# Mediation - indirect effect, total effect and direct effect are negative

#### Jana Doe

##### New Member

I did mediation analyses in Spss and Process program. I took simple mediation, model 4. I got my results. Total effect x on y is -.412 and it is statistically significant. Direct effect x on y is -.278 and it is also stat significant and indirect effect is -.134 (CI 95) lower -.205 upper -.071. My question: is indirect effect significant, because it doesnt include zero? Thanks.

Last edited:

#### hlsmith

##### Less is more. Stay pure. Stay poor.
Given what you have provided, yes. If confidence interval excludes the null values then they are different from 0 at the set alpha level.

#### Jana Doe

##### New Member
Thank you very much! I have dilema because mediation effect is negative. Can I interpret these results that direct effect is bigger but indirect effect even though is smaller it is stat significant, and we can speak about mediation?

Last edited:

#### hlsmith

##### Less is more. Stay pure. Stay poor.
To say one effect is bigger you ideally would compare them some how. You may get away with such a comment if their confidence intervals don't overlap, a pseudo proxy for a formal test. Did you find the negative associations surprising? Also, people should try to not get stuck on the "significant" terminology and pay more attention to the effect sizes and precision of those estimates.

Last edited:

#### Jana Doe

##### New Member
Direct effect is -.278 and it is statistically significant p .000 (Lower -.445 and upper -.112) and indirect effect is -.134 (CI 95) Lower -.205 Upper -.070, and it is also stat sign because excludes zero. I thought because both effects are significant that i can interpretate that I have direct and also indirect effect x on y? I didnt know that i have to look at overlapping of CI.

Last edited:

#### hlsmith

##### Less is more. Stay pure. Stay poor.
You don't have to look at overlapping CI's for that. In post #3, you write, direct effect is bigger. I was responding to whether you were trying to truly say it is bigger or not. If you aren't trying to do that, then ignore the overlap part.

#### Jana Doe

##### New Member
I didn`t understand you. Sorry. But now i am not sure can i say that i have mediation effect? From these results i can see that CI are overlapping in some value.

#### hlsmith

##### Less is more. Stay pure. Stay poor.
If your CIs exclude '0' then you have mediation given your model and alpha level.