Calculating indirect effects for model with multiple mediators of which only SOME are binary

[anny fenton]

Dear All,

Slowly going mad as to how to calculate indirect effects for my mediation model with a continuous outcome and several mediators (some binary, some ordinal, some continuous).

medeff doesn't work because it can't handle multiple mediators

khb doesn't work because, as far as I can tell, it can only use one regression model at a time (e.g. only logit or only linear models)

GSEM doesn't seem to work since there's no way to estimate indirect effects with non-linear pathways

ldecomp doesn't work because it can't handle factor-variables.

Any ideas?

[tanya ogarra]

Hi Anny,

I am in a similar place - did you find a solution? Until I find a better way to tackle this problem, my approach has been to assume linearity of all the variables, including the binary ones, and use SEM - which then allows for indirect effects to be calculated. This is what is suggested in this paper by Robin Gomila https://www.robingomila.com/files/pu..._vs_Linear.pdf.

If you have another way of addressing this, I'd be very glad to know!

Best,
Tanya