On the maximum likelihood estimates for the Goel-Okumoto software reliability model

We show that the maximum likelihood (ML) estimates of the parameters of a w ell-known software reliability model are not consistent as the observation period for observed software failures extends to infinity. Properties of th e ML estimators as the observation period gets long are particularly import ant when the observation period corresponds to the test interval, since ext ending the test interval is the most natural way to improve the reliability of the software prior to its release. In addition to providing insight on how to interpret the ML estimators in actual applications, our result also has pedagogical value as an illustration that asymptotic properties of ML e stimators cannot be taken for granted.