Last season, I took a look at what I call “opponent compatibility” in basketball. Check out the full post for all the details, but I basically looked at the idea of specific strengths and weaknesses affecting the outcome of games. For example, what happens when a good offensive rebounding team plays a bad defensive rebounding team? In my piece from last year I laid out the question using an NCAA tournament matchup:
On one hand, Minnesota should kill UCLA on the offensive boards, possibly creating a huge advantage for Minnesota. On the other hand, Minnesota kills just about everyone on the offensive boards. UCLA wouldn’t be able to stop the lethal Minnesota rebounding attack regardless, so maybe this is a waste of an opponent weakness for Minnesota.
I went through thousands of games over the past five seasons to find an answer. This idea of opponent compatibility is often used by basketball analysts all over, but ultimately my results were pretty clear:
Simply put, the best way to predict the winner of a game appears to be just picking the better of the two teams.
I didn’t find any signs of matchups making a difference in game results beyond just picking the best team, but that didn’t stop me from continuing to look. I think sometimes the idea that a team was just a “bad matchup” for another team (after the fact) is a bit of a cop-out, but it would be silly to suggest matchups play no prominence at all.
In general, betting lines tend to follow Kenpom’s efficiency rankings very closely. A key injury or suspension could change the story significantly, but purely using efficiency rankings is about as far from a market inefficiency as possible these days. What follows is a look at factors beyond efficiency rankings that may or may not influence results. I used every single DI game played thru Sunday (1/19) as my sample. The idea here is to:
1) Predict a score for every game simply using adjusted offensive/defensive efficiencies.
2) Identify a specific section of games where a potential “matchup advantage” is occurring.
3) Compare the actual results of the specific games from step two to the initial prediction of these games from step one.
Before I get into the new research, I ran the numbers for home-court advantage. Obviously home-court advantage is a well-established concept, but I think it serves as a useful reference point for comparison later on. Take a look:
This method is able to isolate home-court advantage entirely from the fact that better teams often play non-conference games at home. These findings are pretty consistent with previous home-court advantages studies. From here forward, I will be using this information in all of my “expected points per possession” calculations below (to account for home-court advantage).
Revenge and trap games are pretty common basketball clichés, but lack specific definitions. I did my best to come up with a reasonable formula for determining these things. First, let’s look at revenge games. My first instinct was to simply look at how a team responds following ANY loss:
As I expected, coming off a loss was completely insignificant in determining the following game’s outcome. However, we can probably come up with a better definition of a revenge game. The trade-off here is lowering sample size, but I identified 844 games played (seemed high, but I triple checked) following a loss to a team 100 spots worse in the KenPom rankings. These are true “revenge” games, following a particularly bad loss:
Defense got ever so slightly better (remember, lower is better on defense) in these games, but not close to enough to support the concept of a revenge game.
Trap games are another basketball narrative difficult to define, but I gave it a shot. I looked at games against teams outside the KenPom top 100 when that same team had a game with a KenPom top 25 team next on the schedule. The sample size isn’t as big here, but there were still 239 games with this scenario:
We know for certain that home-court advantage is very really, but even that only had differences of .02 and -.03. That’s what makes this so tough with a relatively small sample size, but in the 239 games teams have actually played better during trap games. This goes against the conventional wisdom of trap games. I guess the counter-argument here would be that coaching staffs are aware of the idea of a trap game and keep their teams extra focused. This is definitely worth updating as the season progresses and sample size increases.
Probably the first thing I think of in terms of basketball matchups is height. Using KenPom’s average height statistic we can identify games with particularly large height mismatches. Small ball is an increasingly large trend in today’s game and I think that makes this analysis very relevant:
I have to continue to stress the importance of sample size here, but the results are certainly interesting. In the 295 games where teams were 2 to 2.5 inches taller in average height than their opponents, they performed particularly well on defense. In the 111 games where there was a huge (>2.5 inches) height mismatch, the “giant” team struggled offensively but excelled defensively well beyond normal expectations. These games don’t happen all that often (particularly in conference play), but it’s an interesting trend in an admittedly small sample.
Close Games (Experience vs. Talent)
The final basketball cliché I covered was the idea that experienced teams are better at pulling out close games down the stretch. For this idea, I had to take a slightly different approach. It’s intuitive, but easier to explain by first simply looking at the charts:
I think this is pretty interesting stuff. We can really see the randomness involved in close games here. The first thing that stands out is how home-court advantage more or less disappears in close games. When looking at all games (regardless of margin of victory),adjusting for home-court advantage only increased prediction accuracy from 79% to 80%. I think this is because of the nature of non-conference play (good teams don’t play road games). However, adjusting for home-court advantage actually loses prediction accuracy in close games (63% to 61%).
We can see that having experience has little to do with winning in general. However, experience was actually slightly more important in close games. I would be interested to see if anyone has done any work on this subject in the past, because I wasn’t expecting this to hold up.
As the season progresses, I’ll be continuing to monitor these and other matchup factors. If you have any ideas of your own for me to take a look at, you can find me on Twitter @hoopvision68. For more tempo-free analysis as well as x’s and o’s, check out my website at www.thehoopvision.com.