Computer-intensive methods for uncertainty estimation in complex situations

International regulations require the specification of an uncertainty estim ate related to experimental data. In chemistry, the situation that a straig htforward statistical machinery is not available for assessing the uncertai nty of a datum extracted from complex systems often occurs. Non-linearity, non-normality, correlation and other nuisance factors add to the complicati ons. Monte Carlo resampling algorithms, in combination with abundant fast c omputing power, have made techniques feasible that do not require profound mathematical insight, but, nevertheless, are fairly general. Assessment of confidence limits at different levels of correctness is discussed using sta ndard and bootstrap methods. Inferiority of standard normal approaches beco mes evident even in mildly non-linear situations. (C) 2000 Elsevier Science B.V. All rights reserved.