Table IV.

Model performance results from MTFS dataset

Model	AIC	Deviance Loss Rank	Quadratic Loss Rank
Normal	12,759.36	4	3
Lognormal	3886.57	3	4
LGP	1620.20	1	1
NegBin	1646.82	2	2

notes: AIC = Akaike’s Information Criterion, LGP = Lagrangian Poisson, NegBin = negative binomial. Roman numerals I, II, and III refer to different methods of incorporating fixed-effects regression into the model. Deviance loss and quadratic loss were averages from five-fold cross-validation. They are presented here ranked from smallest (1) to largest (4). Deviance loss is −2 times the model’s log-likelihood when all of its parameters are fixed at estimates from the calibration data. Quadratic loss is a squared error of prediction metric (details in text).