Variability sources, statistical validation

An indicator must also be reasonably amenable to measurement. Just what one measures is likely to depend on the facilities one has, and which types of figures are most likely to be complete, representative, accurate, rapidly available and statistically valid [23]. Some types of data can be (and often are) captured in large administrative data bases however these, as pointed out in Chapter 2, may have been designed for uses other than cost containment and policy review, and lack precisely the variables which one now needs. Data from such an information system may therefore prove disappointing when one puts questions which the system was not designed to answer. If one has any reason to doubt the rehability of the data in such a system, some form of peer review at the source will be advisable, for example to ensure that data are being entered consistently and in fine with prescribed procedures. 

Validation of the Component Manufacturing Variability Risk, is essential to CA in determining estimates for component characteristics at the design stage. Collecting component parts from various industrial sources with known statistical histories was central to this. The components were taken from a number... 

In many cases, the assumption of pure Poisson counting statistics is invalid because other sources of variability affect the distribution of measurement results. In these cases, the laboratory should consider determining the critical value and MDA based on a series of blank samples, as in Eqs. (10.18) and (10.19). When the Poisson model is valid but the background level is so low that the Poisson distribution is not approximately normal, other formulas for the critical value and MDA tend to give the best results, as discussed in MARLAP (EPA 2004). 

---