Chasing 72: The Warriors and Variance

Jan 18, 2016; Cleveland, OH, USA; Golden State Warriors guard Stephen Curry (30) celebrates a three-point basket in the third quarter against the Cleveland Cavaliers at Quicken Loans Arena. Mandatory Credit: David Richard-USA TODAY Sports
Jan 18, 2016; Cleveland, OH, USA; Golden State Warriors guard Stephen Curry (30) celebrates a three-point basket in the third quarter against the Cleveland Cavaliers at Quicken Loans Arena. Mandatory Credit: David Richard-USA TODAY Sports /
facebooktwitterreddit
Jan 18, 2016; Cleveland, OH, USA; Golden State Warriors guard Stephen Curry (30) reacts on the bench in the fourth quarter against the Cleveland Cavaliers at Quicken Loans Arena. Mandatory Credit: David Richard-USA TODAY Sports
Jan 18, 2016; Cleveland, OH, USA; Golden State Warriors guard Stephen Curry (30) reacts on the bench in the fourth quarter against the Cleveland Cavaliers at Quicken Loans Arena. Mandatory Credit: David Richard-USA TODAY Sports /

After humiliating the Cavs on MLK, the Warriors are still eyeing a possible record setting season. Disregarding the obvious discussion points like “which team is better”, I was rather  wondering “for which team is the outcome of only losing 10 games more likely”? This is related to a recent Nylon Calculus article by Brian Skinner and one of my own articles from last summer. The idea in short: If you are the favorite, you want to reduce the variance as much as possible, while if you are the underdog, your best chance is to play a high variance game.
As I found previously and unsurprisingly that a higher amount of three point attempts increases your variance significantly, I was sure that the Bulls had the advantage in this regard. But simulating 5000 games for each team and their opponents, using simple team stats from basketball-reference.com, I found the differences to be minuscule, with distributions that are strikingly similar:

comparison Bulls vs Warriors
comparison Bulls vs Warriors /
Bulls playing with Warriors pace
There are two reasons for these basically imperceptible differences. Firstly, the Bulls played at a remarkably slower pace (90’s basketball for the win!), something that will increase the variance. Doing the same simulations and using the Warriors pace for the ’96 Bulls team: /

It’s not easy to see, but the bulls points per possession distribution (and the one of their opponents) becomes a little bit more focused. But that even this difference is not so big has another reason: For my last article, I found that the variance in scoring for one team is around 10 times higher than the difference in variance between the team with the lowest and the team with the highest variance, rendering the part of the variance that can be control through strategy relatively meaningless in terms of the outcome of a contest[1. This might actually be an interesting point, but I forgot my click-bait hat at home today].

So, what does this mean? Are the Warriors more or less likely to get to 72 than than were the Bulls? Considering the Warriors have been quite “skilled” in keeping their opponents to 31.6% from beyond the arc[3. A genuine candidate for regression towards if not all the way to the mean in that regard.], it might not be an easy task. After all, this is what the graph would look like if their opponents would have an average 3 point shooting percentage:

Warriors opponents with average 3P%
Warriors opponents with average 3P% /

So, will the Warriors make it?

My momentary opinion if they will make it or not is clearly ¯_(ツ)_/¯. But for the sake of argument, let us imagine that my complicated coin flip simulation is a good reflection for reality and also that the 3 point percentage defense is legit. Then historically great teams like the ’96 Bulls or ’16 Warriors are expected to lose 10 or less times…

Number_losses
Number_losses /

… *count* around 93 (Bulls) or 88 (Warriors) in 1000 times. The infinitesimal edge for the ’96 Bulls is due to their slightly higher margin of victory. As an example for the effect of decreased variance, the Bulls playing with Warrior-pace would win 72+ games 133 times and the Warriors with average three point shooting opponents only 7 times (less than 1%). But there are several obvious drawbacks to this simulation. For one, my simulated teams always play against their average opponent and second, the teams in this scenario would always play the same combination of starters and bench players, thus never adapting to closer games. It’s hard to say how those two aspects would influence the scenario as a whole. But it seems that you have to be historically great and lucky to get to 72 wins. Which should come as a surprise to nobody.