We were contacted by N8 recently concerning a discussion on blowouts that he was having with derbynerd over on DNN. While I’m not sure I exactly answered his question (sorry N8), it started me on an investigation that I found interesting and I thought I’d share some observations.
The question I started to ask was “how big of a lead has ever been erased, regardless of who eventually won?”. And, “how often does that sort of thing happen?”. So I went back into our stats database to have a look. As of this writing we now have over 1500 statsbooks uploaded, most of which (over 1000) are from this season or last season. There are some notable exceptions like the VRDL vs. Wasatch bout from the Big O earlier this year – I seem to recall close to a 100 point Wasatch lead in the first half that VRDL eventually erased (someone send us that statsbook...please). So I will not claim that this analysis is fully inclusive, but it is still interesting nonetheless.
In the following plot, I show the distribution of every lead (that we know of) that has ever been erased as a function of the maximum score-difference that was achieved before it was erased.
For those not used to looking at a plot like this, the x-axis is showing the score difference (at it’s maximum) for the lead, before it was erased. The y-axis is showing the number of times over all the bouts that we have, that a lead of that score difference has been erased. The smaller inset plot in the upper right is exactly the same, I have just changed the y-axis to use a logarithmic scale, allowing easier observation of the single events out at the extreme right. This plot represents almost 2800 total leads that have been erased. The largest recorded lead to be erased is 90 points from the 5/18/13 bout between Ohio and Atlanta. There are 2 more at 82 and then everything else is below 80.
Now derbynerd had an interesting suggestion that a “blowout” should be defined as a point difference at which “the losing team have a < 1% [chance] of winning” – guessing it was somewhere around 60 points. By integrating the above plot, I find that 99% of the erased leads are at 56 points or below – a commendation to derbynerd's instincts.
Another way to look at the probability, however, is to ask how many jams it takes to erase the lead. In the following scatter plot, I’m now showing the same 2800 leads, but now the y-axis is the number of jams between when that maximum score-difference happened and when the lead was “erased” – either the game was tied or the other team was leading.
I find this plot fascinating. I wonder about those games that took 34 or 35 jams (almost an entire game) to recover from a 30-40 point lead. But more than that, I notice that the vast majority of these leads were erased within 10 jams (even leads as large as 70 or 80 points). In fact, almost 85% of these were erased within only 5 jams.
So that started me thinking about bouts that are uncertain right up to the end. Instead of “blowouts”, the ubiquitous “anybody’s game” comment comes to mind. I think a lead that can be erased in the last 5 jams would be a pretty good representation of that uncertainty. So what does that really look like? Well, the following plot is identical to the first one, but now it’s only showing leads that were erased in 5 jams or less.
As you can see, it’s pretty rare to erase a lead above 40 points in less than 5 jams -- the highest recorded so far is 57 points from this bout. To use derbynerd’s criteria, the 99% inclusion limit would be at 29 points, but I might be more inclined to think about the 90% limit which is at 19 points. So that means that with 5 jams left in the game, there’s only a 10% chance that a 20 point lead could be erased and only a 1% chance that a 30 point lead could be erased. That’s actually much harder than my instincts would have predicted.
So what’s it all mean? I have no idea. And I don’t think I can provide a better definition than anyone else of what constitutes a blowout. But I found it interesting to look at these numbers and I thought you might too.