Improving AMS uncertainties and detection of instrumental error

We emphasize the estimation of uncertainties of AMS results and the detection of instrumental error in the framework of commonly used methods of AMS data-analysis.The conventional methods are fast but raise the following four issues: (1) observed sampling variance has scatter that may lead to some true values being over five standard errors away from the mean. (2) Use of the Gaussian distribution for the end-result is unwarranted in several cases. (3) The standard error of the mean is slightly biased due to sampling and correlations. (4) Detection of instrumental errors could be improved. The Bayesian CAR model that we have introduced carries the calculations out with full probability distributions and uses an overall probabilistic process to describe the instrumental error. The scatter in the uncertainties given by CAR is an order of magnitude smaller than that of the sampling-based uncertainties, resulting in more reliable uncertainties. Compared to previous methods, better detection and estimation of instrumental error is also achieved.