What is the range of research?
It’s going to sound small at first – the last two seasons of college basketball, 2019-20 and 2020-21 – but hear me out. Collectively, these two seasons produced a total of 9,705 games played, per Bart Torvik’s website. That’s a pretty massive sample of games to choose from, and conveniently, it also encompasses the two seasons since the last major rule change in college basketball: the three-point line moving back a little over a foot. Since this change happened, three-point percentage has dropped about 0.8%, three-point attempt rates are down by about 1.2%, and other significant statistical shifts – more turnovers, fewer rebounds – have started to impact the sport, too.
All of this means we’re drawing on nearly five digits’ worth of games for our sample size. With more time, I can always go back and analyze the seasons before this, but it made sense to focus on 2019-20 and 2020-21. We’re living in the present, and coaches will be most curious to find out what’s presently determining wins and losses.
What defines a close game?
This one is simple on its face but requires a small explanation. Any game within six points at game’s end was determined to be a Close Game, just as it’s measured on Bart Torvik’s site and on countless others. The issue here, of course, is that this could exclude games that were within six points with two minutes to go but ended up being 7-12 point decisions. Alternately, a game that was legitimately close and goes to overtime may end up being a blowout in the five-minute period. Both of these are excluded from our research, as I’m only doing what I know I can do with the data.
These exclusions still allowed us an impressive 3,079 games to draw from over the last two seasons. A much higher amount of college basketball games were technically close under this metric than I’d anticipated, and sorting through the data took a good chunk of time. Declaring close games to be six points or less margins was the most effective, simple way to get it across to the average reader. Others may have different definitions; this is the one I’ve found to work the best for me.
What are some of the common stats-unfriendly tropes that can be proven or disproven?
That’s a long one, but I tried to narrow it down to a few that we can actually check into somewhat.
1. Experience matters. I measured this based on correlating Bart Torvik’s Fun Unexplained by Numbers (FUN) metric, which is a cousin of KenPom’s luck metric, with the average experience of every team in college basketball. The theory here is simple: more experienced teams should be able to use their experience to somewhat consistently squeak out close games. This is a favored theory of most broadcasters, who seem to jump to it whenever it’s most convenient to their argument. With our same range of data, the correlation was just +0.15, meaning there was nothing more than a very weak correlation between experience and close game records. The impact of experience on winning close games was actually less than the correlation of winning games with a margin of 7+ points or greater, which sat at +0.25. The difference here is also minimal, but you could reasonably twist these words into saying “experience means less in close games.” Technically, you wouldn’t be wrong, you’d just be reaching.
2. Experience matters, but for coaches. This one took a little deeper work: trying to correlate the amount of years a coach has spent coaching Division I basketball with their record in close games. Shockingly, this had no correlation whatsoever to a better record in close games. Mike Krzyzewski, Cliff Ellis, and Jim Boeheim have a combined 137 years of coaching experience between them; they have combined to go 73-86 in close games in the last five seasons. Coaching experience had even less relevance to the story than player experience; we should immediately throw this narrative out the window.
3. They have more heart…as in they make the ‘hustle’ plays. This one is less easy to measure, but hey, sure, let’s entertain it. I’m defining ‘hustle’ plays as one thing I can actually measure: did this team win the rebounding battle? If so, we’ll give them the ‘hustle’ point. If every rebound is truly a 50/50 ball as commentators claim, surely the team that wins the rebounding battle (measured by if they had a better OREB% than their opponent) makes this the most important stat in a game, right?
(Quick sidebar: the other, more obvious ‘hustle’ play is any time a player dives after a loose ball. Play-by-play doesn’t differentiate between a normal steal and a spectacular/effort-based one, so this became too difficult to appropriately measure. Besides, the number of true loose-ball dives in a game is a very, very small percentage of the overall actions.)
Uh…not really. Actually, not at all. Of the Four Factors in basketball stats – Effective Field Goal Percentage (eFG%), Turnover Percentage (TO%), Offensive Rebound Percentage (OREB%), and Free Throw Rate (FTR) – OREB% had the second-lowest correlation to wins. In fact, it was barely better than a coin flip. In our 3,079 game dataset, the team that won the rebounding battle won 51.8% of the time. Compare this to the fact that teams who won eFG% won 62.5% of games and those who won the free throw battle pulled off a surprising 63% win rate. Even essentially getting no serious push on the offensive boards wasn’t a huge issue; teams that rebounded 15% or less of their misses still went 102-94 (52%) in these close games. (In all games combined, though, they won just 36.5% of the time.) Rebounds are useful, but the data for both blowouts and close fixtures shows that of the four major factors, they’re the third-most important factor to overall efficiency on average.
We’re three questions in and already, all three of these commentator-promoted cliches don’t have serious statistical backing at all. That begs follow-up questions: what actually does matter in close games, and is it a different formula from all other games we’ve observed?
NEXT PAGE: The nitty gritty, of a sort