Accuracy of near infrared spectroscopy in relation to precision of calibration data

To obtain a more precise estimate of the accuracy of near infrared (NIR) spectroscopy in determination of total solids and fat in cheese, the measured standard error of prediction (SEP) should be corrected for imprecision of the reference method. High precision of reference data obtained through replicate analyses may not be the optimal goal in practice for the development and maintenance of calibration equations. In a situation of limited resources, it may be better to accept an increase in imprecision by reducing the number of replicate analyses for each sample and instead introducing additional samples. In the case here reported, it was possible to increase the standard deviation related to precision of reference data from 0.08 to 0.14% for total solids and from 0.08 to 0.16% for fat without any pronounced effect on the accuracy of NIR transmission spectroscopy in the calibration range 50-63% for total solids and 23-29% for fat. The mean SEPs were 0.19% for total solids and 0.18% for fat. if the standard deviation was increased from 0.08 to 0.32% for total solids and from 0.08% to 0.28% for fat, the mean absolute increases in SEPs were 0.02% for total solids and 0.03% for fat.