2022 NCAA Tournament - FULL ANALYSIS

It's finally here. It seems like yesterday I was putting out the introductory article yet here we are. I'm going to post everything in this article, so all the information will be in one spot. I'll also make the section headings larger than normal so you can find what you're looking for while scrolling.

Championship Profile

Let's start with the obvious rules, and work from there. First, National Champs have never been lower than an 8-seed (1985 NOVA), they have never lost the first game in their conference tournament (italics), and in the Advanced Metrics era (2002-present), they have never had double-digit losses (strike-through).

1. GONZ,   ARI,       KU,       BAY
   
2. DUKE,  NOVA,   AUB,      UK
3. TXTC,   TENN,  WISC,     PUR
4. ARK,     ILL,       PROV,    UCLA
5. CONN,  HOU,     IOWA,   STMY
6. ALA,      COST,    LSU,      TEX
7. MIST,    OHST,    USC,     MURR
8. BOST,   HALL,    SDST,    UNC
   

After removing non-power conference teams (STMY, COST, MURR, BOST, and SDST) and hypocritically non-scientifically keeping GONZ, here's our National Champion contenders and their disqualifiers.

You will notice that I've added one extra column (6W). If you follow this link, it will explain the historical significance of this attribute. (TL/DR: An NCAA title is six consecutive wins against power conference quality teams, so if a team has already done this in the regular season, they should be more likely to do it again in the post-season). Using the table, here's our tier list of title contenders (notwithstanding other disqualifiers from other systems).

1. GONZ - The only disqualifier is lack of six straight wins against quality competition. Two national title contenders had this disqualifier: 2004 CONN and 2014 CONN.
2. NOVA, UK, TENN, PUR, PROV. Of these PROV has the hardest two disqualifiers. Unless they have the profile of 2014 CONN (Spoiler Alert: They Don't), they are firmly off the contender list. UK missing both last year's tourney and not having a conference title makes them the next weakest contender. PUR is lacking a conference title and missing PG production badly. NOVA is a system disqualifier (0.3 AST short of the 3.6 cut-off line) or they would be in GONZ's tier. TENN is also a system disqualifier, and if JR Vescovi qualified, they would only have one DQ.
3. The four DQs in yellow are teams that fell in seed-line from last year. Only two National Champions went down in seed-line from the previous year (1998 UK and 2016 NOVA).
4. High-Risk pick: The teams in green are high-risk picks based on a pattern-of-trend in the original article. Since we've ruled out ARK, USC, UCLA, and HOU as lower-tier picks, this leaves GONZ. There is no statistical evidence supporting this method!!! Use at your own risk!!!

Over-seed/Under-seed - Predictive

This should seem relatively familiar, as this information is found on the "Bracket Modelling" page. However, I've condensed the size of it for the sake of being tidy and neat.

First, let's look at the pathing of octets: [1-4-5-8-9-12] and [2-3-6-7-10-11]. This is simply a matching exercise based on the 2022 octet having a historical equivalent based on the notations. This does not mean match the comparative region pick-for-pick/seed-for-seed because there are noticeable differences between the comparisons. The goal is to get a feel for the resistance or passivity of the octet.

• GONZ: Most similar to 2014 ARI, 2016 KU except stronger match-ups or 2021 BAY except weaker match-ups.
• DUKE: Most similar to 2014 WISC and 2016 XAV, also 2018 PUR except weaker mid-seeds.
• BAY: There's honestly nothing like this octet (or this region for that matter). Every team is either the best at their seed-line or the third-best, and the only three over-seeds are the 1-seed, 3-seed and 9-seed. Either 2015 DUKE or 2017 KU best resembles (although weaker match-ups) this octet, but even those results don't match one another.
• UK: Again, there nothing exactly like this octet, having the stronger team at each seed-line except for the 3-seed. 2017 UK is closest, and 2021 HOU is next closest except it is much weaker.
• ARI: 2016 UVA and 2019 UVA
• NOVA: 2014 MICH except with stronger lower-seeds.
• KU: 2015 NOVA or 2015 DUKE, except with a weaker 4-pod.
  
• AUB: 2014 KU and 2017 DUKE with weaker mid-seeds, or 2018 UNC with weaker top-seeds.

Now let's look at some of the groups:

• 3-seeds: Of the five 3-seeds that were over-seeded by at least three seed-lines (6-seed quality or worse), only one made the S16 (2014 IAST with help from a weak pod). Three of these candidates were from the 2021 field (TEX, WVU, and KU).
  
• 4-seeds: Under-seeds like UCLA are 3-1, with the one loss being 2016 UK who ironically had the toughest pod of the four under-seeded fours. No 4-seed has been over-seeded eight lines like PROV, but two have been 8- and 9-seeds in disguise (ironically both won). Four 4-seeds with 6-seed quality have a 1-3 record.
• 5-seeds: All four are either under-seeded or accurately seeded. Only happened in 2018 and all four advanced to R32. Need to examine the 12-seeds before YOLO'ing.
• 12-seeds: Two are 20-seed quality or worse. Of the ten 12-seeds with 16-seed quality or worse, they are 2-8 with both wins coming against over-seeded 5-seeds (none in this tournament). The other two 12-seeds are coin-flips versus their 5-seed counterpart.
• 6v11: This model doesn't give a good reading for this match-up, except for one game in particular: LSU vs IAST. Only once in five tries has a slightly US 6-seed defeated a slightly OS 11-seed, and this happened with 2018 HOU. They were one of the hottest 6-seeds to ever enter a tournament, and we're ultimately defeated on a buzzer-beater by 2018 MICH, who happened to be the hottest 3-seed in that same field. I wouldn't call 2022 LSU "a hot team."
• 7v10: US 10-seeds versus AS/OS 7-seeds are 1-7, with the only win coming from 2017 SCAR, who played the game one hour away from their campus. This is good for both LOYC and SANF.
  
• 8v9: Of the thirteen 8- or 9-seeds that were over-seeded by at least three seed-line, they are 4-9 against their opponent, with three of those four wins coming against an opponent that was also over-seeded. SDST looks like a lock and UNC a 75% lock.

OS/US Model - Poll Based

I should state for the record that "THIS IS A HIGH-RISK MODEL." I would use predictions from this model as a tie-breaker or a contrarian model. If you want to read more on why I feel they way I do about this model, here is a link to the article. Let's look at the results, and I'll point out the most important stuff.

First off, there are 21 OS predictions and there are 9 US predictions. This is not the ideal balance, but it's not completely one-sided like 2018's model (20 OS vs 5 US).

Second off, this model struggles with lightning strikes: Teams that flash in and out of the polls like lightning. These teams are either (1) There at the beginning, take a few Ls, and tread water for the season (MEM & MICH), (2) Show up mid-season when they peak and leave as quick as they appeared (MARQ, WYO, and LOYC) and (3) Make their first appearance in the last few weeks as they start to get hot in March (IOWA and BOST). All of this could be a result of polls being lagging indicators, and the information being used as predictive indicators. While I can't speak much about the first two groups, hot teams usually get signaled as OS but still exceed seed expectations. Teams like IOWA that receive a ranking in the AP poll for at least three weeks at the end of the season win at least one game in the tournament. Teams like BOST that receive a ranking for two weeks or less (one in the case of BOST) usually lose their first game of the tournament.

Third off and finally, I want to emphasize the teams in BLUE and YELLOW. These are OS/US match-ups. If we ignore 2018's results because of imbalanced results, this function of the model predicts with a hit rate of 60% or better. Yes, 60% is 40% below where I want to be, but if you have nothing else to go upon for coin-flip match-ups in R64, it is better than nothing. The UNC-MARQ match-up is listed in yellow because no matter what happens, it will yield a false-positive: Two OS teams where one must win and advance. The other four in blue feature an OS-team versus a US-team, and we always pick US-teams over OS-teams (yielding MEM, MIST, MICH and IAST).

OS/US Model - Conference-Based (Updated 3/17)

If you want to read more about this model, use this link for the full write-up. This model is currently projecting 17 outcomes, which is a safe quantity. When the count gets above 19 projections, the success rates plummet. Based on a quick glance of this model and the prediction-based OS/US, they seem to be contradicting each other with a few teams: UK, AUB, UCLA, ILL, LSU, CREI, and MIA are OS in one model and US in the other.

I can honestly say that this model is a lot easier to figure out when you know the results already. I'll try to work through the logic for each projection.

• Either DUKE fails to make the E8 or UNC & MIA win in R64. With DAME winning the play-in game, I'm not sure if this prediction holds since MIA would be US as part of the group yet between MIA and DAME, MIA would be OS. I don't think any of the years had this conflicting overlap.
  
• If CREI loses in R64, then they can't be US, which means HALL and MARQ must be OS and also lose in R64. If HALL wins in R64, then they can't be OS, which means CREI and MARQ must be US and also win in R64. Possible combinations: 1) All win R64, 2) All lose R64, 3) CREI wins R64, HALL and MARQ lose R64, or 4) CREI and MARQ win R64, HALL loses.
  
• If PUR makes S16, then they can't be OS, which means WISC and ILL must be US, and must also make S16. If ILL fails to reach S16, then they can't be US, which means PUR and WISC must be OS, and must also fail to reach S16. Possible combinations: 1) All reach S16, 2) All fail to reach S16, 3) ILL reach S16, PUR and WISC fail, or 4) ILL and WISC reach S16, PUR fails.
  
• For the IOWA/OHST/RUT/MIST/MICH combo, there are two possibilities. If MICH wins, then OHST must lose and everything works out. If MICH loses, then MIST and IOWA must also lose for everything to work out.
• If UK fails to reach E8, then it takes care of all three. If UK reaches E8, then they can't be OS, which means AUB and TENN must be US. Thus, AUB would reach E8 and TENN would reach S16.
  
• Either USC wins or UCLA falls short of the S16. There is an interesting twist to this pick and it revolves around the DAME/RUT play-in game. If DAME wins, they become an 11-seed, which makes 10-seed MIA an OS. Thus, USC would be US and MIA would be OS, and we take US over OS every time. If DAME loses, then MIA's only two projections are US, which means we have a US 7-seed and a US 10-seed, and the bracket forces one US to lose, which our model can do nothing about.
• For the accuracy of our model, I'm kind of low-key pulling for RUT. MIST might be an OS now that RUT is out without making field of 64 (-1W & +4Sd).

2022 Quality Curve (Final) and 2022 Seed Curve

Brace yourself, because my hunch throughout the year has not materialized. I assumed mean-reversion after an insane record-setting 2021 tournament and the fifth-year (super-senior) rule plus instantaneous transfer eligibility would assist this result. Thus, I assumed a potential 13-15% range on the M-o-M rating levels of insanity, but it does not look like that will be the case. Let's start with the usual month-to-month chart of the QC and work from there.

We know from the earlier two QC analysis that the QC improved due to a rotation/pivot at the 16-spot. This can be seen by looking at how the FEB line deviates from the JAN line. The front of the QC (spots 1-15) is higher and the back of the QC (spots 16-50) is lower. When we look at the deviations between the FEB and the FIN QC, the front of the curve falls back to the JAN QC and the back of the curve stays roughly the same. In fact, the 2016 QC took this same path: Pivoting/rotating from JAN to FEB and then falling backwards from FEB to FIN. Let's see how the FEB and FIN QC fluctuated over this time period by looking at the two QCs in relation to the maximum and minimum values of the QC over that month.

With the exception of a few isolated spots (#13-#14 and #22-#25), most of the FIN QC is located at or below the middle of this range. Considering most of the FEB QC is above the FIN QC, it stands to reason that the last six weeks has taken a toll on quality, with most spots testing the minimum line and barely half of these spots bouncing back towards the middle. This is a general decline in quality with only a handful of teams bucking the trend. Since we know that quality now is not as good as it has been at an earlier point in the season, let's see how the current QC compares to previous QCs at this point in the season.

Yikes!!! From spots #1-#6 and spots #15-#45, the 2022 QC is below all other curves. In fact, the only spot on the QC that shows any comparative strength historically is the #13 spot, which falls in between the the 2017 (highest #13) and 2019 (third-highest #13) QCs. Theoretically, this spot would be the highest 4-seed if teams were seeded according to efficiency metrics. I'm not sure if the identity of the team is significant (IOWA), but they are theoretically under-seeded as a 5-seed. In a weird sort of way, the 2022 QC does something that the 2017 curve does. From the #2/3 spot to the #13 spot, the 2022 QC flat-lines like the 2017 QC does from the #8 spot to the #22 spot. However, the important takeaway is the 2022 QC most resembles the 2018 QC, which is not something to brag about. From all this, I ask two questions, both in regards to upsets per round and M-o-M rating.

1. The 2022 QC entangles itself with the other QCs from the #7-#14 spots. What does this entanglement mean?
2. Generally everywhere else, 2022 is worse. What does this general weakness mean?

Before we can answer these questions though, we have to see how the Selection Committee has altered the QC, and to do this, we model the Seed Curve (SC).

Here's the run-down:

1. The largest alteration is clearly the 3- thru 5-seed group. The 3- and 4-seed groups are significantly below their QC-equivalent whereas the 5-seed group is significantly above its QC-equivalent. By rule, this would project 3- and 4-seeds to fail seed expectations as a group (8 total wins among all teams in the seed group) and 5-seeds to meet or even exceed seed expectations as a group (4 total wins among all teams in the seed group). Since the QC and SC drop significantly after the 12-seed line, they cannot accurately project if the threats to the 3- and 4-seeds are R64 opponents (14- and 13-seeds). Also, the 3-seed group would be significantly higher if not for a 9-seed posing as 3-seed (I guess you can make the same argument for 4-seeds being pulled down by a 12-seed in disguise). As for the 12-seed group, the group average is being elevated (otherwise would be lower) due to two of the five (highest- and middle-ranked) 12-seeds being above average yet playing against one another in the play-in game (Only one can advance to face the awaiting 5-seed). Thus, things look pretty promising for 5-seeds as a whole.
   
2. The other significant alteration is the 7- and 10-seed groups. Not only do the 10-seeds have a higher group average, all four 7-seeds are below the QC-equivalent. In other words, we have 8/9-seeds impostering as 7-seeds while we have actual 7/8-seeds wolfing in 10-seed clothing. When I compare these two groups to 2018's 7- and 10-seed groups (when it looked break-even), it makes an even stronger argument for this year's 10-seed group (and in 2018, the groups split the match-up 2-2). 
3. As for the rest, 6- and 11-seeds look in-line, especially when you account for the two below-average 11-seeds pulling down the SC group average. 8-seeds look to have a slight advantage over 9-seeds but I would study the match-ups before YOLO'ing with a 3-1 advancement of the 8-seeds. 1- and 2-seeds are below the QC equivalent, but technically 1-seeds can never be under-seeded (they either get the four best teams or they get over-seeds). Although the SC doesn't seem to give any clear reads, I do think the overall lack of quality in the field this year suggests 1- and 2-seeds are likely to fall short of seed expectations (16 total wins as a group for 1-seeds and 12 total wins as a group for 2-seeds).

As a final insight, I want to follow-up on something I noticed in the JAN QC.

If you read that article, you should remember this table and what it portends. If the quantity of Top 50 AdjO and Top50 AdjD teams increases from JAN to FIN, it suggests a relatively calm and chalky tournament like 2017 and 2019. But if the quantity decreased over that time span, it suggests a relatively insane and upset-ridden tournament. The quantities fell! Take that data point in connection with the 2022 QC being generally below 2018 and 2021, and I think we might have "a little bit of history repeating" with those two tournaments. For the present moment, I favor outcomes more in-line with the 2018 tournament, but I've got more models to finish, and when I put it all together, I'll give my AM and UBR targets at the very end.

Historical Seed Comparisons

For each seed group, I will detail their historical ranks (from 2008-2022) in the form A,B,C,D (1=best, 56=worst) and provide the years with the closest comparisons with the ranks in the same form. As always, my analysis will follow, and I will try to keep them in bracket order for ease of match-ups.

1-seed: (11,50,51,52) similar to 2021 (3,42,43,46), 2014 (16,31,48,54), and 2018 (14,25,53,55). These are not the three years that we won't to see as matches. Even the QC analysis showed that the top seeds in 2022 were weaker than all other years (2018 and 2021 were in that study). On the bright side, they all had a 1-seed make the F4 with 2014 being the only year of them with only one. All three years also had a 1-seed leave before reaching the S16. Thankfully, none of them match the 2018 UVA profile, but BAY checks off the most boxes. The win totals for the 1-seed group was 10 in 2014, 11 in 2018 and 15 in 2021. Structurally, 2022 looks more like 2018 and 2014.

16-seed: (6,18,22,28,30,63) most similar to 2012 (1,2,8,20,44,72) and 2015 (11,13,15,40,41,74). These are arguably the strongest three years for 16-seeds, but those two comparative years had strong 1-seeds whereas 2022 does not. I stated in the UVA Upset article that all the years of strong 16-seeds and no upset, it took a weak field of 1-seeds to produce the first 16v1 upset. I am a little worried due to historical weakness of 2022's 1-seeds, but I don't see enough similarities to confidently call another 16v1 upset. However, I will not sleep easy until the ARI game goes final. For the field of 64, #18 plays #30 and #28 plays #63 in the play-in games. For the sake of history, here is 2018's crop (4,25,43,70,71,78), and it was #43 that pulled off the upset.

8-seed: (26,38,45,54) most similar to 2015 (20,27,47,55) and 2018 (22,29,32,40). 2018's 8-seeds were historically stronger top to bottom, but they fell 1-3 to the 9-seeds (28,34,37,39) who were weaker at the top, but stronger at the back three (see 9-seed ranks below to compare). 2015's 8-seeds have less deviation from 2022's crop and they went 4-0 against their 9-seeds (24,30,31,45) who were also weaker at the top, but stronger at the back three.With the 8-seeds in-line with these years (5-3 total) and 2022's 9-seeds only better at the top spot, either 2-2 or 3-1 (based on match-ups) in favor of the 8-seeds.

9-seed: (15,38,50,51) most similar to 2019 (21,32,46,52), 2009 (20,35,40,55), and 2017 (22,33,43,48). These groups went 4-0 in 2019 (when the 8-seeds were stronger than 2022's), 2-2 in 2009 and 1-3 in 2017 (when the 8-seeds were even stronger than 2022's). With the 9-seeds in line with these years (7-5) and 2022's 8-seeds being historically weaker than their counterparts in those years, either 2-2 or 3-1 (based on match-ups) in favor of the 9-seeds. Putting both pieces together, it kind of looks like 2-2 might be the play. (Additional Note: 2015's 9-seeds are probably the fourth best comparison to 2022's 9-seeds, but 2015's deviation is three times that of 2017's devation which is why I didn't include them).

5-seed: (3,10,35,48) most similar to 2013 (4,17,22,47), 2016 (5,12,31,39), and 2019 (11,14,42,49). Each of these comparisons have their flaw: 2013 misses by 13 at the 3rd spot, 2016 misses by 9 at the 4th spot, and 2019 misses by 8 and 7 at the 1st and 3rd spot. Of these three years, 2019's 12-seeds (17,24,39,50) are the closest to 2022's 12-seeds (7,30,35,47,52) assuming 7 beats 35 in the play-in game, and 5-seeds were 1-3. 2016's 12-seeds were weaker top to bottom and the 5-seeds went 2-2 against them. 2013's 12-seeds were the strongest top to bottom and the 5-seeds were also 2-2. I would assume these illogical results are probably match-up driven (such as best 12 paired against worst 5), but I don't have my spread sheet set up that way (maybe I need to).

12-seed: (7,30,47,52) closest is 2018 (15,25,42,46), and 12-seeds were 0-4 in that year. The best 12-seed is better in 2022 and they draw the 3rd-best 5-seed, so I'd study that match-up twice, especially since the 12-seed also happens to be the winner of the play-in game. The next two closest are kind of a paired examination: 2009 (6,11,49,57) and 2021 (13,43,48,55). In essence, they are the same in the back two spot, but in the front two spots, 2009 is stronger in the front two spots and 2021 is weaker in the front two spots. 2009 was 3-1 against 5-seeds while 2021 was 1-3 against 5-seeds. The 2nd-ranked 12-seed this year faces the top 5-seed, which appears to favor the 5-seed. Thus, 12-seeds at best look like 1-3 against a really strong 5-seed group this year.

4-seed: (22,36,50,56): By relative ranks, 2022 matches with 2021 (24,34,45,53) which is no surprise given the quality issues of 2022. It also pairs with 2018 (26,33,47,48) and 2008 (19,35,40,55) in relative terms. However, in absolute terms (the actual rating itself), the rating gap between 54 and 55 (remember 56 is the worst relative rank) when added to 54 would produce a relative rank of 30th. Add that gap again, it would produce a relative rank of 3rd. That's how bad the 55th and 56th teams are in the 4-seed group, and 55 (2008 VAN) lost their opening round game. All three years went 2-2 in R64, 1-1 in R32, and 0-1 in S16.

13-seed: (7,12,24,54) most similar to 2009 (6,9,39,53) and 2017 (16,20,29,55). In 2009, the 13-seeds went 1-3 against a much stronger 4-seed group than 2022's 4-seeds. In 2017, 13-seeds went 0-4 against the 2nd-strongest year of 4-seeds. With weaker 4-seeds this year than those two years (especially one of them being the historically worst), 1-3 seems like a certainty with the potential for 2-2.

3-seed: (13,15,32,56) most similar to 2009 (8,17,39,41), 2014 (11,23,38,43), and 2017 (8,15,33,34) if you ignore the outlier (yes, we have another historical worst in this group too). In 2009 and 2017, 3-seeds exceeded win expectations as a group. In 2014, they went 3-1 in R64, 1-2 in R32, and 0-1 in S16, and I would guess the 2022 tourney to be most like the 2014 tourney.

14-seed: (42,49,50,51). No comparisons. Best is 2011 with (21,27,34,56). In fact, there are four years in which their worst 14-seed is still better than the best in this year. Looks good for 3-seeds in the R64, even with the outlier.

6-seed: (16,38,45,54) most similar to 2013 (9,43,48,52), and the next closest is 2011 (19,28,39,51). In 2013, 6-seeds went 3-1 against 11-seeds, and 2013's 11-seeds were relatively stronger than 2022's 11-seeds at every spot (the one loss is probably a match-up). 2011's 6-seeds, who were stronger in the back three spots, went 1-3 against 11-seeds and those 11-seeds (15,17,18,31,72) match the front-half of 2022's while being much stronger in the back. Depending on match-ups, 6-seeds should be looking like 2-2 at worst since the competition at the 11-seed is worse.

11-seed: (16,19,41,50,66) Since the play-in game (41v66) doesn't finish until Wed, this is the best I can do. The closest years are 2017 (12,25,29,57) if 66 loses and 2021 (26,35,40,69) if 41 loses. Both are not very good approximations for this group, which doesn't help in R64 projections.

7-seed: (37,47,50,55) most similar to 2018 (30,34,38,53). 2018 is better at every spot, especially in the middle, and again, this is expected given the lower overall quality of 2022. 7-seeds went 3-1 against 10-seeds (as well as 2-1 against 2-seeds), and historically, 2018's 10-seeds (8,32,46,52) were in-line with 2022's 10-seeds. So, it appears with weaker 7-seeds than 2018, 2-2 is a possibility. I will compare the match-ups for both years. In 2018, the pairings were 3rd(38) vs 1st(8), 1st(30) vs 4th(52), 2nd(34) vs 2nd(32), and 4th(53) vs 3rd(46). The two landslide differentials each went their expected way. The close differentials both broke for the 7-seeds. In 2022, the match-ups are 3rd(50) vs 1st(14), 2nd(47) vs 3rd(49), 1st(37) vs 2nd(37), and 4th(55) vs 4th(50). One landslide differential favors the 10-seed (SANF). The close three differentials, with simple logic, would favor 7-seeds 2-1 for a 2-2 split.

10-seed: (14,37,49,50) most similar to 2018 (8,32,46,52), 2019 (11,38,45,53), and 2009 (9,29,43,47). In 2009, 10-seeds went 3-1 against a much stronger batch of 7-seeds than 2022's batch, but keep in mind 2009 is stronger at every rank. In 2019, 10-seeds went 3-1 against a relatively stronger batch of 7-seeds. By relative comparison, 3-1 looks doable in 2022 due to historically weaker 7-seeds. Consolidating with the 7-seeds analysis, a 2-2 split is also a possibility.

2-seed: (36,37,45,47) most similar to 2011 (30,38,48,56) and 2014 (31,32,39,44). 2014 is historically better than 2022 at every rank: One went to the F4, one went to E8, and two lost in R32 (One to a 7-seed and one to 10-seed). For reference, 2014's 7-seeds were at least 12 ranks better at every spot and its 10-seeds were top-heavy (better at the top spots). 2011 is better at the front half but weaker at the back half: Two went to the E8, one went to S16, and one lost in R32 to a 10-seed. Based on the big picture (2 vs 7/10 winner), if there is a R32 upset, it looks likely to be a 10 over 2, and the best chance (albeit not a strong chance) of this is MIA over AUB (if MIA wins against USC).

15-seed: (11,20,25,48) most similar to 2019 (16,19,28,44). The next closest two are 2009 (7,29,42,52) and 2013 (5,10,18,34). 2013 was much stronger at every rank, and the best 15-seed of that year was somehow paired with the worst 2-seed of that year (and one of the worst historically). This year, the strongest 15-seed is St Peter's and they are paired against the strongest 2-seed UK. I'm saying 2-seeds are 4-0 this year against 15-seeds, especially with last year's upset still fresh in our minds.

2022 Tournament - Returning Participants

2019 was the last time I looked at this information. It counts the number of teams by seed that participated in the previous tournament and received a bid in the current tournament. In short, they are returning participants. Since 2020's tournament was cancelled, there were no returners, and 2021 couldn't return teams from a 2020 tournament that never happened, so 2022 is the first tournament since 2019 that can have returning participants.

First, the yellow box is an outcome dependent on the results of the play-in game. If RUT wins, the total for 11-seeds becomes 3 and the total returning teams improves to 29. If DAME wins, nothing changes since they didn't go to the 2021 tournament. Either way, 2022 will set a record for lowest returning teams in the advanced metrics era.

The key takeaways are this. Only three other tournaments had three returning 1-seeds: 2006, 2010, and 2014. All of which were crazy tournaments:

• 2006: 17.44% M-o-M Rating, 12 total upsets, and no 1-seeds in the F4
• 2010: 17.14% M-o-M Rating, 11 total upsets, and two 5-seeds in the F4
  
• 2014: 21.35% M-o-M Rating, 15 total upsets (at least one upset every round except title game), and a 7v8 title game match-up

If you don't make the tournament in one year, and then the next year you are good enough to claim a 1-seed, it seems as though you accomplished this feat with the help of a weak field. Other models are portending another crazy year, and this data supports that thesis. Only one other year had only one returner in the 2-seed group -- 2002 -- and surprisingly, the only returner of the seed group went to the F4. That bodes well for 2022 NOVA. Other than those two points, I can't seem to find any patterns with the rest of the data, but tournament experience has to be an advantage.

Seed-Group Loss Table (SGLT)

We're going to examine the 1- thru 4-seed groups, both with the Matching method and the Linear Regression method. In the 2- thru 4-seed groups, their numbers are dependent on the outcome of the RUT/DAME play-in game, and I've highlighted their matching years dependent on who wins (Red for RUT, Yellow for DAME, Orange for either). 2020's lack of a tournament and 2021's season of cancelled games/shortened schedules have delayed my implementation of (as well as my work on) this model. This might be one of two projects I focus the most on over the off-season because I really want this tool to work. Anyways, let's start with the 1-seeds.

Matching Method: 2007 is a perfect fit except their 4-seeds were slightly stronger (lower L% of the total field). 2022 has the fourth highest L%, all three above only produced one 1-seed in the F4. Of the nine years with an L% above 4%, only two of those years had a 1-seed fail to reach the S16. What if everyone is so bad this year that the bad of the Cinderella teams just isn't good enough to beat the bad of the top-seeds? In 2013, 1-seeds had an equal number of losses as 2022, they just lost more to tournament quality teams. Once they started facing better tournament teams, they lost, which explains only nine wins (keep in mind win counts in SGLT only go up to the F4 whereas other models add in the wins from the F4 and NC games). In 2018, 1-seeds had an equal number of losses as 2022, they just lost to worse teams, suggesting either they were imposters or they didn't face enough tournament quality teams to get more quality losses. It makes me think I should look at the win-side of the equation too.

Regression Method: Either one or two 1-seeds in the F4, and two or three 1-seeds in the E8. (NOTE: The regression model for 1-seeds in the E8 is historically accurate 50% of the time. In all five years that I highlighted with similar values to 2022, the E8 model was wrong). Let's move to the 2-seeds.

Matching Method: 2018 can be a match no matter who wins. If DAME wins, three other years also match. I included 2013 because the only difference is the quantity and percentage of losses to non-tournament teams. It's the only one that matches 2022 and a 2-seed was upset by a 15-seed (most likely indicated by the higher quantity and percentage of bad losses). By inference, it suggests no 15v2 upset in 2022 since the N quantity and NL% are not as high as 2013's. If RUT wins, only two years match and they both look like floors (no worse that seven total wins through the first four rounds with four of these coming against 15-seeds). Of the ten years with lower L%, only one produced less than eight total wins through four rounds. The table seems to suggest eight to nine is more likely than ten to twelve.

Regression Method: No matter who wins, either zero or one 2-seed in the F4, and one or two 2-seeds in the E8. (NOTE: SGLT has had only one miss in F4 projection and only one miss in E8 projection). Let's see what the 3-seed group looks like.

Matching Method: If RUT wins, only two years come up: 2010 and 2011. Since both are different on the L quantity (and as a result, the L%), I toyed with counts of these years to see what they would like if they had 30 losses. If 2010 had one more L, the L% goes up higher and it is already higher than 2022 at 29 losses. Thus, it looks like it is a floor. If 2011 had three more Ls, the L% goes up too but not as high 2022's L% of 7.673%. Thus, it looks like a ceiling. As a result, I would target six or seven for WTOT if RUT wins. If DAME wins, then 2022 looks more like 2013, and 2010 would probably be a close 2nd place. Both of those years had a 3-seed go down to a 14-seed. I would think either WISC (the historically worst 3-seed) losing to returning 14-seed COLG or TXTC losing to MTST, but only if DAME wins the play-in game.

Regression Method: No matter who wins, either zero or one 3-seed in the F4, and one or two in the E8. (NOTE: SGLT has had only one miss in the F4, which happened in 2004). Let's see the 4-seeds.

Matching Method: If DAME wins, then only 2006 and 2013 matter. If RUT wins, 2010 is a perfect match except for L%. Looking at this closer, 2010's 29 Ls represented 7.8% of the field's total losses whereas 2022's 29 L represented 7.4% of the field's total losses. In short, the rest of the 2022 field lost at a higher rate than the four-seed group, so the 4-seed group should out-perform 2010's results. 2006 and 2013's results show the limit of that out-performance. With similar L%, as the N/L% goes up, out-performance goes down. Thus, 2006 should represent the ceiling of that out-performance since 2022's N/L% is even higher than 2006's N/L%. Of the ten years in which there was nine or more losses to non-tournament teams ("N" column), eight of the ten produced a 13v4 upset. That result looks highly probable in 2022.

Regression Method: If RUT wins, either zero or one 4-seed in the F4 and zero or one 4-seed in the E8. If DAME wins, either zero or one 4-seed in the F4 and zero in the E8 (which would logically force zero in the F4 since you can't be absent in the E8 and reappear in the F4).

META ANALYSIS

I'm not sure what to make of this year or its numbers.

In the Elite-level of each stat (Top 20), 2022 doesn't have the best of the stat of the five years. (NOTE: I'm leaving out 2021's data because of the anomalies of that season.) However, 2022 does have the worst value for Elite-level in two categories: 3P%D and ORBA (offensive rebounding percentage allowed). As for 3P%D, these seven go up against 12 Elite 3P% teams, which is 2nd-highest overall. I personally believe 3P%D is a statistical mirage. Though I haven't done much work to prove/disprove this notion, I believe teams have a better chance at controlling the quantity of 3pt attempts than they do at controlling the percentage of those attempts that go in. As for ORBA, it might be in the same realm as 3P%D. Teams strategically choose to commit to ORBs (UNC under Roy Williams) or get back and set up their defense (usually less athletic teams that don't want to play the game in transition or pack-line man-to-man defenses that want to clog the paint with bodies/traffic). If you play against more teams that crash the offensive boards, you're likely to have a higher ORBA (bad) and if you play against more teams that run back and set up their defense, you're likely to have a lower ORBA (good). For decades, every opponent knew UNC would rebound offensive misses, and it seemed like every year UNC was Top 5 in ORB (one game against them could really distort your numbers). Let's look at the cumulative numbers for the same table.

On this table, I've focused on the Top 60 (simply because there are 64 teams in the field). The numbers get a little better for 2022, but the real objective is to find out what qualities have the best advantage against the rest of the field. Somehow, 2022 has the most Top 60 2P% teams to go against the fewest Top 60 2P%D teams. In the year of the big man, I thought both of these numbers would be best. Apparently, 2022 should be renamed the year of the scoring big man. Either way, I like 2P% teams as a meta-pick. This year also has the fewest TOR of the five years. It's kind of hard to get the ball to your scoring big man when you're turning it over a lot. I've actually notice a lot of teams in the QC that are under-water on TOR. In other words, they turn it over more times per 100 possessions than their opponents, which means more shot attempts for their opponents. It seems like teams with a high TORD would be a safe meta-pick. Finally, this year has the most Top 60 ORB teams (probably a reaction to the 2021 National Championship game as well as the quantity of scoring big men in the game this year). In a year where the field is lacking Elite ORBA, 28 Top-60 ORB teams should have a field day. On the flip side, seven of these 28 teams are located in the ARI octet, and none of the teams in that octet are Top-60 ORBA. I'd make sure if a team is "Elite at ORBA", it is a product of fundamental boxing-out and not a mirage in the numbers. A good TORD is also a meta-pick against ORB: The opponent can't get an offensive rebound if a shot never happens. Teams that score a high 2P% and force a lot of turnovers might be the sneaky meta play this year.

FINAL PREDICTIONS

As much as the 2022 tournament looks like the 2018 tournament according to the majority of models, I think I'm going to rely on the information in the QC and model the U-B-R after the 2010. It's one thing to be lacking quality at the top: That's called parity. When you're lacking quality at every seed-range along the QC, then you're just bad all around. The UBR in 2010 was 5-4-2-0-0-0 (For my personal bracket, I was 5-3-2-0-0-0 because the other models forced particular match-ups that weren't capable of producing a 4th upset in the R32). I don't want to target AMs because I'm not sure how the total lack of quality translates to each round. If I had a guess, I would say 190s-60s-30s-12-6-1, with the last three being my personal picks for those rounds, but don't follow the AM targeting values.

As for model-based picks:

• Profile: GONZ being the only Tier 1 pick. (NOVA did make the F4 being the next closest.)
• Predictive OS/US: Followed odds for all seed-based picks except 5- and 12-seeds.
• Conf OS/US: Followed all rules except USC/UCLA
• Hist Seed Comps: Only followed for R64 picks, but I went 8v9 2-2, 7v10 0-4, 3v14 4-0, 4v13 2-2 (rolled the dice here).
• Returners: Orig article in 2019 predicts BAY not advancing past S16, crazy year M-o-M rating (17-21%) based on three 1-seed returners, and only 2-seed returner advancing to F4.
• SGLT: Went for WTot targets: Eleven for 1-seeds (Avg between 7,7,13,14,14), nine for 2-seeds, Seven for 3-seeds, and six for 4-seeds. Did not follow any of the Regression predictions, but I think 1-seed and 3-seeds will probably be correct.
• Meta: Took the 2P% and TORD special for sleepers: TEX and MIA to the E8 on Meta.

LET'S SEE HOW IT WORKS!!!!!

Let me know if any of the chart/graphics aren't large enough. I do my work on a PC, but I feel most people read on a phone/tablet with smaller dimensions. As always, thanks for reading my work, let's keep the bracket busting to a minimum, and hopefully I'll be able to do this again next year.