My current study has almost 1000 responses. However, for one of the item I am interested in examining, it is not normally distributed (see attached image 1 for your reference). Since participants’ responses are really diverse on that item, so even removing some of the extreme outliers still cannot solve the problem.
I would like to run a moderation analysis using this item as the dependent variable. From different source of information on the Internet, I learned that normality should not be an issue for PROCESS macro as it provides the function to bootstrap.
From one tutorial, I saw that in order to not to care about the normality issue, we should select the “Bootstrap inference for model coefficients” option (see attached image 2 for your reference). However, from the other tutorials I read, they only mention about the number of bootstrap samples, without mentioning that we have to select the “Bootstrap inference for model coefficients” option. I tried to run the analyses with and without this option, and it changed from significant interaction effect of IV and moderator (when this option is not selected) to insignificant (when this option is selected).
I cannot really find the purpose of the “Bootstrap inference for model coefficients” option in Hayes's book or on the Internet, and I am also not really good at statistics. Therefore, I would like your help in the following:
Thank you in advance for your kind assistance.
What is the variable shown in that histogram? I ask because some other type of model may be appropriate, depending on what it is (e.g., a Poisson or negative binomial model if it is a count, etc.). Thanks for clarifying.
Thanks for clarifying. The first thing that occurs to me is that you might use some kind of time-to-event model, the event being the end of the emotional experience. A quick search suggests that you might find relevant examples in the oncology literature. Look for "duration of response". HTH.
Pcl Lee, why do you think it necessary to use the PROCESS macro for (so-called) moderation analysis? If you are using the term moderation the way most people use it nowadays, all you mean is that you wish to include an interaction in your model. PROCESS is not needed to estimate and interpret a model with an interaction in it. This UCLA Textbook Examples page shows some Cox regression models with interaction terms, for example, with no PROCESS macro in sight! ;-)
https://stats.oarc.ucla.edu/spss/examples/asa2/applied-survival-analysis-by-hosmer-lemeshow-and-maychapter-5-model-development/
HTH.
- Badges
- Science topic
- Similar topics
- Psychology