Quote:
Originally Posted by bruce.augenstein@comcast.
Who says the lap times are absolutely dominated by power to weight? There appears to be some correlation, but until you have say, five hundred passes using the data you've been using, or a fewer number (but still large) using data from strictly controlled passes (weather and driver, mostly), you have at best a loose correlation. There are bunches of flyers (some greater than others), and most don't fall on a statistical line.
|
Some clarifications here. The correlation for the Ring times is by no means "loose". If you look at the R^2 and P values, you'll see that it is very well established. That is pretty much as good as a regression will ever get with real-world numbers. (If you haven't seen that thread, you might want to check out the specifics.)
That said, yes, that doesn't establish cause and effect. One could argue that any manufacturer who invests into raising the hp/lb ratio will also invest in the handling and traction of the car, so it might very well be that the real cause might be something else such as handling and traction rather than hp/lb. And the inverse argument would hold water, too.
One way to explore that further would be to add more variables to the regression that would be representative of handling and traction. I believe several people have tried that, and if I remember correctly, they didn't seem to make a major difference? (Someone correct me if I am misremembering here).
But that doesn't mean that you can design an extremely powerful lightweight car, and screw up the suspension, and still post great times. The same goes for designing a car that has amazing handling performance, but no power. Again, the point is that cars are pretty well packaged as a whole in general, that there are intrinsic associations between power, weight, and handling. But the question that is being asked is "which one of those variables has the strongest impact on the outcome?"
As to the variations due to uncontrolled variables, that is why regressions are done to begin with; to see if such variations can overpower any underlying trends. In this case, the regression outcomes suggest that they don't (there will always be outliers, but that doesn't mean the trend is off base). The ideal way to approach this problem would be to list all of the variables that might be associated with the outcome, construct a multiple variable regression model, drop the variables that don't seem to make a contribution to the model, and see if the regression gets stronger and tighter.