Are assists good?
This seemingly basic question with an obvious answer popped into my head while watching a game. Of course it is! By definition, an assist being scored means a basket was made. More assist therefore skews the sample, and means it's better than the alternative.
But, what does it tell us? Are assisted opportunities better than an unassisted opportunity?
The fact that there was an assist does not in itself provide any information about the value added by the pass. Was a good opportunity created due to this pass, one that was better than what was available to the player before it? Some passes find a free player an open layup, but others benefit from the fact that the ball got to another player, and that the player made the (tough) shot.
And although statistics are available on potential assists these days, it is somewhat limited. There is no public statistic available about the location of a potential assist. But, we might be able to look at teams and locations to understand how higher assist rates correlate with accuracy. So let's do just that.
Shot accuracy vs Assists
Let's have a look at a simple scatter plot to see if teams that produce more baskets from assists shoot at a better accuracy.
Look at that. There is a clear relationship between assists per 100 possessions, and field goal accuracy overall. Generally, higher assist/100 means that the team shoots a higher field goal percentage.
For the most part, it also means high points per 100 possessions as well (look at the colours). With a notable exception for Houston, who gets a lot of their points from threes and free throws, so their points per 100 possession is higher than their percentage might suggest.
By areas on the court
We can look at it in more detail, by breaking the information into various areas on the court. This is how I break down the court:
And instead of per 100 stats, we'll replace it with a similar, assists per 100 made shots stat from each zone:
That's quite interesting. It looks like certain zones (like 1 - within 4ft and 4 - Cornder 3s) show higher correlationships between assist rates & FG%. Maybe we should look at the the data in further detail.
Let's take a look at the data in each zone separately. The next charts split up the data by zones, mark each team, and indicate their shot frequency from the zone with size:
Near the rim, the data shows a great correlation between assists and FG%. What does the data look like in the midrange?
There is pretty good relationship between 4-14 feet, but not so much for the long 2s. Other than the 18-19 Warriors with KD and Klay, there really isn't much of a relationship here.
Moving on to corner 3s:
This looks more promising. This definitely looks like a positive relationship. Almost all of the Spurs' corner 3s are assited, and they have the highest accuracy from those spots, while the Pistons and the Rockets rank lowest in assists and very low accuracy. The teams in between show a good correlation as well generally.
The relationships get murkier above the break - I would say that there's very weak correlation, if any. The chart for long 3s essentially look pretty close to a random spread.
These data points show that high assist rates show clearer benefits close range. In my opinion, this is because at close range, the quality of opportunities matter more, and as we get further from the basket, two things happen.
Firstly, the quality of shooters begin to matter more than any additional value from getting better looks, as the shots come from further and further. Secondly, randomness begins to dominate the statistics. Three point shots, especially from long range, are low percentage, high variance shots. As a result, any marginal improvements are obscured by the effects of randomness.
Even then, ultimately the number of assists seem to be good proxies for ball movement, and the quality of looks.
Assists change the game
This next chart probably tells us as much as anything else that we've seen to date. Here, we compare the difference between assisted shots and unassisted shots.
The bulk of unassisted shots come from near the rim, or in the paint, with some from above the break. But take a look at the distribution for the assisted shots. Given that so many of assisted shots are good shots, it's no wonder that higher assists lead to higher accuracy.
Note: If you are interested in how I built these visualizations, I posted about that (and other dataViz work) here on TowardsDataScience.com. Check it out as well as my other coding work on Medium.