search

gertvv / gemtc

GeMTC R package: model generation for network meta-analysis
GNU General Public License v3.0
43 stars 25 forks source link

ISSUE with fixed effect (normal/identity, for mean difference) #66

Open CatherineRoyer opened 3 years ago

CatherineRoyer commented 3 years ago

Hi, I have an issue with the normal/identity:

By default normal/identity (without specifing the model)= random and if "fixed" is specified, the model doesn't take into consideration the "burn in".

example:

model1 <- mtc.model(network, likelihood = "normal", link = "identity") -------> the CODA included a convergence of SD (no SD for fixed effect, I guess it does a random effect)

image

model2 <- mtc.model(network, likelihood = "normal", link = "identity", linearModel = "fixed") results<- mtc.run(model2 , n.adapt = 20000, n.iter = 50000) -------> the 20,000 burn-in iteration is not taken into consideration

please see the "summary": image

CatherineRoyer commented 3 years ago

I realized that by default, it's "random" and not suppose to be "fixed"; so all good for my first question. Still have an issue with the second question: when "random" is specified with normal / identity, "n.adapt" (burn in) is not taken into consideration. Thanks in advance,

CatherineRoyer commented 3 years ago

@gertvv did you have time to verify that ? Is there a way to use your package for an fixed effect model for NMA in mean difference including the burn in iteration?