I had a look at the test data on Sheldon Brown's site
http://www.sheldonbrown.com/rinard/EFBe ... e_test.htmThe published analysis is actually pretty limited; they only discuss enough to make a conclusion. So, I decided to do my own analysis.
To start with, the test is extreme, and cannot be correlated to mileage. However, the manner in which the bikes are loaded seems to appropriately mimic usage. While it is difficult to say from an article such as this, it seems to be a reasoanbly designed accelerated life test. If that is the case, even if the results cannot be directly correlated to bicycle life, the comparisons are valid.
For the analysis, then....
In the analysis I did, I assume a Weibull distribution, which is typical of fatigue. The data are quite limted, so it is difficult to place much confidence in the results, but that is often the case for life data analysis from test results.
Despite a lack of confidence, I actually think that the results are quite interesting....
You can see that the unreliability increases with the number of cycles. This is what we would expect. However, the part that is (to me) interesting is that the welded steel frames and carbon frames exhibit a much more abrupt increase. What that means is that the time to failure is much more consistent than for the other frames.
In the case of the welded steel frames,
Classic fatigue failure with Fondriest: Both chain stays of the light-weight steel frame broke in the weld relatively early. The difference in rigidity between the thin, flexible tube and the substantial bottom bracket shell is probably excessive and provides for stress risers.
Given that the welded steel frames were different brands, it is surprising to me that the times to failure are so consistent to produce this very steep slope on the graph. It is perhaps less surprising for a carbon frame, given that the properties can be more easily controlled.
I didn't include the statistical confidence on the above graph because it made for a confusing mess of lines. In general, with an analysis like this, the only way you can really improve confidence in results is to test more samples. However, when results are very consistent, this also improves confidence. So, you can see here 70% confidence bounds on the results for lugged and welded steel frames...
To me, the only drawback is that which Brucey has described; that the user is less likely to have a warning of imminent failure with CF versus other materials.
However,
“In some ways, it is easier to be a dissident, for then one is without responsibility.”
― Nelson Mandela, Long Walk to Freedom