DataAnalysisToolkit#5UncertaintyAnalysisandError.pdfVIP

Data Analysis Toolkit #5: Uncertainty Analysis and Error Propagation Page 1 General statement of the problem If some quantity of interest z is calculated from other quantities x, y, q, w, etc., z = f ( x ,y ,q, w,L) what is the uncertainty in z, and how is it related to the uncertainties in x, y, q, w, etc? In other words, given the functionf above, what is the corresponding function g s z = g ( sx , s y , s q , s w , L) that permits us to compute the uncertainty in z from the uncertainties in its component parts? Note that we could estimate the uncertainty in the average value of z (the standard error) from the standard errors of the component means (as in the formula above). We could also calculate the uncertainty (or variability) in individual estimates of z (the standard deviation) using the same function g, but in this case the inputs are the standard deviations of the input variables x, y, q, and w, rather than the standard errors of their means: s z = g ( sx , s y , s q , s w , L) Sources of uncertainty 1. statistical error/random variation of replicate measurements 2. spatial and temporal variability 3. systematic error (bias) 4. imprecise definitions or unrepresentativeness of samples 5. uncertainty in the form of the function relating z to x, y, q, w, etc. This toolkit explains methods for quantifying uncertainty that arises from random measurement error and from spatial or temporal variability (where one wants to average over that variability). Uncertainty arising from sources (3)-(5) is not adequately addressed by these methods (or by any other general techniques either). Ways to