Table 2.

Summary of statistics of near-infrared prediction models for colostrum immunoglobulins concentrations

Traita, g/L	n	Mean	SD	Math treatment	LV	RMSEC	R2C	RMSECV	R2CV	RMSEV	R2V	Relative RMSEV
IgG
100%	189	88.39	36.08	MSC 2,5,5,1	14	6.72	0.97	14.22	0.84	25.21	0.63	0.27
80%	152	91.14	36.28	SNV 2,5,5,1	14	5.26	0.98	14.55	0.84	29.15	0.53	0.32
60%	115	91.81	35.04	MSC 2,5,5,1	16	3.20	0.99	15.93	0.79	29.46	0.49	0.32
IgA
100%	153	4.25	2.48	MSC 1,4,4,1	4	1.84	0.45	2.00	0.35	2.23	0.40	0.46
IgM
100%	187	4.71	2.08	DET 2,5,5,1	2	1.58	0.42	1.73	0.31	2.10	0.32	0.41

SD standard deviation, LV number of latent variables, RMSE_C root mean squared error in calibration, R²_C coefficient of determination in validation; RMSE_CV root mean squared error in cross-validation, R²_CV coefficient of determination in cross-validation, RMSE_V root mean squared error in external validation, R²_V coefficient of determination in external validation, Relative RMSE_V relative root mean squared error in external validation, expressed as the ratio between the RMSE_V (this table) and the mean of the reference trait (Table 1).

IgG immunoglobulins G, IgA immunoglobulins A, IgM immunoglobulins M, MSC multiplicative scatter correction, SNV standard normal variate, DET detrend

^aPredicted from NIRS spectra using 100, 80, or 60% of calibration set