Well he clearly succeeds in demonstrating a correlation and that confirms that aero drag (in that speed range) and gear shifting are much less significant factors than power/weight. Factors like those account for the "scatter" that is seen in the plots. Another major source of error is the peak power figure. As far as I know, the test relied on manufacturer's quoted power which can vary significantly from actual power. Acceleration figures come from "Autocar" magazine and the plots I posted come from here http://www.competition-car-engineering. ... Torque.htmrscsr wrote:How can you judge who can judge something. It is still garbage science. He tries to correlate average acceleration over a massive speed differential with peak power. The speed more than doubles and the drag can easily be a problem which is not accounted for and for some cars you need to shift while for others you won't.gruntguru wrote:It is actually quite good data. If you were to fit a trendline with the equation P x t = C (Power, time (30-70mph) and a Constant) I wager you would get a correlation coefficient of 0.9 or better.rscsr wrote: Although he sounded kinda rude, I agree with him. These are just nice pictures with a whole lot of variations. I mean you can get the same acceleration with 125-180 bhp/t and if you squint a bit you can see a slight trend to higher accelerations with higher torque, although with also a lot of variations.
I mean it pretty obvious to me that torque has not much to do with acceleration but this is still kinda bad science.
Anyone who doesn't understand the significance of that, is not qualified to judge "bad science".
Nice work rscsr. Any chance you can post the digitized data?But be glad that you haven't wagered anything. I checked the correlation and guess what. A Pxt=C trendline has a R² of 0.78. And it gets better. A linear correlation gets a R² of 0.84 and a P^(0.7)xt=C gets a R² of 0.89. So what conclusions do you draw from that? A linear corealtion is significantly better than a Pxt=C?
I think I see some errors though. Firstly, my error - the trendline equation should have been of the form (P+a)x(t+b)=c. (The a and b constants simply allow the trendline to be moved vertically and horizontally.) Better still (P+a)^n x (t+b)=c. (Double or nothing R^2 is over 0.9 now.
) Point being, the relationship follows a hyperbola but a few extra constants are needed to help account for some of the real-world losses in the system.Secondly, your error. You didn't include the equation for the P x t = C trendline.
Conclusions I draw?
1. The trendline equation needed at least one more constant to allow a vertical and/or horizontal shift.
2. 0.78 (or 0.84) is still an excellent correlation coefficient indicating a high probability that the variables are related - especially given a sample size of 27 cars. Definitely not bad science. (Good of you to hunt around and find an even better fit - 0.89 is outstanding.)