Tag: stats

Packing it in: the Chad Holderfield Question

Chad Holderfield doesn’t know who I am, and we will probably never meet, but his tweet was the influence for this blog post. Well, Jordan Sperber deserves some credit too. Chad’s tweet was in response to Sperber, who, by the way, is a fantastic follow for any basketball fan.

Sometimes the best ideas come organically while scrolling through social media. Thanks, Jordan and Chad!

Sperber’s original post was about teams that basically disregard the three-point line on defense but are still able to prevent their opponent from scoring. In other words, lockdown defenses that don’t really guard the arc. You can (and should) read the whole piece here. The 18-19 Virginia Tech team was an extreme case of this, and was the focus of his post.

Chad then wondered if these teams are better at defensive rebounding. Presumably these are teams that pack it in and protect the paint, so that would make intuitive sense. Packing it in means staying between your man and the hoop, which should leave you in good rebounding position. Of course, as Sperber pointed out in our thread, zone-exclusive teams like Syracuse are also likely to fit the mold, and the Orange certainly have trouble boxing out.

That being said, man defense is much more common than zone defense, so I still expect teams that give up threes to rebound better on the whole. That probably means we’re heading towards one of those blog posts where I pretty much find out exactly what we thought was true is, in fact, true, and no one learns anything new. (I must’ve been sick the day we learned about hooking the reader in Englsh class.) Nonetheless, curiosity prevails, and besides, I can’t let Chad down!

For starters, let’s define which specific teams we’re talking about here. I let Chad pick the arbitrary cutoffs, so if you don’t like them you can direct all hate mail his way.

We are going to define the lockdown defenses as those that ranked in the top 25 in KenPom’s adjusted defensive efficiency for that season, and teams that don’t guard the arc as teams in the top 150 in highest opponent three-point rate. There have been somewhere between 7 and 12 of those teams every year in KenPom’s database (2002-19) for a total of 160.

Of course, top-25 defenses are going to be better at defensive rebounding than average teams whether they prevent threes or not, so let’s start by comparing the elite defenses that guarded the arc to the elite defenses that didn’t. Because I hate long and awkward-sounding names and I’m too lazy to come up with a better solution than made-up acronyms, I’ll call the arc-guarders AGs and the non-arc-guarders NAGs. Here are the results:

The NAGs did rebound better than the AGs, but the gap wasn’t quite as big as I thought it would be. In fact, in 7 of the 18 seasons the AGs actually rebounded better. Over half of the Syracuse teams in this timeframe are included, but that still only accounts for 10 of the 160 NAGs. The overall average for NAGs was 69.95%, and for AGs it was 69.42%. That is close enough that it’s time for some good ol’ significance testing. A two-sample t-test gives us a p-value of .00007, so the difference is real. If you are new to this blog or to statistics, that means that our sample is big enough that the small difference in rebounding percentages matters quite a bit.

Now that we’ve confirmed that elite defenses that still give up threes really do gain a rebounding edge, let’s forget the arbitrary groupings and look at this relationship among all defenses. The r² value for the correlation between opponent three-point rate and defensive rebounding percentage is just over 0.25, which means over a quarter of a team’s defensive rebounding percentage can be explained by how often it gives up three-point looks. With 6,157 team-seasons in the KenPom database, 0.25 is extremely meaningful. Here is a plot of all 6,157 teams.

The blue line is the trendline, showing the increase in defensive rebounding percentage as teams allow more threes. Those logos are just for kicks and giggles–all 18 of Jim Boeheim’s Syracuse teams are below the trendline, and all of Tony Bennett’s teams are above it. Boeheim and Bennett are probably the most well-known zone and packline coaches, respectively. If I had either a Sperberesque attention to detail or a bunch of interns to do my bidding, I’d go through all of these teams and separate them into bins based on defensive scheme–the zones, the packlines, the denials, etc. Instead I’ll just leave that idea dangling to keep you up at night.

There’s another Bennett I want to spotlight before I wrap up. Randy Bennett of Saint Mary’s stands out as a coach who doesn’t like to choose. He was probably that kid who took one of every color sucker at the bank, because he couldn’t choose just one. While his colleagues are choosing whether to give their opponents open looks from deep or second-chance opportunities, he is taking away both. Here’s the chart with Randy Bennett’s data points on it:

There you have it. Taking away threes does seem to come at a rebounding cost, although it’s certainly possible to be good at both. That tends to be the beauty of basketball. There are endless different styles of play, and they all involve giving up something to improve something else, but it’s always possible to be good at both with the right combination of skill, effort and coaching. While most teams that run their opponents off the three-point line will have to live with the offensive rebounds, there will always be Randy Bennetts that find ways to get the best of both worlds.

Follow me on Twitter at @cwetzel31 for more hoops content.

Back, back, back it up!

If you follow men’s college basketball, you’ve surely heard by now that the NCAA is finally moving back the three-point line. That won’t take effect in D2 and D3 until 2020, but D1 will use the new line this season. The move is from 20’ 9” to 22’ 1 ¾” (or 6.75 meters–the distance for FIBA and other international competition).

There has been some debate and discussion as to whether this move will matter a whole lot. Since it’s summer, and college basketball news is slow, it seems like a good time to dig into the impact this might have on the game. The NCAA used the FIBA line for the last two NITs, so we have 62 games-worth of a sample to look at. I wouldn’t be surprised if the move creates a domino effect and we see lots of subtle changes–creating more spacing could open up the lanes for drives and give us a boost in two-point percentage, for example–but if I looked at all of them we’d be here until the start of the Maui. For now let’s just focus on the two biggies: three-point percentage (3P%) and three-point rate (3PR, the percentage of a team’s field goal attempts that come from behind the arc).

It’s important to note that a million factors affect both of these numbers in any given game or set of games. How many threes are taken and made is partially a function of the defense, the venue, the score of the game, what the players ate for breakfast…you get the idea. Since I’m just one guy, I just looked at these two numbers for all offenses in both the regular season (including conference tournaments) and in the NIT.

Let’s start with the basics.

We can clearly see that teams made less threes in the NIT the last two seasons. Teams took less threes in 2018 as well, but last year the three-point rate remained about the same in the NIT.

But what if the teams that made the NIT in one or both of these years were worse than average teams at shooting threes? What if they were better? Let’s do it again with the regular season stats from just the 32 NIT teams.

The gap between regular season shooting and NIT shooting is actually even bigger when we look at just the regular season stats from the 32 NIT teams. Of course, shooting in the NIT in general is not necessarily going to be the same as regular-season shooting. It is well-documented that percentages go up throughout the year as players shake off the rust, so the NIT percentages probably would be higher with no change in arc distance. Let’s include 2017 (the last NIT with the same three-point line as the regular season) as a control.

Granted, when dealing with 31-game tournaments the sample is small enough to include a lot of noise. But it looks nice, and isn’t that the only thing that really matters? We can see that 3P% and 3PR both went up in the 2017 NIT, and on the whole this increase is fairly typical if we include previous years. So when we look at 2018 and 2019 compared to 2017, we are really seeing a decrease in two quantities that would have been expected to go up. All the caveats considered, it seems like we can expect a slight dip in three-point rate and three-point percentage when the season kicks off in November.

The last time the three-point line moved back was the 08-09 season, and we saw the same thing then. Three-point shooting did go down for a while, but eventually it stabilized to a pre-08 level as players improved and coaches adjusted the offenses they ran. Three-point shooting varies so much from game to game that the effects might not be discernible by what announcers like to call the “eye test,” but it will be fascinating to watch how both offenses and defenses adapt to the added distance and spacing. There may be some growing pains early on, but I foresee teams figuring it out in the next couple of seasons.

Do Teams Create Their Own Luck?

You’ve probably heard an announcer say “sometimes it’s better to be lucky than good” when someone does something crazy like banking in a three at the shot-clock buzzer. Of course, there are also times, more of them actually, where it is better to be good than lucky. But what about teams that are both? Or neither? Can being good make a team lucky? Is it possible to manufacture your own luck?

The What

Anyone familiar with Ken Pomeroy’s website has probably noticed that each team has a luck rating. And anyone not familiar with his website probably has no interest in my blog. For those who have noticed this rating but are unsure what exactly it measures, I was going to type up this obnoxiously long explanation (teaching is my day job–sometimes I can’t turn it off). Then I decided to just let other people do that for me. Dean Oliver invented it, so naturally his explanation is included here, but for something less math-heavy check out this New York Times article.

For those too captivated by my riveting writing to click away, the TLDR version is that close games are basically tossups. Teams should win around 50% of them over time. Teams that win a bunch of close games are considered lucky and vice versa. Teams that win blowouts have better margins of victory than teams that win close games, so luck is just the difference between what a team’s winning percentage “should” be based on its margin and what it actually is.

The Why

After following KenPom for a while, I started to notice that there are a few teams that seem to be either really lucky or really unlucky every year. Let’s look at Kansas. Over the (admittedly arbitrarily selected) last five years, they have ranked 26th, 23rd, 32nd, 75th and 33rd in luck. That’s out of well over 300 teams. The only year they were outside of the top 35 was 2016, when they finished third overall in KenPom’s ratings. In other words, they were supposed to have an extremely good record and their record was still better than it should’ve been.

There’s a reason the Jayhawks won a record 14 straight conference titles in the Big 12. They win lots of close games. Either they are truly a perennially lucky team, or Bill Self is a final-minute wizard. My guess is a little of both, but I wanted to know if teams have any control over their “luck.” I wanted to know if maybe Kansas does something of their own accord to win all these close games–if maybe, just maybe, Self knows something about closing out games.

The How

Here’s where that math stuff comes in. In theory, if KenPom’s luck value is literal luck, then it won’t be caused by anything. While we can’t actually test causation very effectively with this data, we can test correlation. And before you yell at me that correlation doesn’t imply causation, let’s look at which stats correlate with luck and see if we can reasonably hypothesize any causations.

I analyzed the relationship between luck and 70 other stats using KenPom data from 2002 to 2019. If I have enough fun with this, maybe I’ll look at correlations with other data from other sources at some point, but 70 KenPom stats seems like a good starting point. Anyway, 39 of the 70 stats were available for that whole period, while 24 were available since ’07, one since ‘08 and the other six starting in ‘10. This gave us a sample of 6,157 team-seasons for most of the stats, and at least 3,492 for all of them.

I tested the significance of each correlation coefficient using a one-tailed t-test, and found 31 of them to be significant at a p=.01 level. In layman’s terms, 31 of the stats were related strongly enough with luck that we can be pretty sure it’s not just random chance and that there really is some sort of relationship. If 31 out of 70 sounds like a lot, it’s probably because a bunch of the stats are pretty much the same thing. I counted raw tempo and adjusted tempo as two different stats, for example, and there are seven separate categories for height alone (five of them made the list of 31). If you’re judging me for forcing it, you’re not wrong. What’s a good study if not for a little misleading number somewhere in there?

Listing 31 stats seems like a kind of boring exercise that wouldn’t really be worth the effort or space, so I’ll just break down the highlights. I see three categories here–stats that are probably explained by luck, stats that probably explain luck, and stats for which I really have no clue.

Stats Probably Explained by Luck

These are for the most part stats that are affected by the whole foul game at the end of tight games. If you are lucky and winning a lot of close games, then you are probably getting fouled at the end of them, and vice versa. This leads to changes in offensive and defensive free throw rates obviously, but also in a few other areas.

Given that the vast majority of teams score under 1.2 points per possession, any trip to the line from a player shooting 60% or better on free throws is more efficient than other possessions, so offensive efficiency was significant. Interestingly, it lost its significance once adjusted for venue and strength of schedule, but that might just be noise. Additionally, a possession resulting in a quick foul off of the inbound usually takes less than five seconds, making the tempo look faster. Both raw and adjusted tempo were significant, as was offensive possession length. These correlations appear to be a result of luck (winning close games) causing the changes in the other stats, so they aren’t necessarily evidence that teams control their own luck. But…

Stats Probably Explaining Luck

Technically, teams have a tiny bit of control over their opponents’ free throw percentages, pretty much based on who they foul (if you foul Ethan Happ a lot, your opponents will miss more free throws.) While Arizona State may beg to differ, however, teams generally control whether their own free throws go in. If your opponent shoots 3-for-20 from the stripe, you were almost certainly lucky, and if they make all 20 the reverse could be said. Opponent’s FT% is one of the 31, and that seems fairly intuitive.

Some other stats on the list may or may not truly explain luck, but it’s not possible for luck to explain them either, so I am putting them in this category. Random chance is a possibility, but it’s more fun for me to give into confirmation bias and use these stats as evidence that teams have slight control over luck. The most notable ones are height (both minutes-weighted and non-minutes weighted, as well as SG, PF and C height) and experience (as well as minutes continuity, which is measuring something similar). Height is actually negatively correlated, meaning shorter teams are luckier.

It is pretty safe to say that getting lucky does not cause a team’s players to either shrink or become older, so either it’s random chance or the causation goes the other way. Experience makes some sense, considering that luck is pretty closely related to performance in end-game situations. Height, on the other hand, well, if you have any theories I welcome them in the comments.

Stats I Have No Clue About

There are a few stats that correlate with luck whose causations could go either way. As with height, I invite you to share your theories in the comments. Having better two-point percentages and effective field goal percentages as well as getting less shots blocked were all things that related to luck. Another one was two-foul participation, both raw and adjusted percentage as well as total two-foul time on the court. This means teams that play guys with two fouls more often in the first half tend to be slightly luckier. Since all three of those stats were significant, this one doesn’t seem like a fluke, but I’m not sure why first-half events would have any relationship with luck. Then again, I haven’t racked my brain yet trying to figure it out, although something tells me that’s coming the next time I’m super bored.

Final Thoughts

This whole study started with conversations with my brother about how hitting free throws in the last minute of undecided games is big, and therefore may have something to do with luck. He’s the kind of guy that would tell you to get better at free throws if you shot 90%, so he was hoping that we would find team FT% to have a significant correlation with luck. Unfortunately, we had no such luck (eye-rolling is acceptable here). We did, however, find that there were several other predictors of luck, and although none of them had a very strong relationship at all, our sample was big enough that they did still mean something. Not that we didn’t already know this, but it may be officially okay to say that Bill Self’s late-game coaching prowess makes the Jayhawks lucky. It does appear that some teams might be able to create their own luck. That is, if luck is still what you want to call it. Next time you hear that it’s better to be lucky than good, however, know that they might not be mutually exclusive.