Might Casey Model

If you've ever heard of or read the Earnest Lawrence Thayer poem "Casey at the Bat," then you're on the right track for the subject of the current article. Mighty Casey is batting with two runners in scoring position and facing an 0-2 count. While we all know that Mighty Casey strikes out swinging, I want to know how many tournament teams have faced the same situation (an 0-2 count) and how they performed. In college basketball terms, an 0-2 count will be defined as back-to-back tournament defeats in the R64 and facing the threat of a strike out (a third tournament appearance).

It is very rare for teams to make three consecutive NCAA tournaments. Of the 70+ Power-6 conference teams since 2002 (the start of the Advanced Metrics era and the start of all comparable data for this study), only one of those teams have made the field of 64 for every single tournament: KU. In this twenty-tourney span (2020 did not have a tourney and disrupted this three consecutive factor), only nine more teams have made at least 15 tournaments, which is the idealized threshold (a team makes it three years and misses the fourth and repeats this pattern for 20 tourneys), though this is not the case for any of these nine teams. As rare as it is to make the tournament three times in a row, it is even rarer to lose two R64 games in the first two of these three games. Only sixteen instances fit our Mighty Casey scenario. If it so rare, then why am I studying this phenomenon? Two teams likely to make the 2023 tourney also fit this criteria. So, let's see what these sixteen historical Caseys can teach us about the two potential 2023 Caseys.

ATTRIBUTES

Let's take a look at the tournament attributes of the teams and try to find some patterns.

The teams are sorted by Wins in Year 3 (Column W3). The first column group is the respective tourney years (Year1, Year2 & Year3). The third column group is the seed for each respective year (S1 = Seed in Year 1, etc). The fourth column group is Margin of Victory for each of the R64 losses in the first two years. The fifth column group is the Coaching Tourney Success up to the point of the third year (Logically, it's the same for all three years since the first two years do not add any wins to the tournament win total of the coach). Yellow highlighting indicates a team played in a play-in game. Blue highlighting indicates a MOV result from an OT game.

Here's what I see:

• General seed improvement: Teams that advance to the S16 in Y3 have their best seeding in year 3. Of the eight S16 teams, seven fit this rule. (Only FSU breaks it). Of the eight non-S16 teams, six fail this rule, with the two exceptions being GTOWN and MARQ. For the nine total teams that fit this rule, four of them improved the seed line by at least five seeds from Y1 to Y3, and all four of these advanced to the S16.
  
• At least a 4-seed: This should come as no surprise, teams with better seeds should have easier paths. Of the eight S16 teams, only two were seeded worse than a 4-seed (FLST and NCST break this rule). Of the eight non-S16, seven fail this test, leaving GTOWN as the lone exception.
• Close losses: Losing close games should be emotionally stirring. It should be motivation to end the bad streak. Of the eight S16 teams, seven played a game within 4 points, and the eighth lost by five in OT (which implies the end of regulation was tied). Of the eight non-S16, four lost both games by 5 or more points each.
• Tourney Winning Coach: A coach that has won previous tournament games should be a calming factor in breaking bad streaks. Of the eight S16 teams, seven had a coach with at least seven tournament wins prior to the third game (again FSU bucks this trend). Of the eight non-S16 teams, seven had a coach with less than seven tournament wins prior to the third game (again GTOWN breaks this rule).
• As for our two notable exceptions throughout (FLST and GTOWN), I will point out that both took place in fairly wild years. 2011 had 13 upsets and a 19.85% M-o-M rating whereas 2012 had an average-matching 9 upsets to go along with a 17.14% M-o-M rating. Early indications (solely based on the ETM and my personal interpretations of current advanced metrics data) predict a 2023 tournament in-line with these measurements. In other words, some of the above rules may break, producing exceptions to the rule.
  

Let's take a look at 2023's two contenders.

Let's apply our four rules and one guideline (the fifth rule).

• Both teams are likely to have their best seed line in Y3. CONN is projecting as a 1-seed and VT as a 6-seed. If CONN achieves a 2-seed or better or if VT achieves a 5-seed or better, then both will have improved their seed-line at least five seeds, which is a 100% hit rate for the S16. It must be noted that VT would be the first team to improve five-seeds from Y1 to Y3 with a worse seed in Y2 than Y1 (the four paragons never fell in seed line from Y1 to Y2).
• CONN passes the 4-seed or better, VT fails this test. If VT fails this test and still advances to the S16, it would be the third exception, joining fellow ACC teams FLST and NCST.
• CONN fails the close loss test. If they advance to the S16, they would be the first to fail this test and still achieve that feat. VT passes this test as it was a close overtime loss.
• Both CONN and VT fail the tourney winning coach test. If either achieve the S16 feat, it would lower the bar set by FLST at 3.
• The fact that the four rules produce a both pass, a CT-pass but VT-fail, a CT-fail but VT-pass, and a both fail, seems to indicate that the fifth rule above might be the most important.

MEASURABLES

Since basketball games are played by and decided by players who produce statistics, let's see what the statistics say about the Casey scenario.

(Note: If the table is too small, let me know in the comments section). This is a table of Returning production (noted by the R in the column heading) for each of these stats: Minutes (M), Points (P), Field Goals Attempted (FG), Free Throws Attempted (FT), and Steals (S). Each stat is the percentage returned from the previous team, so the 2nd year team returned from the 1st year team (denoted by the 2) and the 3rd year team returned from the 2nd year team (denoted by the 3). Thus, R-M2 is the percentage of minutes played by the Y1 team returned to the Y2 team (etc. for each change in notation). Finally, the change is the difference between Y2 and Y3. If Y3 returned a larger percentage than Y2, change will be positive, and it will be negative if Y2 returned a larger percentage than Y3.

Though there's a lot of information, there's not many identifiable patterns, but here's what I see:

• Teams with at least +12.81% Change in Returning Points win at least one game (9 out of 9).
• Teams with at most -30% Change in all categories fail to advance to the S16 (3 out of 3).

What do our two current Caseys look like?

Two words: Not promising. CONN looks like the three teams that failed to reach the S16, and if this happens to be the case, none of the rules in the attributes section seem to indicate if it would be 0 or 1 win. On the bright side, VT has similar comparisons to 2015-2017 BAY, albeit within 2-5 percentage points of the four offensive categories. However, VT doesn't fall into either of the two rules.

CONCLUSIONS

While there is nothing conclusive from this model at the current moment, I thought the intellectual exercise would provide some benefits. First, several tourneys like 2023 had multiple Casey candidates (2011, 2012, 2015, and 2017). I thought maybe this model could provide comparable insights into the quality of the current year, such as the insight identified in fifth attribute rule. Second, sports pundits and analysts in March will likely do what they do best and try to make a storyline out of CONN's short-comings in the last two tourneys (while simultaneously over-looking VT's similar situation). I'll have already done some of the work. Thus, if they don't identify the previous 16 teams in their work, you'll know their work isn't thorough. Likewise, they may identify categories and rules that I haven't thought of, and I can double-check them as well as add them to my model. Third, rules and exceptions to rules always seem to be the Achilles' heal of the bracket scientist. A model that can aid in identifying what the exception might look like and when/how to apply it is the next level in bracket science. In short, it's the difference between what happens and why it happens. As always, thanks for reading my work, and hopefully, we'll get more clarity on these two teams in March as a result of the Mighty Casey Model.