Hi all, I am hoping to do a moderated mediation using Hayes's PROCESS Macro. The model is as follow:
IV: Binary 2 levels
DV: Binary 2 levels
W (mediator): Continuous 2 subscales
M (moderator): Binary 2 levels (Moderating the W-DV path and IV-DV path in the moderated mediation model).
I am intending to use Model 15 for my analysis.
In the case my DV has 4 levels (Not seek help, Formal Help Seeking, Informal Help Seeking, Both), is there any available model to handle this?
Appreciate your advice on this !!
If your DV has only 2 levels, PROCESS will automatically use logistic regression.
In case of your 4 levels example, neither OLS regression, nor logistic regression are appropriate. You may think about cumulative probit regression for example, to handle the ordinal DV.
Rainer Duesing Thank you very much Rainer, from looking at my model, do you think it is appropriate to use Hayes PROCESS Macro if I only assess the binary outcome? I am not an expert and most model I watched on Youtube are with Continuous IV/DV. Also, how can I plot the effect of moderator if my moderator only have 2 levels (as PROCESS has the option of plotting -1SD, Mean, + 1SD i believe)
1) I cannot say if it is appropriate. PROCESS will work with a dichotomous DV, yes, BUT you should not adjust the data to your model, but the model to your data. If your DV is multinominal or ordinal, you should use an appropriate model. PROCESS will not work here, but PROCESS is just ONE tool you can use and you are not obligated to stick to it.
2) If you are not an expert, then you should get some help from a local statistician maybe. Logistic, multinominal or ordinal regression have a quite different interpretation of the regression coefficients, which is not always intuitive, if you are used to OLS regression. Just recently I mixed it up and wrote something wrong here in RG. Therefore, if you are not sure what you are doing, how much would you trust your results? Can you be sure that you didnt miss any pitfalls? And there may be other aspects to consider, if you do not use OLS regression.
3) In case of dichotomous moderators, PROCESS automatically recognizes this adjusts the tables and figures. In case of multicategorical variables, you need to specify this separately.
Rainer Duesing Thank you very much Rainer, I will definitely get in touch with a stas expert to get help.
Hello Hau Nguyen. Rainer Duesing has given you some excellent feedback, IMO. However, I am not entirely convinced that the DV should be treated as ordinal. You wrote:
In the case my DV has 4 levels (Not seek help, Formal Help Seeking, Informal Help Seeking, Both), is there any available model to handle this?
I suppose that Rainer is ordering the categories this way: no help sought < informal help < formal help < both informal & formal help. But without knowing more about the nature of the informal & formal help, I am not certain about the ordering of the informal and formal help groups. Do you see them as being ordered along some continuum, Hau Nguyen?
- Badges
- Science topic
- Similar topics
- Social Science
- Sociology