Turbine D: "As an engineer, we tend to think in terms of probabilities of failure, e.g., 10⁻⁹ or thereabouts, and designs are formulated and tested on a probability basis. What often is not thought of is "possibility" of failure. Possibilities are real."
That is very interesting. I work on prediction of response, or failure to respond (a.k.a. "resistance"), to anticancer drugs either in clinical development or approved by regulatory agencies and on the market (various tumor types). We look at various parameters (often called "biomarkers") that we think might have clinical utility in assessing risk. So we run stats on the data and come up with positive and negative predictive value, sensitivity, specificity, hazard ratios (with confidence intervals) and so on.
Like you in a sense, we are also interested in trying to understand and model or predict "failure", which in our case translates to progression of (usually metastatic) disease and death. In your case it's death by another means I suppose.
In coming up with models, you must have to assign different priorities or "weights" to the many different types of variable, no? I mean, all technical failures that can happen shouldn't be treated the same when trying to build a model I assume - or are they? Do you weight them differently, according to some estimate of criticality? I'm not a statistician but I run a fair bit of stats analysis for a living - sounds like you do as well. Sometimes you can't make real progress without some fairly heavy stats, at least in my experience.