BudhaSlug wrote:
With all due respect, your stats prof sounds terrible. The first day in my class, I explain that statistics is NOT a field of taking some numbers in and popping out some other numbers on the other side. Statistics is a way to tell a story of what data means in the real world. Heck, in my classes we don't even do manual calculations of any statistics because computers are good at that part.
As has been pointed out by others, there are better distributions to use than Gaussians. They are supported by what the raw data actually looks like. As I said, I have never worked on yaw, wind speed, bike speed, or anything related using statistics, so Normal was my default distribution suggestion. Using a more appropriate and data supported distribution that can be parameterized is obviously better and more accurate. It is definitely NOT arbitrary here, it is supported by the data of how these variables actually are distributed. As with any parametric statistic, the output is only as good as the fit to the assumptions. If you use Mean and SD and assume a Gaussian distribution but actually have a skewed or otherwise non-normal distribution, then your statistics will spit out flawed conclusions. The closer the data is to the shape of the actual distribution, the closer the output will be to the real world. The only reason you pretty much have to use a distribution in cases like this is that you need to parameterize the data to ease the calculations and because you generally lack sufficient data for all the situations you are trying to model. It obviously incurs errors, but statistics is all about limiting the amount of error and knowing where it will be and how big it will be. Kiley already knows that since he was already planning on using error bars (i.e. measures of errors in statistical estimates) on his graphs. Oh, and he never asked, but I vote for 95%CI error bars rather than something like SE error bars.
One other note. The debate over Yaw-at-front-wheel and Yaw-behind-steering-axis is probably irrelevant here for modeling the wind tunnel data. In the tunnel, I assume that the wheel is kept straight with the rest of the frame, so all data that is coming out of the tunnel (and is the input to these statistical equations) is stuck in a single wheel-frame relationship. I don't think you can model how the drag would change in other conditions unless you already had 2 other pieces of information:
1) How drag changes in a tunnel when the wheel and frame are not perfectly aligned
2) What the distribution of wheel-frame relationships are out on the road
But, again, if you had both of those pieces of information, you could these factors to your estimates and come up with an even better "real world" estimate of drag in the real world for a given set of wind-tunnel data.
"Me thinks the Lady doth protest too much" :-)
While I said that statistics is chosing the formula to get the number you want, I was being somewhat facetious of course. However, what I insinuated was choice by the statistician, you (to paraphrase) "fit to the assumptions". I guess I'm contending that statistics is an art, when it appears to be a science. Regardless of what your data set represents, there are a multitude of ways it can be
interpreted and the validity of those ways can be argued by statisticians until the cows come home!
In the case of selecting bicycle for triathlon, we here at ST understand that the aeroness (drag) is a very important factor when considering what bike to purchase. We look to the testers for answers on which bike is most aero. the vast majority of us are looking for one number to tell us which bike is best. Most of us don't look at yaw angle performance and compare that to the wind direction (and apparent yaw we will experience on our bikes), we just want to know which bike is best. For that, we need one number. While statistics has the ability to pop out a single number at the end of the formula, it has the ability to pump out lots of single numbers that all represent the data set depending on your interpretation.
In this particular case I think the headtube mounted sampling apparatus gives the most accurate wind yaw information, I can also see the validity of the fork mounted apparatus since we wobble our way down the road, even in still air! I will leave it to others to argue the validity of each case, because it's almost beer O'clock here in New Zealand, but just like we anticipate that the bikes will all end up being dangerously close to each other in drag, which testing protocol results in the right answer will also be dangerously close. Potato, Potato :-)
TriDork
"Happiness is a myth. All you can hope for is to get laid once in a while, drunk once in a while and to eat chocolate every day"