1- significant model,
2- non-significant effect of X and W on Y.
3- non significant interaction between X and W on Y,
Model : 1
Y : GHQ
X : ERI
W : OC
Sample
Size: 246
**************************************************************************
OUTCOME VARIABLE:
GHQ
Model Summary
R R-sq MSE F df1 df2 p
,4756 ,2262 134,2708 23,5784 3,0000 242,0000 ,0000
Model
coeff se t p LLCI ULCI
constant 12,6936 8,4220 1,5072 ,1331 -3,8962 29,2835
ERI -4,2675 8,8092 -,4844 ,6285 -21,6199 13,0849
OC ,3929 ,5635 ,6972 ,4863 -,7170 1,5028
Int_1 ,7137 ,5082 1,4046 ,1614 -,2872 1,7147
Product terms key:
Int_1 : ERI x OC
Test(s) of highest order unconditional interaction(s):
R2-chng F df1 df2 p
X*W ,0063 1,9728 1,0000 242,0000 ,1614
When conducting a linear regression analysis, it was observed that both variables ( X and W ) exhibited a significant effect. i expected that the main effects of variables X and W would also be significant in a moderation analysis.
How should I interpret these result?
Many thanks!
Hello Mehdi,
On the face of it, a significant model (with X, W, X*W) generally would imply that at least one of the individual predictors would be significant as well.
Given the sample size, I don't believe that the difference in sampling distributions: F(3, 242) vs. F(1, 242) would account for the apparent discrepancy.
Comparing ordinary MLRs of model 1 (X & W as predictors) vs. model 2 (X, W, and X*W as predictors) should address the moderation question. Your report of noting X and W as significant predictors is consistent with the process report of a significant overall (3 df) model.
One possible explanation is that the individual predictor test results you list were based on bootstrap estimates rather than parametric (asymptotic) estimates.
I suppose other explanations might be model mis-specification (wrong PROCESS set-up), or variable mis-codings (e.g., incorrectly coding categorical variable/s).
Good luck with your work.
David Morse Finally, when I mean centred the for construct of products, both X and W became significant. However, the interaction term is still not significant. therfore i suppose that we can not interprete the Conditional effect of focal predictor at values of the moderator, although it shows a trend. and actually the trend it's what it is expected. I genuinely appreciate your valuable response – thank you!
Mehdi Berber, what you observed about the interaction term is expected: The highest order interaction term in a model is unaffected by centering. IIRC, it is often described as "invariant" to centering. If you don't believe it, estimate your model after centering on different values. E.g., Center your variables on:
- Their minimum values
- Their maximum values
- Their medians
- Their means
As you will see, the results for the interaction term will be the same in every model. (But note that you must re-compute the product term each time using the centered variables.) HTH.
- Badges
- Science topic
- Similar topics
- Philosophy
- Esthetics
- Interpretation