计数的统计误差.ppt

LLD Estimate the background intensity from measurements on non blank samples LLD Calculated in SuperQ/MA LLD Calculated in SuperQ/SS Regression Analysis Multi-linear Regression Analysis Calculates by minimising errors: A. slope intercept of calibration line B. inter-element correction factors (a, b, d, g) C. line overlap and background factors (L) using concentration count rate data from “standards” Multi linear Regression Analysis If errors are independent from their concentration level (are constant) and no error-weighting is applied, then the regression software minimizes and calculates the quality of regression expressed as: where n is the number of samples and k is the number of constants determined Errors reported in calibration Chemical composition of standards Best commercially available standards have a composition determination error of 0.1% relative; often worse. e.g. SiO2 content 0.5% which means: between 0.45% and 0.55% (i.e. ± 10% relative !) e.g. Ni content 10.2% which means: between 10.15% and 10.25% (i.e. ± 0.5% relative !) Multi-linear Regression Analysis The regression software is very powerful but large errors in the constants calculated are often caused by: Too many correction constants calculated simultaneously Poor data fitted with “fictitious” calculated correction constants Deleting standards that contain valuable information such as interference from line overlap Incomplete data on standards that must be used to correct for spectral interferences and matrix effects Human Errors during calibration Multi-linear Regression Analysis Systematic errors can be minimised by: use a minimum of 3*k well analysed standards make sure that the standards cover all the elements, ranges, line overlap factors and matrix effects to be expected in the analysis of unknown samples later the use of theoretical correction coefficients (fundamental parameters) for homogeneous specimens isolating each line overlap effect by careful choice o