For the past few years, Iāve engaged in a narrowly specific and superfluous activity of simulating the 3-Point Contest. Much of All-Star Weekend will be chaotic and meaningless, and even the most organized or entertaining moments will be ones of aesthetics or pure awe-inspiring joy.
There are few ways an analytic-minded person can approach the weekend, save for compiling the ridiculous All-Star Game stats or something similar. But the 3-Point Contest makes sense; it has clear rules and players repeat the same activities. However, as Iāve learned from past contests, itās less predictable than Iād like, and there are few discernible patterns.
Letās get some background first. Iām sure everyone has a favorite pet theory explaining why some players do well in the contest, but Iāve found that various other factorsĀ āĀ like height or usage or even free throw percentage ā donāt matter. The only variables with a high level of significance areĀ a multi-year 3-point percentage and a dummy variable for the final round, as players typically shoot better during it.
Read More: Official guide to All-Star Weekend prop bets
In fact, I even tried using past performance, and even that wasnāt useful ultimately [1.]. You can see some graphs I created with older dataĀ (nothing has changed since then.) If you think a certain type of player has an advantage, check out the history of the contest and youāll find plenty of counter-examples. Chaos reigns here.



As for the simulation itself, itās pretty simple, and itās similar to the one I used last season. It uses beta regression based on every contest and round from 2000 to 2016. The only tweaks I made were to the 3-point percentage variable by slightly changing the weighting for past seasonsĀ and regressing the percentage to 40 percent [2.].
You can see histograms below of the simulation results compared to historical data ā I used histograms with older data because recent contests use more 2-point āmoneyballs.ā If youāre wondering what the point of a model is when you can just rank shooters by their percentages, itās to figure out the spread of the odds. And from what Iāve learned in past contests, there should be no heavy favorites and the field is usually a safe pick.


Finally, letās get to the results for this yearās contest. You can see the contestants here and the odds various markets have for them. Klay Thompson is the obvious favorite and, lo and behold, heās the favorite through the simulation too. But heās not the runaway leader; no one should be. Heās only three times as likely as, say, Nick Young, whose win would be a metaphor for the Los Angeles Lakers clinging to superficial fixes and vanity ā and weād probably get a great celebration too.
Here are the odds for 2017 3-Point Contest:
Player | Odds |
Klay Thompson | 21.9 |
CJ McCollum | 17.0 |
Eric Gordon | 12.8 |
Kyle Lowry | 11.4 |
Nick Young | 9.4 |
Kyrie Irving | 8.8 |
Wesley Matthews | 8.2 |
Kemba Walker | 7.8 |
The interesting thing about a simulation is that you can analyze outliers andĀ estimate the chances of certain extreme events. For example, a score of 31 or higher, accomplished by Thompson, Wesley Matthew and CJ McCollum in certain simulation runs, happens about every 2000 contests, while a score of 30 or higher happens once every 300, roughly. But the more interesting feature of a simulation you wrote is that you can use any input you want. Wouldnāt this contest be better with Kristaps Porzingis and Joel Embiid (and without Kemba Walker and Wesley Matthews)?
Here are the odds for coolestĀ 3-Point Contest:
Player | Odds |
Klay Thompson | 23.6 |
CJ McCollum | 18.1 |
Eric Gordon | 13.3 |
Kyle Lowry | 11.9 |
Nick Young | 10.2 |
Kyrie Irving | 9.4 |
Joel Embiid | 7.5 |
Kristaps Porzingis | 6.1 |
Yes, there are multiple universes out there where Embiid is healthy and winning the 3-Point Contest. (Just donāt tell Kyrie Irving about the multiverse theory; heās having enough trouble with a spherical Earth.) If you want to question the veracity of a model where someone like The Process can win, Iāll just point to past contests with strange results. In fact, if you see an anomalous result this weekend, which has been happening with greater frequency, seemingly, in the sports world and beyond, perhaps itās just more evidence thatĀ we are part of a grand simulation. (I suppose one could ask what the point would be of simulating a universe, but itās probably stranger to code a 3-Point Contest simulation.)
Next: Do shooting fouls increase the value of mid-range shots?
For now, we can pretend that the universe isnāt an unrelenting hellscape dependent on the whims of a bored programmer, and I for one will be curious as to who will win the 3-Point Contest. Because history has shown us, consistently, that itās not predictable ā remember that last season Stephen Curry, greatest 3-point shooter ever in the midst of the greatest shooting season ever, didnāt win ā and thatās why I keep watching.
[1. The past performance variable did have a positive correlation and it wasnāt completely insignificant.Ā Itās possible that with more data, or just fine-tuning, it could be included and may actually improve prediction. Thereās just not a lot of great data with the shortened line from 1995 to 1997 mucking things up and the limited number of rounds.]
[2. The pseudo R^2 is now 0.156, which is actually an improvement.]