Statistical models for analyzing count data: predictors of length of stay among HIV patients in Portugal using a multilevel model

Table 2 Tests and Fit Statistics (see the table in: Predicting length of stay from an electronic patient record system: a primary total knee replacement example)

PRM		BIC = 445,377.671	AIC = 445,042.083	Prefer	Over	Evidence
vs	NBRM	BIC = 198,352.015 AIC = 198,008.241 LR^a	dif = 247,025.656 dif = 247,033.841 prob. = 0.000	NBRM NBRM NBRM	PRM PRM PRM	Very strong p = 0.000
vs	ZIP	BIC = 432,948.578 AIC = 432,277.401 Vuong^b	dif = 12,429.093 dif = 12,764.682 prob. = 0.000	ZIP ZIP ZIP	PRM PRM PRM	Very strong p = 0.000
vs	ZINB	BIC = 198,103.519 AIC = 197,424.156	dif = 247,274.153 dif = 247,617.927	ZINB ZINB	PRM PRM	Very strong
NBRM		BIC = 198,352.015	AIC = 198,008.241	Prefer	Over	Evidence
vs	ZIP	BIC = 432,948.578 AIC = 432,277.401	dif = −2.346e+ 05 dif = − 2.343e+ 05	NBRM NBRM	ZIP ZIP	Very strong
vs	ZINB	BIC = 198,103.519 AIC = 197,424.156 Vuong^b	dif = 248.497 dif = 584.085 prob. = 0.000	ZINB ZINB ZINB	NBRM NBRM NBRM	Very strong p = 0.000
ZIP		BIC = 432,948.578	AIC = 432,277.401	Prefer	Over	Evidence
vs	ZINB	BIC = 198,103.519 AIC = 197,424.156 LR^a	dif = 234,845.060 dif = 234,853.245 prob. = 0.000	ZINB ZINB ZINB	ZIP ZIP ZIP	Very strong p = 0.000

PRM Poisson Regression Model, NBRM Negative Binomial Regression Model, ZIP Zero-inflated Poisson, ZINB Zero-inflated Negative Binomial
^a Verified with a likelihood ratio test
^b Verified with the Vuong test

ISSN: 1472-6963