Quote:
Originally Posted by byroncheung
I have put in about 70 records into my spreadsheet so far, this is the histogram of failure mileage... Obviously part of it is driven by the fact that there are probably more cars at mileage round 45-60k mile... I need to adjust for that fact somehow if i want the plot to represent failure rate / probability... But it does feel like if a car managed to survive past 60k mile, the chance of a failure after 60k is lower than before 60k...
|
There's more to the story than simply tallying up the frequency of failure at an indicated milage. And you're going to have a tough time with the limited data to generate any kind of stats that can be predictive. You would need to account for a few handfuls of variables (driving styles, maintenance, modifications, environmental, gas, etc) and then perform some kind of factor analysis of mixed data. I doubt anyone has the time or care to do a proper study here and the final stats are going to be crude and blunt at best.
My work involves performing statistical analysis to determine the susceptibility, or risk, that children have of developing a leukemia. You can think of RB failure as a complex disease of a car much like we think of leukemia. Like leukemia in kids, there are many risk factors that predispose the M3 to to this complex state. To save everyone a lecture on statistical power, (appropriate) statistical significance testing, relative risk ratio and odds ratio - if you lack a sufficient population sampling and are missing many critical variables that play into RB failure then your data can only be descriptive (not predictive and certainly not able to make any associations). We can only take what you showed (or will show) with a huge grain of salt. You def cannot say that when you hit a certain mileage that your risk suddenly drops - whether the risk is high or low for a particular car at mile 0, the risk only goes up with time/mileage (increased wear, aging parts, etc) save for cars that just sit on display. I think a more appropriate inference from your histogram is that although RB failure has the potential to occur at any mileage interval on your x axis, the frequency is highest in mid mileage cars and the reason for that is poorly understood. As these cars continue to live and die we will continue gathering data, and maybe at some point we could have enough to do a proper retrospective study.
I'm glad someone is putting the data into a figure though. Seeing a visual representation of the available data is certainly more compelling that just seeing the total number of events.
I'm sure there's a dissertation on this somewhere, and if not then soon lol (or maybe for another car).
