MLE and Least Squares in RG

[hazarcon@pinball.com]

[Originally Posted: 11/15/99-- Transferred by ReliaSoft Moderator]

I'm testing 2 identical systems having a total of 4 failures. An improvement was made after the second failure and raw data indicates there is improvement. When using RG, MLE shows downward MTBF plot while least squares show an upward plot indicating improvement. Why is this?What's the difference btwn the 2 in this case? Thanks

[David J. Groebel]

[Originally Posted: 11/15/99-- Transferred by ReliaSoft Moderator]

The discrepancy arises due to the fact that you have only four failure points. Rank Regression will minimize the distance from the line to the points, while MLE will maximize the Likelihood function. Therefore, given your points, Rank Regression seems to follow the data better than the MLE based on the graph. However, it does not maximize the Likelihood function. Although, you will notice that both methods seem to converge at time=2.4E+6 on the plot. But the reason that you get different results is due to the fact that there are only four failure points.