Good day,
For my bachelor degree project I have to do a 2 moderator design. I've already done the calculation and reported the results, however, I have yet to come up with an withstanding conclusion to the outputs of the statistical calculations.
There are 2 moderators influencing a a simple regression (basically the macro model 2 of PROCESS).
I have prepared a diagram with the names of the variables and the relationships. Note that the pseudonyms in brackets are meant for the shortened version of the names used in SPSS.
Basically, I used Haye's PROCESS 4.0 and SPSS to do the statistical work. I will post a pastebin link to the full outputs of the analyses. However, in the meantime, here is a short summary of my findings:
I did 3 separate tests. One with the full model, and 2 others where I "discarded" one of the two moderators to see the changes.
- First test - Full model (2 moderators and a predictor on a criterion)
Nvr↓ Hstr↓
Adzv------------------>Agr
17% of variance explained
Adzv, Nvr, (adzv x nvr) not significant
Hstr and Hstr x adzv significant ----> 1/2 interaction variables significant
- Second test - Hstr is omitted
Nvr↓
Adzv------------------>Agr
16.45% of variance explained
Nvr and (adzv x nvr) not significant ----> 0/1 interaction variable significant
Though Adzv becomes significant in this scenario
- Third test - Nvr is omitted
Hstr↓
Adzv------------------>Agr
11.97% of variance explained
Adzv not significant, but Hstr x Adzv is -----> 1/1 interaction variable significant
Now, Adzv and Agr have a .335 correlation (effect size 11%)
And Hstr and Adzv have a .237 correlation (effect size 5%)
Basically what I would like to ask is how to report the first test (is it statistically viable, having 1/2 interaction variables significant? Is it only partially viable?) and how to take in these changes that happen when one or the other moderators is omitted? From my very limited knowledge in statistics, I suspect there may be a suppressor variable at work here. Namely, Nvr might just not be a significant moderator, while Hstr might be suppressing Adzv, while also being a significant moderator for the model.
Thanks a lot in advance for all the good help! And I'm sorry for the shortened names of the variables, though they may not make sense in english, they are the shortened version of the romanian words used to describe them, hence why there may be confusion for "Adzv".
pastebin:
https://pastebin.com/2CSSwCZX
All the best!
Hello Sabin,
The full model shows support for moderation due to the Hstr variable, but not for the Nvr variable. The two single-moderator models essentially show the same thing: Hstr (Histrionism) singly or jointly shows moderation, whereas Nvr Neuroticism) does not.
The fact that Adzv does (when combined with Nvr only) or doesn't show significance (when combined with Hstr) as an IV suggests that Hstr shares variance with Adzv.
The difference in R-squared (between Adzv & Hstr model vs. full model) would likely be statistically significant, given your sample size, suggesting that Nvr (and the second interaction term) does bring something to the table. As you propose, this could well be an example of a suppressor effect.
Good luck with your work.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS