The choice between parameter estimation methods in combination with confidence bounds methods are commonly determined in a nonmethodical manner. Especially for small sample sizes, a nonmethodical approach may result in highly biased reliability metrics. This paper addresses the problem of choosing the best performing combinations of estimators, bias-corrections, and confidence bounds based on data properties (e.g. censoring share and sample size). For this purpose, Monte Carlo simulation (MCS) studies are conducted for uncensored and type II right-censored data to evaluate the performances of biased and bias corrected maximum likelihood estimators.
The second objective is to propose the best suited confidence bounds method for the same type of data depending on the estimator, sample size and censoring share. To this end, a score based on the distance between the confidence limits and the number of times the confidence limits contain the true B1, B5, B10 and B63.2 lives serves as the key metric.