I want to test the following mediation model:
X: stigma
M: social support
Y: medication adherence
I found earlier by regressing medication adherence on social support that this relationship is nonlinear. This was hinted by the scatterplot, so I included a quadratic term in the multiple regression model (social support x social support) which was significant.
Now I want to test this mediation model, but obviously the assumption of linearity is violated. How can I still test this? I was going to use model 4 with the PROCESS tool in SPSS. I have been thinking about doing parallel mediation with the quadratic term added but I am not sure whether this would be correct.
I also considered structural equation modelling, but I'm not sure how to incorporate this relationship. Could I for example do this by letting the errors/residuals of the two endogenous variables (corresponding to the linear and the quadratic term of social support) correlate?
Any suggestions?
Thanks a lot!