"My rule was I wouldn't recruit a kid if he had grass in front of his house.
That's not my world. My world was a cracked sidewalk." —Al McGuire

Saturday, January 14, 2012

Pomeroy correctly predicted our record preseason - and now has MU on pace for 2nd place

Preseason Big East Player of the Year Ashton Gibbs leads Pitt into the Bradley Center at 1 p.m. CST as the Panthers seek an upset for their FIRST Big East win. Like Dwight Buycks last year, Gibbs has been forced to play out of position at point guard and the Panthers have the worst offense (93.2 points per 100 trips) and worst shooting percentages on both 2-pointers (42.3%) and 3-pointers (22.7%) of any team in Big East play so far.

Marquette is exactly on pace this year - having lost exactly as many games (4) as Pomeroy predicted in the preseaon prediction. A loss to Pitt would be a major blow, while a win could be the second in a 10-game stretch of being the favorites during which an 8-2 mark would keep MU on pace for 2nd place in the Big East.

Focusing on Pomeroy’s predicted conference finish gives by far the clearest picture of where all of the teams stand today. Marquette is 2-2 in conference play – big deal – we’ve played Georgetown and Syracuse to single digits on the road and if we keep playing like this we finish 12-6.

How accurate is Pomeroy? I saved his predictions before the season started, and pulled them back up tonight when a colleague said he really didn’t like the fact that Pomeroy uses a “cumulative” predictor. Here is what he published BEFORE the season started:

Pomeroy PreseasonPrediction
Mount St. Mary's94%
Norfolk St.96%
Wisconsin Green Bay91%
Northern Colorado93%
Louisiana St.73%
Wisconsin Milwaukee89%
St. John's85%
Total 11-4 (CORRECT)1072%

Adding up the percentages gives the best prediction of what a team’s record will be, and in this case they added up to 10.7 wins and 4.3 losses – rounding to the actual record of 11-4 (prior to the season there was no prediction for the two unknown opponents in tournament play, so Marquette is 13-4 overall with those two games not on this list).

Conference predictions even more accurate now; USF win over Seton Hall big
Of course, predictions get even more accurate as more and more games are played, so it is much more important that Pomeroy predicts Marquette will be one of four teams to go 12-6 in conference play to finish in a 2nd place tie behind 16-2 Syracuse.

Cincinnati and Notre Dame are currently tied in 2nd in the conference at 3-1 apiece, but they are projected to finish 10-8 and 9-9 respectively, so we should be more worried about Seton Hall, Georgetown and West Virginia right now, who have much easier schedules and better results to date and are likely to be battling MU for a top spot.

If MU is upset no more than two times in the 10-game stretch that started with the win over St. John's, then MU could pass Cincinnati while the Bearcats are playing at UConn, West Virginia and Rutgers and at home against Syracuse after today’s much easier game against Villanova. Meanwhile Notre Dame gets UConn twice, Syracuse and road games at Seton Hall and Rutgers for their next five. If Cincy and Notre Dame win most of their five in that tough stretch then the prediction will adjust, but right now MU is in better position than either team.

The fact is that right now MU is on course to finish ahead of traditional powers UConn, Louisville and of course Pitt. Things can change, but this is the much better snapshot of the “standings” than just looking at current records:

TeamWLTo stay on pace
Seton Hall126win at Nova Wed
Georgetown1265 wins in next 6
West Virginia126win 2 of 3 before Cuse
Marquette1267 wins in next 9
Cincinnati108after Nova, 1 of next 4
Connecticut108win 2 or 3 of 4 before at Gtown
Notre Dame992 upsets in next 7 games
South Florida810 
St. John's612 

So of the four evenly matched teams, the key will be to see if any can beat one of the other three on the road to shoot ahead. Marquette came the closest so far with the narrow loss at Georgetown, while the other two home teams dominated. Now Marquette needs to hold home court against Seton Hall and Georgetown, and try to steal a road upset at West Virginia.

The potential miniconference tie-breakerat Seton Hall 67, West Virginia 48
at Georgetown 73, Marquette 70
at West Virginia 74, Georgetown 62
PredictedJan. 31: at Marquette 73, Seton Hall 66
Feb 21: at Seton Hall 63, Georgetown 62
Feb. 24: at West Virginia 74, Marquette 71
March 3: at Marquette 70, Georgetown 67

Why is Pomeroy’s predictor so much more accurate than a game-by-game approach?
The idea of going by actual game-by-game predicted wins rather than cumulative percentages could be less accurate than flipping a coin to predict overall records.

Here is why. Let's say that a team has an pretty clear edge in it's next six games, a 2-to-1 chance of winning in each of those games. This is the same odds of rolling a 3, 4, 5 or 6 on a single die (a win) instead of rolling a 1 or 2 (a loss).

The suggestion that predictions should be the result of simply assuming wins for every team that is favored over the course of a season would indicated that the team should be predicted to go 6-0 over those games because they are clearly better than all 6 opponents.

There is only an 8% chance that a team goes 6-0 over those games, and a 92% chance that they play "below expectations" if that is the bar. While there is a 66% chance of winning the first game, there is only a 44% chance of winning the first two, a 29% of winning the first 3, etc.

So if someone bets such a team will go 6-0 against a guy who flips a coin and bets they will go 3-3, then the coin-flipper is more likely to prove the better predictor.

But Pomeroy's cumultive total is much more likely to be the most accurate than either going game-by-game or flipping a coin - correctly predicting that a team in this situation will have the greatest chance of going 4-2, though certainly there is also a good chance of going either 5-1 or 3-3 - just not 6-0.

The reason it is NOT immediately obvious that going with game-by-game is such a terrible system is that most teams have games they are favored in and games they are not in any stretch, so it just doesn't hit you that this is what is happening.

Also, I'd refer to my piece earlier in the week where I documented that every year Marquette - and every other team - play a game well below their ability once every six games and play a game well above their ability once every six games, and during the hypothetical 6-game stretch it is very likely they will play one terrible game, and it is very unlikely they will win that game, so game-by-game is terribly flawed and Pomeroy is exactly right.

It's only in cases like Murray State, where their last 11 games are all against teams that are ranked #200 or lower and they have between an 81% and 97% chance of winning, that a game-by-game approach has any chance of being more accurate than the cumulative prediction model Pomeroy uses. There the game-by-game prediction of a 28-0 record is not quite as likely as Pomeroy's prediction of a 27-1 mark, but there you have a fighting chance.

Stop setting unrealistic expectations! 7-2 in next 9 would be strong
This isn't just an academic exercise - this is where so many fans get unfairly down on the team. For example, if you look at Marquette's next 9 games you can say, "Marquette should win EACH of these games." True, we have the edge in all nine matchups. But many fans - and a game-by-game predictor model - misunderstand this to think it means that "Marquette should win these next nine straight to be 23-4."

No they shouldn't. It is VERY unlikely Marquette wins all 9, so at this point the expections are so detached from reality that if the team goes 7-2 during the stretch fans view it as a disappointing stretch because they were upset twice and didn't beat anyone they weren't supposed to beat. 9-0 becomes PAR in fans minds, and even if the team goes undefeated - like the early stretch the Three Amigos had to get to #8 one year, and people don't realize the accomplishment. in fact, MU should lose at least two of those games, and possibly three, and has only a 7% chance of winning all 9, so fans should be happy with 7 of 9 wins, very happy with 8, and unbelievably ecstatic if we win 9 more in a row. It should only be if the team goes 6-3 or worse that fans should be disappointed in the stretch.

So expect to have Marquette upset twice in the next 9 games – and only panic once we’ve been upset for a third time. A 7-2 keeps Marquette on pace for a potential 2nd place finish.

No comments: