Three Pointers Made Props in the NBA

A threes made line of 2.5 looks like a question about shooting. It is mostly a question about attempts, asked of a small count distribution where the average sits in an awkward place and a single miss moves the result. Streaks make it feel more predictable than it is.

By SlatelinePublished
A row of bars forming a distribution of outcomes, with a dotted line marking a threshold near its peak

Few markets are as easy to form an opinion on as threes made. You watch a shooter get hot, you see the line at 2.5, and the decision seems obvious. Few markets punish that instinct as reliably. The count is small, the percentage behind it is noisy, and the result often turns on a single shot at the end of the third quarter that nobody could have forecast.

Under the official scoring rule, a successful field goal from outside the three point line counts three points. The prop does not care about points, though. It counts makes, and makes are the product of two numbers that behave very differently. This piece takes them apart, then looks at the shape of the resulting distribution, what an integer line does to it, and why the hot hand is the wrong lens for all of it.

Attempts times percentage, and attempts matter more

The Basketball Reference glossary defines three point percentage simply as threes made divided by threes attempted, and the NBA.com stat glossary defines attempts as the number of three point field goals a player attempted. Turn it around and a player's makes on a given night are his attempts multiplied by his conversion rate on those attempts.

The two factors are not equally knowable. Attempts are a role. They come from how many minutes he plays, how the offense uses him, and whether he is the player the defense leaves open or the one it chases off the line. Those things are decided by coaching and roster construction, and they carry over from game to game. Percentage on a single night is close to a string of weighted coin flips. Over a season it settles, but on any one night it swings widely.

  • Estimate attempts from recent games in the current role, with minutes checked first. The minutes and blowouts piece covers why a lopsided game can strip attempts from starters.
  • Estimate percentage from a long sample, ideally more than one season of attempts, and move it only slightly for recent form.
  • Check what changed around him. A new primary ball handler, a teammate who draws help defense, or a teammate out injured can shift both his volume and the quality of his looks.
  • Treat opponent scheme as an effect on attempts first. Some defenses give up threes to protect the paint, and some chase shooters off the line.

The raw material is public. Every box score lists threes made and attempted for each player, so a game log gives you attempts per game and, with minutes beside them, attempts per minute. That per minute rate is the cleaner signal, because it separates how often he shoots from how long he played. When his minutes are about to change, carry the rate over and rebuild the attempts from the new minutes rather than reusing the old per game number.

The shape of a low count

If a player takes a fixed number of attempts and each one goes in with the same probability, independently, his makes follow a binomial distribution, with a mean of attempts times probability. Real shooting is not that tidy, but the binomial is a useful first picture, and it shows at once why threes lines are hard.

Example: Seven attempts at 36 percent, made up inputs

Suppose a made up wing takes exactly 7 threes and makes each with probability 0.36. His average is 2.52 makes. The binomial gives roughly these chances: 0 makes 4.4 percent, 1 make 17.3 percent, 2 makes 29.2 percent, 3 makes 27.4 percent, 4 makes 15.4 percent, and 5 or more about 6.3 percent. Against a line of 2.5, the over needs 3 or more, which happens 49.1 percent of the time. His average sits above the line and he still clears it less than half the time. The inputs are invented, and the arithmetic is exact for them.

That result is the key idea for threes props. When outcomes can only be whole numbers and the mean sits between them, the mean alone cannot tell you which side of a half point line is favored. Reading a distribution covers the general version: the average is a summary of the outcomes, and at low counts it can mislead about where the probability actually sits.

Now add the fact that attempts are not fixed. The same wing might take 5 on a night the offense runs through someone else and 9 when he is the focus. Spreading attempts across a range, even around the same average, widens the distribution and puts more weight on both tails. In the same made up setup, mixing nights of 5, 7, and 9 attempts in a one to two to one ratio leaves the average at 2.52 but lowers the chance of 3 or more to about 48 percent. Uncertainty about volume is uncertainty about the whole prop.

Integer lines and the push

Some threes lines are posted on a whole number rather than a half. That changes the question, because the exact number becomes a third outcome. In the seven attempt example, a line of 2 gives 49.1 percent over, 29.2 percent exactly 2, and 21.7 percent under. Nearly three nights in ten land on the number. How that outcome settles is a platform rule, not a convention: it might return the stake, count as a loss, or be handled some other way.

What a push costs walks through why this matters so much. The short version is that on a low count stat the probability of landing exactly on the number is large, much larger than on a points line, so the settlement rule can move the value of a side substantially while the projection stays the same. Confirm the rule on the platform before estimating anything, and do not compute the under as one minus the over when a push is possible.

Why shooting streaks mislead

Shooting is where the hot hand idea was born. For many years researchers found no evidence for it and called it a fallacy, starting with a 1985 paper by Thomas Gilovich, Amos Tversky, and Robert Vallone. Later work questioned that conclusion, and some recent studies find evidence for the effect in some settings while others do not. Where it appears, the evidence suggests it is limited to a small subset of players and is small in size.

For prop research the practical reading is the same either way. Even if a small hot hand exists within a game, it does not justify treating last week's five for eight night as a new shooting level. Three point percentage over a handful of games is dominated by randomness, and extreme stretches tend to be followed by less extreme ones. A shooter who makes a high share of threes for a week has mostly shown what variance looks like on a few dozen attempts.

Streaks mislead in a second way too. A hot game often comes with more attempts, as the player looks for his shot and teammates look for him. If the attempts rose because his role changed, that is real information. If they rose because he was hot, the extra volume is likely to fade with the percentage. Separate the two before reading a hot stretch as anything more than a hot stretch.

The same caution applies to recent hit rates on the prop itself. A player who has cleared 2.5 threes in seven of his last ten games is describing ten draws from a distribution like the one above, and those ten draws say little about where the line should sit. Hit rate versus projection explains why a short hit rate is a weak guide compared with a projection built from volume and a long run percentage.

The limits of researching threes

Even with the best volume estimate and an honest long run percentage, a threes prop remains a small count with a wide range. In the made up example above, a correct projection still leaves nearly a coin flip at 2.5. Research can tell you where the line sits against a realistic distribution. It cannot remove the randomness of whether the fourth attempt rims out.

The binomial picture is also a simplification. Shots are not identical: a corner three and a pull up from well beyond the arc have different chances, and a player's mix of them shifts with his role and the defense. Attempts are not independent of makes either, since a player who starts cold may shoot less. Use the model to understand the shape, not as a precise answer.

The working order: estimate attempts from the current role, anchor percentage on a long sample, look at the whole distribution instead of the average, check how the exact number settles, and read every streak as a question about volume before it is a question about shooting.

References

Follow NBA on Slateline

The public NBA page lists each market with its graded record where one is published and, when a slate is open, the props the free plan opens. No account is needed to look.

Open the NBA page

Not ready to sign up? Get the weekly note.

The Slate

One weekly note: what published, what the graded record shows, what changed. No picks, and nothing is sent until you confirm your address.

Keep researching

Three Pointers Made Props in the NBA · Slateline