Pomeroy's projected records

[JasonEvans]

I am not sure how many of you know this, but on Ken Pomeroy's excellent site (http://kenpom.com/rate.php), you can click on a team and find out how his models project that team will finish the year in terms of wins and losses. I am not certain exactly how he does it but it is more sophisticated than saying, "Duke will be at least a slight favorite in all their remaining games, so Duke will run the table." If a team is playing 5 games where he projects them with just a 60% chance of winning, his program will probably say they will go 4-2 in those games. Make sense?

Anyway, I looked at how he projects the ACC (http://kenpom.com/conf.php?y=2009&c=ACC)and thought it would be worth copying here--



Team Current Projected
Duke 18-1 (5-0) 28-3 (14-2)
Wake 16-1 (3-1) 25-4 (12-4)
UNC 17-2 (3-2) 26-4 (12-4)
Clem 17-2 (3-2) 24-6 (10-6)
VaT 14-5 (4-1) 19-11 (9-7)
Miami 14-5 (3-3) 19-10 (8-8)
FSU 16-4 (3-2) 21-10 (8-8)
BC 15-6 (3-3) 19-12 (7-9)
Mary 13-6 (2-3) 17-13 (6-10)
NCSU 10-7 (1-4) 14-15 (4-12)
Virg 7-9 (1-4) 10-17 (4-12)
GaT 9-10 (0-6) 11-18 (2-14)

-Jason "Look at that a bit and then lets chat about it ;)" Evans

[Billy Dat]

From your computer simulators' results displays to the ears of the basketball gods, Ken Pom!

I wonder if his machines could predict the ebb and flow of debate on this board if those predictions came true?

I'll attempt a NY Post Headline capturing the zeitgeist:
"OH NO, GERALD'S GOING PRO!"

[Cell-R]

Whew. Thank God Georgia Tech will win at least one conference game, it says they will win TWO! I was starting to get worried. :p

Oh and there is NO WAY we will be 14-2 in conference play if they don't start giving more minutes to that Gerald Singler guy... He's going to forget how to play basketball at this rate.

[Mr Blue Devil]

I was just thinking about Duke's remaining schedule and figure we split with Wake and UNC, giving us 3 total losses on the season.

[Jumbo]

What's really amusing is he gives Duke a 9.81% chance of going unbeaten. Uh .... ?

[KyDevilinIL]

14-2 seems a hair optimistic to me, but hardly impossible. Still, I wouldn't put money on that as Duke's final ACC record.

EDIT: If you look at the projected Duke schedule, it has us going 29-2, yet it still says we finish 28-3. I must be misreading or misunderstanding something.

[gumbomoop]

Interesting projections, to be sure, and so far Duke has been the most impressive, if by a slim margin over Wake and UNC. My concern is with the wacky unbalanced ACC schedule. Other posters will remember better than I how evenhanded over time this unbalanced thing balances out, but I gotta say, I don't remember Duke ever getting one of the easier schedules. Probably selective homer-memory on my part, so if anyone can enlighten me on recent history, do so.

And this year, playing UNC, Wake, and Va Tech twice, plus Clemson on road, looks pretty tough for Duke. Especially since UNC and Wake are finished with each other, and both UNC and Va Tech get UVA twice. I can't remember who gets Ga Tech twice, but it ain't us.

So we have left lots of tough road games: UNC, Wake, and Clemson (talent), Va Tech (some talent plus revenge), MD (vitriolic revenge), and BC (who knows). Plus tough home games with UNC, Wake, Miami, maybe FSU. Only obvious win is UVa at home. To go 9-2 the rest of the way with this unbalanced schedule would be truly impressive, and a deserved #1 seed.

Will Carolina lose 2 more? Away to Duke (let's hope)? At Va Tech? At NC St? (surely not) Will Wake lose 3 more?

[dukelifer]

I am not sure how many of you know this, but on Ken Pomeroy's excellent site (http://kenpom.com/rate.php), you can click on a team and find out how his models project that team will finish the year in terms of wins and losses. I am not certain exactly how he does it but it is more sophisticated than saying, "Duke will be at least a slight favorite in all their remaining games, so Duke will run the table." If a team is playing 5 games where he projects them with just a 60% chance of winning, his program will probably say they will go 4-2 in those games. Make sense?

Anyway, I looked at how he projects the ACC (http://kenpom.com/conf.php?y=2009&c=ACC)and thought it would be worth copying here--



Team Current Projected
Duke 18-1 (5-0) 28-3 (14-2)
Wake 16-1 (3-1) 25-4 (12-4)
UNC 17-2 (3-2) 26-4 (12-4)
Clem 17-2 (3-2) 24-6 (10-6)
VaT 14-5 (4-1) 19-11 (9-7)
Miami 14-5 (3-3) 19-10 (8-8)
FSU 16-4 (3-2) 21-10 (8-8)
BC 15-6 (3-3) 19-12 (7-9)
Mary 13-6 (2-3) 17-13 (6-10)
NCSU 10-7 (1-4) 14-15 (4-12)
Virg 7-9 (1-4) 10-17 (4-12)
GaT 9-10 (0-6) 11-18 (2-14)

-Jason "Look at that a bit and then lets chat about it ;)" Evans

Given the NCAA's infatuation with computers- I say we have enough data now to declare Duke the ACC champion. There is probably enough data for all other conferences as well. So let's stop all this regular season business - with unbalanced schedules which mean nothing- and get right to the NCAA tourney next week. Duke is in good shape and playing well right now. What are we really going to learn in the next few weeks anyway? Also, the longer we wait the greater the chance that someone will get sick or injured or worn out from playing too many minutes. I say play now and expand the field to the every team in Division 1! Let's get on with the February Frenzy.

[JasonEvans]

Team Current Projected
Duke 18-1 (5-0) 28-3 (14-2)
Wake 16-1 (3-1) 25-4 (12-4)
UNC 17-2 (3-2) 26-4 (12-4)
Clem 17-2 (3-2) 24-6 (10-6)
VaT 14-5 (4-1) 19-11 (9-7)
Miami 14-5 (3-3) 19-10 (8-8)
FSU 16-4 (3-2) 21-10 (8-8)
BC 15-6 (3-3) 19-12 (7-9)
Mary 13-6 (2-3) 17-13 (6-10)
NCSU 10-7 (1-4) 14-15 (4-12)
Virg 7-9 (1-4) 10-17 (4-12)
GaT 9-10 (0-6) 11-18 (2-14)

If the season were to turn out this way -- and I know it will not, but this is as good a guess as any you will find right now -- what does it say for A) NCAA bids and B) coaching futures.

A) Duke would be a #1 seed. I am not so sure about Carolina and Wake-- if one of them were to win the ACC tourney they could likely also get a #1 seed. Neither would be lower than a #2, I think.

Additonally, Clemson would get a very high seed, probably about a #4. Va Tech would make the dance, likely in the #7 or #8 seed kind of range. Miami and FSU would probably need either a signature win (over one of the top 3 studs in the league) or would need a win in the ACc tourney to make the dance. There is no question that if the standings play out like this, the ACC could be looking at only 5 bids-- which would be a shame.

B) I think Paul Hewitt might not get to coach Derrick Favors if he goes 2-14 in the conference. That is really bad. Similarly, I wonder if Lowe can survive a losing record at State. 6-10 would be a bad ACC record for Gary Williams, but I think he would survive another year.

--Jason "I know Dino is doing a great job, but if we go 14-2 then K better win COY" Evans

[gumbomoop]

Interesting projections, to be sure, and so far Duke has been the most impressive, if by a slim margin over Wake and UNC. My concern is with the wacky unbalanced ACC schedule. Other posters will remember better than I how evenhanded over time this unbalanced thing balances out, but I gotta say, I don't remember Duke ever getting one of the easier schedules. Probably selective homer-memory on my part, so if anyone can enlighten me on recent history, do so.

And this year, playing UNC, Wake, and Va Tech twice, plus Clemson on road, looks pretty tough for Duke. Especially since UNC and Wake are finished with each other, and both UNC and Va Tech get UVA twice. I can't remember who gets Ga Tech twice, but it ain't us.

So we have left lots of tough road games: UNC, Wake, and Clemson (talent), Va Tech (some talent plus revenge), MD (vitriolic revenge), and BC (who knows). Plus tough home games with UNC, Wake, Miami, maybe FSU. Only obvious win is UVa at home. To go 9-2 the rest of the way with this unbalanced schedule would be truly impressive, and a deserved #1 seed.

Will Carolina lose 2 more? Away to Duke (let's hope)? At Va Tech? At NC St? (surely not) Will Wake lose 3 more?

[pfrduke]

14-2 seems a hair optimistic to me, but hardly impossible. Still, I wouldn't put money on that as Duke's final ACC record.

EDIT: If you look at the projected Duke schedule, it has us going 29-2, yet it still says we finish 28-3.

It's the difference between what's projected in each individual game and what the likely won/loss record will be. So we may have a greater than 50% chance of winning our all but one of our remaining games, but similarly our most likely record at the end is 28-3.

A very important note to make is that the Pomeroy statistics are not really predictors of future performance. What they are is reflective of past performance. The "prediction" function on his site essentially asks, if Duke and Wake play at Wake, and Duke plays to the same level it has all season, and Wake plays to the same level it has all season, and home court advantage is factored in, here's the most likely final score. It makes no effort to "predict" in a more sophisticated fashion whether Duke will be able to play at the same level against Wake (or vice versa).

In that way, it's wholly separate from something like Baseball Prospectus' projected records over the course of the season (if you're at all familiar with them), which factor in expectations about how we thought a team would play along with actual performance thus far. To be more clear (or perhaps more confusing), Pomeroy accepts past performance at face value - a team with a 110 offensive efficiency and a 90 defensive efficiency is just that. It makes no effort to determine whether that team should be performing at a 110 offensive efficiency and a 90 defensive efficiency; that is, whether a team has overperformed or underperformed expectations.

[gumbomoop]

Forgot to add: maybe Miami will give UNC a game in Miami?

[CameronCrazy'11]

14-2 seems a hair optimistic to me, but hardly impossible. Still, I wouldn't put money on that as Duke's final ACC record.

EDIT: If you look at the projected Duke schedule, it has us going 29-2, yet it still says we finish 28-3. I must be misreading or misunderstanding something.


He thinks we should be favored in every regular-season game except 1 (at Carolina), but that we'll still probably lose two more.

Let's say you believe there's a 60% chance for Team A to win in each of its next 10 games, meaning that it would be the favorite in each of those games. Now if I asked you how many of those games you thought Team A was going to win, you wouldn't say all 10 (the chances of this happening are less than 1%). The best guess is probably that they'll win 6 or 7 out of the 10, even though they're favored in each individual matchup. Pomeroy is doing something like that to predict Duke's record.

[bdh21]

What's really amusing is he gives Duke a 9.81% chance of going unbeaten. Uh .... ?

... In conference play!

[JasonEvans]

If you look at the projected Duke schedule, it has us going 29-2, yet it still says we finish 28-3. I must be misreading or misunderstanding something.

The way it works is that he projects each game individually, but projects all the results as a group. So, like I was saying at the stat of the thread, if he projects Duke will beat UNC 60% of the time then he is going to list Duke as the winner every time they play. But, over the course of many games, that 60% is not going to hold up every time so he would likely project Duke to go 4-2 against UNC in a 6-game series.

Does that make more sense?

If you look only at each game as an individual event it is very different than looking at the whole season and projecting a record. It is sorta like how, in the preseason, folks were saying that Carolina would be favored in every single game they played and yet most of us felt the notion of them going undefeated was insane.

-Jason "hope that helps" Evans

[Bluedog]

Another thing I find weird is the breakdown of Player Stats. It has Singler as the only major contributor, Plumlee as a significant contributor, Ewill and Pocius as role players, Paulus under Limited Role, and McClure as Nearly Invisible. Huh? How it is possible that Plumlee is two rungs above Paulus and three above McClure? And Ewill and Pocius ahead of Paulus and McClure is also ridiculous. Although certainly McClure doesn't stuff the stat sheet (since great defense doesn't show up in stats), neither do Ewill, Plumlee, and Pocius. Paulus certainly has a larger role than 3 people ahead of him in this rung system, and arguably 4 (LT). And Olek and Johnson are missing completely from the list! ;)

It also has Duke's bench ranked 32nd in the country. Only behind Minnesota, NC St, Missouri, Georgia, TN, Miami FL, and K st among BCS teams. Btw, UNC is 144th.http://kenpom.com/height.php?s=BenchRank

Overall, though, I like the statistical analysis. Certainly very interesting and no bias since it's all based on numbers.

[pfrduke]

Another thing I find weird is the breakdown of Player Stats. It has Singler as the only major contributor, Plumlee as a significant contributor, Ewill and Pocius as role players, Paulus under Limited Role, and McClure as Nearly Invisible. Huh? How it is possible that Plumlee is two rungs above Paulus and three above McClure? And Ewill and Pocius ahead of Paulus and McClure is also ridiculous. Although certainly McClure doesn't stuff the stat sheet (since great defense doesn't show up in stats), neither do Ewill, Plumlee, and Pocius. Paulus certainly has a larger role than 3 people ahead of him in this rung system, and arguably 4 (LT). And Olek and Johnson are missing completely from the list! ;)

Overall, though, I like the statistical analysis. Certainly very interesting and no bias since it's all based on numbers.

The titles (Major Contributor, Significant Contributor, etc.) are based solely on the percentage of offensive possessions used by each player when on the court. Essentially, it's a measure of usage, and it's broken out in ranges. Anyone who uses 24% of possessions or above is labeled a Major Contributor, 20-24% is a Significant Contributor, 16-20% is a Role Player, 12-16% is Limited Role, and below 12% is Nearly Invisible. The titles refer only to the role the player plays in the offense while in the game, and have nothing to do with minutes played, defensive contribution, etc. And also, it captures positive and negative contribution equally - Plumlee's role as a Significant Contributor, for example, comes largely from the fact that he has committed a lot of turnovers in limited minutes, which ups his usage (usage comes from offensive rebounds, shots taken, free throws taken, assists, and turnovers).

I think it's not the best system of titles to use, because it causes confusion like yours (e.g., why is Plumlee a "Significant Contributor" while Paulus is just "Limited Role"?).

[JasonEvans]

Another thing I find weird is the breakdown of Player Stats. It has Singler as the only major contributor, Plumlee as a significant contributor, Ewill and Pocius as role players, Paulus under Limited Role, and McClure as Nearly Invisible. Huh? How it is possible that Plumlee is two rungs above Paulus and three above McClure? And Ewill and Pocius ahead of Paulus and McClure is also ridiculous. Although certainly McClure doesn't stuff the stat sheet (since great defense doesn't show up in stats), neither do Ewill, Plumlee, and Pocius. Paulus certainly has a larger role than 3 people ahead of him in this rung system, and arguably 4 (LT). And Olek and Johnson are missing completely from the list! ;)

It also has Duke's bench ranked 32nd in the country. Only behind Minnesota, NC St, Missouri, Georgia, TN, Miami FL, and K st among BCS teams. Btw, UNC is 144th.http://kenpom.com/height.php?s=BenchRank

Overall, though, I like the statistical analysis. Certainly very interesting and no bias since it's all based on numbers.

It is worth noting that, BASED ON NUMBERS, Duke's bench performance in greatly skewed by one aberrant game. That game was UNC-A when Smith, Scheyer, Henderson, Singler, and Zoubek all came off the bench. They combined to score 64 points in the game with 20 rebounds and 12 assists -- HUGE numbers from your bench. In fact, that is probably what the Duke bench ordinarily gives us in 3 or 4 games.

So, while it was just a one-game occurrence, that game may have severely thrown off our bench statistics.

--Jason "our bench's strength is its versatility, not how prolific it is" Evans

[pfrduke]

It is worth noting that, BASED ON NUMBERS, Duke's bench performance in greatly skewed by one aberrant game. That game was UNC-A when Smith, Scheyer, Henderson, Singler, and Zoubek all came off the bench. They combined to score 64 points in the game with 20 rebounds and 12 assists -- HUGE numbers from your bench. In fact, that is probably what the Duke bench ordinarily gives us in 3 or 4 games.

So, while it was just a one-game occurrence, that game may have severely thrown off our bench statistics.

--Jason "our bench's strength is its versatility, not how prolific it is" Evans

Actually, the "Bench" numbers are minutes played, not points contribution. And for Duke, it's not skewed by the UNCA game. Tracking who starts v. who comes off the bench every game for all 341 (or however many there are now) D-I teams is too tall a task. Pomeroy estimates, as described here:


Bench Minutes: Computed by assuming that the starters’ minutes are accounted for by the five players with most minutes played. Bench minutes are then assumed to be the minutes played by the remaining players, among those that play at least 10% of their teams minutes. This figure is divided by the starters’ minutes plus bench minutes.

Actually, for Duke, this inaccurately captures our average bench minutes, since his system treats Paulus (who plays the 5th-most minutes) and not Zoubek as a starter.

So in reality, we're probably a touch higher than 32nd.

[ArnieMc]

Interesting projections, to be sure, and so far Duke has been the most impressive, if by a slim margin over Wake and UNC. My concern is with the wacky unbalanced ACC schedule. Other posters will remember better than I how evenhanded over time this unbalanced thing balances out, but I gotta say, I don't remember Duke ever getting one of the easier schedules. Probably selective homer-memory on my part, so if anyone can enlighten me on recent history, do so.We have the easiest schedule this year and most years. This year, every other team has to play the best team in the country at least once, and some have to play them twice. We never have to play Duke.

[juise]

We have the easiest schedule this year and most years. This year, every other team has to play the best team in the country at least once, and some have to play them twice. We never have to play Duke.


Not so fast! No ACC team has ever really beat Duke... Duke only beats itself. So, in a sense, Duke plays itself every game.

(He was before my day, but I always appreciate a good Makhtar reference.)

[pfrduke]

Not so fast! No ACC team has ever really beat Duke... Duke only beats itself. So, in a sense, Duke plays itself every game.

(He was before my day, but I always appreciate a good Makhtar reference.)

"Ademola Okulaja beats himself"

(you got the wrong Heel - clearly before your day).

[rasputin]

Not so fast! No ACC team has ever really beat Duke... Duke only beats itself. So, in a sense, Duke plays itself every game.

(He was before my day, but I always appreciate a good Makhtar reference.)

Ademola Okulaja beats himself. (clap, clap) Ademola Okulaja beats himself.
(clap, clap) Ademola Okulaja, Ademola Okulaja, Ademola Okulaja beats himself. ...

Yikes; just saw that Okulaja is battling cancer. Best wishes AO.

[Chitowndevil]

I love Ken's site. My main criticism of his system is the probabilities it generates for games between top teams. They are too far from 50% in general. For example, no way do we have a 76% chance to beat UNC at Cameron. I wonder how much shrinkage there is in his estimates (that's Bayes, not George Costanza).

I also think the home/road spreads are excessive. For example we have a 76% chance to beat UNC at Cameron but only a 46% chance at Chapel Hill. This leads to rather silly situations like Pitt being 73% to beat UConn at home but only 63% to beat Villanova on the road. Home/road has been shown to be a strong predictive factor overall but I'm convinced it makes much less of a difference for the #1 vs #2 teams than say for the #100 vs #101 teams.

[juise]

"Ademola Okulaja beats himself"

(you got the wrong Heel - clearly before your day).

Doh! I knew it was one of the African-born heels (i.e. - I don't know how to spell their name without looking it up), but I thought that surely it was the more obnoxious one of the two. Thanks for the correction.

[bird]

I recently ordered tickets to Greensboro relying, at least in part, on the idea that Duke has a great shot at winning the regular season which should, IMO, guarantee us Greensboro, with the second spot going to either Wake or UNC if one of those two wins the ACC tourney.

[ACCBBallFan]

I love Ken's site. My main criticism of his system is the probabilities it generates for games between top teams. They are too far from 50% in general. For example, no way do we have a 76% chance to beat UNC at Cameron. I wonder how much shrinkage there is in his estimates (that's Bayes, not George Costanza).

I also think the home/road spreads are excessive. For example we have a 76% chance to beat UNC at Cameron but only a 46% chance at Chapel Hill. This leads to rather silly situations like Pitt being 73% to beat UConn at home but only 63% to beat Villanova on the road. Home/road has been shown to be a strong predictive factor overall but I'm convinced it makes much less of a difference for the #1 vs #2 teams than say for the #100 vs #101 teams. I agree with you.

Pomeroy is actually picking Duke as favorite in all but the last game at UNC, but the partials add up to a second ACC loss:

98% UVA
96% FSU and St. John's
95% Miami FL
91% BC and MD
89% VA Tech
86% WF home
76% UNC home
75% Clemson
63% WF road
47% UNC road

[dcarp23]

One interesting stat from Pomeroy's Duke page is the following:

Bench Minutes: 38.9% [32] 31.4%

Disclaimer: without knowing the specifics of the statistic, I assume that means that Duke's bench plays 38.9% of the game, 32nd most in the country.

[Wander]

One interesting stat from Pomeroy's Duke page is the following:

Bench Minutes: 38.9% [32] 31.4%

Disclaimer: without knowing the specifics of the statistic, I assume that means that Duke's bench plays 38.9% of the game, 32nd most in the country.

I also don't know the specifics, but there are two things that I imagine skew this stat:

1. Really good teams are more often in blowouts, so they're playing their bench at the end of games.
2. That one game where we started our 2nd string.

[dcarp23]

I also don't know the specifics, but there are two things that I imagine skew this stat:

1. Really good teams are more often in blowouts, so they're playing their bench at the end of games.
2. That one game where we started our 2nd string.

Really good point on number 2. Likely another big skewer of the stat is that BZ has a number of games were he ends up sitting a lot more than playing.

It appears, however, that either the blowouts don't affect stats all that much or that Duke blows teams out a lot more than others. BCS schools that rate ahead of Duke in bench minutes are limited to the following:

Minnesota, N.C. State, Missouri, Georgia, Tennessee, Miami and Kansas State.

[tbyers11]

One interesting stat from Pomeroy's Duke page is the following:

Bench Minutes: 38.9% [32] 31.4%

Disclaimer: without knowing the specifics of the statistic, I assume that means that Duke's bench plays 38.9% of the game, 32nd most in the country.

The specifics of this stat are covered in this post (http://www.dukebasketballreport.com/forums/showpost.php?p=248340&postcount=19) a little bit up this thread.

Short answer: The game where all the starters came off the bench doesn't affect it at all, but the fact that we have a bunch of blowouts might skew it slightly in our favor.

[dcarp23]

The specifics of this stat are covered in this post (http://www.dukebasketballreport.com/forums/showpost.php?p=248340&postcount=19) a little bit up this thread.

Short answer: The game where all the starters came off the bench doesn't affect it at all, but the fact that we have a bunch of blowouts might skew it slightly in our favor.

My bad...completely missed that series of posts

[SeattleIrish]

If a team is playing 5 games where he projects them with just a 60% chance of winning, his program will probably say they will go 4-2 in those games. Make sense?

The short answer is, "No, it doesn't make sense."

4-2 in 5 games is tough to sell, even for Ken Pom!

s.i.

[wisteria]

The projections from kenpom's site change when more games have been played. For example before wake's loss and our win over UMD, it was projected that we'll lose @wake. But now the projection is we will win.

[BlueintheFace]

...only 3 losses in this ACC season would have me pretty psyched. 4 would leave me a bit anxious about consistency and final four hopes, but still pretty positive considering my original expectations for the season. 5 and anything worse would be pretty bad. Anything better would be AWESOME!!

Just one man's opinion...