Computes model weights and a Bayes factor by comparing two groups of models based on their Information Criterion (IC) values. The function works with either numeric vectors or data frames containing multiple IC measures (e.g., MD, BPIC, DIC).
Value
A data.frame with the following columns:
wForThe aggregated weight of the models in favor.
wAgainstThe aggregated weight of the models against.
FactorThe Bayes factor (ratio of
wFortowAgainst).
If IC_for is a data frame, a matrix with rows corresponding to each IC measure is returned.
Details
When provided with numeric vectors, it computes the weights for the two groups by first
converting the IC values into relative weights and then normalizing them. When provided with
a data frame, it assumes that the data frame is the output of a call to compare
and applies averaging to each IC metric
Examples
# First set up some example models (normally these would be alternative models)
samples_LNR2 <- subset(samples_LNR, length.out = 45)
samples_LNR3 <- subset(samples_LNR, length.out = 40)
samples_LNR4 <- subset(samples_LNR, length.out = 35)
# Run compare on them, BayesFactor = F is set for speed.
ICs <- compare(list(S1 = samples_LNR, S2 = samples_LNR2,
S3 = samples_LNR3, S4 = samples_LNR4), BayesFactor = FALSE)
#> DIC wDIC BPIC wBPIC EffectiveN meanD Dmean minD
#> S1 -615 0.210 -579 0.197 35 -650 -657 -686
#> S2 -615 0.223 -580 0.223 35 -650 -657 -686
#> S3 -615 0.282 -580 0.295 35 -651 -657 -686
#> S4 -615 0.285 -580 0.285 35 -651 -657 -686
# Model averaging can either be done with a vector of ICs:
model_averaging(ICs$BPIC[1:2], ICs$BPIC[2:4])
#> wFor wAgainst Factor
#> 1 0.3431633 0.6568367 0.5224484
# Or the output of compare:
model_averaging(ICs[1:2,], ICs[3:4,])
#> wFor wAgainst Factor
#> BPIC 0.4197186 0.5802814 0.7233018
#> DIC 0.4324931 0.5675069 0.7620930