We use Deviance Information Criterion (DIC) and the posterior mean of the total residual deviance to assess Bayesian model fit. how to assess the frequentist model fit under network meta-analysis?
Hi Mona,
Under frequentist framework, the akaiki information criterion (AIC) is the alternative choice. I advise to read this book which gives you the criteria used either in case of Bayesian or frequentist situations.
Biondi-Zoccai G (2014) Network Meta-Analysis: Evidence Synthesis with Mixed Treatment Comparison. Hauppauge, NY: Nova Science Publishers.
Hi Mona
DIC is a model comparison/selection criterion and not a goodness-of-fit criterion. There are frequentist analogues such as the AIC as Safaa suggested.
For model fit, perhaps you refer to deviance residuals. Deviance and deviance residuals are not limited to a Bayesian framework. You may have them on a frequentist framework as well, in which you would get point estimates instead of a posterior distribution (and possibly its mean)
In my opinion, it is more important to focus on (in)consistency, which is a fundamental assumption of the NMA model, and heterogeneity rather than DIC and deviance
Best wishes
Dimitris
Thank You Safaa K. Kadhem and Dimitris Mavridis
Your replies are really helpful for me.
For consistency in a frequentist framework, you may have a look on papers here
https://onlinelibrary.wiley.com/toc/17592887/3/2
and
Article Visualizing inconsistency in network meta-analysis by indepe...
https://academic.oup.com/ije/article/42/1/332/699485
and R package netmeta and Stata network package have some options
https://cran.r-project.org/web/packages/netmeta/netmeta.pdf
https://www.stata-journal.com/article.html?article=st0410
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis