I am trying to find what the best method it is to estimate indirect effects of an experimental intervention on an outcome variable that follows a Poisson distribution, through continuous mediators.
Would OLS and thus PROCESS macro work well?
The direct effect models experience over-dispersion that cannot be fixed, so I am fearing that no estimation will be reliable.
Any advice would be much appreciated
I suspect you'd a full SEM package to model this. However, one alternative would be piecewise SEM which is more flexible in terms of the models than can be used for X->M and X->Y and M-Y paths (as long as you don't have latent variables). I'd fit this in R either using
https://cran.r-project.org/web/packages/piecewiseSEM/vignettes/piecewiseSEM.html
or brms if the model has serious estimation challenges (as the latter is a Bayesian approach that would fit the model using MCMC methods).
However, piecewiseSEM is pretty easy to use and supported by some good online tutorials. For the over dispersion you can either us a negative binomial or quasipoisson model (neither of which I think is supported in PROCESS).
It might be that you would find use for the graphical model-evaluation tools that are well-developed for OLS regression. In that case you could use the standard square-root (variance-stabilisation) transformation before using OLS.
- Badges
- Science topic
- Similar topics
- Biological Science
- Biochemistry
- Enzymology
- Enzymes