Wednesday, November 19, 2008

Sample moderation/mediation APA write-up

Looking through the notes, I notice that we do not have a sample for how to write up a moderation or mediation results section like we had for Multiple Regression/Linear Regression. From the notes, I know what moderation and mediation mean, I just don't know how to write them up. I know how to use APA format, I just do not understand how to say what I need to say and what statistics are appropriate for moderation/mediation.


Justin said...

For a moderated model, you want to make it clear to your readers that the effect of one predictor on the outcome is dependent upon another predictor. You can say it pretty much just like that, assuming of course that your moderated model was statistically significant.

As for which statistics to include, that is going to depend to some extent on whether your predictor and moderator are categorical, continuous, or one of each. See slide 7 in the handout, entitled "Testing of Moderation."

If your predictor is continuous and your moderator is categorical, you will be regressing the outcome on the predictor twice (or more), once for each level of the moderator. The standard statistics that you have been reporting all along need to be reported for both (or however many) of these regressions. Then compare the unstandardized Bs of the two regressions. Tell your reader that because they are very different (or not), there is evidence for moderation (or not).

If your moderator is continuous, you need to run a regression using the predictor, the moderator, and the moderator by predictor (moderator times predictor) term on the outcome. Include all of the typical statistics you have been required to thus far for this regression. If the interaction term is significant, tell your readers that moderation is occuring.

For mediated models, make it clear to your readers that you are suggesting that a certain predictor influences the outcome variable through a second variable. For example, the amount of calories in an ice cream cone affects its perceived tastiness which then affects the weight of a person. Put another way, the amount of calories in an ice cream cone affects the weight of a person through its effect on perceived tastiness.

You will be running three regressions in order to demonstrate mediation. See slide 11, "Testing Mediated Models." You need all three pieces of information in order to demonstrate moderation, so include all of the typical statistics for each regression you ran.

Mari said...

You might find the last paragraph in the results section (on p. 297 of the pdf) of the following article useful to you IN REGARD TO TALKING ABOUT A TEST OF MEDIATION.
(note that you will need to paste this link into a word processor so that it is all on one line before pasting it in your web browser...alternatively, search for papers jointly authored by Stanley, Whitton, & Clements.)

As Justin said, you will want to include all the typical statistics for regression for the mediated model.

I'll look for a good source for the text for describing a moderated model, but if I do not find one, note that Justin's advice above is quite sound.