BMC Medical Research Methodology

Table 2 Prediction criteria results of the gamma scale and gamma shape and gamma shape/scale models for noninstitutionalized veterans in fiscal year 2017

From: Better performance for right-skewed data using an alternative gamma model

		Population
Criterion	Model ^a	Overall	GEC Cohort	HBPC	JFI < 6	JFI ≥ 6
N		4 838 736	330 358	69 649	4 427 618	411 118
R²	Scale	Neg^b	0.40	0.45	Neg^b	0.37
R²	Shape	0.57	0.62	0.61	0.41	0.57
	Both	Neg^b	0.51	0.50	Neg^b	0.45
RMSE	Scale	118 526	52 099	44 439	25 483	55 478
RMSE	Shape	18 592	41 585	37 501	11 973	45 826
	Both	108 444	46 881	42 301	22 458	51 790
Max Error	Scale	-56 164	-46 181	-24 676	-17 232	-46 239
Max Error	Shape	8 369	9 198	9 789	4 675	9 436
	Both	-49 436	-27 233	-14 737	-13 753	-32 706

^aScale = gamma scale model; Shape = gamma shape model; Both = gamma shape/scale model with covariates impacting both shape and scale parameters
^bNeg denotes negative R square values, which is possible if the regression model does not match the true conditional expectations

Back to article page

ISSN: 1471-2288

Contact us

General enquiries: journalsubmissions@springernature.com