CODEWITHSUNDEEP
rglmmtmb
Hannah Miriam Dancy
Hannah Miriam Dancy

Reputation: 25

AICc(): Error with logLik, cannot apply to object of class 'logical'

I'm trying to get an AICc table for a few of my models which were constructed using package glmmTMB. The model gives a logLik value but no AICc. When I put the model into AICc():

a <- print(AICc(model, trace = TRUE, 
                rank = "AICc", REML = FALSE))

I get this error:

Error in UseMethod("logLik") : 
  no applicable method for 'logLik' applied to an object of class "logical"

I have used other functions like AICtab() and have gotten the same error, so I believe it is in the model itself. If anyone is able to interpret this error here please let me know, thank you.

Edit:

Minimal dataset and model used:

ID <- c("A","B","C","A","B","C","A","B","C","A","B","C") 
#random effect
Sesh <- c("A1","B1","C1","A2","B2","C2","A3","B3","C3","A4","B4","C4") 
#nested random effect
Stim <- c("Old","New","Old","New","Old","New","Old","New","Old","New","Old","New")
Temp <- c(75, 76, 72, 80, 71, 65, 69, 60, 76, 80, 81, 60)
Total <- c(0,1,5,6,3,10,2,1,0,0,4,6)
z <- data.frame(ID, Sesh, Stim, Temp, Total)

m <- glmmTMB(
  Total ~ Stim + Temp + (1|ID/Sesh),
  ziformula = ~1,
  data = z,
  family = nbinom2)

Upvotes: 1

Views: 2412

Answers (2)

Kamil Bartoń
Kamil Bartoń

Reputation: 1562

Your model does not have likelihood (logLik(m) is NA), so it is impossible to calculate any likelihood-based criterion from it. This is presumably due to small sample size for a model with zero-inflation (the same model without ziformula gives logLik).

Also note that AICc (I assume it is MuMIn::AICc) does not have arguments trace, rank nor REML, hence the error. I believe you confused the command with dredge.

Upvotes: 1

NelsonGon
NelsonGon

Reputation: 13319

We can extract it manually(see NOTE):

summary(m)$AICtab
     AIC      BIC   logLik deviance df.resid 
      NA       NA       NA       NA        5

To directly get the AIC:

summary(m)$AICtab[[1]]
[1] NA

To get the AICc(I have not encountered this criteria in my studies at the time of writing):

MuMIn::AICc(m)
[1] NA

It is however the same output as above.

NOTE

  • It seems the developer(s) did not implement an AIC method for glmmTMB models so using AIC fails.

  • The above AIC is NA likely due to insufficient data. This answer is just to show how to extract the AIC manually. From the docs of AICc:

Calculate Second-order Akaike Information Criterion for one or several fitted model objects (AICc, AIC for small samples).

Upvotes: 0

Related Questions