Interpreting a data set that looks like it contains different failure modes

[Tony M.]

I have a data set that has a population of over 2 million (17 months worth of data). Needless to say, I have an excess amount of right censored incomplete data. When I use the distribution wizard it determines that my data follows an exponential distribution. Knowing my product (electro mechanical device), I know it does not have a constant failure rate so I disregard that suggestion. I know that I should be using a 2P Weibull, but I choose to entertain a mixed population (2 population, 5 distinct data pts.). This distribution fits the data really nice (graphically). But this is where my dilemma comes in. At month 12 I have a somewhat valid cumulative failure rate, but at month 24 it calculates over 90% failure rate! Is this really the right way to analyze this data set (mixed population)?

[David]

Are you using the mixed population because you believe the data to contain multiple populations or because it seems to give you a really nice fit? Just as you did with the exponential distribution, you should only use a mixed population if you know or have reason to believe that the data set contains a mixed population. Using the Mixed Weibull distribution will typically give you a better fit because it has more parameters.

I can be more specific if you wish to email me (david.groebel@reliasoft.com) the Weibul++ data file.