Moderation at some levels of moderator but not in overall model

#1
Hello

I have three continuous variables one of which I have hypothesised to be a moderator. In the overall regression model there is no significant result for the product of the IV and predicted moderator, R2 change = .00, F change (1, 118) = 0.46, p >.05. That is, there is no overall moderating effect. However, when I use the PROCESS plug in to test for possible combined effects at different levels of the hypothesized moderator, using both the pick a point and Johnson Neyman technique a moderator effect can be seen at certain levels of the moderator.

My question:
1) Is it justifiable to proceed to test for combined effects at different levels of the moderator once you have shown there is no overall moderator effect?
2) Are there published papers which explain this approach?

I have searched the forum and noted a few other people have posted similar questions but to date there have not been any suggestions as to the best way to proceed.

Any advice would be appreciated.
 
#2
Hello

I have three continuous variables one of which I have hypothesised to be a moderator. In the overall regression model there is no significant result for the product of the IV and predicted moderator, R2 change = .00, F change (1, 118) = 0.46, p >.05. That is, there is no overall moderating effect. However, when I use the PROCESS plug in to test for possible combined effects at different levels of the hypothesized moderator, using both the pick a point and Johnson Neyman technique a moderator effect can be seen at certain levels of the moderator.

My question:
1) Is it justifiable to proceed to test for combined effects at different levels of the moderator once you have shown there is no overall moderator effect?
2) Are there published papers which explain this approach?

I have searched the forum and noted a few other people have posted similar questions but to date there have not been any suggestions as to the best way to proceed.

Any advice would be appreciated.

If a joint significance test (I'm assuming what you mean by "overall significance") of the terms relating to a particular moderator is not significant, it doesn't make much sense to test any of the terms individually. The joint test answers the question "Is at least one of these terms useful for predicting the DV, all else constant?" If it's nonsignificant, you can see that the answer is no and you should only followup on a significant test.

I hope I've understood your question.
 
#4
Yes I think you have understood my question and I appreciate you taking the time to share your advice.
Glad you thought it was useful. Always remember to think of the hypothesis tested at each step. This can help you contextualize the result of a given test and help you proceed to the next step.