Wednesday, March 20, 2013

South bracket: log5 prediction

As I did last year and have done in the past, I will be running log5 analysis based on the Pomeroy ratings for at least as long as the Hoyas are in the field.

Ken Pomeroy has run the official log5 odds for the tournament as a whole on his website. Rather than duplicate his what I will try to do here is provide supplementary information.

Log5 in Other Regions

The first question I want to answer is to what extent the Hoyas were unlucky or lucky to be the #2 seed in the South region as opposed to the #2 seed in some other region. To answer that question, I ran the log5 odds for the Hoyas in each region. Here are the results:
Bracket   2nd Round  Sweet 16  Elite 8  Final 4
South       89.6     60.7       20.0     10.2
East        88.9     63.3       44.5     18.4
Midwest     91.3     55.8       31.4     12.0
West        86.0     57.2       35.1     15.5
Based on that analysis, the South region is of only average difficulty when it comes to making it to the Sweet 16, but getting to the Elite 8 is a particularly difficult challenge. The reason is the presence of the Florida Gators, the top team in the Pomeroy ratings thanks to their obliteration of the rest of the SEC and some very strong non-conference performances. Should they manage to get past Florida though, they are a slight favorite to advance. Kansas is the weakest #1 seed by the Pomeroy ratings, #4 seed Michigan rates as a tossup, and the Hoyas would be a 60% or greater favorite against any other team from the top half.

Top 100 Performance: South Region

In last year's log5 post, I wondered whether looking at performance solely against teams in the top 100 of the Pomeroy ratings would produce more accurate predictions. Given no team the Hoyas are likely to meet other than Round of 64 opponent FGCU will be ranked outside the top 100, it seemed like measuring how teams fare against tournament-quality teams would be a better predictor of tournament success. To that end, I calculated how every NCAA tournament team fared against the top 100. Please note these are raw averages, not adjusted for opponent or venue. Here is what the South region log5 looks like based on top 100 performance.
Seed  Team           2nd Round  Sweet 16  Elite 8  Final 4
1     Kansas            98.6      88.4     68.8     36.4
16    Western Kentucky   1.4       0.1      0.0+     0.0+
8     North Carolina    55.2       6.8      2.2      0.3
9     Villanova         44.8       4.6      1.3      0.2

5     VCU               63.1      33.6      9.7      2.8
12    Akron             36.9      15.0      3.1      0.6
4     Michigan          77.6      45.0     14.1      4.3
13    South Dakota St.  22.4       6.5      0.9      0.1

6     UCLA              59.7      10.5      4.5      1.2
11    Minnesota         40.3       5.3      1.8      0.4
3     Florida           91.9      81.0     66.5     44.5
14    Northwestern St.   8.1       3.2      1.0      0.2

7     San Diego St.     62.3      25.2      5.4      1.5
10    Oklahoma          37.7      11.4      1.7      0.3
2     Georgetown        91.1      61.9     18.9      7.2
15    FGCU               8.9       1.6      0.1      0.0+
Note I listed a couple teams in bold in the bracket. The Hoyas were bolded to highlight their odds. The other teams I bolded are the result of one of the problems inherent in the method, namely that not every team in college basketball plays enough top 100 teams to get a good grasp of how good they are. Like Ohio last year, Akron presents a particular challenge. By virtue of a season sweep of the Bobcats, the 79th-ranked team in the Pomeroy ratings, the Zips come out as a well above-average squad in their seven games against top 100 competition. This seems very implausible to me, so I therefore manually adjusted their rating to reflect average performance against the top 100. This still leaves them an above-average 12th seed and better than North Carolina and Villanova, but does not break the system.

The Effect of Greg Whittington (and Pitt)

As devoted Hoya fans, we are aware the Hoyas experienced of Greg Whittington, which resulted in changes in how the team played. Cognizant of that, I broke down the Hoyas' performance in three ways.
Sample      Off Eff.   Def Eff.
All Games     96.8       89.4
Post-Whitt   100.7       87.7
W/o Pitt      98.1       87.2
I used the ratings for all games to produce the above odds. Keep in mind that even the least flattering sample size of all games has the Hoyas as an very good team, coming out 11th in the field. While that would not leave them as a #2 seed, they are comparatively much more deserving than last year's #3 seed. They are the best defensive team in the field and have a profile similar to, but better than, last year's #4 seed Louisville team that made it to the Final 4.

The most natural split would be before and after Whittington's suspension. This is natural for a number of reasons. First, Whittington will not be returning this season. Second, the Hoyas played 11 games against top 100 foes without Whittington, so we do not have a small sample size problem. Third, as Hoya fans, we recognize the current version of the Hoyas is better than the version of the Hoyas we saw in November and December, so we want to think of them at their best.

While I want to adopt this approach, particularly for the third reason, I am not fully comfortable with it for a number of reasons. First, the Hoyas still played those games and 80% of the contributions to those games came from people who will be playing in March. Second, the Hoyas are not the only team whose characteristics have changed over the course of the season. Judging them off post-Whittington ratings reflects a trend-based approach I am otherwise eschewing in my analysis. (If you're curious, Dan Hanner ran those numbers.) Third, they had one particularly anomalous game that is skewing their overall numbers. Half of the effect of removing Whittington comes from removing this one game. Given that I believe college athletics is prone to extreme games resulting from events extrinsic to the on-court (or -field) action, I am in some ways more comfortable with throwing out just one game than I am six.

With that in mind, here's how all three incarnations of the Hoyas fare in the analysis.
Team           2nd Round  Sweet 16  Elite 8  Final 4
All Games         91.1      61.9     18.9      7.2
Post-Whitt        94.9      75.3     31.9     16.5
W/o Pitt          93.8      71.0     27.0     12.7
In doing this analysis, it's worth noting the Hoyas without Whtitington come out as the fourth-best team in the field. That they still only have a 16.5% chance to make the Final 4 is because Florida and Kansas are the two best teams in the field.

The Problems Inherent in the Method

I have already discussed the problem of small sample sizes resulting in teams that appear to be much better than their overall body of work indicates. Akron is perhaps the most extreme example of this, but Memphis is another. The Tigers come out slightly better than the Zips, but they did not fare particularly well in their two games against teams ranked in the top 70. I am not comfortable declaring Josh Pastner's squad to be the 16th-best team in the country.

Since this is the second year I have done this, I can also look at last year's results and see which teams overachieved and underachieved relative to how their performance against top 100 teams would lead you to predict. The big overachievers all came from one of the prototypical power conferences. Some of the teams, like Louisville, did fairly well in conference, while others, including Cincinnati, South Florida, and, yes, NC State, fared poorly. Underachievers came from everywhere, but the biggest cause seemed to be mid-majors with gaudy performances against top 100 but not elite foes. Teams in this category included Wichita State, San Diego State, and St. Mary's.

It is worth noting that on the whole Ken's official log5 projection, based on all games, outperformed my raw, unadjusted top 100 performance last year when it came to predicting how the tournament ensued. If this method does not do better this year, I most likely will not be running it next year.

Top 100 projection: East region

Having gotten my disclaimer out of the way, I wanted to quickly run through projections for the other regions.
Seed  Team           2nd Round  Sweet 16  Elite 8  Final 4
1     Indiana           97.3      69.2     51.5     28.9
16    LIU/James Madison  2.7       0.2      0.0+     0.0+
8     NC State          56.7      18.7     10.7      4.2
9     Temple            43.3      12.0      6.1      2.0

5     UNLV              52.9      28.2      9.1      3.0
12    Cal               47.1      23.9      7.1      2.2
4     Syracuse          84.1      45.0     15.3      5.4
13    Montana           15.9       2.9      0.3      0.0+

6     Butler            53.2      31.1      9.2      3.4
11    Bucknell          46.8      25.9      7.0      2.4
3     Marquette         60.4      28.4      7.5      2.5
14    Davidson          39.6      14.6      2.9      0.7

7     Illinois          40.4       8.5      4.1      1.3
10    Colorado          59.6      16.4      9.4      3.8
2     Miami-FL          93.9      73.9     59.8     40.0
15    Pacific            6.1       1.2      0.3      0.0+
NC State-Temple is one of several intriguing mid-major vs. power conference games. Syracuse, Marquette, and Illinois are all good test cases for whether a stronger adjustment for strength of schedule would help the methodology, as all three performed poorly relative to their seed in strong conferences. Illinois-Colorado is a very interesting matchup in that regard.

Top 100 projection: Midwest region

Moving on to the other half of the bracket, we see the NCAA's top overall seed is a very good team, but perhaps not quite the strongest team in the field.
Seed  Team           2nd Round  Sweet 16  Elite 8  Final 4
1     Louisville        99.0      67.1     42.5     24.5
16    N. Carolina A&T    1.0       0.0+     0.0+     0.0+
8     Colorado St.      55.3      19.4      9.4      4.0
9     Missouri          44.7      13.6      5.9      2.2

5     Oklahoma St.      51.2      19.4      6.4      2.3
12    Oregon            48.8      18.0      5.8      2.0
4     St. Louis         79.3      54.9     28.3     15.2
13    New Mexico St.    20.7       7.7      1.7      0.4

6     Memphis           65.2      36.6     13.6      5.6
11    St. Mary's        34.8      14.5      3.6      1.0
3     Michigan St.      86.7      46.8     16.9      6.7
14    Valparaiso        13.3      2.1      0.2      0.0+

7     Creighton         72.7      28.5     16.8      7.8
10    Cincinnati        27.3       5.8      2.1      0.5
2     Duke              93.9      64.8     46.6     28.0
15    Albany             6.4       0.9      0.1      0.0+
Creighton-Cincinnati is perhaps the most interesting game of the round of 64. In addition to the mid-major vs. power conference battle, it also is a contrast of styles, as the Blue Jays are an outstanding offensive team with a weak defense while the Bearcats have a mediocre offense and a strong defense. I have already noted Memphis; my subjective opinion is that this system overrates their chances of a Sweet 16 trip relative to Michigan State's. Yes, St. Louis does seem to be that good, though the official log5 prediction likes Oklahoma State much more than this does. Duke's projection is based on their performance with Ryan Kelly. With Kelly, they come out as the fifth-best team in the field, narrowly behind the Hoyas. In all games, they are the ninth-best team in the field and have a 16.8% chance to make the Final 4.

Top 100 projection: West region

Seed  Team           2nd Round  Sweet 16  Elite 8  Final 4
1     Gonzaga           99.3      70.7     54.7     39.8
16    Southern           0.7       0.0+     0.0+     0.0+
8     Pitt              54.0      16.6      9.9      5.4
9     Wichita St.       46.0      12.7      7.1      3.6

5     Wisconsin         44.9      23.8      6.6      2.9
12    Ole Miss          55.1      32.0     10.2      5.0
4     Kansas St.        58.6      27.9      7.9      3.5
13    Boise/La Salle    41.4      16.2      3.6      1.3

6     Arizona           63.9      34.7     22.0      9.6
11    Belmont           36.1      14.7      7.3      2.3
3     New Mexico        75.1      42.9     28.1     12.8
14    Harvard           24.9      7.7      3.1      0.7

7     Notre Dame        47.1      20.0      6.5      1.7
10    Iowa St.          52.9      23.8      8.4      2.3
2     Ohio St.          81.4      51.1     23.6      8.9
15    Iona              18.6       5.2      0.9      1.1

I manually adjusted Iona's odds to put them more in line with the other 15 seeds in the field. They played only four games against top 100 competition and went 2-2, beating Denver and Georgia and losing to La Salle and St. Joseph's. Ohio State by this method is by far the weakest #2 seed, coming out just behind Ole Miss, Missouri, and NC State. They still stand a decent chance of getting to the Elite 8, as Arizona and New Mexico are both likely to be overrated by this methodology while the Buckeyes are underrated.

Finally, a gentle reminder: efficiency ratings are not destiny.

No comments:

Post a Comment