Bench Production Across the League

Across Liberty Ballers comment sections, articles, and other publications such as The Ringer and ESPN, there have been a multitude of articles on how the Sixers need to get bench help. Of course you’ve heard national broadcast teams mention Hall of Famers Ersan Ilyasova (2018 BPM: -0.3) and Marco Belinelli (2018 BPM: -0.9), and their absence as the main reason the Sixers have not won every single game. My thought was to take a look at bench production through a variety of metrics across the NBA to see where the Sixers fit in amongst their peers.

Production

One of the tricky parts about measuring production is that its team dependent and situation dependent. However, by using a large number of data points across multiple seasons, my hope is that the situational concerns relating to when and how bench players get run is smoothed out. In the most simplistic manner, we’ll use points scored and minutes played as our two pillars of production.

Absolute vs. Relative Production

In the most recent game against the Houston Rockets, the Sixers scored 121 points. The bench scored 44, approximately 36% of the total. The absolute bench production was 44 in this case whereas the relative production was 36%. Let’s add another layer to the points production by factoring in minutes played, as the bench played 98 minutes, roughly 40% of the total 240 minutes. I will combine the points and minutes data into pairs of numbers as seen below.

Relative Metric Pair (pts, min): 0.36, 0.40 — think of this as an (x,y) pairing

Relative Metric Ratio (pts/min): 0.36/0.40 = 0.9 — a single number to describe relative bench productivity

What the ratio provides us is an indication of what kind of points to minutes production the bench players had in total. This is also so that a bench unit who is (0.1, 0.1) has the same relative productivity as a bench unit of (0.5, 0.5).

Now, let’s take a look at some plots. Figure 1 presents the absolute data, and Figure 2 presents the relative data. Of note, Figure 1 makes a great case that Thibs creates OSHA violations with his rotations.

In the below plot, the blue lines are averages across both seasons, and the red line is a line of equivalence, showing where a 1:1 relationship exists. I think I will call it the Sweet Lou Line.

I mean, those are moderately interesting, or at least are to me, but they lack a certain “so what”.

Winning & Bench Production

Let’s take a look at how (if) bench production relates to winning in the next couple figures.

Essentially, Figure 6 is the most important in my opinion. The slopes are consistent across the two seasons with similar confidence intervals and there is some difference between quality teams which is nice to have as well.

However, its difficult to tell which direction causality moves (if at all obviously). Does winning create a scenario where your bench might have less relative production? Or does too much bench production indicate your starters are not that good causing you to lose? Or in an entirely possible situation, is this just an artifact of something else entirely? Using data going back to the 2008 season, Figure 7 and Figure 8 binned data into playoff seeding to take a look at similar information.

Of note, with both figures, the 1-4 Seed data was statistically significantly different from the No Playoff data. Again, the same issues of causality occur here, but I find that the binning is certainly visually useful.

Marco & Ersan

Now, I think it’s fairly well established that Marco and Ersan both had a valuable role in last year’s end of season streak which maaaay have skewed some of the expectations for this year. Figure 9 shows the Sixers bench production compared to the NBA average across the last two seasons.

The Marco & Ersan boost is quite obvious, whereas the JJ/JB move is more murky. However, and this is important to mention, we saw earlier that an elevated relative bench production is correlated with losing, not winning. This to me is the best argument that the 10,000 foot view of the scatterplots is useful in general, but not necessarily applicable to specific unique circumstances.

2017 vs. 2018

Next up, we will compare 2017 to 2018 with two different tools. The first, Figure 10, plots our relative points and relative minutes across each season - it’s a bit of a mess, so we will also use a table shortly after.

Table 1, presented below presents four columns of the percentage based metrics that you are now familiar with. The final two columns require some explanation. The “Euclidean Dist. Change” is the distance between (x17, y17) and (x18, y18) for each team - it is the length of the line connecting the points in Figure 10. The final column “Relative Production Ratio”, is relative production 2017/relative production 2018.

Using the combination of the two, we can get a good picture of how production changed from 2017 to 2018. The larger the distance, the larger the difference in the pair of parameters. It does a nice job of showing how a team that changes from (0.3, 0.3) to (0.5, 0.5) is actually a change in play style, but not relative production, where the ratio would still be 1.0. For fun, I put in the Sixers pre buyout and post buyout as well as the overall for 2017 in comparison to 2018.

Conclusions

Well - I’m not entirely sure what to make of this information. There is clearly a correlation between winning and lower bench productivity on a macro level, due to what believe is just higher quality starters. This also might make you slightly skeptical of playoff success of bench heavy units like the Pacers. However, I am also the person who wrote an article about potentially just stalling for 23.9 seconds per possession when Joel is off the court, so I clearly understand there are certain bench deficiencies. Maybe they purely manifest on the defensive end, or that season long averages are too zoomed out to really provide any information of value. Regardless, hopefully you found this somewhat enlightening.