What combinations might Kentucky try?
UK's staff is turning to some advanced stats to find lineups and combos that might work. What might they find?
On Monday evening, I noticed an interesting tweet on my timeline:
Obviously, a comment like this is catnip to a numbers geek like me. I cracked my knuckles, opened my laptop, and pulled up the numbers to see what he might be talking about. And lo and behold, I found some stuff. But before I get to that, I wanted to talk a bit about lineup/combo stats.
I’ve been doing data-driven analysis on Kentucky and Louisville basketball for 8 seasons now. When I started, my main goal was to look at how different lineups played and find good ones (and bad ones). But I realized that looking at the data for specific lineups just isn’t very insightful.
There are a few big reasons:
1) 5-man lineups don’t play very much together
Over a full 32-40 game season, a college team will play about 2500-3000 possessions. The most common lineup might play 250-500 possessions together. Past that, there might be 4-5 lineups with 50 possessions or more. It is very rare for a college team to play a single lineup more than 20-25 possessions in a game. Most lineups stats don’t start to stabilize until 50 or more possessions, and they aren’t very predictive even at 200 possessions. There’s just not enough of a sample size to be able to use historical data to predict future performance. This is even a problem in the NBA, where the most common lineups play over 1,000 possessions together.
2) Lineup data doesn’t account for context
Just because Lineup A has better offensive numbers than Lineup B doesn’t mean that Lineup A is better at offense. Lineup A could be playing much more often against bench lineups, or have played more against weaker opponents, or have played against teams who are worse on defense. It is incredibly hard to adjust for any of these factors and honestly it’s probably not worth the effort because of point #1 above. The best that can be done reasonably is to drill down and see if there are any contextual factors to consider.
3) Overall data doesn’t tell you if the lineup is consistent
Imagine that two lineups both average 1.25 points per possessions. That’s an amazing number! But what if one lineup averages that number each game for 10 games in a row, while another averages 0.75 in one game, 1.50 for the next 3 games, 1.00 for the next 2 games, 0.85 for 3 games, and 1.40 for the 10th game? You’d probably rather rely on the more consistent lineup. But just looking at aggregate numbers misses that context.
4) College kids are very random
When in doubt, remember that it’s really hard to predict what a group of 19 year olds will do next. It’s even harder when they are competing against another group of 19 year olds. Players, lineups, and even entire teams have good nights and bad nights. A college coaching staff observing their players in practices every day is probably better equipped to identify the best combinations than a spreadsheet…assuming that coaching staff is actually trying to test, measure, and evaluate combinations.
With all that being said, I try to stick to a few key principles:
Don’t just focus on 5 man lineups; look at 2, 3, and 4 man combos
Look at the underlying fundamental stats to see why combos might be playing well or not
Look at context around who combos have played against and how consistent they are in playing well or not
So, let’s see what Kentucky might find when looking for combinations that are playing well!
#1: Dontaie Allen, Keion Brooks, Isaiah Jackson
I’m pretty confident this is the combo Calipari was talking about in the tweet above. The very best offenses in the country averages about 1.25 points per possession, and when this trio plays together Kentucky is averaging 1.30 points per possession. They are shooting very well (52% from 2, 75% from 3, 62% effective Fg%) and snaring 69% of their own misses. They are scoring wel despite a high turnover rate (23% of possessions), which is a sign that their performance is not qildly unsustainable. The 75% 3 point shooting is an outlier, but it’s 3 makes on 4 shots, so I’m not wildly concerned. This group is also elite defensively, allowing 0.67 points per possessions. Opponents have a 38% eFG% and are turning it over on 23% of possessions, and this group allows almost no free throws.The big issue is they’ve only played 30 possessions total, so we’re in super small sample size territory here. They have played together each of the last 3 games (12 poss vs Florida, 12 vs Alabama, 6 vs Auburn) and outscored each opponent. They have scored at least 1.08 points per possession in each game, and allowed 0.83 or less. If UK is looking to shake things up, play this group more. They have played in 5 different lineups and had success with each other duo (Mintz/Askew, Mintz/Boston, Mintz/Toppin, Askew/Boston, Askew/Toppin), but I’d probably go with Mintz and Askew for ballhandling. Interestingly, this group has NOT played with Olivier Sarr at all!
#2 Davion Mintz, Devin Askew, Dontaie Allen, Olivier Sarr
This group has seen more time together than #1, with 55 possessions played. 54 of those have come since the new year, beginning with the comeback win against Mississippi State where these 4 played a total of 29 possessions. They also played 16 against Vanderbilt, but then played 4 possessions against Florida, 5 against Alabama, and none against Auburn. In the 54 possessions since Allen joined the rotation, this group has outscored opponents by 22 points, including +10 vs Mississippi State, +7 vs Vanderbilt, and +8 vs Florida in 4 possessions. They have hit only 33% of their twos, but 56% of their threes, and have largely avoided turnovers (15% of possessions). Overall this group has scored 1.15 points per possession and allowed 0.74, thanks to opponents hitting 29% of their twos and turning it over on 19% of possessions. This group could do damage against a team that allows a lot of threes, but asking them to sustain 56% shooting from deep is a lot.
Now, that’s a couple combos that you can build a rotation around…let’s look at another group that has some good superficial numbers but might not be reliable, and end with a combo that should be avoided!
#3 Davion Mintz, Brandon Boston, Terrence Clarke, Olivier Sarr
This group won’t see any time until Clarke returns from injury, but they’ve been very good offensively (for Kentucky at least!), scoring 1.09 points per possession. They’ve done so with a variety of other players in that 5th spot (Isaiah Jackson, Lance Ware, Jacob Toppin, even Cam’ron Fletcher) although they haven’t had any time with Keion Brooks or Dontaie Allen. Looking deeper, however, we see that there are a lot of signs that this group may not be great going forward. First, this group’s performance is inflated by scoring 16 points in only 7 possessions against Morehead State. Against competition with a pulse, they’ve averaged just under a point per possession. This group has also been poor defensively, allowing 1.03 points per possession against non-Morehead teams.
#4 Davion Mintz, Lance Ware, Jacob Toppin
Kentucky fans may like the effort of Ware and Toppin, but the fact is that they are very challenged offensively. UK’s offense craters when those two share the court with Davion Mintz, as the Wildcats score a paltry 0.64 points per possession across 77 possessions. They’ve played together in 9 games and have been outscored in 7 of them, posting a +2 in each of the other two games. UK has hit 38% of their twos and a miniscule 5% of their threes, which has completely negated any benefit of a relatively low turnover rate (16% of possessions). Their defense has been about average by UK standards this season, but their complete lack of offense has doomed the Cats. UK’s coaching staff probably regrets the 18 possessions they gave this trio against Auburn, as they were outscored by 6 points and scored just 0.67 points per possession; during the remainder of the game UK was outscored by only 1 point and scored a more reaosnable 0.90 points per possession….not good, but not disastrous.
So, what would a revamped UK rotation look like featuring better lineups? They would probably feature 2 main lineups:
Askew/Mintz/Allen/Brooks/Jackson
Askew/Mintz/Allen/Brooks/Sarr
while working in Brandon Boston in lineup #1 for Askew/Mintz and Toppin/Ware in lineup #2 for Brooks. You’d probably end up with 30-32 minutes each for Askew, Mintz, Allen, and Brooks, 20 each for Sarr and Jackson, 15 for Boston, and 8 or so each for Toppin and Ware. When Terrence Clarke returns, he could get 15-20 minutes in place of Allen/Mintz/Askew, and Cam’ron Fletcher could get some spot minutes as well in place of Ware or Toppin.
I’d expect Kentucky to shake up the rotation majorly. If Kentucky turns to Askew/Mintz/Allen/Brooks/Jackson heavily, they are really leaning in to using the data to try to jumpstart the offense. If they turn to other lineups, they’ve likely got their reasons other than the data. Whatever changes happen, hopefully the players react well and use it to jumpstart the season. If not, I don’t know what other moves are left.