|
Abstract:
|
The goal of this paper is to model, and compare, the lifetime distribution of various hard drive brands. The data has many features that make a standard maximum likelihood estimation approach problematic. Many hard drive brands have few failures, which makes stable parameter estimates difficult to obtain. There are also two modes of failure (infant mortality and wearout) that are present in the data, although the cause of failure is not reported. To address these challenges we propose a hierarchical model that allows us to pool information across brands and obtain more precise estimates: especially for brands with few failures. In addition, we estimate the two different failure modes using a competing risk model, which models each failure mode as a coming from a separate Weibull distribution.
|
