Modeling count data for health care utilization: an empirical study of outpatient visits among Vietnamese older people

Table 2 Results of log likelihood, information criteria and goodness-of-fit measures

Model	K	Log (L)	AIC	BIC	RMSE	MAPE
PRM	34	− 8549.928	17,167.856	17,364.852	1.856	0.531
NB1	35	− 6085.813	12,241.626	12,444.416	1.680	0.433
NB2	35	− 5942.007	11,954.016	12,156.806	1.469	0.372
HNB1	69	− 5751.954	11,641.908	12,041.694	0.491	0.184
HNB2	69	− 5698.735^a	11,535.471^a	11,935.257^a	0.405^a	0.160^a
HPM	68	− 8314.001	16,764.001	17,157.993	1.818	0.485
ZIP	68	− 8332.979	16,801.958	17,195.95	1.931	0.518
ZINB2	69	− 6218.109	12,574.219	12,974.005	1.667	0.426
LCNB2	72	− 5769.265	11,682.53	12,099.698	0.905	0.253

PRM stand for the Poisson regression model, NB is the negative binomial, HNB is the hurdle negative binomial, HPM represents the hurdle Poisson model, ZIP the zero-inflated Poisson, ZINB is zero-inflated negative binomial, and LCNB the latent class negative binomial. K is the number of parameters estimated for each model, Log(L) denotes log likelihood, and RMSE and MAPE stand for root mean square error and mean absolute prediction error, respectively
^aindicates the preferred model

ISSN: 1472-6947