This is just an Excerpt from a larger document, click here to view the entire document.Confidence Levels of Predictions
In general a reliability prediction cannot be linked to a specific confidence interval, as might be done with a demonstration test or when measuring failure rates from field returns (References 9, 10, and 11). The primary reasons for this inability to define a confidence interval are:
Reliability prediction models, including 217Plus™, FIDES Guide and MIL-HDBK-217, are typically based on part data gathered from a variety of sources. Complete models are not usually developed from a single data source.
In some cases, while it might be possible to calculate a confidence interval for some basic part failure rate, it is practically impossible to predict the confidence interval for all of the modifying parameters, even when they are based upon well known and widely used physical acceleration laws, e.g., Arrhenius or the Inverse Power Law.
In addition to the variability associated with developing the models, there is also human variability involved in making prediction assumptions, analyzing the data, counting of field failures, and even in the failure definitions themselves.
Thus, because of the fragmented nature of the part and environmental data and the fact that it is usually necessary to interpolate or extrapolate from available data when developing new models, no statistical confidence intervals should be associated with the overall model results for any given prediction.