mixed model results

[Tommaso Salvitti]

Hi to everybody When reporting mixed model post hoc results, do you report the observed means or the estimated means (i.e. obtained from margins)...The "Contrasts" of the post hoc are based on the estimated means so I would be for the latter but I can't find a bibliographic reference. What do you recommend? thanks a million in advance Tommaso

[Tiago Pereira]

I am not sure if there is a specific recommendation. In health sciences, it is common to report the linear predictions (e.g., mean and 95% CI), specially in repeated measurements with missing data.

[Tommaso Salvitti]

Thank you...is there any bibliographic reference?

[Roman Mostazir]

I don't think you need a reference here. The observed mean is only useful as a descriptive statistic, something that is not generalisable in the wider population as the the raw mean is confounded. The model estimated mean is adjusted for confounders, given the model is adjusted with possible confounders, and this sample mean has the properties of generalisability. Further, if you don't use the model estimated mean, the choice of fitting the model become useless!

[Tiago Pereira]

Some suggestions to start:

https://journals.plos.org/plosone/ar...one.0112653#s6

https://bmcmedresmethodol.biomedcent...874-019-0876-8