MODMED in PROCESS generates insignificant interaction term yet significant conditional effect: how to interpret? I have a moderated mediation model with a strong mediation effect, but with confusing results for the moderating effect (using PROCESS). Although the interaction effect in the regression model is insignificant, the conditional indirect effect does show that the mediated effect is conditional on the moderator (values above the 10th percentile are significant).
To get a better understanding of whats going on I also used the MODMED macro which enables me to use the Johnson Neyman Technique. What this shows is that the moderator has a significant effect at values of the moderator higher than 3,38 (i'm using a 5 point scale). Yet again, the interaction in the regression model is insignificant. 
I've been going through the literature but am unable to find an answer. Would you be able to help me?
 A: I'm not as familiar with PROCESS but I had a similar issue using MODMED. This is the answer I found on Andrew Hayes' MODMED documentation on his website: 
http://www.afhayes.com/public/modmed.pdf
MODMED will print conditional indirect effects even if the interaction corresponding to the moderated
effect is not statistically significant. Traditionally, in the absence of a statistically significant
interaction involving a moderator, one would interpret this to mean that the path is not moderated,
and so indirect effects should not be conditioned on the proposed moderator. It is recommended
that conditional indirect effects be interpreted as such only when there is evidence that the path or
paths specified in the model are actually moderated, using a statistically significant interaction as
the criterion in the model of that variable for making the decision. In the absence of evidence of
moderation, use the SOBEL or INDIRECT command for estimating unconditional indirect effects.
GOOD LUCK!
