Problems with correlated data

摘要

A misunderstanding exists in the PRA (probabilistic risknassessment) field over what constitutes correlated data. This reportnclarifies the applications that fit the initial intent of thendefinition. In addition, even when used as intended, current theorynappears to give an overly conservative answer. A more realistic answer,nwhich is still conservative (e.g., overestimates the failure frequencynof the group being estimated), is obtained by assuming the data arenuncorrelated. Definition: “correlated” data (as used in thisnpaper) are data linked by a common data-distribution; i.e., if twonseparate components derive their failure rate from this samendistribution, they are “correlated”. This is not the same asnstatistically correlated data wherein the data can always benstatistically correlated (e.g., race, sex, age, and education of poornpeople)