I'm using the popular mediation function (model 4) in PROCESS (Hayes, 2017). Given: 1 control, an IV, a DV, and one mediator, why would my direct and indirect paths be INSIGNIFICANT, but the total effect WAS? Perhaps because the control was significant with a large effect size? (it was). I mean-centered and standardized all items BEFORE analysis. Any help is appreciated.
************** TOTAL, DIRECT, AND INDIRECT EFFECTS OF X ON Y **************
Total effect of X on Y
Effect se t p LLCI ULCI c_cs
.0948 .0446 2.1274 .0345 .0070 .1827 .1446
Direct effect of X on Y
Effect se t p LLCI ULCI c'_cs
.0664 .0458 1.4500 .1485 -.0239 .1566 .1012
Indirect effect(s) of X on Y:
Effect BootSE BootLLCI BootULCI
.0284 .0165 -.0003 .0638
Unfortunately this does happen from time to time.
You can see that your total effect (.0948) is correctly decomposed to be the sum of your direct (.0664) and indirect effect (.0284).
You can see that your indirect effect is marginally significant at the .10 level (t about 1.72; p about .08 depending on your sample size). Likewise the direct effect is marginally insignificant.
Unfortunately these two mean estimates separately combine as an additive mean but with a lower standard error thereby a significant total effect at the .05 level bc in one model you have just one variable and the other you have the data working harder to infer he impact of two variables.
If you had either stronger effects (eg via cleaning) or lower errors around the estimates (via more data) you may get a clearer result. Alternatively, but unlikely to appease a reviewer, is to report your results as is at the .10 rather than .05 level. I would advise against that but does depend on the context.
Hope that helps and gives you some options to explore.
Paul
Thank you, Paul F Burke Paul. This was helpful, but I defended the dissertation last week. For the sake of the discussion, here is what I reported:
I definitely used the conventional a priori alpha = .05 and power = .80 and a frequentist, NHST approach.
I focused on the point that the paths were not significant:
1-no direct relationship between the IV and the DV, given the control.
2-M did not mediate the IV and DV, given the control.
I was adequately powered (n=399); I think I am dealing with small effects and noisy data.
Thanks again for your help!
Congrats on defending. Hope it went well. I think being honest about the results is appropriate and encouraged. It does sound like a reasonable sample.
One thing to then ask is why was there no effect? Could you hit the respondent harder on the head with your manipulation for example or was the mediator slightly off in terms of what you were intending to measure. Research is an evolving process by which we learn something evey time so taking that time to stop and reflect is a valuable endeavour.
Good luck with your research
Paul
Thanks, it went well. I think several limitations contributed. The measures lacked granularity, the mediator was theoretically derived, and the sample was unusual. I'll continue to familiarize myself with the PROCESS Macro and work toward a future study on the topic. Thanks again, Paul F Burke
hi, is it normal not to have an effect (b) in this scenario:
(LLCI=-.074 ULCI=.044) where there is no indirect effect?
versus the presence of an indirect effect:
(LLCI=.144 ULCI=.324, b=.224)
thx
Abdulaziz,
When there is no significant direct effect and significant indirect effect, you have complete mediation
When the direct and indirect are both significant you have partial mediation.
Both are found in many studies along with the case of no mediation.
Paul
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling