MBB Dork Polls/Stats, 2015-16 Edition

[Troublemaker]

Guess who is #1 in KenPom's preseason rankings?

http://kenpom.com/

[SilkyJ]

Love it!

I'm sure this question has been answered in the past, but how does KenPom do this in the preseason given how little (or zero) data there is? I think of the dork polls as using data/stats to backup their rankings, but no one's played any games yet! Clearly he's using stats on players/teams from the prior year, but I'm curious what sort of assumptions and leaps he's making in order to create his model.

[sagegrouse]

Love it!

I'm sure this question has been answered in the past, but how does KenPom do this in the preseason given how little (or zero) data there is? I think of the dork polls as using data/stats to backup their rankings, but no one's played any games yet! Clearly he's using stats on players/teams from the prior year, but I'm curious what sort of assumptions and leaps he's making in order to create his model.

Because his subscribers don't want to wait until mid-January to get rankings and analysis.

[SilkyJ]

Because his subscribers don't want to wait until mid-January to get rankings and analysis.

Well the man's gotta get paid I suppose, but your point answers "why" he does it, while I'm asking "how" he does it.

[Troublemaker]

I'm sure this question has been answered in the past, but how does KenPom do this in the preseason

Not sure, but he was pretty darn accurate about Duke in the preseason last year. I just checked my archives. KenPom had Duke ranked #1 last preseason (edging out #2 UK), and Duke did end up earning a #1 seed and went on to win the national championship. His projections did underestimate Duke's regular season record, as it predicted a regular season of 25-6 (14-4 ACC) when, in reality, Duke went 28-3 (15-3 ACC), but that just goes to show how well that team played in big games that usually would be losses (e.g. @Wiscy, @UVA).

Hopefully KenPom is as accurate about this year's team as he was about last year's team, and we stay as healthy as that team did.

[Duke95]

Well the man's gotta get paid I suppose, but your point answers "why" he does it, while I'm asking "how" he does it.

This is his explanation:

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

[SmartDevil]

UNCheaters at 7.

[Troublemaker]

UNCheaters at 7.

Maryland at #24, haha

[RobbieG]

Champions Classic looks stacked again as usual. That game will be fun to watch.

[Ichabod Drain]

Looking a little deeper he predicts Duke to have the #1 offense. Not surprising considering he uses the last five seasons of data and Duke has had a top ten offense all five of those years and top five offense four out of five years.

On the other side of the ball he predicts Duke to have the #9 defense. Now as we all know, Duke's defense has been a lot more unpredictable than the offense over the last five years. In fact if we did finish with the #9 defense that would be our best finish since 2010 (#8). I think we have a lot of defensive potential this year but we are still young so it may take some time to find our groove.

I think it's certainly possible that our defense may end up rated higher than our offense, which would be the first time since 2008.

[Nosbleuatu]

Maryland at #24, haha

For what it's worth, Kenpom doesn't include transfers in the "personnel" part of the equation.

His recent stats from Duke probably wouldn't add much to the ranking, though, even if it was included.

[MCFinARL]

Looking a little deeper he predicts Duke to have the #1 offense. Not surprising considering he uses the last five seasons of data and Duke has had a top ten offense all five of those years and top five offense four out of five years.

On the other side of the ball he predicts Duke to have the #9 defense. Now as we all know, Duke's defense has been a lot more unpredictable than the offense over the last five years. In fact if we did finish with the #9 defense that would be our best finish since 2010 (#8). I think we have a lot of defensive potential this year but we are still young so it may take some time to find our groove.

I think it's certainly possible that our defense may end up rated higher than our offense, which would be the first time since 2008.

This may have something to do with the fact that our returning players (who, according to KenPom's explanation, are weighted more heavily than new personnel) are all known to be fairly tenacious defenders--but of course that is just my guess.

[sagegrouse]

This may have something to do with the fact that our returning players (who, according to KenPom's explanation, are weighted more heavily than new personnel) are all known to be fairly tenacious defenders--but of course that is just my guess.

Uhhh... This is educated-guess time. The best teams are easy guesses -- Duke and Kentucky will be real good, as will Virginia and NC, who return a lot of players. But he has 300+ teams to rate!

[Troublemaker]

Luke Winn / SI uses a statistical system to project all Div-I teams (http://www.si.com/college-basketball/2015/11/03/every-team-college-basketball-ranked-north-carolina-kentucky-kansas)

This system projects Duke to be a 1 seed (ranked #4 overall) with #3 offense and #37 defense

[Troublemaker]

Luke Winn / SI uses a statistical system to project all Div-I teams (http://www.si.com/college-basketball/2015/11/03/every-team-college-basketball-ranked-north-carolina-kentucky-kansas)

This system projects Duke to be a 1 seed (ranked #4 overall) with #3 offense and #37 defense

In the video accompanying this article, Luke Winn explains that even though UNC is the #1 team in the system this season, their team would only have been ranked #8 if inserted into last season's field.

A little bit more grist for the "college basketball is weak this season" mill

[MChambers]

In the video accompanying this article, Luke Winn explains that even though UNC is the #1 team in the system this season, their team would only have been ranked #8 if inserted into last season's field.

A little bit more grist for the "college basketball is weak this season" mill

Interesting that UNC's defense is projected as #45.

[luvdahops]

Luke Winn / SI uses a statistical system to project all Div-I teams (http://www.si.com/college-basketball/2015/11/03/every-team-college-basketball-ranked-north-carolina-kentucky-kansas)

This system projects Duke to be a 1 seed (ranked #4 overall) with #3 offense and #37 defense

Interesting. That seems somewhat counter to the expectations of most on this board (e.g. we had an entire thread dedicated to "Where Will the Points Come From?", along with numerous posts alluding to this squad's defensive potential).

[Billy Dat]

The CBS web guys (Parrish, Norlander, Vecenie) had a podcast last week where they mentioned the predictive accuracy of the AP Preseason Poll.

http://sports.cbsimg.net/swf/audio/AudioPlayer.swf

[Troublemaker]

Interesting. That seems somewhat counter to the expectations of most on this board (e.g. we had an entire thread dedicated to "Where Will the Points Come From?", along with numerous posts alluding to this squad's defensive potential).

For sure. And Pomeroy's system predicted Duke to have a better offense than defense as well, albeit without such a drastic spread (#1 offense, #9 defense in KenPom.)

I would guess these predictive systems heavily weigh recent seasons in their formulae, and so if the current narrative (defense > offense) is very different from what has occurred in recent seasons, it's almost impossible for these statistical systems to match the narrative.

You know, if I HAD to bet the house on one or the other, i.e. choose between (offense > defense) or (defense > offense), I think I would lean towards Duke finishing this season with a higher ranked offense. I think it has a chance to be very close, but it's tough to go against history. Duke under Coach K over the past decade or so has had killer offenses.

[MChambers]

I think it has a chance to be very close, but it's tough to go against history. Duke under Coach K over the past decade or so has had killer offenses.
That and no so great defenses, with the exception of 2010. Of course, last year's team got the defense going in the NCAA tourney, so that was fine.

[flyingdutchdevil]

That and no so great defenses, with the exception of 2010. Of course, last year's team got the defense going in the NCAA tourney, so that was fine.

The personnel makes this a defensive team from the get go, however. We return 2 of our top 3 defenders from last year (AJ and Jones) and add two players who have the length and athleticism to be disruptive from day one. Hell, even Derryck comes in with a defensive reputation.

I'd argue, as would half the team (through interviews), that this is a defensive team right now. As many have commented, that could easily change and probably will. But I think it's a safe assumption that our D is ahead of our O right now.

[MChambers]

I've had that thought, and then I remember that I've thought it several times in recent years and been wrong. So I'm from Missouri on this team's defensive chops. Show me.

Hope you're right and I'm wrong.

[flyingdutchdevil]

I've had that thought, and then I remember that I've thought it several times in recent years and been wrong. So I'm from Missouri on this team's defensive chops. Show me.

Hope you're right and I'm wrong.

Oh I agree. It certainly remains to be seen. But on paper, this team should be good (maybe even great) on defense (and suspect on offense). However, paper also told us that Winslow would be a good Swiss Army Knife, but a good-looking legend with a James Harden-esque Euro step and the defensive chops to also stop James Harden.

[ChillinDuke]

It appears StatSheet.com, which used to list a multitude of interesting stats/metrics across all of D-1 hoops teams, has been acquired and/or is no longer free to peruse.

Does anyone know of a different site which lists similar information (pts per possession, possessions per game, etc)? Preferably free being the key. But if the price is reasonable I'd consider paying.

Thanks!

- Chillin

[Troublemaker]

It appears StatSheet.com, which used to list a multitude of interesting stats/metrics across all of D-1 hoops teams, has been acquired and/or is no longer free to peruse.

Does anyone know of a different site which lists similar information (pts per possession, possessions per game, etc)? Preferably free being the key. But if the price is reasonable I'd consider paying.

Thanks!

- Chillin

Sports-Reference (http://www.sports-reference.com/cbb/schools/duke/2016.html) is your best bet for free advanced stats for college basketball now that StatSheet is gone. (If someone is aware of a better site, please inform us, of course.)

I do recommend paying $20 for a full year of KenPom, though.

[ChillinDuke]

Sports-Reference (http://www.sports-reference.com/cbb/schools/duke/2016.html) is your best bet for free advanced stats for college basketball now that StatSheet is gone. (If someone is aware of a better site, please inform us, of course.)

I do recommend paying $20 for a full year of KenPom, though.

Thanks, Trouble. Much appreciated.

On a different note, we bumped up a bit in KenPom's defense to #28 in all the land. Seems to make sense to me given how we looked last game.

I'll offer a third note, KenPom currently lists the Top 5 offenses as ACC schools. Pretty interesting. Pitt is the next ACC school on the list at #14. Then no one til L'Ville at #32 and FSU at #40. By comparison, the only ACC defense in KP's top 10 is UVA at #10. Then no one til Cuse at #22 (then a smattering from #26-41).

- Chillin

[Olympic Fan]

Just for fun, I visited ESPN's RPI site earlier today.

It's fascinating to see how this is shaping up -- wa-a-ay too early to start taking this seriously. But, oddly, Duke leaped from No. 27 to No. 11 in the RPI after beating Utah State.

On the other hand, the top three teams in the rankings at the moment are 1. Valparaiso; 2. Davidson, and 3. Louisville. UNC is No. 8 and Kentucky is No. 16. Our next opponent is Indiana, which currently ranks No. 87 (although they may move up a bit after beating Alcorn tonight ... a bit, not much).

Of course, none of this matters at the moment. The RPI will change a lot in the next two-three months. Even then it is flawed, but it's still the committee's No. 1 resource. Here's the link (I think it is free):
http://espn.go.com/mens-college-basketball/rpi/_/sort/RPI

There's another link on that page to a pay article titled Bracket projections. Can't link, but it explains that its projections are based on the formula used by teamrankings.com.

The current projections are crazy -- Kansas is the current favorite to win the national title (10.3 percent chance), which isn't that bad. But they have Kentucky at No. 11 (4.6 percent), just ahead of No. 12 Duke (4.2 percent) -- and behind such teams as Gonzaga (6.1 percent), Purdue (5.4 percent), West Virginia (5.1 percent) and Butler (4.8 percent)

Now, I think 4 percent is a reasonable projection of Dukes chances ... but I gotta think Duke (and Kentucky) are better bets than West Virginia or Butler!

FWIW, the current Vegas odds favorite Kentucky (5-1), but Duke is the second choice (9-1), followed by Maryland, UNC and Virginia (all at 10-1). These change every day (as of Sunday, Duke was in the 10-1 group):

http://www.vegasinsider.com/college-basketball/

[Troublemaker]

From Luke Winn's power rankings (http://www.si.com/college-basketball/2016/01/07/power-rankings-kansas-oklahoma-michigan-state-north-carolina) column today:


...the teams atop this year’s kenpom.com (http://www.si.com/college-basketball/2016/01/07/kenpom.com) efficiency rankings are lagging well behind their counterparts from 2014–15. The top four teams from Jan. 6 of last season—Kentucky, Virginia, Duke and Wisconsin—ere all stronger, efficiency-wise, than the No. 1 team as of Jan. 6 of this season, Villanova.

The advanced stats support the theory that there are no elite teams this season, although the eye test suggests that Kansas, Oklahoma and Michigan State, at their peak levels, are still pretty damn good.

http://www.si.com/sites/default/files/2016/01/07/pr_1_7_16_michstate.gif

[House P]

There were a few KenPom-related comments in the UVA and UNC game threads. I wanted to reply with a fairly dorky response, so I thought it might be better to use the “Dork Polls” thread to respond.


Does anyone recall what our worst defensive efficiency kenpom ranking has been this season? I believe it was somewhere in the 140's? Right now we're sitting at 94. Loving the trajectory right now. Of course sustaining it may may hinge a lot on Jones' ankle.


for what it's worth...duke gave up a 105.7 on defense...which adjusted is...

the third straight game in a row duke has done <90 on defense.

duke has played the best defense in the country 3 games in a row.


Many thanks to uh_no for pointing this out this trend. Hopefully the trend continues despite Matt’s injury.

That being said, a sub 90 adjusted defensive efficiency for the UVA and Louisville games may be a bit optimistic. Apologies if I am nitpicking, but it looks like uh_no didn’t factor home court advantage or normalize for the average D1 team’s efficiency. For a road game, like Duke @ UNC, these adjustments essentially cancel each other out. But for a home game, these extra adjustments result in an efficiency a bit higher than what uh_no calculated.

Based on my understanding of KenPom’s method (and my understanding could certainly be wrong, see the bottom of this post for more info), Duke’s adjusted defensive efficiency over the past 3 games is approximately

90.6 @ UNC
94.9 vs UVA
94.3 vs Louisville

That averages to 93.3, which is still excellent as uh_no noted, but would “only” be good enough for #15 overall.

For kicks, here’s an estimate of Duke’s game by game adjusted efficiency this season.



Game
Game AdjD
Rank
Season AdjD


Siena
94.2
19
94.2


Bryant
115.2
342
104.7


Kentucky
97.4
55
102.3


VCU
93.5
16
100.1


Georgetown
117.9
350
103.6


Yale
90.5
4
101.4


Utah St.
74.5
1
97.6


Indiana
109.9
297
99.1


Buffalo
84.6
1
97.5


Ga Southern
94.9
27
97.2


Utah
89.0
1
96.5


Elon
87.6
1
95.8


Long Beach St.
104.5
197
96.4


Boston College
96.7
52
96.4


Wake Forest
103.4
173
96.9


Virginia Tech
80.4
1
95.9


Clemson
98.5
71
96.0


Notre Dame
114.7
342
97.1


Syracuse
107.1
243
97.6


NC St.
106.2
226
98.0


Miami FL
111.3
318
98.7


Georgia Tech
93.5
16
98.4


NC St.
109.0
281
98.9


Louisville
94.3
20
98.7


Virginia
94.9
27
98.5


North Carolina
90.6
4
98.2



Some explainations:

"Game AdjD" is an estimate of Duke's adjusted defensive efficiency for that game, based on each opponent's Feb 18 KenPom rating.

"Rank" is where a particular game's adjusted D would rank nationally based on KenPom's Feb 18 overall ratings.

"Season AdjD" is the average of the individual AdjD values through that game. For example, the Season AdjD for the Buffalo game (Amile's last game) is 97.5. That means that Duke's average adjusted defensive efficiency for games 1-9 was 97.5.

A Couple Notes:

- As uh_no pointed out, Duke’s adjusted defensive efficiency has been excellent over the past 3 games. Looking back a bit further, it has been pretty solid (96.5) since the Ga Tech game (extra credit to Coach Capel?). A season long rating of 96.5 would rank in the top 50 in the country. Other teams with similar defensive ratings are 20-6 Iowa, 20-6 Gonzaga, and 21-5 South Carolina. A team with a top 50 defense and the #2 offense, Duke’s current season long rating, would a top 5 team overall.

- Duke’s worst stretch in terms of adjusted defensive efficiency was the 4 games from ND-Miami. For these games, Duke’s adjusted efficiency was 109.7. A season-long rating that poor would be somewhere around #300 in the country. Teams with a current 109.7 season long rating include 6-20 Delaware State, 9-18 Kennesaw State, and 5-22 Mississippi Valley State. Yikes.

- Prior to the UNC game, Duke’s best defensive performance against a NCAA tourney caliber team was probably the loss to Utah. This was also Duke’s worst offensive performance from an adjusted efficiency performance.



we gave up a 105 [defensive efficiency] for the game, which doesn't sound great, as it's over 100 and our current average...but UVA is currently a 117+ offense. That means that our performance was fantastic to keep them so far below their average. How fantastic?

105*100/117 = 89


My understanding (which could be wrong and I would be happy to be corrected) is that KenPom makes adjustments based on something like the following formula

Adjusted Defensive Efficiency = DE/OppAdjO * HCA * AvgD1Efficiency

Where,

DE = a team’s unadjusted defensive efficiency for a game (104.3 for Duke vs UVA)

OppAdjO = the opponent’s adjusted offensive efficiency for the season (117.2 for UVA)

HCA = the home court advantage (~1.03 for a home game, ~1/1.03 for a road game, and 1 for a neutral site)

AvgD1Efficiency = the average efficiency across D1 for the season (103.5 for the 2015-16 season)

Thus, Duke’s adjusted defensive efficiency vs UVA would be approximately

(104.3/117.2) * 1.03 * 103.5 = 94.9

KenPom also adjusts for blowouts vs weak teams and places greater weight on recent games. I don’t have a good sense for how he makes these adjustments, so they aren’t included in the table above. However, this almost certainly means that KenPom’s adjusted efficiency would be higher than what I list for Duke’s early season blowouts. It also may explain why the current season long AdjD listed above, 98.2, is a bit lower than the 99.6 rating KenPom currently lists for Duke on his site.

Early in the season, KenPom also incorporates data from previous seasons. That isn’t relevant now, but does explain why our early season defensive rating on KenPom's site wasn't too bad despite several games (Georgetown, Bryant) that look pretty bad in retrospect.

Anyone know if I am missing anything else?

Finally, for anyone interested, here is what KenPom says about his method: http://kenpom.com/blog/index.php/weblog/entry/ratings_explanation