PDA

View Full Version : The Dork Polls 2015



dukelion
10-27-2014, 12:41 PM
I know this dosen't mean much but interesting nonetheless.


http://kenpom.com

hurleyfor3
10-27-2014, 12:47 PM
Ken probably noticed we care about dork polls more than any other fanbase and threw us a bone.

CDu
10-27-2014, 12:49 PM
This is sort of like Lunardi doing his Bracketology in May. Hard to take much stock in this without any actual game data to work with.

flyingdutchdevil
10-27-2014, 12:53 PM
What stats is Kenpom using? His top 20 is like a blue chip parade with Wichita St thrown in there.

Also, come on! Wichita St! Again? Sorry, but I'm probably the one dude in America who loves the BCS teams more than the mid-majors.

uh_no
10-27-2014, 01:04 PM
What stats is Kenpom using? His top 20 is like a blue chip parade with Wichita St thrown in there.

Also, come on! Wichita St! Again? Sorry, but I'm probably the one dude in America who loves the BCS teams more than the mid-majors.

he describes it here:

http://kenpom.com/blog/index.php/weblog/entry/pre_season_ratings_2014

I'm sure he tweaks it year to year, as it seems way too much of a difference on defense to NOT have taken incoming frosh into account.

Des Esseintes
10-27-2014, 01:10 PM
What stats is Kenpom using? His top 20 is like a blue chip parade with Wichita St thrown in there.

Also, come on! Wichita St! Again? Sorry, but I'm probably the one dude in America who loves the BCS teams more than the mid-majors.

Some explanation here (http://kenpom.com/blog/index.php/weblog/entry/a_look_back_on_preseason_ratings) and here (http://kenpom.com/blog/index.php/weblog/entry/pre_season_ratings_2014). He uses a combination of past five years' program performance, returning player performance, and the rankings of incoming recruits. Looks as though it works as well as any other system, including the AP poll.

ETA: uh no beat me to it. On D, my guess is that last year's defensive disaster is treated as an aberration from an otherwise very strong defensive profile. 2012 was also not great, but otherwise we are a good program defensively.

flyingdutchdevil
10-27-2014, 01:13 PM
Some explanation here (http://kenpom.com/blog/index.php/weblog/entry/a_look_back_on_preseason_ratings) and here (http://kenpom.com/blog/index.php/weblog/entry/pre_season_ratings_2014). He uses a combination of past five years' program performance, returning player performance, and the rankings of incoming recruits. Looks as though it works as well as any other system, including the AP poll.

Yeah, but the beauty of Kenpom is his use of actual stats rather than relying on bunch of hunches. I understand his methodology, but I don't really understand the necessity to have a "pre-season" ranking based on hardcore metrics/actual stats. Just my two cents.

Bay Area Duke Fan
10-27-2014, 01:16 PM
Interesting that he ranks unc 20th; much lower than others have ranked the heels. Maybe he deducted points for ethics standards.

subzero02
10-27-2014, 01:18 PM
This early nod of respect is likely due to a phenomenon loosely referred to as the OIE... Or the Okafor Intimidation Effect. Although not entirely understood or even universally accepted, particularly in regions that practice equine worship, many experts have already accepted the OIE as law.

devildeac
10-27-2014, 01:20 PM
Interesting that he ranks unc 20th; much lower than others have ranked the heels. Maybe he deducted points for ethics standards.

Not to worry. They can get extra points for submitting a paper.:rolleyes:

Kfanarmy
10-27-2014, 01:40 PM
Not to worry. They can get extra points for submitting a paper.:rolleyes:

Actually they get the extra points by having someone else say they submitted a paper.

sagegrouse
10-27-2014, 01:58 PM
Yeah, but the beauty of Kenpom is his use of actual stats rather than relying on bunch of hunches. I understand his methodology, but I don't really understand the necessity to have a "pre-season" ranking based on hardcore metrics/actual stats. Just my two cents.

It's a necessary business decision. His "in-season" analysis is widely admired and cited. Unhappily, it's not useful until about January. But he has paying subscribers plus followers in the press and all, so he has to do something. So, he has a stopgap pre-season analysis and then he uses a hybrid model (the pre-season expectations and actual performance) until he can go "full production," when he has enough data to rely solely on his model (or combination of models).

MChambers
10-27-2014, 02:10 PM
It's a necessary business decision. His "in-season" analysis is widely admired and cited. Unhappily, it's not useful until about January. But he has paying subscribers plus followers in the press and all, so he has to do something. So, he has a stopgap pre-season analysis and then he uses a hybrid model (the pre-season expectations and actual performance) until he can go "full production," when he has enough data to rely solely on his model (or combination of models).
I think he's doing it to see if he can do a better job than the preseason polls. It is a laudable goal, although, like you, I'm not at all convinced that it's worth very much.

mr. synellinden
10-27-2014, 08:36 PM
*First, I wonder if we should rename this thread 2014-2015 Dork Poll Tracking.

ESPN has Duke at #4 in its preseason Power Ranking (http://espn.go.com/espn/feature/story/_/id/11754505/college-basketball-top-25-power-rankings).

Excerpt:

Early speculation held that Cook would be the odd man out of the starting lineup with the arrival of freshman point guard Tyus Jones. But Cook used his summer wisely too, improving and showing consistency. Now, don't be surprised if Jones and Cook both start in the backcourt.

Freshman Justise Winslow will challenge junior Rasheed Sulaimon, who played some point guard last season, at small forward.

-jk
10-28-2014, 09:14 AM
*First, I wonder if we should rename this thread 2014-2015 Dork Poll Tracking.

ESPN has Duke at #4 in its preseason Power Ranking (http://espn.go.com/espn/feature/story/_/id/11754505/college-basketball-top-25-power-rankings).

Excerpt:

Early speculation held that Cook would be the odd man out of the starting lineup with the arrival of freshman point guard Tyus Jones. But Cook used his summer wisely too, improving and showing consistency. Now, don't be surprised if Jones and Cook both start in the backcourt.

Freshman Justise Winslow will challenge junior Rasheed Sulaimon, who played some point guard last season, at small forward.

That has to be the worst presentation of a poll I've ever seen...

-jk

COYS
10-28-2014, 09:35 AM
That has to be the worst presentation of a poll I've ever seen...

-jk

That's what I was thinking, although I don't necessarily disagree with ranking Duke at 4.

At any rate, I hope KenPom is prescient.

Troublemaker
10-28-2014, 11:25 AM
Kenpom #1, but only projected to go 25-5 (Overall) / 14-4 (ACC) on the team page.

We have a tough schedule this year apparently. Let's hope we're not too far off from #1. If we're #8, we might end up with 7 losses or more.

uh_no
10-28-2014, 11:31 AM
Kenpom #1, but only projected to go 25-5 (Overall) / 14-4 (ACC) on the team page.

We have a tough schedule this year apparently. Let's hope we're not too far off from #1. If we're #8, we might end up with 7 losses or more.

that was our record in 2010, if i recall :)

AncientPsychicT
10-28-2014, 11:34 AM
that was our record in 2010, if i recall :)

26-5, 13-3 (http://en.wikipedia.org/wiki/2009%E2%80%9310_Duke_Blue_Devils_men's_basketball_ team#Schedule)

Close, but not quite. Notably, we only played 16 conference games that year, not 18.

MChambers
10-28-2014, 02:22 PM
Some good stuff here:

http://kenpom.com/blog/index.php/weblog/entry/preseason_ratings_2015

"The system does not give any special consideration to new players entering the program. There is some credit given for high-profile recruits, but the poor performances in 2012-13 of UCLA and Kentucky, among others, in recent years have tended to mute the impact of recruits in the model. Recruiting rankings are useful, but the impact of high-level prospects on their respective teams as freshman can vary wildly."

Makes me even more optimistic

uh_no
10-28-2014, 02:37 PM
Some good stuff here:

http://kenpom.com/blog/index.php/weblog/entry/preseason_ratings_2015

"The system does not give any special consideration to new players entering the program. There is some credit given for high-profile recruits, but the poor performances in 2012-13 of UCLA and Kentucky, among others, in recent years have tended to mute the impact of recruits in the model. Recruiting rankings are useful, but the impact of high-level prospects on their respective teams as freshman can vary wildly."

Makes me even more optimistic

IMO, duke is one of the teams positively affected by "high-profile recruits"....otherwise we're not returning enough to justify our ranking, even with a relatively high baseline. if okafor and perhaps tyus are not "high-profile" not sure who would be. Previously he hinted at having used just a top 100 ranking list....so wouldn't doubt he just used that again.

Philadukie
10-29-2014, 10:07 PM
Is there historical data on how his pre-season rankings compare to season end rankings?

Dukehky
10-31-2014, 02:35 PM
What stats is Kenpom using? His top 20 is like a blue chip parade with Wichita St thrown in there.

Also, come on! Wichita St! Again? Sorry, but I'm probably the one dude in America who loves the BCS teams more than the mid-majors.

Right there with ya boss. If you wanna play against and beat the big boys move conferences like TCU did.

jipops
10-31-2014, 02:51 PM
Right there with ya boss. If you wanna play against and beat the big boys move conferences like TCU did.

Somebody should tell that to Lehigh and Mercer.

Kedsy
10-31-2014, 03:39 PM
Somebody should tell that to Lehigh and Mercer.

Sure, because Duke's the only team, ever, in the history of the NCAA tournament to lose to a low-major team.

For what it's worth, Duke was the only ranked team Mercer played last season and Lehigh played just one other ranked team besides Duke. I'm fairly confident that sort of schedule wasn't what Dukehky was talking about.

sagegrouse
10-31-2014, 03:48 PM
It's a necessary business decision. His "in-season" analysis is widely admired and cited. Unhappily, it's not useful until about January. But he has paying subscribers plus followers in the press and all, so he has to do something. So, he has a stopgap pre-season analysis and then he uses a hybrid model (the pre-season expectations and actual performance) until he can go "full production," when he has enough data to rely solely on his model (or combination of models).


I think he's doing it to see if he can do a better job than the preseason polls. It is a laudable goal, although, like you, I'm not at all convinced that it's worth very much.

Would you accept, MC, that it is possible that both can simultaneously occur: (a) he decides that he must modify his methodology so he can sell subscriptions from the beginning of the season and (b) he decides to do the best job he can with the early-season projections?

MChambers
10-31-2014, 03:57 PM
Would you accept, MC, that it is possible that both can simultaneously occur: (a) he decides that he must modify his methodology so he can sell subscriptions from the beginning of the season and (b) he decides to do the best job he can with the early-season projections?
I can accept that duality.

bob blue devil
11-15-2014, 08:17 AM
duke's 69 point victory over presbyterian apparently not as impressive as louisville's 13 point win over minnesota. things that make you go 'hmmmm'.

does anybody know what our pythagorean rating was prior to the game? i wonder how much the game moved it (now is 0.9470).

vick
11-15-2014, 08:28 AM
duke's 69 point victory over presbyterian apparently not as impressive as louisville's 13 point win over minnesota. things that make you go 'hmmmm'.

does anybody know what our pythagorean rating was prior to the game? i wonder how much the game moved it (now is 0.9470).

Kenpom reduces the effect of huge victory margins, a change made before last season (http://kenpom.com/blog/index.php/weblog/entry/pomeroy_ratings_version_2.0). So there isn't really any way for Duke to help its rating in a game against a very low-ranked team.

uh_no
11-15-2014, 09:52 AM
duke's 69 point victory over presbyterian apparently not as impressive as louisville's 13 point win over minnesota. things that make you go 'hmmmm'.

does anybody know what our pythagorean rating was prior to the game? i wonder how much the game moved it (now is 0.9470).

9466. and it bumped back down to 9469. most of the bump is because of offense, which went from 118.4 to .5

I wouldn't be concerned about 1 part in 10,000 separating us and UL.

bob blue devil
11-16-2014, 09:30 AM
take that louisville! our 50 point beat down of fairfield was far more impressive than your night off. 0.9484 and counting.

going back to '02, ken pom's end of year #1 has never been below 0.95. our preseason 0.9466, if translated to end of year, would have us finishing #4, #3, #4, #2, #3, #4, #5, #5, #2, #3, #5, #3, #5 going backwards (last year to '02). naturally, it would be hard for the highest rated team in the country by the end of the year to not have outperformed expectations, but it's a nice reminder that championships aren't handed out at the start of the season.

Troublemaker
11-16-2014, 11:11 AM
I wonder if that one day Louisville spent atop the kenpom rankings is going to prevent Duke from going wire to wire #1.

davekay1971
11-16-2014, 11:14 AM
I'm eagerly awaiting the banner hanging in Cameraon: National Champions, November 2014, Kenpom Rankings

Oh wait, we don't do that. That's the "school" 8 miles down the road.

bob blue devil
11-16-2014, 11:42 AM
I'm eagerly awaiting the banner hanging in Cameraon: National Champions, November 2014, Kenpom Rankings

Oh wait, we don't do that. That's the "school" 8 miles down the road.

why do the heels need to be #1 in kenpom to hang a banner? i would have guessed they've identified a nice spot to hang a november 2014 kenpom top 25 banner by now. but that does raise the question over whether ken would require them to vacate it if their players are found to have been ineligible...

barjwr
11-16-2014, 05:52 PM
If they had any academic rigor over there at all, they'd have already calculated HelmsPom ratings from the 1920s and 1930s for even more banners.

kAzE
11-19-2014, 11:59 AM
After their complete and total annihilation of Kansas, UK is now ranked the #2 offense (behind us), and #1 defense, bumping them to the #1 overall ranking.

Kedsy
11-19-2014, 12:36 PM
After their complete and total annihilation of Kansas, UK is now ranked the #2 offense (behind us), and #1 defense, bumping them to the #1 overall ranking.

Are we really doing this in mid-November? KenPom doesn't really mean too much until January or so.

kAzE
11-19-2014, 12:59 PM
Are we really doing this in mid-November? KenPom doesn't really mean too much until January or so.

It's just as relevant as the preseason KenPom rankings that started this thread. At least we have SOME data now.

hurleyfor3
11-19-2014, 01:23 PM
Are we really doing this in mid-November? KenPom doesn't really mean too much until January or so.

I gave up on this battle. You can continue it if you want.

uh_no
11-19-2014, 01:32 PM
Are we really doing this in mid-November? KenPom doesn't really mean too much until January or so.

I'm caching kenpom on a daily basis this year with the goal of analyzing what the standard deviation from the end of the year a ranking for a team is on a given date during the season :)

Kedsy
11-19-2014, 02:04 PM
I'm caching kenpom on a daily basis this year with the goal of analyzing what the standard deviation from the end of the year a ranking for a team is on a given date during the season :)

I don't know if you're serious, but that's a cool idea.

hurleyfor3
11-19-2014, 02:11 PM
I'm caching kenpom on a daily basis this year with the goal of analyzing what the standard deviation from the end of the year a ranking for a team is on a given date during the season :)

Pomeroy did midday updates last year at least during the conference tournaments. You might have to be reloading more often.

uh_no
11-19-2014, 02:17 PM
I don't know if you're serious, but that's a cool idea.

yes i'm serious :)

[kevin@kevin-fedora kenpom]$ ls
nov18 nov19
[kevin@kevin-fedora kenpom]$ cat nov18 | grep Duke
<tr><td>1</td><td style="text-align:left;"><a href="team.php?team=Duke">Duke</a> <span class="seed"></span></td><td><a href="conf.php?c=ACC">ACC</a></td><td>2-0</td><td>.9483</td><td class="divide">118.6</td><td><span class="seed">1</span></td><td>92.1</td><td><span class="seed">7</span></td><td class="divide">70.4</td><td><span class="seed">78</span></td><td class="divide">+.000</td><td><span class="seed">54</span></td><td class="divide">.1757</td><td><span class="seed">276</span></td><td>94.9</td><td><span class="seed">273</span></td><td>108.5</td><td><span class="seed">278</span></td><td class="divide">.1757</td><td><span class="seed">276</span></td></tr>
[kevin@kevin-fedora kenpom]$

uh_no
11-19-2014, 02:18 PM
Pomeroy did midday updates last year at least during the conference tournaments. You might have to be reloading more often.

he is doing them most days now. The frequency of the datapoints sholuldn't matter though...only the difference between the final ranking and the ranking on a given date.

I'm usually taking them at like 9:30 am.

ChillinDuke
11-19-2014, 02:52 PM
he is doing them most days now. The frequency of the datapoints sholuldn't matter though...only the difference between the final ranking and the ranking on a given date.

I'm usually taking them at like 9:30 am.

Hats off. I like this.

- Chillin

grad_devil
11-19-2014, 02:54 PM
[kevin@kevin-fedora kenpom]$ cat nov18 | grep Duke

Funny. I often use cat | grep instead of

grep Duke nov18
Maybe it's a Duke thing :)

bob blue devil
11-19-2014, 03:00 PM
I'm caching kenpom on a daily basis this year with the goal of analyzing what the standard deviation from the end of the year a ranking for a team is on a given date during the season :)

Sounds like fun. Tricky part would be how to interpret the data given that none of the observations are independent. You could maybe do some momentum analysis too.

uh_no
11-19-2014, 03:51 PM
Sounds like fun. Tricky part would be how to interpret the data given that none of the observations are independent. You could maybe do some momentum analysis too.

well the the question is, "how much of a predictor of final ranking is ranking on day X," and that should be independent of the system which generates the ranking on day X.

KenPom talks in his blog all the time about people who move around and what bad predictions are before the graph is "connected"...but what i'm interested in is say, if we are rated X on day Y, what is the % chance we move as high as Z by the end of the year. I'm sure kenpom runs those kind of calculations all the time....but it's also nice to record it to look at our movement over the course of the year.

Kfanarmy
11-19-2014, 04:07 PM
well the the question is, "how much of a predictor of final ranking is ranking on day X," and that should be independent of the system which generates the ranking on day X.

KenPom talks in his blog all the time about people who move around and what bad predictions are before the graph is "connected"...but what i'm interested in is say, if we are rated X on day Y, what is the % chance we move as high as Z by the end of the year. I'm sure kenpom runs those kind of calculations all the time....but it's also nice to record it to look at our movement over the course of the year. In this context, what is "Z?"

uh_no
11-19-2014, 04:40 PM
In this context, what is "Z?"

whatever you want it to be....

for instance "duke is ranked 7 on nov 28, what are the chances they get to #1 by end of year?"

bob blue devil
11-19-2014, 06:56 PM
well the the question is, "how much of a predictor of final ranking is ranking on day X," and that should be independent of the system which generates the ranking on day X.

KenPom talks in his blog all the time about people who move around and what bad predictions are before the graph is "connected"...but what i'm interested in is say, if we are rated X on day Y, what is the % chance we move as high as Z by the end of the year. I'm sure kenpom runs those kind of calculations all the time....but it's also nice to record it to look at our movement over the course of the year.

and that's great if that relationship interests you - it is fairly interesting to me as well. i'm excited to see what you come up with. however, i'd have a little trouble extrapolating a meaning from that since the final ranking is partially defined by the data that defines the observation (i'm guessing you get this). the value, i guess, is doing things like predicting our odds at a kenpom championship or top 5 finish, etc. but i can't think of a direct way to tie that information to saying the team is going to perform as a top 5 kenpom team during the remainder of the season (because we don't have a ranking that measures performance and excludes the information in the observation). if i remember right, kenpom emphasizes more recent results, so perhaps that mitigates the challenge by making the end of year ranking effectively uninfluenced by the data used in the early season rankings.

Tappan Zee Devil
11-19-2014, 08:28 PM
I'm caching kenpom on a daily basis this year with the goal of analyzing what the standard deviation from the end of the year a ranking for a team is on a given date during the season :)

Are you assuming then that a team's final ranking is where they "should" have been at the beginning?
But doesn't this exercise assume that teams don't change (grow or fall apart) during the year.

uh_no
11-20-2014, 01:21 AM
Are you assuming then that a team's final ranking is where they "should" have been at the beginning?
But doesn't this exercise assume that teams don't change (grow or fall apart) during the year.

oh, i think that's equally as interesting....obvioulsy in a perfect system a team would start from A and travel linearly to B (in terms of their rating and ranking....but that's obviously not true...

the point is that kenpom has proven to be a very good predictor of tournament success...certainly not perfect, as teams like uconn showed last year....but pretty darn good. All the time in november/december/january, we see exchanges like "we can't win because we're only kenpom ranking XYZ"...."you can't trust kenpom because it's not fully connected!!!"....i'm interested in knowing how much can you trust kenpom? obviously the answer is not 0, and obviously it's not 100%...kenpom indicates sometime in january is when his ratings start to have meaning....but we also know that his preseason rankings are generally decently accurate....i'm interested in how the accuracy in predicting the final rankings changes over the course of the year....analyzing the cause of why they deviated from their initial ranking is somewhat orthoganal.

grad_devil
11-20-2014, 07:03 AM
....analyzing the cause of why they deviated from their initial ranking is somewhat orthoganal .

Anybody that uses the word orthogonal in an Internet forum is an immediate friend of mine. I'd spork you, but, alas, I must first spread the love.

Skitzle
11-20-2014, 07:10 AM
yes i'm serious :)

[kevin@kevin-fedora kenpom]$ ls
nov18 nov19
[kevin@kevin-fedora kenpom]$ cat nov18 | grep Duke
<tr><td>1</td><td style="text-align:left;"><a href="team.php?team=Duke">Duke</a> <span class="seed"></span></td><td><a href="conf.php?c=ACC">ACC</a></td><td>2-0</td><td>.9483</td><td class="divide">118.6</td><td><span class="seed">1</span></td><td>92.1</td><td><span class="seed">7</span></td><td class="divide">70.4</td><td><span class="seed">78</span></td><td class="divide">+.000</td><td><span class="seed">54</span></td><td class="divide">.1757</td><td><span class="seed">276</span></td><td>94.9</td><td><span class="seed">273</span></td><td>108.5</td><td><span class="seed">278</span></td><td class="divide">.1757</td><td><span class="seed">276</span></td></tr>
[kevin@kevin-fedora kenpom]$


Two quick notes.

1) I highly doubt that you would ever see this post on ANY other college basketball message board
2) I doubt even more highly that there would be anyone on that board who would read the post, understand the post, and comment on the post.

I love this. Sporks for you and grad_devil for his comment!

MChambers
11-21-2014, 01:05 PM
I'm following Northwestern this year, since my daughter is a freshman there, and since Collins is coach.

Right now, in Pomeroy, the Wildcats are #92. Right below Northwestern is, wait for it, Northeastern. Kind of cool.

Edouble
11-21-2014, 06:53 PM
Anybody that uses the word orthogonal in an Internet forum is an immediate friend of mine. I'd spork you, but, alas, I must first spread the love.

I slouch too much at my desk, so I've been trying to keep my hips and spine in an orthopedically correct, orthogonal positioning.

(I will take my sporks now, thank you :o )

-jk
11-21-2014, 07:37 PM
I slouch too much at my desk, so I've been trying to keep my hips and spine in an orthopedically correct, orthogonal positioning.

(I will take my sporks now, thank you :o )

Where've you been, greybeard?

-jk

uh_no
12-21-2014, 08:14 PM
wanted to point out that after UVA's thwomping of harvard, they jumped us in kenpom...dumping us to #3

their best win is a decent maryland team, but they also have wins over VCU and harvard....there's a very good chance IMO they come into the duke game undefeated....about 47%....duke only has an 11% chance of getting their undefeated

MCFinARL
12-22-2014, 09:25 AM
wanted to point out that after UVA's thwomping of harvard, they jumped us in kenpom...dumping us to #3

their best win is a decent maryland team, but they also have wins over VCU and harvard....there's a very good chance IMO they come into the duke game undefeated....about 47%....duke only has an 11% chance of getting their undefeated

Well, sure, because they don't play Louisville until after that game and Duke plays Louisville before that game. Don't get me wrong, I think Virginia looks formidable, and I suppose it might give them even more of psychological edge if they come into the game undefeated, especially since it is a home game for them. On the other hand, it could take a bit of pressure off of Duke to come in as the underdog. But either way, in the abstract I'm not sure how important it will be that Virginia might be undefeated then and Duke might not.

gumbomoop
01-07-2015, 12:05 PM
Questions for Kenpom experts --

(1) It appears that generally the greater the difference between a team's AdjO and AdjD, the higher that team's rating in Kenpom. Yes?

(2) If generally yes, I nevertheless notice that it doesn't hold precisely. For here are the numbers for top 4 as of this morn:

UK 115.9 (O) - 84.0 (D) = 31.9
UVa 117.8 - 86.3 = 31.5
Duke 122.0 - 90.0 = 32.0
Wisc 119.0 - 89.7 = 29.3

(3) I also noticed that the gap between the #1 AdjO (Duke) and #2 (ND) is considerably larger (1.9) than is the difference (0.2) between the #1 AdjD (UK) and #2 (UL). Ditto for gap between Duke and AdjO #3 Wisc (3.0) compared to gap between UK and AdjD #3 Okla (1.6). Are these difference-differences statistically significant? And if so, how so? For example, does it mean that so far, Kenpom-wise, Duke's O is even more impressive than UK's D?

Kedsy
01-07-2015, 12:35 PM
Questions for Kenpom experts --

(1) It appears that generally the greater the difference between a team's AdjO and AdjD, the higher that team's rating in Kenpom. Yes?

Pomeroy uses a "pythagorean model" to determine his ratings, the basis of which looks something like this:

Expected winning pct = ((points scored)^2) / (((points scored)^2) + ((points allowed)^2))

which can also be expressed as:

1 / (1 + (points allowed / points scored)^2)

Except instead of an exponent of 2, he uses a different exponent (I think 10.25). He also makes adjustments for home/away and for difficulty of schedule, and I think there's some sort of diminishing returns principal, and who knows what else.

So, since I'm neither a mathematician nor a statistician I can't say for certain, my guess is his complicated rating formula doesn't always come out to an equivalent of oRtg - dRtg.

uh_no
01-07-2015, 12:53 PM
Pomeroy uses a "pythagorean model" to determine his ratings, the basis of which looks something like this:

Expected winning pct = ((points scored)^2) / (((points scored)^2) + ((points allowed)^2))

which can also be expressed as:

1 / (1 + (points allowed / points scored)^2)

Except instead of an exponent of 2, he uses a different exponent (I think 10.25). He also makes adjustments for home/away and for difficulty of schedule, and I think there's some sort of diminishing returns principal, and who knows what else.

So, since I'm neither a mathematician nor a statistician I can't say for certain, my guess is his complicated rating formula doesn't always come out to an equivalent of oRtg - dRtg.

That happens when he calculates efficencies. The pyth should be a direct calculation from those.

Answering the questions of diminishing returns: This makes sense given that offense is a positive stat, and defense a negative one.

If an offense puts up 100, and another offense is "twice" as good, they'll put up 200. However, if a defense is 100 and another defense is "twice" as good they'll put up a 50. Of course those numbers aren't realistic, but demonstrate how since "good defense" is bound at zero, regardless of how much better you get, you will see less and less difference in an opponents performance.

Olympic Fan
01-07-2015, 01:38 PM
Kentucky's narrow escape at home against Ole Miss had an interesting impact on Pomeroy.

Before the game, he gave Kentucky almost a 30 percent chance to finish the record season undefeated.

This morning, their chances were listed at 17.7 percent.

Still remarkably high for this time of year, but a 12-point drop after one close call is pretty remarkable.

Henderson
01-07-2015, 02:26 PM
Kentucky's narrow escape at home against Ole Miss had an interesting impact on Pomeroy.

Before the game, he gave Kentucky almost a 30 percent chance to finish the record season undefeated.

This morning, their chances were listed at 17.7 percent.

Still remarkably high for this time of year, but a 12-point drop after one close call is pretty remarkable.

Interesting and not surprising given the OT thingy. But we don't need Pom to tell us that the Ol' Miss game indicates that KY could lose. KY showed us all in that game. Still, the numbers are interesting to follow.

Kedsy
01-07-2015, 04:43 PM
That happens when he calculates efficencies. The pyth should be a direct calculation from those.

Yes, that makes sense. But when I try to reproduce his numbers, it doesn't come out just right, so either I don't entirely understand the calculation or there's more to it than just the two efficiencies.


(1) It appears that generally the greater the difference between a team's AdjO and AdjD, the higher that team's rating in Kenpom. Yes?

Upon further reflection, the answer to this question is no. It's not the difference (i.e., not AdjO - AdjD), it's the quotient (i.e., AdjO/AdjD) that should predict the order for KenPom's overall rankings.

Des Esseintes
01-07-2015, 05:11 PM
Yes, that makes sense. But when I try to reproduce his numbers, it doesn't come out just right, so either I don't entirely understand the calculation or there's more to it than just the two efficiencies.



Upon further reflection, the answer to this question is no. It's not the difference (i.e., not AdjO - AdjD), it's the quotient (i.e., AdjO/AdjD) that should predict the order for KenPom's overall rankings.

Right, because beating a team 117.8 to 86.3 (UVA) is winning by a greater percentage (36.50%) than beating them 122 to 90 (Duke, 35.56%), even if the raw point differential is more in the second game. The difference is tiny but enough to put UVA ahead of us for the time being.

vick
01-07-2015, 05:38 PM
Yes, that makes sense. But when I try to reproduce his numbers, it doesn't come out just right, so either I don't entirely understand the calculation or there's more to it than just the two efficiencies.



Upon further reflection, the answer to this question is no. It's not the difference (i.e., not AdjO - AdjD), it's the quotient (i.e., AdjO/AdjD) that should predict the order for KenPom's overall rankings.

Use exponent 11.5, per this update (http://kenpom.com/blog/index.php/weblog/entry/pomeroy_ratings_version_2.0) (yes it contradicts elsewhere in the site).

Here's an example of how I think you get the predicted score and odds for any two teams using Duke @ Wake as example (note there can occasionally be rounding errors).

Duke O: 122.0
Duke D: 90.0
Duke Pace: 68.0
Wake O: 99.5
Wake D: 95.0
Wake Pace: 69.5
Average O and D: 100.3
Average Pace: 65.9
Home court effect: 1.4%

Predicted Duke efficiency: 100.3*((98.6%*122.0)/100.3)*((98.6%*95.0)/100.3) = 112.3
Predicted Wake efficiency: 100.3*((101.4%*99.5)/100.3)*((101.4*90.0)/100.3) = 91.8

So Duke winning percentage = (112.3^11.5)/((112.3^11.5)+(91.8^11.5)) = 91%

Predicted pace = 65.9*(68.0/65.9)*(69.5/65.9) = 71.7

Predicted Duke score = (112.3/100)*71.7 = 81
Predicted Wake score = (91.8/100)*71.7 = 66

Hope I haven't miswritten a number, but I think this is the math.

Kedsy
01-07-2015, 05:45 PM
Use exponent 11.5, per this update (http://kenpom.com/blog/index.php/weblog/entry/pomeroy_ratings_version_2.0) (yes it contradicts elsewhere in the site).

Thanks, 11.5 seems to work. So the "pythagorean" rating is entirely derived from AdjO and AdjD.

AdjO and AdjD are derived using game-by-game results, including home court factors, opponent strength, possibly diminishing returns if the score is run up too high, and perhaps other factors.

uh_no
01-07-2015, 05:54 PM
Thanks, 11.5 seems to work. So the "pythagorean" rating is entirely derived from AdjO and AdjD.

AdjO and AdjD are derived using game-by-game results, including home court factors, opponent strength, possibly diminishing returns if the score is run up too high, and perhaps other factors.

If I recall, the exponent is determined by what number gives best predictive value to the model...and he likely has to adjust it as he a) has more years worth of data, and b) tweaks the efficiency, most recently likely due to the changes to blowout calculations

uh_no
01-10-2015, 09:35 PM
couple notes after todays games, including yet ANOTHER subpar performance from kentucky...

kentucky's offense has dropped out of the top 10, to 12.
Duke's defense has risen into the top 10...at 10

That leaves duke and virginia as the only two teams whose rankings are in the top 10 for both offense and defense.

kentucy is only .0022 ahead of UVA overall....continuing to perform poorly against the mediocre SEC will cause UK to fall out of the top spot, especially if duke and UVA perform decently against their much tougher competition (regardless of w/l)

With UVA, duke, UL, UNC, and ND, the ACC has 5 teams in the top 15, easily the most of any conference.

the rest of the rundown is:
big 12: 3
big 10: 2
pac 10: 2
sec: 1
BE: 1
gonzaga: 1

bedeviled
02-26-2015, 08:21 PM
This will be the thread to re-emphasize that the past National Champions have NOT been top 20 in offense and defense prior to the tournament.

For now, I would like some help in understanding his rankings. To me, it appears like his calculations put too much weight on the "who did you lose to" argument with much less effect of "who can you beat." While this certainly helps rank who are the most consistent teams during the year, it doesn't do a marvelous job of ranking according to my conceptualization of the 'best team.'

As Andre Buckner Fan stated in the VaTech post-game thread (http://forums.dukebasketballreport.com/forums/showthread.php?35395-MBB-Duke-91-VPI-86-(OT)-Postgame-Thread&p=784090#post784090), "Whenever Duke is playing a team we're not playing their record; we're playing their pure potential." This is a theme that, theoretically, becomes applicable to everyone at tournament time - when all teams are motivated to perform at their peak for every game. In that case, I don't want to know who was able to prevent a loss to a sub-100 team; I want to know which team can beat the others. Yes, lapses occur and teams lose games that they shouldn't, but that's not how I think of "best team." Looking at Pomeroy's rankings this year, I think I'm finally cured of putting so much emphasis on kenpom when filling out my brackets!

Here is the most obvious example that caught my eye:



Utah
Duke


Record
21-5
25-3


Conference record
11-3
12-3







Kenpom rank
7
8


Kenpom SOS
57
22


vs Kpom 1-10
0-1
2-0


vs Kpom 11-25
1-1
4-1


vs Kpom 26-50
3-2
2-1


vs Kpom 51-100
3-1
9-1


vs Kpom sub-100
14-0
8-0



In what world does Utah rank ahead of Duke when Duke has the harder schedule (per kenpom) and better record? The only thing I can think is that Utah was better at winning the games they were expected to win (ie Duke's losses were worse. We lost to kenpom 20,33,62 while they lost to 3,11,26,42,52).

Is that right? Or is it some black-box interconnected calculation that can't really be explained? I can't think of another explanation. Most irritating is that we weren't rewarded sufficiently for our high-end wins. BUT, high-end potential is what I'm most excited about in March/April.

For the curious, here's the same info using RPI stats instead of kenpom



Utah
Duke


RPI rank
10
5


RPI SOS
39
11


vs RPI 1-10
0-2
2-0


vs RPI 11-25
1-1
2-0


vs RPI 26-50
1-2
5-2


vs RPI 51-100
6-0
6-1


vs RPI sub-100
13-0
10-0

MarkD83
02-26-2015, 08:26 PM
This means if Duke can get past the first round team that is ranked between 51-100 Duke is in good shape.

uh_no
02-26-2015, 08:36 PM
This will be the thread to re-emphasize that the past National Champions have NOT been top 20 in offense and defense prior to the tournament.

For now, I would like some help in understanding his rankings. To me, it appears like his calculations put too much weight on the "who did you lose to" argument with much less effect of "who can you beat." While this certainly helps rank who are the most consistent teams during the year, it doesn't do a marvelous job of ranking according to my conceptualization of the 'best team.'

This demonstrates to me you have a fundamental misunderstanding of how his rankings work. For starters, wins and losses play ZERO part in determining the rankings.


Looking at Pomeroy's rankings this year, I think I'm finally cured of putting so much emphasis on kenpom when filling out my brackets!
Welp, kenpom ratings consistently perform better than other rating methods at predicting winners in the tournament...so do so at your own risk....


In what world does Utah rank ahead of Duke when Duke has the harder schedule (per kenpom) and better record? The only thing I can think is that Utah was better at winning the games they were expected to win (ie Duke's losses were worse. We lost to kenpom 20,33,62 while they lost to 3,11,26,42,52).

You make it sound like kenpom is doing making a subjective analysis. He's only examining their efficiencies...nothing else. As I mentioned, there is absolutely no weight on wins and loses. literally absolutely none. We could have lost by one to UNC the other week and there would be virtually no effect on the rankings.



Is that right? nope


Or is it some black-box interconnected calculation that can't really be explained?
is it some interconnected calculation? yep
Is it a black box that can't be explained? nope I had a post where I laid it all out...trying to find it...EDIT. here it is! http://forums.dukebasketballreport.com/forums/showthread.php?34579-Check-out-how-not-obsessed-we-are-with-Kentucky&p=765680#post765680

Read around his site...try to understand whats going on before you rip it.

Wander
02-26-2015, 08:52 PM
"Whenever Duke is playing a team we're not playing their record; we're playing their pure potential."

So, Notre Dame's pure potential is to lose by a million?

JNort
02-26-2015, 09:29 PM
Read around his site...try to understand whats going on before you rip it.

Well since most people don't have unlimited time to look through the threads and read all the posts I don't see anything wrong with him having a discussion on it.

Not sure if I'm just doing somthing wrong but this site has the worst search function to find what you are looking for. Always comes up with 0 results even if what I search for is on the 1st page

uh_no
02-26-2015, 09:38 PM
Well since most people don't have unlimited time to look through the threads and read all the posts

I don't have a problem with not knowing every constant and equation that goes into kenpom....but I do take issue with people who write with enough confidence as to rip a model while having so little understanding of it.

the existing thread is also http://forums.dukebasketballreport.com/forums/showthread.php?34409-The-Dork-Polls-2015/ should probably be merged methinks.

Wahoo2000
02-26-2015, 09:42 PM
Well since most people don't have unlimited time to look through the threads and read all the posts I don't see anything wrong with him having a discussion on it.

Not sure if I'm just doing somthing wrong but this site has the worst search function to find what you are looking for. Always comes up with 0 results even if what I search for is on the 1st page

He obviously put a ton of time into writing that post. If he had invested that time instead into reading the Kenpom blog, or researching HOW Kenpom works, his issues would've been answered.

Kenpom isn't a perfect tool, but it's the best one (as far as statistical analysis) we have to predict results of college bball contests. Sure, some shmuck is going to beat it in a tourney pool every year. But MOST won't. That's the point.

Kenpom's biggest failings are the inability to factor in injuries/suspensions/etc in predicting upcoming results, and the inability to discern that some teams can "take a night off" here and there but really flip the switch when it matters (in bigger games and come post-season time). That second one seems to apply to you guys this year. If I had to guess whether Duke will under- or over-perform its kenpom rank in the postseason, I'd guess over-.

Wahoo2000
02-26-2015, 10:08 PM
Maybe it's just me, but if I were pomeroy, I'd take any/all of these as "rips" in context of the overall tone of the post (which I'd classify as "kenpom is dumb/wrong because it doesn't rank Duke as high as I think it should"):

"In what world does Utah rank ahead of Duke"
"some black-box interconnected calculation that can't really be explained"
"Most irritating is that we weren't rewarded sufficiently"
"his calculations put too much weight on"...

Atldukie79
02-26-2015, 10:09 PM
Regardless of one's understanding of Kenpom (or interest in doing research to understand it), the charts presented in the original post are interesting and worthy of a discussion. Based on the evidence shown, I would think Duke likely to beat Utah. (Of course I feel that way anyway)

In any case, I appreciate the original post and enjoy the discussion of legitimate merits of various methods to rank teams.

hurleyfor3
02-26-2015, 10:51 PM
Geez, people, it doesn't take much for you to turn a fair (yet highly debatable) topic into a mess, does it?

Everything that was extracurricular to bedeviled's original post has been deleted. The converse is, if I didn't touch your post it was fine.

I'll merge this into the Dork Polls thread later, but for now will keep it separate for clarity.

bedeviled
02-26-2015, 11:40 PM
I am definitely embarrassed that I posted before thinking through the Pomeroy stuff. If I had thought through it, I would have couched my thoughts in terms of 'performed well / performed poorly' instead of 'win/lose.' That would have changed my comment about Duke lost to #20,33,62 and Utah lost to #3,11,26,42,52 - Instead, I should have made a list of underperforming games rather than simply listing losses.

Nonetheless, my question still stands.
What is it about Utah's history that makes it rank above Duke? Or Duke's history that makes it rank below Utah?


is it some interconnected calculation? yep
Is it a black box that can't be explained? nope I had a post where I laid it all out.This is what I was afraid of....that it was a bunch of little numbers that make the final calculation come out in a particular way but without a real-world concept that I could wrap my mind around. While the method of his calculations is helpful, I'm really interested in a real-world pattern or examples.

Ultimately, the numbers are based off of some real-world happening and subsequently predict a real-world happening. Pomeroy writes (http://kenpom.com/blog/index.php/weblog/entry/ratings_explanation) that he sees
the philosophy of the system as this: it looks at who a team has beaten and how they have beaten them. Same thing on the losses, also. Yes, it values a 20 point win more than a 5 point win. It likes a team that loses a lot of close games against strong opposition more than one that wins a lot of close games against weak opposition. So, even he thinks of it in terms of wins and losses. http://kenpom.com/blog/index.php/weblog/entry/ratings_explanation

It DOES appear that the concept of win vs loss is not part of the calculations, but I'm not entirely convinced that the concept isn't hidden in the numbers somewhere (though probably not in a binary form). Similarly, I believe his SOS is not considered to be factored into the ratings. Yet, the ratings are adjusted according to how good teams are and who they've played, so some reflection of SOS is in the numbers.

At the very least, comparisons between teams are made (not just comparison to a raw average of all teams). Otherwise, efficient teams in a crappy, no-name conference would rate as highly as efficient teams in the ACC. Unfortunately, I don't know what adjustments he makes to his raw efficiencies.

He states (http://kenpom.com/blog/index.php/weblog/entry/pomeroy_ratings_version_2.0) that big upsets get the most weight in his system, while expected lopsided wins are minimized. This suggests that poor teams get lots of credit for beating good teams, but good teams get little credit for beating poor teams (or, if you prefer, performing efficiently against poor teams). Importantly, he does not make a comment about the inverse:
In the big upset, is there more weight to the penalty given to the good (but underperforming) team???
And, when a team loses to a highly ranked team, is the penalty of the losing team's performance minimized??

In addition, when an elite team plays another elite team, the predictive difference between the teams is negligible and the resulting difference (win) is attributed to luck, right? So, when Duke beats an elite team, but not in some way significantly different than the normal efficiencies, Duke doesn't get credit in the ratings for the big win (though they do in the luck column). Indeed, Pomeroy does state that 'lucky' teams tend to be rated lower than their records suggest. Thus, I think Duke's big wins aren't rewarded in the Pomeroy model.

My point is that the ratings are not just raw efficiencies. He factors in opponents strength and other factors, including some kind of element of diminishing returns (and when some results are diminished, others are effectually emphasized). So, what returns are diminished and what returns are emphasized?
My question is: Is the combination of opponent strength + diminishing returns (in the form of 1. severe penalty for poor performances against bad competition and 2. minimal rating boost from elite-quality wins) the reason why Duke is ranked lower than Utah?.


Welp, kenpom ratings consistently perform better than other rating methods...so do so at your own risk....For a stretch of years (ending years ago), I actually filled out brackets based on kenpom, AP, "Power Rankings," analysts, Sagarin, RPI, etc predictions and found none that stood out.....except for Vegas, but Vegas doesn't give out odds for the full bracket ahead of time :(

* I agree with the merge to Dork Poll thread idea, and mod says it will likely happen (Sadly, I had searched for "geek" instead of "dork").

bedeviled
02-27-2015, 12:02 AM
Here is the second example that jumped out at me and led me to pull the trigger on the Duke-Utah post: Pomeroy ranks Texas as the 21st best team in the nation, while their RPI says that they are 1 of 11 versus top 50 teams!!!. It's funny, yet quite interesting given that Texas is currently slated to be in the tournament while Joey doesn't even have Miami and Pittsburgh in the first four out. The disparity between kenpom rankings for Texas and Miami/Pitt is quite significant and appears, to me, to devalue big wins and to punish unexpected underperformance against poor teams (again, maybe not directly as win vs loss, but as some systematic adjustment/weighting of raw efficiencies).




Texas
NCState
Syracuse
Miami
Pittsburgh


Record
17-11
17-11
18-10
18-10
19-10


Conference record
6-9
8-7
9-6
8-7
8-7










Kenpom rank
21
33
56
62
75


Kenpom SOS
18
23
66
52
49


vs Kpom 1-10
0-4
1-2
0-2
1-1
0-2


vs Kpom 11-25
1-4
2-3
2-1
0-3
2-2


vs Kpom 26-50
3-3
1-2
1-1
2-1
0-3


vs Kpom 51-100
1-0
6-3
3-5
4-3
5-1


vs Kpom sub-100
12-0
7-1
12-1
11-2
12-2



It looks like the ACC teams are punished for their bad losses (er, underperformance) but do not get credit for the big wins (er, overperformance). Or, any wins for that matter, lol. It's as if Texas' losses to all the #25-ranked Big12 teams are ignored while the ACC teams have better winning percentages in all levels of top competition.
4810
Is it simply a matter of 'luck,' like Texas is efficient on O & D but just keeps losing anyway?....which is conceivable - if the whole conference is basically equal "strength of opponent" then maybe it's all a Pomeroy toss up and there is minimal penalty for them performing worse than their peers. Is the Pomeroy system set up with adjustments/weighting that devalues the ACC teams' big wins and/or, more likely to me, emphasizes the penalty of underperformance against poor teams?

And, for the sake of comparison, here's the RPI data:



Texas
NCState
Syracuse
Miami
Pittsburgh


RPI rank
41
36
62
65
38


RPI SOS
13
2
70
79
35


vs RPI 1-10
0-2
1-2
0-2
1-1
0-2


vs RPI 11-25
1-6
2-2
1-1
0-3
1-3


vs RPI 26-50
0-2
2-1
1-3
1-1
1-2


vs RPI 51-100
4-1
3-5
2-3
4-1
3-1


vs RPI sub-100
12-0
9-1
14-1
12-4
14-2


My apologies if I'm too obtuse to understand or am beating a dead horse. Something just seems off with the ratings compared to what I would expect, and I'd like to understand it in terms of real-world patterns/scenarios if possible. My guess is that "bad losses" (or, underperformance against poor teams) are driving the oddities I perceive.

uh_no
02-27-2015, 12:47 AM
What is it about Utah's history that makes it rank above Duke? Or Duke's history that makes it rank below Utah?
For duke, it is consistently poor defensive play. Ever since the new year, duke's defense has not been good. The system works by limiting the error between what the system would predict would happen and what actually happened. When we got absolutely anihilated on D by miami, not only did that game hold the most weight (because it was most recent), but it would have introduced a massive amount of error if our ranking DIDN'T drop. Miami's offense correspondingly shot up. I'd pull the actual numbers before and after the game, but that comptuer is inaccessible ATM.


It DOES appear that the concept of win vs loss is not part of the calculations, but I'm not entirely convinced that the concept isn't hidden in the numbers somewhere (though probably not in a binary form). I'm not sure how you mean. Since a teams offense is only considered against the opponent defenses, and a team's defense against others' offenses, there's no way for wins and losses to come into account. He even has a metric "luck" which represents the deviation between his rankings and a team's actual wins and losses. A team 30-0 team with all 1 point wins and a 30-0 team with all 1 point losses (given similar schedules) would have almost identical rankings, but their luck would be wildy divergent. Obviously you'll win more games if you have better efficiencies, but the theory is that efficiencies determine wins and losses (probabilisticly), not that wins and losses determine his efficiencies.


Similarly, I believe his SOS is not considered to be factored into the ratings. Yet, the ratings are adjusted according to how good teams are and who they've played, so some reflection of SOS is in the numbers. It's included in the whole "error" thing i mentioned before. If a team you're playing has a better defense, then you're expected to play worse offense against them. So it's included on a game by game basis (or more accurately, an offense/defense split in each game basis), and the SOS number he lists (like most of the numbers on the front page) is derived from that, but is not used to calculate the rankings or efficiencies.



At the very least, comparisons between teams are made (not just comparison to a raw average of all teams). Otherwise, efficient teams in a crappy, no-name conference would rate as highly as efficient teams in the ACC. Unfortunately, I don't know what adjustments he makes to his raw efficiencies.
You're exactly right. you're rating on offense and defense is the value that minimizes the error between what you would be predicted to do in each game and what you actually did. And as I said, how you would be predicted to do on offense/defense in each game is scaled based on your opponent. I believe it's the geometric mean of your offense and the opponents defense, scaled a couple points for home court advantage.


big upsets get the most weight in his system, while expected lopsided wins are minimized. This suggests that poor teams get lots of credit for beating good teams, but good teams get little credit for beating poor teams (or, if you prefer, performing efficiently against poor teams). Importantly, he does not make a comment about the inverse:
In the big upset, is there more weight to the penalty given to the good (but underperforming) team???
And, when a team loses to a highly ranked team, is the penalty of the losing team's performance minimized??

There's a few things going on here.
1. In ANY model, results which contradict the model's expectation are always going to result in bigger changes than results that confirm the model. For duke, which started highly rated on offense or defense, that can only mean underperforming. Teams at the bottom of the rankings would only ever see changes when they overperformed against good teams, which is where I think some of your intuition breaks down, since you only see what happens to duke.
2. In kenpom's model, there are diminishing better performances against overmatched teams. I do not believe, however, that big upsets are given more weight than regular games. What I mean is that our loss to NCSU would probably have the same weight as our win over Syracuse (aside from the fact that syracuse was more recent). It would, however, hold much more weight than our 20 point win over army.
3. You shouldn't think of the model as "rewarding and penalizing" It's effectively recalculating from scratch after every game. It doesn't have memory for what duke was ranked yesterday, only the results duke had in the past.



In addition, when an elite team plays another elite team, the predictive difference between the teams is negligible and the resulting difference (win) is attributed to luck, right? So, when Duke beats an elite team, but not in some way significantly different than the normal efficiencies, Duke doesn't get credit in the ratings for the big win (though they do in the luck column). Indeed, Pomeroy does state that 'lucky' teams tend to be rated lower than their records suggest. Thus, I think Duke's big wins aren't rewarded in the Pomeroy model. Yup. Duke's win over UNC, for instance, will show up in "luck" far more than the efficiencies. You can either agree with that sentiment or not. It's not perfect...there is certain value to having high IQ to run smart plays down the stretch. THe thought is that the effect of such is far outweighed by other random factors such as whether you happened to knock down one more three or not.



My point is that the ratings are not just raw efficiencies. He factors in opponents strength and other factors, including some kind of element of diminishing returns (and when some results are diminished, others are effectually emphasized). So, what returns are diminished and what returns are emphasized?
My question is: Is the combination of opponent strength + diminishing returns (in the form of 1. severe penalty for poor performances against bad competition and 2. minimal rating boost from elite-quality wins) the reason why Duke is ranked lower than Utah?.
Spot on again. As I said, it all comes down to minimizing error between actual results and what the model would predict. The error is amplified for more recent games. It is reduced for performing exceptionally well against overmatched opponents (and exceptionally poorly against far superior ones), and for home court advantage.

Remember that opponent strength is not a single number, but analyzed as your offense vs their defense, and vice versa. There is not an intrinsic severe penalty for poor performance against bad competition. It's just as I pointed out, that a top team has nowhere to go but down, and it's going to go down when the model is most contradicted, which is as you point out. For Grambling state, they'd see little effect for big losses to terrible opponents, but huge gains for good performance over almost any opponent. In either case, it's not really "penalizing" it's the model saying "OOPS! I was wrong the whole time and didn't have enough data yet....this is a more accurate evaluation of your strength"

So the ultimate question is "why is utah ranked ahead of duke?". The easy answer is "because that's what the numbers say!" I think the answer you're looking for is a more intuitive grasp. The answer is that duke's defense is really bad for a team ranked in the top 10.

You can go http://www.scacchoops.com/duke-mens-basketball-roster and see our defensive efficiency over the course of 2015 has averaged a decent amount over 100. This is in line with kenpom's 97.3, when you scale for an offense we play against. If you take miami for instance, a middling offense (for the ACC) they put up a 109. That means they should be expected to score 103 or so against us, which is right in line with the average of what we're giving up to most teams.

Utah, on the other hand, has consistently held their opponents to far lower efficiencies, including holding kansas to barely over 100, and many teams far under 100 (especially recently...which is important for kenpom).

The big problem with kenpom, and with College ball as a whole, and kenpom freely admits, is that at the end of the regular season, any OOC play is so long and varied ago as to be useless. Now, we sit here and say " Duke has played harder teams in the ACC so we should get credit!" But the relative strength of the teams in the ACC vs the PAC10 is very hard to establish. Ultimately, the answer I think you're looking for is that yes, duke has played a better schedule, and yes Utah has played better defense against a weaker schedule....but the numbers the model pops out say the error is minimized if we say that Utah has played good enough defense against their poorer opponents to account for the fact that they're are not as good as Duke's.

What we'd all love is if some pac 12 teams could play some ACC teams...then if the ACC teams blew them out, the model would say "the PAC12 teams are really not good, so even though utah's defense has been pretty good, the opponents have been really REALLY bad." on the contrary, if the pac12 teams played the ACC teams tight (and again, the model would properly account for whether a good PAC12 team played a bad ACC team etc), then it would say the PAC12 teams would get more credit and utah would have even MORE credit for playing good defense against teams that are actually pretty good!

I think it's important to point out that the way the pyth is calculated will favor a balanced team over one with a super good offense and poor defense (and vice versa). This is due to a huge exponent (something like 12 now or something...he adjusts it for better predictive value). Our offense, though stupendous, has passed the point of diminishing returns for being meaningful. If you look at the pyth graph in terms of x (offense) and y (defense)



and set x and y from like 85->120, you'll see a flat range when x goes really large. The problem is we're on that flat part...so our offense could be 150, and it would hardly make a huge difference. The only way we'll get much gain is from improved defense. Utah, on the other hand, is much more on that slope, so marginal gains on either offense or defense will help them.




For a stretch of years (ending years ago), I actually filled out brackets based on kenpom, AP, "Power Rankings," analysts, Sagarin, RPI, etc predictions and found none that stood out.....except for Vegas, but Vegas doesn't give out odds for the full bracket ahead of time

I'd have to go hunting, but someone had an article on the prediction % over time, and kenpom was a few % points ahead. He adjusts certain parameters in the model over time to try to come up with better predictions (for instance he stalled adding the diminishing returns for blowout wins for a long time until he could demonstrate better predictive ability). It's a fine balance between improving the model and avoiding overfitting the data.

ChillinDuke
02-27-2015, 01:15 AM
Uh_No, you clearly have a great handle on KenPom and statistics in general.

This was a pleasure to read.

Please bear with the rest of us who don't as easily follow the concepts behind the model. I, for one, am a big KenPom believer even though I admit to not fully grasping the intricacies of his theory/model.

Again, thanks.

- Chillin

bedeviled
02-27-2015, 07:20 AM
Thanks so much for the thoughtful explanations. They definitely helped me think about some stuff, though I still have some issues. There are some unimportant things I'd like to beat into submission, like the idea of win/loss and strength of schedule being part of the system in epiphenomal and iterative types of fashion. But, I should probably refrain from embarrassing myself again.

Honestly, I hear your points and see how they could generate the Pyth trends in question. Yet, I still think there is more to it....the results are slightly too outlandish for me to accept as natural in the way you described. (I'll buy the trends but not the degree)

Overall, I'm stuck on the secretive adjustments made to the observed efficiencies. Here's my chief supporting evidence :)
Pomeroy states (bold type is mine) (http://kenpom.com/blog/index.php/weblog/entry/pomeroy_ratings_version_2.0):

In those days, I had a method to give variable weight to games in my otherwise elementary least squares system. The weight was based on three ingredients - how close the game was expected to be, how close the game actually was, and when the game was played.

The result is that games perceived by the system as big upsets get the most weight, while the influence of expected lopsided wins is minimized..[SNIP]..

So I’ve dusted off that algorithm, spent some time tuning the various parameters, and applied it to the efficiency model to improve the predictive power of adjusted offense and adjusted defense.
I interpret this to mean that he does specifically include a factor for weighting individual game results. That is, not all games are treated equally (aside from the expected blow-outs of overmatched teams, which we agree are diminished).
I understand that this is done to correct error in an effort to refine predictive power. So, if it does improve predictive power, then it is helping define the truth of Utah having a higher Pyth than Duke. But, this quote, about tweaks he is making to his system, is from late Oct 2013. So, I wonder if 2014-15 is the first year incorporating it for the season (or, if it was used in 2013-14, if there were ratings that should have struck my same nerve last year). If so, there is a chance that the current form of this tweak has a deleterious effect on some prediction ratings. (I'm sure he's tested it on data, though he comments that he reluctantly went back and applied it to previous years for kicks, which suggests that he might not have tested it on empirical, full season, or full Div I data when designing it).

There is not an intrinsic severe penalty for poor performance against bad competition
Hahaha, I believe we've come to an impasse.

Here are the 3 building blocks to my conspiracy, lol -
1. Adjusted efficiencies incorporate strength of competition (they are not just straight up efficiencies, but are adjustments made based on results against your opponent, whose own adjusted efficiencies factored in their opponents, and so on and so on - in effect, a strength of competitor and their SOS)
2. Expectations are made based on these adjusted efficiencies (so, strength of competitor factors into the expectations)
3. Games in which results that are more different from expectations are weighted more than those with results closer to expectations

Here are the tenets of my conspiracy (yes, I realize the things I describe are numbers, but indulge me) -
- I think the system provides for characterizing competition on a continuum of good/bad in the form of adjusted efficiencies
- I think the system uses its characterization of a team (as good or bad) in developing its expectations for a game
- I think games between more disparate teams (very good vs very bad) have potential for greater difference between expectations and actual result
- I think that games with results greater in difference from expectations are weighted more heavily than games closer to expectations.
- I think very good teams are disproportionately punished for poor performances against poor competition (ie the adjustment is relatively more negative than would be expected naturally)

On a positive note, I think we agree that Duke's big wins are reflected in the 'luck' column instead of in the Pyth rating.

Another positive note (because your post was really good and I feel bad for being obstinate), I especially enjoyed your discussion about how the exponent interacts with the OEff and DEff to effect the Pyth. I hadn't considered that effect. Pretty cool.

AIRFORCEDUKIE
02-27-2015, 07:25 AM
There is no world where Utah is better than Duke, and any rankings system that tries to make the argument that they are is clearly dork polls made by analytics folks who can't really play basketball so they make up stats and polls and other BS so they can be a part of the game..:cool:

Sorry Charles Barkley took over my keyboard and forced me to type this.

bedeviled
02-27-2015, 07:46 AM
I probably should have finished my last post with:
The proof is in the pudding. If all his adjustments result in a system with good predictive power for tournament teams playing each other (as those are the only ones I'm concerned about), then I retract everything! :D There is no room for fairness complaints when predictions match actual results.

-jk
02-27-2015, 07:52 AM
I think sagegrouse's Fool's Errand post has bearing here, too. If KP can't compare conferences on data after the new year, any biases run up in the fall will continue until the NCAA/NIT tournies.

Teams evolve over the season: teams with a heavy underclass component, teams with injuries, transfers, what have you.

I can't see how any stats based model could account for these changes.

-jk

77devil
02-27-2015, 07:56 AM
There is no world where Utah is better than Duke, and any rankings system that tries to make the argument that they are is clearly dork polls made by analytics folks who can't really play basketball so they make up stats and polls and other BS so they can be a part of the game..:cool:

Sorry Charles Barkley took over my keyboard and forced me to type this.

I'm with you Chuck.

p.s. Pls keep calling out Kenny Smith whenever he promotes the cheaters.

Karl Beem
02-27-2015, 08:15 AM
There is no world where Utah is better than Duke, and any rankings system that tries to make the argument that they are is clearly dork polls made by analytics folks who can't really play basketball so they make up stats and polls and other BS so they can be a part of the game..:cool:

Sorry Charles Barkley took over my keyboard and forced me to type this.

I'm with ya, Sir Charles!

bluedev_92
02-27-2015, 08:42 AM
Well since most people don't have unlimited time to look through the threads and read all the posts I don't see anything wrong with him having a discussion on it.

Not sure if I'm just doing somthing wrong but this site has the worst search function to find what you are looking for. Always comes up with 0 results even if what I search for is on the 1st page

Agreed. Some posters seem to take things a little too seriously & maybe should spend a small amount of time thinking about being civil before posting.

Wander
02-27-2015, 10:17 AM
bedeviled, to say to what uh_no said in a simpler way... you're making a mistake by comparing win-loss records between Duke and Utah. You should be comparing margin of victories. Duke and Utah are both 12-3 in conference. All 12 of Utah's conference wins have come by double digits, 8 have come by 20 or more, and 2 have come by 30 or more. Duke only has 5 ACC wins by double digits, 2 by 20 or more, and 1 by 30 or more. The ACC is better than the Pac-12, but still.

Anyway, I agree Duke should be ahead of Utah in the polls and the selection committee - which is why we have these things in the first place.

bedeviled
02-27-2015, 11:34 AM
bedeviled, ...[SNIP]...You should be comparing margin of victoriesYeah, thanks. I didn't appreciate the size of their differentials until this late morning when I checked to see that Utah thoroughly embarrassed ASU (9 1st half points!) and me in one fell swoop. But, I had already posted a billion times to this thread, so I was going to spare y'all the picture of my devil tail between my legs. :(

I am fully on-board now - the calculations could naturally account for the trend and the degree for Utah.
But, I'm still going to give Pomeroy the ol' middle pitchfork - even if he is right, his Duke (and Utah) rating stands out against the consensus (http://www.masseyratings.com/cb/compare.htm).

Kedsy
02-27-2015, 11:41 AM
I think sagegrouse's Fool's Errand post has bearing here, too. If KP can't compare conferences on data after the new year, any biases run up in the fall will continue until the NCAA/NIT tournies.

Teams evolve over the season: teams with a heavy underclass component, teams with injuries, transfers, what have you.

I can't see how any stats based model could account for these changes.

Going even further, there is NO statistical system that can predict all winners in a 68 (or even 64) team tournament. Because there is a variability in sporting events. Last year, Pomeroy's system gave Duke an 84.5% chance of beating Mercer. That's a huge favorite, but it still meant Mercer had a 15.5% chance to win. Any statistical system that predicted Mercer would beat Duke would have been silly, but by the numbers Mercer still had about as much chance to win as you would to roll an eight with a pair of dice -- which if you roll dice enough will happen fairly often.

In other words, just because a team is better doesn't mean it'll win. Suggesting that a system designed to measure which team is better should always (or even almost always) be able to predict a winner is a fallacy.

Some systems are better than others, sure, and will have a higher decree of predictive accuracy. Many computer systems have better predictive accuracy than most people's eyes. But the level of predictive accuracy that a lot of people expect from a computer system is probably not realistic or attainable.

Listen to Quants
02-27-2015, 12:14 PM
I think sagegrouse's Fool's Errand post has bearing here, too. If KP can't compare conferences on data after the new year, any biases run up in the fall will continue until the NCAA/NIT tournies.

Teams evolve over the season: teams with a heavy underclass component, teams with injuries, transfers, what have you.

I can't see how any stats based model could account for these changes.

-jk

KP does not do a good job of adjusting for injury, BPI (the ESPN statistical index) does better (ESPN has begun to assimilate stats people). Vegas does great.

Other changes over a year, which show up strongest in March, would be included in the calculation IF (big if perhaps) the conferences showed similar such dynamics (similar to each other), so that the changes would be incorporated in the individual teams vs. others the conference. Thus, is some freshman loaded team were in fact getting much better, say defensively, that would show up [Duke fails there both in the stat measures and to my eyeball]. There are a fair number of total out of conference games played post Jan 1 if one looks across an entire conference and that would serve to stem any drift amongst conferences. Computationally it's reminiscent of an 'average of averages' calculation which gets the same result as a single global average. (KP is not a linear system so it won't be fully equivalent.)

Seattle Hoo
02-27-2015, 12:52 PM
1) One of the best criticisms of KP I have seen is that his "luck" category should really be called "the system's error". So if a team performs better than KP predicts it should, that team is LUCKY? That seems to be what it says. If so, I will self-translate "luck" into "hubris" when I peruse his statistics.

2) Having seen Utah play and being very impressed with Delon Wright and the freshman center, I think Utah is in the class of teams that can beat any other team out there on an equal basis. No, I don't think they are as good as Duke, but I do think it's pretty close.

3) Does it seem to anyone else that in the OOC season we saw that:

a) Other than Arizona and Utah, nobody in the Pac-Twelve is very good at all;
b) Other than Villanova, nobody in the Big East is that good;
c) Other than Kentucky the SEC pretty much stinks; and
d) The Big 12 has a lot of good teams but no great team?

4) Given 3), does it seem that these other conferences are having their non-elite teams become way over-valued as potential NCAAT teams, such that we are hearing talk of six or seven teams from some of these conferences while the ACC could wind up with only FIVE teams in? We have five teams that are on the same level as Arizona, Utah, Villanova, Kentucky and Kansas, and a much smaller spread than all of those conferences other than the Big 12 between the top tier and the next tier, yet our mid-tier teams are ranked behind those other conferences' teams in the bracketology? If any team from the SEC other than Kentucky gets in, and any team from the Big East other than Villanova, and any team from the Pac 12 other than Arizona and UCLA gets in while NC State, Miami and Pitt go to the NIT, then the ACC teams are getting screwed. Those three teams are at least as good as any of the non-Top 10 teams from those other conferences and belong in the NCAA Tournament.

hurleyfor3
02-27-2015, 12:59 PM
merge bump

Wander
02-27-2015, 01:48 PM
a) Other than Arizona and Utah, nobody in the Pac-Twelve is very good at all;
b) Other than Villanova, nobody in the Big East is that good;
c) Other than Kentucky the SEC pretty much stinks; and
d) The Big 12 has a lot of good teams but no great team?


Your standard for a good team is way too high; it seems to be "serious Final Four contender," which, yes, out of those conferences, that would include Arizona, Utah, Villanova, and Kentucky. Teams like Arkansas and Providence are perfectly good teams, just not elite.



Given 3), does it seem that these other conferences are having their non-elite teams become way over-valued as potential NCAAT teams, such that we are hearing talk of six or seven teams from some of these conferences while the ACC could wind up with only FIVE teams in? We have five teams that are on the same level as Arizona, Utah, Villanova, Kentucky and Kansas, and a much smaller spread than all of those conferences other than the Big 12 between the top tier and the next tier, yet our mid-tier teams are ranked behind those other conferences' teams in the bracketology? If any team from the SEC other than Kentucky gets in, and any team from the Big East other than Villanova, and any team from the Pac 12 other than Arizona and UCLA gets in while NC State, Miami and Pitt go to the NIT, then the ACC teams are getting screwed. Those three teams are at least as good as any of the non-Top 10 teams from those other conferences and belong in the NCAA Tournament.


First of all, I guarantee* the ACC will get more than 5 teams in. I'm not sure who yet, but at least one of NCSU/Miami/Pitt will get in. But you're VASTLY overrating those teams if you think they're all "at least as good" as any non-Top 10 teams from those conferences. Providence, for example, is 3-1 against the ACC (including a victory over Miami) and has a player better than anyone on NCSU/Miami/Pitt. Oklahoma is far better than all these teams by any conceivable metric.

*Edited to say that "guarantee" is probably a bit too strong, but it's highly unlikely that all 3 teams collapse, especially since two of them play each other still

I completely agree on Pomeroy's "Luck" metric, though. I'd rather it be called "Error" or something.

Seattle Hoo
02-27-2015, 02:26 PM
First of all, I guarantee* the ACC will get more than 5 teams in. I'm not sure who yet, but at least one of NCSU/Miami/Pitt will get in. But you're VASTLY overrating those teams if you think they're all "at least as good" as any non-Top 10 teams from those conferences. Providence, for example, is 3-1 against the ACC (including a victory over Miami) and has a player better than anyone on NCSU/Miami/Pitt. Oklahoma is far better than all these teams by any conceivable metric.

I excluded the Big 12 from the part of the post to which you responded, but I can see how you would miss that. It would require a more careful parsing of my post than reasonable to expect in order for you to catch it. Yes, the Big 12 has about five teams that *do* deserve to be in the field before our next three.

As for Providence, I fail to see how the ACC Three are not on the same level as Providence. I mean, Providence lost to both Boston College and Marquette, last place teams in the respective conferences. They just got hammered by Villanova. They've got 6 losses in a weaker conference than the ACC. They're in fourth place in a ten-team conference. I don't see how they are an better than NCSU/Miami/Pitt at this point in the season. Of course, introducing "at this point in the season" into the discussion calls into question Louisville and even UNC....

sagegrouse
02-27-2015, 02:45 PM
KenPom is a highly refined model that collapses all the data about a forthcoming basketball game into five numbers: offensive and defensive efficiency for both teams and an adjustment for home court (if needed). How could it possibly produce reasonable and reasonably accurate results in all cases?

Basketball involves many, many more factors than can be captured in five numbers: competing styles of play, player match-ups, motivations of teams and individuals, injuries, cyclical performance of players, particular home court advantage, etc., etc. some teams run up the score; some really good teams seem to slack off with a big lead -- both affecting efficiency measures.

KenPom numbers are summary statistics of the past experiences of the teams.

Aside from the observation that in such a simple model (five output numbers, although millions of calculations) anomalies are inevitable, the PAC-12 plays relatively few games against other major conferences (2.5 per team versus about 4.0), presumably because of geographical distances.

hurleyfor3
03-08-2015, 09:08 PM
We somehow jumped Utah and Gonzaga and are now up to #6 Pomeroy. We had the #1 offense for a few hours today before Ken plugged Wisconsin's game into his HP-35.

toooskies
03-09-2015, 12:31 AM
On luck:
It is not error. It is a literal statistical description of how often a team concentrates good possessions in close games. Effectively, it is a measure of winning close games While not running up the score against weaker opponents.

uh_no
03-09-2015, 12:46 AM
On luck:
It is not error. It is a literal statistical description of how often a team concentrates good possessions in close games. Effectively, it is a measure of winning close games While not running up the score against weaker opponents.

The rest of what you say is spot on, but not sure that's quite precise.

It's the literal statistical description of the difference between a team's expected record when calculated from efficiency, and their actual record. I'm not sure how the concentration of good possessions factors into it.

Luck is based on efficiency, and efficiency is calculated separately from O and D, so the expected winning percentage takes no account for closeness of games. Further, your overall record takes no account for close games, and certainly does not care how efficient you were.

Skitzle
03-09-2015, 04:52 AM
I was looking at these polls the other day and cringing at Duke's 65th ranked defensive stat.

Then something jumped out at me. Our opponents average adjusted Offense is 14.

So I ask the people who know better than me. Is it possible that our Defensive stat looks worse, because we actually been playing the better offensive teams all season, or is this all standardised out and we're really not that good on defense

OppO Ranking in the top 6





Rank Team OppORank
1 Kentucky 46
2 Arizona 75
3 Virginia 49
4 Wisconsin 38
5 Villanova 59
6 Duke 14

Troublemaker
03-09-2015, 07:44 AM
Is it possible that our Defensive stat looks worse, because we actually been playing the better offensive teams all season, or is this all standardised out and we're really not that good on defense

Yes, kenpom is schedule-adjusted, so theoretically, Duke's defensive ranking should be very close to the same even if we hadn't played against so many good offensive teams.

As for Duke being "really not that good on defense" -- maybe. Duke's shown some recent signs of having a good defense. I think we've played very well on defense in 4 of the past 5 games. Because kenpom is a reflection of what Duke's been doing all season rather than just the recent games, the defensive ranking could still be underrating Duke as a defensive team right now. But we'll see. If Duke has truly improved, the defensive ranking should climb as we play more games.

budwom
03-09-2015, 08:53 AM
just for grins I think I"ll track KenPom vs Sagarin vs consensus point spread (yes, i know, not a prediction as much as a bet balancer) to
see how they fare this post season.

uh_no
03-09-2015, 10:00 AM
Yes, kenpom is schedule-adjusted, so theoretically, Duke's defensive ranking should be very close to the same even if we hadn't played against so many good offensive teams.

As for Duke being "really not that good on defense" -- maybe. Duke's shown some recent signs of having a good defense. I think we've played very well on defense in 4 of the past 5 games. Because kenpom is a reflection of what Duke's been doing all season rather than just the recent games, the defensive ranking could still be underrating Duke as a defensive team right now. But we'll see. If Duke has truly improved, the defensive ranking should climb as we play more games.

Two points.

Point 0: you're spot on...the idea is that the rating is SOS invariant.
Point 1: despite attempting to account for the "wisconsin" affect, it seems to me there still is some bias towards teams with weaker schedules. One of the issues here is the non-normality of the distribution, at least with duke. We've been able to blow mediocre teams out of the water on defense recently (clemson syracuse wake), but still not done great stopping better teams (UNC x2) (obviously it's not a perfect correlation). The correlation between duke's opponents offensive prowess and the difference between how an opponent actually performs EVEN when normalized for their offensive prowess seems to be quite high, even though it should be 0 in theory.

Point 1: I tend to agree we've played some better games, so it's very hard to tell. on the whole, though, once duke dropped to the 50/60 range, most defensive performances have been in line with that new ranking (including UNC). And the only games we kept opponents significantly below that average were clemson, cuse, and wake. ALl of which were at home, two of which were repeat games.

http://www.scacchoops.com/duke-mens-basketball-roster

everyone else has been a bit over one PPP.

So we'll see, but maybe not until the sweet 16 when we run into a really good team that we haven't played before.

ChillinDuke
03-09-2015, 10:57 AM
Yes, kenpom is schedule-adjusted, so theoretically, Duke's defensive ranking should be very close to the same even if we hadn't played against so many good offensive teams.

As for Duke being "really not that good on defense" -- maybe. Duke's shown some recent signs of having a good defense. I think we've played very well on defense in 4 of the past 5 games. Because kenpom is a reflection of what Duke's been doing all season rather than just the recent games, the defensive ranking could still be underrating Duke as a defensive team right now. But we'll see. If Duke has truly improved, the defensive ranking should climb as we play more games.

If I were to advance a theory that Duke's defensive efficiency is not overly good because (a) we are short a couple rotation players, so can't afford foul trouble and (b) our offensive efficiency is so good that we can afford less efficiency on D in order to address (a) and essentially dare teams to keep up, would I be missing a key statistical concept that KenPom employs?

Hey, I'm trying here.

- Chillin

Listen to Quants
03-09-2015, 11:33 AM
On luck:
It is not error. It is a literal statistical description of how often a team concentrates good possessions in close games. Effectively, it is a measure of winning close games While not running up the score against weaker opponents.


The rest of what you say is spot on, but not sure that's quite precise.

It's the literal statistical description of the difference between a team's expected record when calculated from efficiency, and their actual record. I'm not sure how the concentration of good possessions factors into it.

Luck is based on efficiency, and efficiency is calculated separately from O and D, so the expected winning percentage takes no account for closeness of games. Further, your overall record takes no account for close games, and certainly does not care how efficient you were.

If a good team had a tendency to relax and play less hard against lesser teams, although playing hard enough to win. the dork polls would rate that good team lower than it actually was and its results against other good teams (if it played hard) it would outperform the dork poll expectation. That effect would dump into the 'luck' factor which is indeed as uh_no says "the difference between a team's expected record when calculated from efficiency, and their actual record." In effect it would be a concentration of better than expected possessions into a game where they are 'needed.'

jv001
03-12-2015, 05:53 PM
Looking at Pomeroy's site and a few things jump out at me.
1) Gonzaga rated 6 and Duke rated 7. I just don't get it. The Zags play in a weak conference and we play in a tough conference.
2) Florida with a 16-16 record is rated #44. It seems this site values the SEC teams more than they deserve.

Just my thoughts. GoDuke!

vick
03-12-2015, 06:32 PM
2) Florida with a 16-16 record is rated #44. It seems this site values the SEC teams more than they deserve.

It's less that the SEC is overvalued (he has it 5th of the 6 power conferences) as that he ignores wins and losses and focuses entirely on efficiency, and Florida has lost a ludicrous number of close games--six games by one or two points.

Kedsy
03-12-2015, 11:54 PM
If I were to advance a theory that Duke's defensive efficiency is not overly good because (a) we are short a couple rotation players, so can't afford foul trouble and (b) our offensive efficiency is so good that we can afford less efficiency on D in order to address (a) and essentially dare teams to keep up, would I be missing a key statistical concept that KenPom employs?

Hey, I'm trying here.

- Chillin

I'm not sure if you're missing any statistical concept, but I don't buy your theory because we aren't really "short a couple rotation players." Duke never plays more than an 8-man rotation and we have eight guys. If we suffer an injury or consistently get into undue foul trouble, then we can entertain the idea that we don't have enough guys. In the mean time, we certainly have enough players to run Coach K's system.

jv001
03-13-2015, 06:52 AM
I'm not sure if you're missing any statistical concept, but I don't buy your theory because we aren't really "short a couple rotation players." Duke never plays more than an 8-man rotation and we have eight guys. If we suffer an injury or consistently get into undue foul trouble, then we can entertain the idea that we don't have enough guys. In the mean time, we certainly have enough players to run Coach K's system.

I agree that in game situations, 8 is usually enough for Duke because Coach K doesn't go much deeper than that and sometimes doesn't go past 7. But it could affect Duke by not being able to practice like most Duke teams. Coach K has even commented on that. So, in that sense, having only 8 rotation players could affect our defense. GoDuke!

Ps: we passed the Zags and went up to 6 in Pomeroy ratings. GoDuke!

toooskies
03-13-2015, 11:23 AM
If a good team had a tendency to relax and play less hard against lesser teams, although playing hard enough to win. the dork polls would rate that good team lower than it actually was and its results against other good teams (if it played hard) it would outperform the dork poll expectation. That effect would dump into the 'luck' factor which is indeed as uh_no says "the difference between a team's expected record when calculated from efficiency, and their actual record." In effect it would be a concentration of better than expected possessions into a game where they are 'needed.'

This is actually Pomeroy's logic when talking about the Adam Morrison Gonzaga team that "broke" the system. They had a tendency to severely slack off on defense when they had a big lead, but played better against good teams.

A more elaborate explanation of how I view the "luck" stat:

Getting a blowout win tends to decrease your "luck". The basic logic here is: Your efficiency in this game was way higher than needed. If the results of possessions are random, you clustered too many "good" possessions into a single game. Your efficiency goes up by more than your W/L record.

Getting a close win tends to increase your "luck". The basic logic here is: your efficiency was very close to the minimum needed to get a W in the game. If the results of possessions are random, then you "luckily" just got enough good ones in the game to win. Your efficiency goes up by less than your W/L record, and can even go down relative to past results.

Getting a close loss tends to decrease your "luck". The basic logic here is: your efficiency was just under the minimum needed to get a W in the game. If the results of possessions are random, then you were "unlucky" not to get a few more good ones in the game. You get an L despite being nearly even with your opponent. Your efficiency goes down by less than your W/L record, and can even go up relative to past results.

Getting a blowout loss tends to increase your "luck". The basic logic here is: Your efficiency was awful in this game. If the results of possessions are random, you clustered all of your "bad" possessions into one game. Your overall efficiency goes down by more than your W/L record.

Now, there's every reason to believe that the results of possessions are not random. Particularly in close games, teams can have different levels of confidence about winning, and the poise of each team can be on display. Tyus Jones seems to have a late-game gear where he succeeds repeatedly.

If you believe a team controls late-game situations and "wins the close ones", you're increasing a situation where you increase your luck and decreasing a situation where you decrease your luck. You are getting the maximum number of wins from the minimum efficiency.

jv001
03-13-2015, 11:52 AM
This is actually Pomeroy's logic when talking about the Adam Morrison Gonzaga team that "broke" the system. They had a tendency to severely slack off on defense when they had a big lead, but played better against good teams.

A more elaborate explanation of how I view the "luck" stat: Getting a blowout win tends to decrease your "luck". The basic logic here is: Your efficiency in this game was way higher than needed. If the results of possessions are random, you clustered too many "good" possessions into a single game. Your efficiency goes up by more than your W/L record.

Getting a close win tends to increase your "luck". The basic logic here is: your efficiency was very close to the minimum needed to get a W in the game. If the results of possessions are random, then you "luckily" just got enough good ones in the game to win. Your efficiency goes up by less than your W/L record, and can even go down relative to past results.

Getting a close loss tends to decrease your "luck". The basic logic here is: your efficiency was just under the minimum needed to get a W in the game. If the results of possessions are random, then you were "unlucky" not to get a few more good ones in the game. You get an L despite being nearly even with your opponent. Your efficiency goes down by less than your W/L record, and can even go up relative to past results.

Getting a blowout loss tends to increase your "luck". The basic logic here is: Your efficiency was awful in this game. If the results of possessions are random, you clustered all of your "bad" possessions into one game. Your overall efficiency goes down by more than your W/L record.

Now, there's every reason to believe that the results of possessions are not random. Particularly in close games, teams can have different levels of confidence about winning, and the poise of each team can be on display. Tyus Jones seems to have a late-game gear where he succeeds repeatedly.

If you believe a team controls late-game situations and "wins the close ones", you're increasing a situation where you increase your luck and decreasing a situation where you decrease your luck. You are getting the maximum number of wins from the minimum efficiency.

I guess Ken Pom thinks the SEC teams(outside of KY) are real lucky. I have to agree they're lucky not to be in the ACC. :cool: GoDuke!

Kedsy
03-13-2015, 12:23 PM
I guess Ken Pom thinks the SEC teams(outside of KY) are real lucky. I have to agree they're lucky not to be in the ACC. :cool: GoDuke!

Kentucky is considered to be by far the best team in the country (by people and by computer systems, including Pomeroy). With that in mind, here are some SEC games from this season:

UK by 3 at home over Mississippi in OT
UK by 6 @Texas A&M in 2OT
UK by 8 at home over Vanderbilt
UK by 11 at home over Georgia
UK by 7 @Florida
UK by 2 @LSU
UK by 8 @Georgia

That's 7 of UK"s 18 conference games, a fairly hefty percentage (39%). So either those SEC teams are "real lucky," or they're better than you think they are.

ChillinDuke
03-13-2015, 12:34 PM
I'm not sure if you're missing any statistical concept, but I don't buy your theory because we aren't really "short a couple rotation players." Duke never plays more than an 8-man rotation and we have eight guys. If we suffer an injury or consistently get into undue foul trouble, then we can entertain the idea that we don't have enough guys. In the mean time, we certainly have enough players to run Coach K's system.

We all know your view on this, Keds. And I subscribe to it.

What I meant by short a couple rotation players is not who normally plays but who can play. We only have 8 guys. That's not the same as only playing 8 guys. If we were to be in foul trouble, we physically wouldn't have a 9th or 10th man that could log emergency minutes. Perhaps rotation was the wrong word.

- Chillin

Bluedog
03-13-2015, 12:41 PM
Interestingly, kenpom has tonight's game as semi-home against Notre Dame. Not sure if that's really true considering UNC and UVa fans will likely all be cheering for ND although I certainly realize it's much closer to Durham than South Bend (and hopefully there will be way more Duke fans that ND). It was neutral vs. NC St. With the semi-home caveat, has us winning 74% of the time -- projected score of 81-74. It being semi-home, though, if we beat ND, it doesn't help our kenpom ranking as much as a neutral win would be (relatively small difference, though). But decisions like that in aggregate can certainly affect a team's ranking a bit.

Kedsy
03-13-2015, 01:02 PM
We all know your view on this, Keds. And I subscribe to it.

What I meant by short a couple rotation players is not who normally plays but who can play. We only have 8 guys. That's not the same as only playing 8 guys. If we were to be in foul trouble, we physically wouldn't have a 9th or 10th man that could log emergency minutes. Perhaps rotation was the wrong word.

Sorry, I didn't mean to offend. But I thought your theory was that our defensive ranking was lower than it should be because we have so few players? If so, that theory would only make sense if we'd have had the injury or foul trouble in past games, the games which have contributed to our defensive ranking, and the injury/foul situations were deleterious to our defensive showing. But that hasn't happened, or actually it did happen, but in the game Jahlil sat out I think our defensive ranking got better, which runs counter to the theory.

Listen to Quants
03-13-2015, 01:10 PM
Interestingly, kenpom has tonight's game as semi-home against Notre Dame. Not sure if that's really true considering UNC and UVa fans will likely all be cheering for ND although I certainly realize it's much closer to Durham than South Bend (and hopefully there will be way more Duke fans that ND). It was neutral vs. NC St. With the semi-home caveat, has us winning 74% of the time -- projected score of 81-74. It being semi-home, though, if we beat ND, it doesn't help our kenpom ranking as much as a neutral win would be (relatively small difference, though). But decisions like that in aggregate can certainly affect a team's ranking a bit.

Does anybody know what constitutes home court advantage? Is it familiarity with the arena, supporting fans, the psychological boost of more family/friends/home-things around? Some of that could be sorted out by looking at home-away records with teams near each other (say Wake-Duke ... or perhaps Duke-UNC although that one might we unusual), or when students are not on campus and such tricks. Has anyone looked?

ChillinDuke
03-13-2015, 01:20 PM
Sorry, I didn't mean to offend. But I thought your theory was that our defensive ranking was lower than it should be because we have so few players? If so, that theory would only make sense if we'd have had the injury or foul trouble in past games, the games which have contributed to our defensive ranking, and the injury/foul situations were deleterious to our defensive showing. But that hasn't happened, or actually it did happen, but in the game Jahlil sat out I think our defensive ranking got better, which runs counter to the theory.

No offense taken at all.

I think my theory has a slight difference from this - slight but important. Because we have fewer players, we proactively try to avoid foul trouble which we can afford to do at least somewhat because of our efficient offense. To avoid foul trouble, we don't challenge on defense as hard as a top defense would be expected to do. Since we don't challenge quite as hard, our defensive efficiency is lower than I (we) would like / expect based on how could we can play defense in spurts (when freely challenging/defending without regard to fouls).

I'm not saying it's perfect or even correct. But it's my theory.

- Chillin

ArtVandelay
03-17-2015, 10:46 AM
Does anyone out there save KenPom's pre-tourney rankings? I typically do, but just realized that I seem to have lost 2013. If anybody has them and is willing to share, please send me a message.

Looking solely at this year's ratings, a few things jump out in terms of first-round match-ups:

- Kansas drew by far the toughest #2 in N. Mexico St. (#88). This was around where Lehigh was ranked a few year's ago, which doesn't necessarily mean anything, but the consensus seems to be that KU is the weakest #2 seed. 2/15 upsets are by nature unpredictable, but if I were to bet on one happening this year, this one seems most likely. OTOH, N. Mexico St. has beaten nobody good this year.

- Baylor got the toughest #14 in Georgia St. (#71). The other #14s look comparatively weak. I don't know much about Baylor this year, but they've played a very tough schedule. Like N. Mexico St., Georgia St. also hasn't beaten anyone of note. Anyone has any insights?

- Harvard's string of tournament successes seems unlikely to continue against Carolina. This is a much weaker Harvard squad than in past years (#122), and Carolina was probably under-seeded as a 4 (#8).

- People seem high on Maryland, but the computer only has them at #33. Seems like they could be vulnerable against Valparaiso (#64) in the first round. Where is Bryce Drew when you need him?

- Lots of folks are down on Georgetown, and it wouldn't surprise me for them to be a fashionable upset pick due to their recent track record of losing early in the tournament. This type of logic has never made sense to me. This is a different team than in year's past. Unless you think Thompson has some sort of coaching flaw that makes G'Town particularly vulnerable in the tournament, I don't see why you would think they would lose to a weak #136 Eastern Washington team. It feels weird to say that it almost feels contrarian to pick G'Town, but I like them to advance.

uh_no
03-17-2015, 11:06 AM
Does anyone out there save KenPom's pre-tourney rankings? I typically do, but just realized that I seem to have lost 2013. If anybody has them and is willing to share, please send me a message.
He has them on his site.

Reilly
03-17-2015, 11:11 AM
Does anybody know what constitutes home court advantage? Is it familiarity with the arena, supporting fans, the psychological boost of more family/friends/home-things around? ...

Wasn't there a book a couple of years ago that posited it was the home crowd's influence on the refs that made up the home court advantage?

I would think it's all the things you mention -- a familiarity and feel and psychological boost, generally had at home. Of course, I also believe there are certain gyms/arenas that certain players just feel better in, and those may not necessarily be in their home gym/arena. Friendly rims, good lighting or sight lines, etc ... might be found on the road and give the same sort of psychological advantage.

And, maybe that folks generally believe there is some sort of home court advantage is enough to help cause one.

Reilly
03-17-2015, 11:13 AM
I posted this on the AP Top 25 thread but it probably belongs here:

I started with the AP Top 25, and compared where those teams were ranked in the USA Today Coaches Top 25 poll, the “SRS” at sports-reference.com, in Kenpom, and via NCAA seed.

Team .... AP Top 25 ... USA Today Coaches Top 25 .... sports-reference.com SRS rank .... Kenpom rank ... NCAA rank based on seed

KY ... 1 ... 1 ... 1 ... 1 ... (1-4)

Villanova ... 2 ... 2 ... 5 ... 5 ... (1-4)

Wisconsin ... 3 ... 3 ... 2 ... 3 ... (1-4)

Duke ... 4 ... 5 ... 4 ... 7 ... (1-4)

Arizona .. 5 ... 4 ... 3 ... 2 ... (5-8)

Virginia ... 6 ... 6 ... 6 ... 4 ... (5-8)

Gonzaga ... 7 ... 7 ... 7 ... 6 ... (5-8)

ND ... 8 ... 8 ... 13 ... 10 ... (9-12)

ISU ... 9 ... 9 ... 14 ... 13 ... (9-12)

Kansas ... 10 ... 11 ... 12 ... 11 ... (5-8)

Northern Iowa ... 11 ... 9 ...42 ... 12 ... (17-20)

Maryland ... 12 ... 12 ... 29 ... 33 ... (13-16)

Oklahoma ... 13 ... 15 ... 11 ... 9 ... (9-12)

Wichita State ... 14 ... 13 ... 24 ... 14 ... (25-28)

UNC ... 15 ... 14 ... 8 ... 16 ... (13-16)

Baylor ... 16 ... 17 ... 15 ... 15 ... (9-12)

L’ville ... 17 ... 16 ... 16 ... 18 ... (13-16)

SMU ... 18 ... 19 ... 35 ... 19 ... (21-24)

Utah ... 19 ... 18 ... 9 ... 8 ... (17-20)

WVU ... 20 ... 21 ... 20 ... 25 ... (17-20)

Arkansas ... 21 ... 20 ... 25 ... 29 ... (17-20)

G’town ... 22 ... 24 ... 21 ... 22 ... (13-16)

Mich State ... 23 ... 22 ... 17 ... 17 ... (25-28)

Butler ... 24 ... 23 ... 19 ... 23 ... (21-24)

VCU ... 25 ... 26 ... 37 ... 30 ... (25-28)

Oregon ... 26 ... 25 ... 52 ... 46 ... (29-32)

********************

These four teams were not in the AP Top 25 but were in the SRS top 25:

OHIO STATE = 10 in SRS, 30 in AP, 34 in Coaches; 21 in Kenpom; (37-40) in NCAA
TEXAS = 18 in SRS, unranked in AP, 38 in Coaches; 20 in Kenpom; (41-44) in NCAA
IOWA = 22 in SRS, 40 in AP, unranked in Coaches; 24 in Kenpom; (25-28) in NCAA
XAVIER = 23 in SRS, 28 in AP, 33 in Coaches, 26 in Kenom; (21-24) in NCAA

Things that stuck out to me:

1. The AP Top 25 and USA Today Coaches Top 25 are in lockstep: Oklahoma (13, 15) and G’town (22, 24) are the only schools ranked more than 1 slot apart, and they’re only two spots apart.

2. The computers (SRS, Kenpom) would’ve given Arizona a #1 seed over Duke.

3. Virginia’s #2 seed seems right.

4. Gonzaga is for real – with uniformity in how it is considered by humans (AP, USAT, seed) and computers (SRS, Kenpom): 7, 7, 7, 6, 5-8

5. The humans (poll voters; seeders) love ND a smidge more than the computers do.

6. The NCAA seeders love Kansas more than the poll-voters or the computers.

7. The humans love MD a good bit more than the computers do.

8. The computers really love Utah (8 and 9), but not the humans.

9. G’town got a better seed than the poll voters or computers would’ve given it.

10. Northern Iowa was not given as much credit by the seeders as human voters and Kenpom would have given them; yet SRS ranks UNI even lower (42).

11. Humans think much more highly of Oregon than the computers (52, 46) do.

12. The computers like Ohio State (10, 21) but they are out of the polls and have a 37-40 NCAA rank.

Bluedog
03-17-2015, 11:30 AM
Does anyone out there save KenPom's pre-tourney rankings?


He has them on his site.

He has the post-tournament rankings on his site, not pre-tournament. Not really fair to see how strong a predictor Kenpom rank is on NCAA tournament success when he has the NCAA tournament as a major input to the final rankings. I think Kedsy (??) has them (or I could be recollecting incorrectly...).

Kedsy
03-17-2015, 11:52 AM
He has the post-tournament rankings on his site, not pre-tournament. Not really fair to see how strong a predictor Kenpom rank is on NCAA tournament success when he has the NCAA tournament as a major input to the final rankings. I think Kedsy (??) has them (or I could be recollecting incorrectly...).

I think Pomeroy has the pre-T rankings now behind his paywall, though I also think he recalculated them based on his new-ish formula.

ArtVandelay, I do have the Pomeroy pre-T rankings from 2013 (and every year back to 2009). If you want me to send them to you, send me a PM with your e-mail and I'll be happy to oblige.

rasputin
03-17-2015, 12:00 PM
Wasn't there a book a couple of years ago that posited it was the home crowd's influence on the refs that made up the home court advantage?

I would think it's all the things you mention -- a familiarity and feel and psychological boost, generally had at home. Of course, I also believe there are certain gyms/arenas that certain players just feel better in, and those may not necessarily be in their home gym/arena. Friendly rims, good lighting or sight lines, etc ... might be found on the road and give the same sort of psychological advantage.

And, maybe that folks generally believe there is some sort of home court advantage is enough to help cause one.

Reilly, I read that book too (can't remember the title or author . . .). One of the things the author did to make his case that it was the home crowd's influence on the refs, was to take out of the equation a lot of the other listed items. For example, not having to travel, sleeping in your own bed, etc., would often be cited, but the author said that the slight statistical advantage of home field/court still held in those games (like Yankees vs. Mets, Duke v. North Carolina, etc.).

Bluedog
03-17-2015, 12:10 PM
I think Pomeroy has the pre-T rankings now behind his paywall, though I also think he recalculated them based on his new-ish formula.

ArtVandelay, I do have the Pomeroy pre-T rankings from 2013 (and every year back to 2009). If you want me to send them to you, send me a PM with your e-mail and I'll be happy to oblige.

Ah, nice! Did not notice he added those. Very nice of him. I thought everybody says we were #1 pre-tournament in 2010. Looks like we were #2 behind Kansas.

Kedsy
03-17-2015, 12:22 PM
Ah, nice! Did not notice he added those. Very nice of him. I thought everybody says we were #1 pre-tournament in 2010. Looks like we were #2 behind Kansas.

No, we were definitely #1. I have the numbers. But when he changed his formula a year or two ago, he recalculated all his past data too, and the recalculation may have moved Kansas ahead of us in retrospect.

COYS
03-17-2015, 12:24 PM
Ah, nice! Did not notice he added those. Very nice of him. I thought everybody says we were #1 pre-tournament in 2010. Looks like we were #2 behind Kansas.

I think we were pre tourney number one but he modified his formula in the coming years and that created enough of a difference to move us to number 2.

Bluedog
03-17-2015, 12:34 PM
No, we were definitely #1. I have the numbers. But when he changed his formula a year or two ago, he recalculated all his past data too, and the recalculation may have moved Kansas ahead of us in retrospect.


I think we were pre tourney number one but he modified his formula in the coming years and that created enough of a difference to move us to number 2.

Got it, makes sense -- thanks. I guess we were "kenpom champs" (after the NCAA tournament) in 2002, 2004, and 2010 (and "pre-tourney champs" in 2006). ;) According to his new formula...

-jk
03-17-2015, 01:40 PM
Does anyone out there save KenPom's pre-tourney rankings? I typically do, but just realized that I seem to have lost 2013. If anybody has them and is willing to share, please send me a message.



Is this (https://web.archive.org/web/20130318161448/http://kenpom.com/) it?

-jk

Kedsy
03-17-2015, 01:41 PM
Is this (https://web.archive.org/web/20130318161448/http://kenpom.com/) it?

-jk

Yep, you got it. Gotta love that way back machine.

budwom
03-17-2015, 01:46 PM
Because of all the excellent discussions (some more heated than others) I have periodically checked out how KenPom predictions might differ from Las Vegas opening lines (yes, I know, balance the bets) and from
Sagarin.....and basically what I found on a very consistent basis was that all three methodologies came up with pretty consistent predictions, almost always within a point or two of each other.

So outside of arguments about whether Gonzaga should be ranked above Duke or not (for example), I'm not seeing a lot of meaningful difference.

I guess I'd like to see a case or cases where there is a substantial difference but I'm not sure that's going to happen.

bedeviled
03-18-2015, 01:36 AM
all three methodologies came up with pretty consistent predictionsIn my past experiments, no particular prediction/ranking system stood out for making NCAA tourney brackets. I decided to ask my friend Google, and here's what he said:

According to Team Rankings (http://www.teamrankings.com/blog/ncaa-basketball/ncaa-tournament-seeding-impact), from 2004 to 2014, the coefficient of determination (r-squared) when investigating the correlation between Kenpom rating and NCAA tourney wins was 0.32. For the correlation between tournament seed and NCAA tourney wins, r-squared was 0.36.

Google went on to point out that, when predicting the full bracket prior to the tournament start, the percentage of games predicted correctly looks like this:


Source
Years
KenPom
RPI
Sagarin
Seed


1*
2007 to 2012
66%
63%




2
2004 to 2012
66%


67%


3
2010-2011
62%

59%



1
2007 to 2011
65%






Here are the percentages if winners are chosen on a game-by-game basis (instead of filling the complete bracket out prior to the beginning):


Source
Years
BPI
KenPom
RPI
Sagarin
Sag ELO
Sag Pred
Massey
Seed
AP Poll
AP Preseason
LRMC
Vegas Favorite


2
2007 to 2012

73%





73%






4
2007 to 2011
74%

72%
73%










5*
2003 to 2011
70%
73%
70%
71%
72%
71%
70%
71%
70%

75%
72%


6
2003 to 2010

73%
71%




72%

72%




7
2004 to 2007

73%
71%
72%


73%
73%






8*
unk


73%







72%




For details and methodology, here are the sources:
1 The Harvard Sports Analysis Collective (link (https://harvardsportsanalysis.wordpress.com/2013/02/19/rpi-prediction/), link (https://harvardsportsanalysis.wordpress.com/2012/03/14/survival-of-the-fittest-a-new-model-for-ncaa-tournament-prediction/)) *kenpom was much better than RPI at predicting F4 and Champion
2 BracketScience.com (link (http://wp.bracketscience.com/?p=418))
3 Toutkoushian paper (link (http://kb.osu.edu/dspace/bitstream/handle/1811/48883/EmilyToutkoushian.pdf?sequence=1))
4 ESPN (link (http://espn.go.com/mens-college-basketball/story/_/id/7561413/bpi-college-basketball-power-index-explained))
5 LRMC (link (http://www.sloansportsconference.com/wp-content/uploads/2012/02/14-MIT-Sloan-Paper_BrownKvamNemhauserSokol_2012-Extended-Abstract.pdf)) *BPI estimated based on 3 year performance
6 FiveThirtyEight (link (http://fivethirtyeight.blogs.nytimes.com/2011/03/11/in-n-c-a-a-tournament-overachievers-often-disappoint/?scp=1&sq=pomeroy%20ratings&st=cse&_r=0))
7 Cripe senior project (link (http://users.manchester.edu/Student/adcripe/math%20senior%20project%20paper.pdf))
8 Net Prophet (link (http://netprophetblog.blogspot.com/2012_10_01_archive.html)) *RPI includes a home court advantage given to the higher seed

Of special note, Ms. Toutkoushian (source #3) did regressions of a ridiculous number of variables on data from 1986 to 2009. She then used the resultant equations to predict the 2010 and 2011 tournaments. She surprised herself when her "full equation" correctly predicted 62 out of 63 games in 2010 (apparently completed as a bracket, not a game-by-game prediction)!!! She only missed Kansas' loss to N. Iowa. For the curious,

The Full equation (R^2= .852) had nine significant variables and its general form, without β and coefficients was: Success = - (Wins vs. ranked opponents) + - (Win%) + - (Total final fours) + Percent of wins away + Percent of wins home + - (Junior rebounds) + - (Freshmen blocks) + - (Sophomore rebounds) + - (Senior rebounds). Notable in this equation is the lack of any ranking variables (Seed was excluded on purpose).

Reilly
03-18-2015, 08:40 AM
In my past experiments, no particular prediction/ranking system stood out for making NCAA tourney brackets. I decided to ask my friend Google, and here's what he said:

According to Team Rankings (http://www.teamrankings.com/blog/ncaa-basketball/ncaa-tournament-seeding-impact), from 2004 to 2014, the coefficient of determination (r-squared) when investigating the correlation between Kenpom rating and NCAA tourney wins was 0.32. For the correlation between tournament seed and NCAA tourney wins, r-squared was 0.36.

Google went on to point out that, when predicting the full bracket prior to the tournament start, the percentage of games predicted correctly looks like this:


Source
Years
KenPom
RPI
Sagarin
Seed


1*
2007 to 2012
66%
63%




2
2004 to 2012
66%


67%


3
2010-2011
62%

59%



1
2007 to 2011
65%






Here are the percentages if winners are chosen on a game-by-game basis (instead of filling the complete bracket out prior to the beginning):


Source
Years
BPI
KenPom
RPI
Sagarin
Sag ELO
Sag Pred
Massey
Seed
AP Poll
AP Preseason
LRMC
Vegas Favorite


2
2007 to 2012

73%





73%






4
2007 to 2011
74%

72%
73%










5*
2003 to 2011
70%
73%
70%
71%
72%
71%
70%
71%
70%

75%
72%


6
2003 to 2010

73%
71%




72%

72%




7
2004 to 2007

73%
71%
72%


73%
73%






8*
unk


73%







72%




For details and methodology, here are the sources:
1 The Harvard Sports Analysis Collective (link (https://harvardsportsanalysis.wordpress.com/2013/02/19/rpi-prediction/), link (https://harvardsportsanalysis.wordpress.com/2012/03/14/survival-of-the-fittest-a-new-model-for-ncaa-tournament-prediction/)) *kenpom was much better than RPI at predicting F4 and Champion
2 BracketScience.com (link (http://wp.bracketscience.com/?p=418))
3 Toutkoushian paper (link (http://kb.osu.edu/dspace/bitstream/handle/1811/48883/EmilyToutkoushian.pdf?sequence=1))
4 ESPN (link (http://espn.go.com/mens-college-basketball/story/_/id/7561413/bpi-college-basketball-power-index-explained))
5 LRMC (link (http://www.sloansportsconference.com/wp-content/uploads/2012/02/14-MIT-Sloan-Paper_BrownKvamNemhauserSokol_2012-Extended-Abstract.pdf)) *BPI estimated based on 3 year performance
6 FiveThirtyEight (link (http://fivethirtyeight.blogs.nytimes.com/2011/03/11/in-n-c-a-a-tournament-overachievers-often-disappoint/?scp=1&sq=pomeroy%20ratings&st=cse&_r=0))
7 Cripe senior project (link (http://users.manchester.edu/Student/adcripe/math%20senior%20project%20paper.pdf))
8 Net Prophet (link (http://netprophetblog.blogspot.com/2012_10_01_archive.html)) *RPI includes a home court advantage given to the higher seed

Of special note, Ms. Toutkoushian (source #3) did regressions of a ridiculous number of variables on data from 1986 to 2009. She then used the resultant equations to predict the 2010 and 2011 tournaments. She surprised herself when her "full equation" correctly predicted 62 out of 63 games in 2010 (apparently completed as a bracket, not a game-by-game prediction)!!! She only missed Kansas' loss to N. Iowa. For the curious,

Is this sort of like the eye test?

budwom
03-18-2015, 08:47 AM
impressive work, which at quick, non deep analytical glance would seem to support the notion that despite all the layers of (interesting) detail, the
deep analytics do not predict victory any more than the other methodologies. So I guess it's not worth getting lathered up
about whether, in his ranking, KenPom has Gonzaga ahead of or behind Duke....(for example).

MChambers
04-03-2015, 11:33 AM
Interesting article about this Kentucky team, its apparent dominance, 1976 Indiana, and other teams.

http://fivethirtyeight.com/features/this-years-kentucky-team-is-more-dominant-than-indianas-undefeated-1976-squad/

cato
04-03-2015, 11:39 AM
Interesting article about this Kentucky team, its apparent dominance, 1976 Indiana, and other teams.

http://fivethirtyeight.com/features/this-years-kentucky-team-is-more-dominant-than-indianas-undefeated-1976-squad/

Boy, is it time for basketball or what. That article started off by saying how they could compare IU to Kentucky, but that they didn't really have time to input all the data. Was that going to stop them from writing a piece? Heck no!

NSDukeFan
04-03-2015, 12:02 PM
Interesting article about this Kentucky team, its apparent dominance, 1976 Indiana, and other teams.

http://fivethirtyeight.com/features/this-years-kentucky-team-is-more-dominant-than-indianas-undefeated-1976-squad/

I thought the chart was interesting showing how few of the most dominant teams according to this statistical model actually won the tournament. This makes some sense in that you are just about always better off picking the field.

vick
04-03-2015, 12:23 PM
Interesting article about this Kentucky team, its apparent dominance, 1976 Indiana, and other teams.

http://fivethirtyeight.com/features/this-years-kentucky-team-is-more-dominant-than-indianas-undefeated-1976-squad/

Interesting stuff. Also, we rightly celebrate 7 Final Fours in 9 years, but the 98-02 run is just ridiculous.

yancem
04-03-2015, 12:42 PM
How is Duke 1992 not on that list???

vick
04-03-2015, 01:15 PM
How is Duke 1992 not on that list???

It is at +24.8 on SRS, and the cutoff was +26. The 1992 team gave up a load of points, both because it often got leads and relaxed, but also not infrequently in other games. You give up 93 points in regulation to a team whose second leading scorer is John Pelphrey, that's going to hurt your metrics.

MChambers
04-03-2015, 01:25 PM
How is Duke 1992 not on that list???
I had the same reaction. 1992 Duke was an awfully good team, regardless of what the stats say.

Listen to Quants
04-03-2015, 01:44 PM
Interesting article about this Kentucky team, its apparent dominance, 1976 Indiana, and other teams.

http://fivethirtyeight.com/features/this-years-kentucky-team-is-more-dominant-than-indianas-undefeated-1976-squad/


How is Duke 1992 not on that list???


It is at +24.8 on SRS, and the cutoff was +26. The 1992 team gave up a load of points, both because it often got leads and relaxed, but also not infrequently in other games. You give up 93 points in regulation to a team whose second leading scorer is John Pelphrey, that's going to hurt your metrics.

Maybe we could merge some of these ideas with the "stall ball" threads? If K is more dedicated to trading points for time than most coaches then Duke wins by fewer points than it would with less stall-ball. Hence some reduction in power calculated. By this argument, Duke's real measure should show up best on an ELO or similar win-based system.

MChambers
04-03-2015, 02:07 PM
Maybe we could merge some of these ideas with the "stall ball" threads? If K is more dedicated to trading points for time than most coaches then Duke wins by fewer points than it would with less stall-ball. Hence some reduction in power calculated. By this argument, Duke's real measure should show up best on an ELO or similar win-based system.
I agree with this point, but I also think that K, more than most coaches, stresses all out hustle for 40 minutes, if the roster is reasonably deep, and that may make for a lot of blowouts.

Listen to Quants
04-03-2015, 05:57 PM
I agree with this point, but I also think that K, more than most coaches, stresses all out hustle for 40 minutes, if the roster is reasonably deep, and that may make for a lot of blowouts.Right you are. These, and perhaps other effects, show up later in games. I've always thought it would be interesting if one of the successful quantitative systems (measured in prediction accuracy) like Sagarin/KenPom/BPI would be repeated but with the score at the 30 or 35 minute mark substituted for the final to see if anything much different fell out.

grateful_duke
04-04-2015, 02:32 AM
Pretty amazing that only 7 of those teams actually won the National Championship.

So many "ughhh's" for Duke on that list.

'98
'99
'02

Some of the greatest college basketball teams ever that didn't win the N.C.

juise
04-05-2015, 12:49 AM
Duke at 4th in KenPom (3rd on O, 12th on D). If only we'd had a really good season like #1 UK or #2 AZ. We'll get a shot at #3 on Monday.

uh_no
04-05-2015, 01:00 AM
Duke at 4th in KenPom (3rd on O, 12th on D). If only we'd had a really good season like #1 UK or #2 AZ. We'll get a shot at #3 on Monday.

the turn around since the loss to ND is historically great. Win or lose, K has redefined the degree to which a team can turn it around. Where our defense was headed, and where it ended up are like Ardbeg .