mediation in SPSS

[AryaStark]

Advertisement - Members don't see this ad

Does anyone know how to do a mediation in SPSS? I don't know if there's a specific point and click method or a syntax used... This may be an incredibly dumb question... er... but really..

 

[deleted176373]

Does anyone know how to do a mediation in SPSS? I don't know if there's a specific point and click method or a syntax used... This may be an incredibly dumb question... er... but really..

It's not "point and click" but this might help.

http://www.comm.ohio-state.edu/ahayes/sobel.htm

Mark

 

[Ollie123]

There is no point and click. Mediation is not really a specific type of test, but rather a technique you can use to infer mediation using existing techniques.

You will need to be more specific about what you want to do: baron & kenny, sobel, various bootstrapping techniques for single and multiple mediators, sem & path analysis, ways to infer causation using a variety of other techniques (some of these I'm not a huge fan of...).

 

[IOPsych]

you typically use regression to test for mediation, although you could use anova.

1. The IV should predict the DV.
2. IV should predict the mediator.
3. the Mediator should predict the DV.


and then you add both the IV and the mediator into a multiple regression equation and the IV should no longer significantly predict the DV (for full mediation).

 

[erg923]

I agree with Ollie here! Many diferent ways to do mediations. You can PM if you want, as I just did one last month for a friends project that I am gonna be on.

 

[cmuhooligan]

you typically use regression to test for mediation, although you could use anova.

1. The IV should predict the DV.
2. IV should predict the mediator.
3. the Mediator should predict the DV.


and then you add both the IV and the mediator into a multiple regression equation and the IV should no longer significantly predict the DV (for full mediation).

This is an old convention and not currently accepted among quantitative psychologists. Look at the works of Preacher, Hayes, Mackinnon, and Shrout for modern approaches that move beyond the Baron and Kenny method.

 

[DynamicDidactic]

This is an old convention and not currently accepted among quantitative psychologists. Look at the works of Preacher, Hayes, Mackinnon, and Shrout for modern approaches that move beyond the Baron and Kenny method.

For Preacher
http://www.people.ku.edu/~preacher/medn.htm

 

[jcam17]

My understanding is that Baron and Kenny is still an acceptable way to demonstrate mediation. The problem with it is that it is conservative. If I recall correctly, the newer models allow for finding mediation when the direct IV-DV relationship is not necessarily significant. Therefore, while Baron & Kenny is still an acceptable way to demonstrate mediation, you will be more likely to find mediation using a different method.

Feel free to correct me if I am off base!

 

[cmuhooligan]

My understanding is that Baron and Kenny is still an acceptable way to demonstrate mediation. The problem with it is that it is conservative. If I recall correctly, the newer models allow for finding mediation when the direct IV-DV relationship is not necessarily significant. Therefore, while Baron & Kenny is still an acceptable way to demonstrate mediation, you will be more likely to find mediation using a different method.

Feel free to correct me if I am off base!

The problem is that is it underpowered to detect mediation...in monte carlo studies it is the worst performer. So yes, of course one can use it, but why would one want to? What benefit does it provide over modern methods?

Fritz, M. S., & MacKinnon, D. P. (2007). Required sample size to detect the mediated effect. Psychological Science, 18(3), 233-239.

Also see this: http://www.comm.ohio-state.edu/ahayes/CM2009.pdf for a great overview of mediation methods.

 

[IOPsych]

This is an old convention and not currently accepted among quantitative psychologists. Look at the works of Preacher, Hayes, Mackinnon, and Shrout for modern approaches that move beyond the Baron and Kenny method.

while one criticism of Baron and Kenny is that it often lacks power to detect mediation even if it is present, if and when it does detect mediation one can be extremely confident mediation does in fact exist, correct?

So why would quantitative psychologists not accept mediation tested in this way? If anything it is a conservative test for mediation, greatly lowering your chances of a type I error.

I simply mentioned a common way to test for mediation using SPSS, not that it was the only way.

 

[cmuhooligan]

while one criticism of Baron and Kenny is that it often lacks power to detect mediation even if it is present, if and when it does detect mediation one can be extremely confident mediation does in fact exist, correct?

So why would quantitative psychologists not accept mediation tested in this way? If anything it is a conservative test for mediation, greatly lowering your chances of a type I error.

I simply mentioned a common way to test for mediation using SPSS, not that it was the only way.

Another criticism is that is does not actually directly test what is of interest--the indirect effect, which is why it is often supplemented by the sobel test (which has its own problems, namely assumptions of normality that are rarely met). Because it does not actually test the indirect effect, but rather involves a number of "causal" steps, you increase your experimenter-wise error by running superfluous tests of various paths. As I mentioned earlier, check out Andrew Hayes' paper for a nice walk through. I didn't mean to offend you though, but I did want the OP to know that the field is slowly turning away from the Baron and Kenny method (although I still see it published quite a bit). Just my two cents.