Reading Reliability Datasets That Were Never Designed for Analysis: Honest Interference from Imperfect Records (Missing Data)
The reliability data the engineer actually receives is rarely the data the engineer would design. Asset records begin where the enterprise system began, not where the asset’s life began. Failure dates are missing for assets withdrawn from service before record-keeping started. Censoring, missing data, is recorded when convenient, and omitted when it is not. The dataset that arrives on the analyst’s desk is a partial, time-bounded, intermittently informative window into a process that has been running, in many cases, for decades. This white paper sets out the discipline of analysing such datasets honestly: the structural categories of missing data, the analytical methods that accommodate them, the analytical methods that silently misrepresent them, and the practitioner’s responsibility to know which is which.