Started at Wisconsin. The Presbyterian and Fairfield games would destroy the correlation, and I think it's fair to omit them since a freshmen-led Duke team was almost definitely focused for their first two college game. Then I figured I might as well start at Wisconsin, where the phenomenon began.

My conclusion is that we should avoid playing teams ranked between #50 and #75 in KenPom at all costs.

Seriously, though, with the exception of Elon, you could have a pretty nice V-shaped graph with a quadratic fit.

That's an interesting twist--perhaps Duke overlooks teams that are actually good enough to do damage to them, but knows how to put away a real bum.

I'm getting "The extension csv is not allowed." when I try to attach the .csv file.

Below is the data for Kenpom ranking as of game date, which is the only annoying part. Everything else is on the Statsheet.com page for Duke schedule (spreads and scores): http://statsheet.com/mcb/teams/duke/schedule

Presbyterian 321
Fairfield 202
Michigan State 18
Temple 108
Stanford 34
Furman 311
Army 156
Wisconsin 4
Elon 255
Connecticut 37
Toledo 97
Wofford 92
Boston College 131
Wake Forest 120
NC State 55
Miami 74
Louisville 8
Pittsburgh 91
St. John's 36
Notre Dame 12
Virginia 2
Georgia Tech 83
Notre Dame 14
Florida State 128

We can't ignore Elon, though. That's the exact kind of game we're most interested in - terrible opponent right after the biggest game of the season to that point (Wisconsin). I also ran the data starting with Michigan State, so only omitting Presbyterian and Fairfield, and the r-squared still seemed significant; it was slightly greater than 0.2, compared with the 0.29 starting with Wisconsin.

These posts display the entire depth of my stats ability, so I'm sure others here can take this further. I had to Google "what is a good r^2?"

SeanMayTriedToEatMe said:

We can't ignore Elon, though. That's the exact kind of game we're most interested in - terrible opponent right after the biggest game of the season to that point (Wisconsin). I also ran the data starting with Michigan State, so only omitting Presbyterian and Fairfield, and the r-squared still seemed significant; it was slightly greater than 0.2, compared with the 0.29 starting with Wisconsin.

These posts display the entire depth of my stats ability, so I'm sure others here can take this further. I had to Google "what is a good r^2?"

Click to expand...

The games following the 4 "big wins" were Elon (Wisc), Pitt (L'Ville), Gtech (UVA), FSU (ND).

I think we would all agree (and the numbers back it up), that Duke has played 6 bad games this year- Elon, Wake, Gtech, FSU, State, Miami. So three of the bad six were after big wins. This is probably not coincidence at this point.

I don't really think losing to ND by 4 on the road should be lumped in with those other 6. I'm pretty sure that was a 2 point loss prediction at the time on KP, regardless of the spread moving in Duke's favor before gametime.

Are you using current KP ratings or ratings at the time of the game?

Any way I see a flaw in how the study is designed. First of all if you're trying to measure the team's perception of an opponent, it has to be at the time of the game. 2ndly I'm not sure the team is basing it's perception of an opponent based on their KP ratings, so a national ranking, program prestige factor probably plays more into in rating than KP.

The dumb eye test narrative is even more startling, so if someone like Goodman wanted to get a lot of clicks in a parallel universe where everyone loved Duke, he could write a great dumb article about this.

Duke has been the underdog in 3 games this season, its 3 biggest games: at Wisconsin, at Louisville, at Virginia. Look at how they won those games, and you can spin a good story. Dominated Wisconsin and Louisville start to finish. Dominated Virginia at the start of the game, then hit a lull, and then when their backs were against the wall with something like a 1% win probability, they played to their maximum potential and could not be stopped even by an elite defense.

Duke lost their 4th biggest game of the season, at Notre Dame, when Duke clearly appeared to be the better team. Tons of missed non-difficult layups that 65%+ layup shooters don't miss. Tons of circus shots for Notre Dame going in. The Duke players must have been pissed. Then, not too long afterward, with Duke focused on revenge, they did terrible things to Notre Dame that no other good team has been subjected to this season.

The games that felt big but turned out not to be were against Michigan State and UConn, and Duke dominated from wire to wire in those games. Add in all the game-after-big-game stuff STPFS said. Assume Presbyterian and Fairfield are anomalies because Duke's Big Two Freshmen plus Okafor had never played a college game before and wanted to live up to the hype.

There's your click-bait column in a Duke-loving world.

LastHearth said:

Are you using current KP ratings or ratings at the time of the game?

Any way I see a flaw in how the study is designed. First of all if you're trying to measure the team's perception of an opponent, it has to be at the time of the game. 2ndly I'm not sure the team is basing it's perception of an opponent based on their KP ratings, so a national ranking, program prestige factor probably plays more into in rating than KP.

Click to expand...

1. I don't know how much more clear I could've been about using Kenpom ratings at the time of the game. It's in the chart itself as a label of the axis.

2. Of course there are flaws. If you have a way to accurately measure "program prestige factor" or "perception of opponent by Duke players," I will use that data instead of Kenpom ratings at the time of the game. Using AP/Coach's polls rankings at the time of the game is going to give pretty much identical results, except there will be about 10 data points, since not all 300+ teams are ranked, unlike on Kenpom.

LastHearth said:

Are you using current KP ratings or ratings at the time of the game?

Any way I see a flaw in how the study is designed. First of all if you're trying to measure the team's perception of an opponent, it has to be at the time of the game. 2ndly I'm not sure the team is basing it's perception of an opponent based on their KP ratings, so a national ranking, program prestige factor probably plays more into in rating than KP.

Click to expand...

I don't think KP prediction data can be viewed retroactively. In my post on how pace has been correlated with performance relative to expectations a couple of pages ago, I used a known Vegas spread prior to the game as the "how Duke should perform" baseline. I think this is the best you can do in order to approximate overall perception of opponent quality/how the game should go at the time of tipoff. And I actually think this is what SM did, based on his Notre Dame (road) baseline. It usually doesn't deviate from KP by more than a basket, and it's usually less.

There's obviously no way to account numerically for the intangibles that you reference above though. Even if SM used current KP predictors, I don't think it would move the needle all that much and change the overall point. I can't think of a team on here that's likely to have moved drastically in the rankings (maybe ND downward like 6-7 spots)?

Retract part of Point #1 based on intervening SM post, I read quickly and didn't really look carefully at the axes.

Okafor being 18th in BPM is actually great. The annoying part is Marshall being 7th and getting no playing time unless K is trying to protect Okafor from foul trouble, because K believes Okafor is so much more valuable to the team than Marshall.

This should be good next season, though, when Marshall gets 30 minutes over a non-defensively-aware freshman, Jeter, and a random fat guy, Obi. Full confidence in this happening.

NBA top 10 in BPM this year is Russ, Curry, Harden, Davis, Paul, LeBron, Lillard, Gobert, Marc, Chandler. Danny Green is 16th, Korver 18th. Looks about right to me.