How to report on model selection

Question

I have run model selection and have been advised to use the BIC value to determine the best model for my purposes.

However, having looked around online I cannot seem to find a standard way to report results, even published papers have different ways of reporting them. Could someone please tell me how to report the BIC value and what other values I need to report? Does anyone know how to generate a table of the BIC results in r?

Thanks

Answer 1

I haven't used BIC, so there may be some differences, but I have used AIC a lot. I will generally report (some variant of) AICcmodavg::aictab results. Exactly what you report will depend on your audience, e.g. different journals may have different reporting requirements, or you may want a very brief summary if it's for a non-technical audience.

An example from the help package for aictab gives

> aictab(cand.set = Cand.models, modnames = Modnames, sort = TRUE)

Model selection based on AICc:

      K   AICc Delta_AICc AICcWt Cum.Wt     LL
mod 2 9  89.98       0.00    0.8    0.8 -35.18
mod 1 7  92.77       2.79    0.2    1.0 -38.89
mod 5 6 115.52      25.54    0.0    1.0 -51.39
mod 4 5 121.02      31.04    0.0    1.0 -55.25
mod 3 4 138.14      48.16    0.0    1.0 -64.90

Where

• the row name is a description of the model,
• K is the number of parameters,
• AICc is the small-sample AIC (you'd use BIC here),
• Delta_AICc is the difference between the minimum AICc and that particular model's AICc
• AICcWt is the "model probability" out of the candidate set
• CumWt is the cumulative sum of the model probabilities
• LL is the log-likelihood of the model