Raise your hand if you’ve ever heard a friend or announcer say that a team was “coming in hot” to the tournament.
Yeah, us too.
So we thought we’d explore NCAA data to paint a more detailed picture of what 🔥 might really mean. What, if any, relevance does it have in predicting tournament success, and what, if anything, might it tell us about Villanova and Michigan tomorrow night?
It’s not whether you win or lose…
As you might expect, we had to start by getting from the qualitative to the quantitative and define our terms. People often equate hotness with streakiness (i.e., how many games in a row has a team won or lost), as if a streak alone might be indicative of future continued success. (In fact, statisticians have long studied and debated the existence of a “hot hand” — if a player is more likely to make a successful shot if their previous shot was successful — with mixed results.) But since there’s a host of factors that contribute to whether a team pulls out a win, we wanted to look at broader metrics that could be more indicative of a team’s upward or downward trend than just the W-L column.
We started by investigating offensive and defensive ratings. An offensive rating (or efficiency) is an estimated number of points per 100 possessions, where possessions are approximated by the sum of the number of field goal attempts, turnovers, and 0.475 for each free throw attempt, but minus each offensive rebound. Defensive rating is the opponents’ total points divided by the opponents’ number of possessions * 100.
- ORTG = 100 * (Total Points / (Field goal attempts — Offensive Rebounds + Turnovers + (.475 * Free Throw Attempts)))
- DRTG = 100 * (Opponents’ total points / (Opponents’ field goal attempts — Opponents’ Offensive Rebounds + Opponents’ Turnovers + (.475 * Opponents’ Free Throw Attempts)))
The higher the offensive rating and the lower the defensive rating, the better. And by subtracting defensive rating from offensive rating, we got a single number for the team’s net rating (+/- per 100 possessions).
With these values set, we built an interactive dashboard of the 68 teams in the NCAA Tournament to help visualize and contextualize a team’s ratings over the course of the 2018 season. Generally speaking, positive values for net rating indicate a win, and negative values indicates a loss. As you might have heard a few times by now, Villanova has the highest offensive efficiency in the NCAA, which you can see in the dashboard. Michigan’s trails a bit, but is still pretty high.
But the data here is richer than showing whether a team is simply on a winning streak — it’s showing how well a team is playing, and therefore, we think, a better indication of a team’s hotness.
Let’s see if there’s anything to this.
How would you go about testing whether net rating over time is related to anything predictable in the tournament?
Glad you asked.
The first step was to express hotness as a function of a team’s net rating in each game. Satisfying as it is to visualize net rating plotted out per game, we realized it would be more meaningful to view the average net rating of a team’s last five games compared to the rest of the season. So that’s what we did.
We isolated for the last five games played by each team entering the NCAA tournament in the last five years, and found each team’s average net rating. We then did the same thing over the entire season without the last five games. The resulting difference between average net rating in the last five games and the average net rating in all games before that is a single value that encapsulates a team’s 🔥 at the time of entering the tournament (aka, Selection Sunday).
On the whole, this hotness metric is nice, but not a particularly strong indicator of tournament success — the hotter team has won 51.5% of the 334 games in our dataset (from the 2014 tournament to the present). But like many things, it gets a bit more interesting as you dive a bit deeper into the data. In general, the closer the teams’ seeds are in a matchup, the more often the hotter team has won.
The hotter team has won 76% of the 25 matchups since 2014 between evenly-seeded teams, and 64.9% of the 117 games played between teams up to and including a three-seed difference from the other. In fact, the hotter team has won more often than not in matchups of every seed difference up to six. (And if you’re curious about the dip to a win % of 0 in games where there is a 10 seed difference between opponents, it’s because there have only been two such matchups in our dataset).
We’ve already covered a bit about how ‘mad’ this March has been, but it’s also one of the hottest: 66.7% of games between teams within three seeds of each other have been won by the hotter team, the highest since 2014 when it was 82.6%. (Incidentally, 2014 was also a pretty wild year, coming in right behind 2000 and 1999 overall, and second only to this year in the timeframe under consideration.)
While hotness is by no means restricted to upsets, if you’re looking for your team to pull one off in the tournament, it helps if they’re coming in hot. Of the 88 observed upsets in our dataset, 63.6% were achieved by the hotter team entering the tournament, including all three upsets of double-digit seed differences, and 83.3% of the 18 upsets with at least a seven-seed difference.
That said, Michigan is at a disadvantage here: Villanova is by far the hotter team. On March 11, Michigan’s hotness score was 3.267. Villanova? 9.223. (And given how the teams have been playing in this tournament, if anything, Villanova has been heating up even further.)
Still, this is only one way to look at hotness. By freezing 🔥 at Selection Sunday, this measure definitely favors teams that go deep in their own conference tournaments. It also benefits teams who didn’t play particularly well early in the season, since if you did play exceptionally well early on, it’s hard to reach your early season average net rating again — by our metric, Texas Southern, which started the season 0–13, actually came in hottest to the tournament at 25.692. (Villanova’s 9.223 was #7 out of the 68, and Michigan’s 3.267 was #13. For additional perspective, Kansas came in cold, at -5.7, and Loyola-Chicago was only slightly more lukewarm.)
In any case, as we already know, there’s always some kind of chance to pull out that last W. Come up with your own hotness rating using team efficiencies, and join us on Monday night for the Final!