Video: Why Wells Report is Wrong and Actually Exonerates Patriots

 

 

 

 

Below is the table presented at the end of the video summarizing the findings of the time-based permutations of each scenario. Note, this video uses Exponent’s “wet” curve projection. Unadjusted data can be seen the Deflategate science wiki.

Deflate Gate Summary

3 thoughts on “Video: Why Wells Report is Wrong and Actually Exonerates Patriots

  1. I like the clear explanation here (and also your earlier piece about cognitive bias), but can you clarify your source for the transient curves, and what you understand Exponent’s handling of the curves to be?

    You show a “13.0 dry ball” curve here, which ranges from 11.8 psi at 0 minutes to 12.9 psi at 13.5 minutes. Exponent shows two different experimental curves, and yours doesn’t quite match either…

    In figures 21-22 of Exponent’s report, the graph (and closeup of the graph) is said to be based on a simulation involving inflating balls to 12.5 and 13.0 when equilibriated and measured at 69 degrees, then exposed to the usual “on field” conditions, and then brought back to a simulation of the official’s locker room at 72-73 degrees and 20% relative humidity. [ If not a typo in the report, the humidity is a surprisingly low value, and one which would encourage faster drying of the “wet” balls (which also may not have been wet enough in Exponent’s simulation to form a true lower bound of natural causation)]. In this graph, the 13.0 psi dry ball curve appears to have a value just under 11.9 at 0 minutes in the locker room and 13.0 at 13.5 minutes. The 12.5 psi dry ball curve seems to be consistently offset exactly 0.5 from the higher curve throughout the time interval. The wet ball curves in this graph don’t show the same consistency in offset, but reach about 12.70 and 12.25 after 13.5 minutes.

    However in figure 24 and later figures, (including the actual “visual proofs” Exponent offers), they use new experimentally derived curves. The stated parameters of the experiment vary just a little from the previous one. In generating this curve, the pre-game temp is raised 2 degrees to 71, the locker room temperature is apparently set the same (“between 72 and 73”), and humidity is not mentioned. However in this seemingly nearly identical experiment, a substantially flatter curve is shown in which return to equilibrium pressure is much delayed. Here for example the “13.0 dry curve” extrapolates to about 11.85 psi at 0 minutes, and only reaches about 12.6 psi after 13.5 minutes. The wet curves show greater variance than in the earlier experiment, but all 4 curves in figure 24 are about .35 to .45 further away from equilibrium after 13.5 minutes than in the corresponding curves shown in figures 21-22. I think the 2 degree difference in starting temperature between the two experiments only explains about 0.1 psi of that difference… Is the experimental error that large, or is there something else going on in the 2nd experiment which I haven’t understood?

    Exponent provides one other version of a transient curve, with the rubbed ball. In that experiment with a presumably dry ball (figure 16), rubbing raises the pressure about 0.68 psi (12.54?-13.22?) , and after 13.5 minutes about .60 has been lost, with the ball back to about 12.62 psi. Then it takes 16.5 minutes to shed that last 0.08 of psi and fully equilibriate. Whether equilibriating up or down, my naive expectation is that if temperature and humidity of the equilibrium environment are similar, then we’d see similar approaches to equilibirium after 13.5 minutes.

    Your demonstration is more in line with Exponent’s 1st set of “generic” curves, and maybe also with the rubbed ball curve, but how do we objectively conclude that Exponent’s first set of curves, or your curve, are more right than the 2nd set? If there aren’t definitive grounds to reject that more slowly equilibriating set of curves, is it too much trouble to update your simulation to show those too?

    • Great questions and comments.

      First, as I’ve said before, it’s hard to get incredible precision here, by nature. Second, I started this process trying to demonstrate that *based on Exponent’s other experiments* they reached the wrong conclusion because of their analytical methods. With that said…

      You are actually asking about the area that I’m the least confident in the model, given that I based it off figure 22 from Exponent. I took their curve and reverse engineered the temperature increase they used (by using the IDL to solve for temperature). I’m more concerned if Exponent’s experiment is valid/repeatable — I’ve actually been trying to find other sources who have produced transient results to test this. As a rough approximation, the YouTube channel “MIT Pats Fan” brought brought an 11.05 ball back into a 74 degree room for it to return to 12.5 PSI.

      -After 5 minutes, it re-calibrated by 50%. After 10 minutes it was about 69% of the way back.
      -He then did a 13.5 PSI ball which re-calibrated by 43% in the first 5 minutes and after 10 minutes was about 58% of the way back.
      -Exponent’s Fig. 22 is ~52% in the first 5 minutes, and after 10 minutes 79%.
      -Exponent’s Fig. 24 assuming an 11.32 start (predicted by IDL) is 43% after 5 minutes and 60% at 10 minutes.

      Needless to say, this is why I’m the least confident in this area — there’s some inconsistency here and a nice little range depending on the scenario. It’s also why Exponent is so darn bad at science — the right way to do this would be to run a bunch of simulations to see what kind of consistency can be observed (and modeled) with regards to rate of recalibration from temperature change. I digress.

      So how much would these different rates change the results of what was presented in the video? Let’s use Exponent’s Fig. 24, the flattest of our scenarios, to do the calculations done in the video:

      LOGO GAUGE average: -0.17 PSI below Colts (avg. p=0.471)
      NON-LOGO GAUGE average -0.36 PSI below Colts (avg. p = 0.170)

      That is for Patriots inflate first, then measure 4 Colts balls (which I hope we can all agree is highly likely, despite Exponent running their experiment in the contrary manner). Worst-case scenario is still not statistically significant. Also, the Colts balls are also EXACTLY where we’d expect them based on IGL in every single scenario using a flatter curve like that (on the Logo Gauge), which means if the Colts balls were dry this is a more likely transient curve (according to IGL, assuming 71 degrees pre-game is accurate). But as you can see, this won’t change the general results. The most likely explanation for the halftime readings is that the Patriot balls behave as expected based on the environment and the small difference is (as expected) due to wetness differences and some clear variability in measurements. (The 2nd, 4th, 10th and 11th balls were probably most exposed to the rain, while the higher balls never left the equipment bag.)

      • Thanks for the quick reply and analysis.
        I’ve seen this mentioned somewhere but don’t remember if you mentioned it (Exponent certainly ignored it): one further source of some variability in the Patriots’ balls is that the sample of 11 more probably than not included both balls re-inflated by Walt Anderson before the game, perhaps with a different target than Jastremski (12.5 vs 12.6), with a gauge which probably had different accuracy than the Patriots’ gauge, probably under slightly different atmospheric conditions. So that could put couple balls in the sample which might vary by .2 or.3 from the starting point of the other 9…

Leave a Reply

Your email address will not be published. Required fields are marked *

*