Sorry for the lack of post-BCW interaction (IRL/Work has been interesting). Anyways, I thought I would give a recap of my processes and my results for those interested. I will do a more detailed analysis of the 2021 tourney itself in my Welcome Back article in November (what a teaser).
MY BRACKET / MY RESULTS
MY PROCESSES
First off, I only do one bracket for obvious reasons. I'm aiming for the perfect bracket so ten different brackets means nine at minimum will always be wrong. Second, I probably shouldn't admit this considering how much I derail it in my blog: I once again used the curve-fitting methodology for my competitive bracket.
Curve-fitting is the method that aims for a specific amount of upsets per round. In other words, I have to be exactly right twice. First, I have to predict the right curve (which I am 0 for 5 counting this year). Second, I have to predict the right upsets because if the predicted curve is correct but the predicted upsets are wrong, I will miss two games instead of one. More in-depth detail in this article under the section "The Failures of Curve-Fitting." Despite it's riskiness, I believe it is the best-approach.
From the bracket above, here's what I predicted: 4 - 2 - 1 - 0 - 0 - 0 round-by-round upsets because of the pod structuring and seed-curve strength/weakness distribution. Let me explain what I mean by both of these. I have explained pod-structuring before, but I'll quickly recap it here. Only certain match-ups can produce upsets because the seed differential must be four or greater (3v6 = NO, but 3v11 YES). I label these as upset-potential match-ups (UPMs). When I looked at the seed-curve distribution of strengths and weaknesses, they were disparately distributed through the pod structuring. For example, strengths were mainly in 1-seeds, 8-seeds, 9-seeds, with slight strength at 5-seeds and 6-seeds. Weaknesses were distributed among 3-seeds, 4-seeds, 7-seeds and 12-seeds. 2-seeds and 10-seeds at even par (which I guess should be implicitly bad for 2-seeds but I have to check the seed-history for that assumption). Anyways, 1-, 8- and 9-seeds all play each other and the entire pod (1-8-9-16) is a UPM in R64 and R32 no matter who wins. I, of course, picked the wrong one and it cost me two games (MICH/LSU and ILL/LOYC, but I probably should have taken the in-state rival upset). 6-seeds were slightly strong while 3-seeds looked like victims, but this is not a UPM in R32, so I couldn't account for too many here. As you can see, I didn't think the seed-group strengths and weakness were distributed in a manner that would be conducive to upsets, and of course I picked all the wrong ones in R64 (while additionally missing two on the curve).
I was worried about the high-rising tail of the 2021 QC as an omen of later-round upsets. Since I couldn't find many UPMs, I tried to offset the lack thereof with higher aggregate values (this was a good adjustment, as I'll explain in the Welcome article). Thus, I tried to blow-up the MICH region and I learned from the 2017 tournament (USC over SMU) not to advance high-efficiency major conference teams against lower-seeded R64 power conference teams (so I picked SYR over SDSU but for some illogical reason I didn't do the same for UCLA over BYU). I punished all of the 3-seeds in the R32, which added 25 points to my S16 AV (another good choice), but only added one upset to my curve (11 SYR over 3 WVU).
FINAL WORDS
Nonetheless, the high-rising tail of the quality curve was the key to understanding the outcomes of the 2021 tournament (along with a few other factors). The seed-curve and seed displacement (over-seed vs under-seed) were not. I look forward to November when I get to talk more about all of this at length. As always, thanks for reading my work, and I'll be back then.
No comments:
Post a Comment