Hello everyone, what does it mean when the overall model (model summary) in PROCESS in SPSS is significant, however all of the effects of the predictors and their interactions are not significant? How do I interpret this?
I attach my PROCESS output for reference.
Thanks in advance!
Hello Irina,
What I see from your results is:
1. The only statistically significant effect/IV in the model is FFConsAv, whatever that represents.
2. It is most likely the influence of this variable that is driving the observed significance for the overall model.
3. As you note, neither main effects of TYPE or NFC_Av, nor their interaction is statistically significant, based on your data set.
4. There isn't anything paradoxical about the outcomes of the individual effect tests (four 1 df tests) vs. the test of the overall model (a single 4df test). The two sets do, however, have different sampling distributions.
Good luck with your work.
Your adjusted R2 is way lower than unadjusted R2 suggesting that you have a small sample and a lot of parameters. What's the N here?
Ronán Michael Conroy: Yes, the sample size is small (N=102) and I believe that this is a main limitation.
David Morse: FFConsAv is significant, however, this is a covariate in the model. Is it appropriate to state that this predictor only significantly affects the IV? Also can you please elaborate a little bit more on your 4th point?
I would like to thank both of you for your responses! It has been a great help to me!
PS: How would you suggest I interpret the results? I am a bit confused whether I must argue that the covariate (FFConsAv) is the only significant predictor.
With small sample size, you should be very careful of complex models such as interactions or mediation effects. In these, you are asking what happens if you hold some predictors constant and change others. The problem is that if you hold a lot of variables constant there may be little variation left in the the remaining predictor variables to use in your model. Consequently the uncertainty around the estimated effect goes up a lot.
It's always worth looking at univariate associations to see if the effect size (the coefficient) changes when you subsequently include the variable in a complex model, or if the SE goes up significantly. If the effect size is similar but the SE has gone up, you may well have a problem that the sample size is too small to estimate the effect with any precision.
Hello again Irina,
The FFConsAv (covariate) having a significant coefficient in the model, is shown to be related to the DV (not the IV/s). For your data set, it is the only one (of four) variables to have demonstrated a significant relationship with scores on the DV.
Is it possible that one or more of your three target variables would or could, in fact, have a non-zero impact on the DV? Yes, but apparently it would require a substantially larger sample size, as the observed effect sizes are not large enough (given the sample N) to show a statistically significant relationship.
Good luck with your work.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis