 |
RE: [RC] LD/50 times - David LeBlancDiane said: Truman did a study on speeds and deduced from the records that the top LDs rode more slowly than the top 50s. What did he use for this study? The AERC ride result records. Are these records accurate? That last is the question I was addressing. My point was that if there are SOME records so badly out of whack that one can spot them knowing nothing about the ride in question, how many more are equally wrong but just harder to spot? What we have here is a common problem. Data are noisy. Any sample will have data that's either in error, or is affected by something you may not be able to analyze for. So what you can do is first weed out the data points which are clearly in error. For example, if I look at the speed distribution for LD rides for a few hundred rides, I'll see a bell curve, but there's a spike on the end where everyone was reported as taking 6 hours. To do things right, that should be documented, and you then don't use those rides (or the corresponding 50's) for the analysis. I did that. For instance, according to the record, the Tidioute Challenge 25 on 08-Jul-06 was won in 4:16. WON! It is not impossible that the winning time was so slow but when the winner of a 25 mile ride goes only 5.86 mph, a reasonable person will think, "Can that be right?" So you look at the 50 that was run concurrently. That ride was won in 5:34, an average of 8.98 mph. That is slow too but not outlandishly so. I think Truman said the 50s run about 15% faster than the LDs. In this ride the record says they ran 53% faster. This could happen. For example, my wife was in the middle of the top-10 on a 30 this year, and it wasn't because she was going fast, it was because the really fast people just weren't in the LD on that ride. We could also have a ride where the hardest loop is only done by the LD riders, or the hardest loop is only done by the 50's. The expectation is that over 100's of rides, this will average out. This is why I wanted hundreds of samples. That's statistics for you - we see that LD vs. 50's explains some of the variability in winning ride speeds. There's more that isn't explained - possible factors include terrain, temperature, one ride being easier or harder on average than the other, and who happened to show up that day. And indeed some of the differences may be explained by ride manager error, but from my experience, this is rare. No matter how carefully you do a study, if the data is flawed, the study results are unreliable. In the case of the AERC LD times, the question is not IF the data is flawed, it is HOW MUCH is the data is flawed. Data are always flawed, and results of an analysis are always unreliable. The key is to understand how data may be flawed, and how unreliable the analysis is. In statistics, we're normally happy if we're 95% confident in the result. There are various tests we can do to determine if 2 populations are really different, or if it's just chance. So for example, the math that allows me to design an aircraft wing has some compromises (for reasons I won't go into). The net result is that the predicted lift and other characteristics are within about 10% of actual behavior, which is good enough to get us into the wind tunnel where we can find out how it really works. Now that we have better computer simulations, we can do better than that, so this is just an example. A good indicator that my analysis is correct is that it was performed on data collected in the late 90's. Two more recent analysis's conducted on newer data come to the same conclusion, and believe me, Heidi isn't an especially enthusiastic fan of LD, to put it mildly. It's pretty rare for me to be riding LD these days, too. =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= Ridecamp is a service of Endurance Net, http://www.endurance.net. Information, Policy, Disclaimer: http://www.endurance.net/Ridecamp Subscribe/Unsubscribe http://www.endurance.net/ridecamp/logon.asp Ride Long and Ride Safe!! =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
|