Calibration sample number determined by theory of sampling provide threshold for multivariate model building

Calibration model building is composed of suitable number of samples and multivariate regression techniques. The accuracy of prediction is determined by both factors (steps). In these two steps, the multivariate regression step is influenced by too many factors, making it impossible to determine the number of samples. However, sample number in sample collection step can be used to ensure population representation in statistics. The sample number is the cornerstone of the robustness of model that should be concentrated on; however, few instructions but some empirical expressions have been given up until now. The factors affecting the sampling accuracy include confidence level, relative standard errors, and relative representation requirements. The required number can be calculated by statistical parameters of population and the required representation. The relative standard error is an important factor related to the statistical parameters of the sample set. For general instructions, the calibration kit should use 100-150 samples, the more the better, but it is not recommended to use more than 200. These suggestions would help guide the operator by selecting an appropriate calibration sample number in spectroscopy.