Mediation Analysis- What does zero means?

[v.codec]

Hi

Can I check if my understanding of mediation analysis is correct.

For example, I have found the ab (indirect effect) is 0.01, can I assume that's very close to zero, therefore the size of the mediation effect is rather large? (almost full mediation?)

Does it also mean X is not really related to Y, the only reason that "X is related" to Y is because of the mediation effect.

My research question is whether X and Y has direct relationship. So can I reject the hypothesis that X is related to Y?

Any advice is appreciated. Thanks

[chiro]

Hey v.codec.

If you are testing an interaction effect with a statistical program, it should give you some sort of p-value which you can use to reject whether an interaction is present or not present.

It would be nice to know if you have used such output and what that output is.

[v.codec]

Hey v.codec.

If you are testing an interaction effect with a statistical program, it should give you some sort of p-value which you can use to reject whether an interaction is present or not present.

It would be nice to know if you have used such output and what that output is.

Hi

Correct me if I'm wrong. I think interaction analysis is to check for moderation instead of mediation?

Anyway, this is what I have done:

1. Regression to check for the interaction and it's not significant, p>.05.
2. Mediation Analysis using Hayes et al., 2012 (PROGRESS software).
Total Effect (c) - Unstandardised Beta = .02, p<.05
Direct Effet (c') - Unstandardised Beta = .01, p>.05
Bootstrap Test (Indirect Effect, ab) - Effect = .01, SE = .00, LLCL = .00, ULCL = .02 (Confidence Level = 95%, Boot Strapping Sample Size = 5000).

Is my above interpretation correct?

Thanks.

[chiro]

I haven't personally seen these test statistics before, do you have a link to them somewhere?

In general though, if you can clearly state what each test statistic (and it's region) corresponds to in terms of it's hypotheses, then you will understand in detail what these numbers are actually supporting with the data.

If you are unsure of the explicit nature of the hypothesis, you should give us more information about these statistics so the readers can see what the exact hypotheses are being tested.

[v.codec]

I haven't personally seen these test statistics before, do you have a link to them somewhere?

In general though, if you can clearly state what each test statistic (and it's region) corresponds to in terms of it's hypotheses, then you will understand in detail what these numbers are actually supporting with the data.

If you are unsure of the explicit nature of the hypothesis, you should give us more information about these statistics so the readers can see what the exact hypotheses are being tested.

Thanks Chiro- here are some background of the moderation and mediation

http://psych.wisc.edu/henriques/mediator.html

http://davidakenny.net/cm/mediate.htm

and this is the program that I use for the mediation- http://www.afhayes.com/spss-sas-and-...-and-code.html

Thanks for looking

[chiro]

I just took a look at the paper and what they are trying to do (the author with his procedure).

Basically, what is going on is that there is a regression model and one component specifies the direct effect. I am including this for your own understanding.

In a regression if we had a model say Y = B0 + B1*X + B2*Z + e, then if we had evidence that B1 was statistically significant (i.e. not to 0), then we would do a hypothesis test testing H0: B1 = 0, H1 B1 != 0, and this is pretty much what is going on with these kinds of effects.

Now if its the way I think it is, then what is happening is that a p-value for the direct effect being < alpha means that there is evidence that the direct effect is non-zero and thus significant.

In your case, this seems not to be the case (i.e. we fail to reject that the direct effect is statistically significantly zero) and thus infer no direct effect (which is what you have come up with as well).

It would be nice to get some other feedback on this as well, but just remember for the future that when you do calculate a test-statistic and do a hypothesis on it, you want to know what the hypotheses are referring to in english and in mathematical symbols.

When you have this, it means you can say exactly what you are testing which means you don't have to rely on rules that you don't know what's going on (i.e. ones that you trust blindly). You can see exactly what is being tested and how that relates to what you're trying to do.

If you want to understand my own interpretation, download the PDF document on that authors website and look at the regression model that involves the direct effect c'. You'll notice that this is what they call the direct effect and it means if this is zero, you can chuck the term out and it will affect the regression model in a specific way.

[SiBorg]

It would be nice to get some other feedback on this as well, but just remember for the future that when you do calculate a test-statistic and do a hypothesis on it, you want to know what the hypotheses are referring to in english and in mathematical symbols.

When you have this, it means you can say exactly what you are testing which means you don't have to rely on rules that you don't know what's going on (i.e. ones that you trust blindly). You can see exactly what is being tested and how that relates to what you're trying to do.

That's good advice which I am going to try to make use of...

[richieboy]

Hi SiBorg, did you ever solve your issue? Currently I am trying to test mediation in SPSS, but coming a little unstuck.

[v.codec]

Hi SiBorg, did you ever solve your issue? Currently I am trying to test mediation in SPSS, but coming a little unstuck.

what do you need to do?

[SiBorg]

I wasn't the OP here....