Steals, Regularized Adjusted Plus-Minus, and Ignoring Position


There has been a lot of discussion regarding the findings of this FiveThirtyEight article praising the value of the steal in the evaluation of players (in the piece, a steal is even claimed to have the value equivalent of 9.1 points in terms of irreplaceability). The biggest question regarding the analysis appears to be this: Are players that go for steals leaving their teams in worse positions defensively?

If we were just looking at box score categories or stats like PER, it would be nearly impossible to answer this question with any definitive answer because defensive value is very difficult to quantify. Instead, we can use a different measure of defensive skill, that is Defensive Regularized Adjusted Plus Minus, or DRAPM for short. This statistic is very similar to the new “Real Plus-Minus” stat that ESPN just rolled out.

Succinctly, this stat finds the value in points that a player adds or subtracts over replacement over 100 possessions by adjusting (through many iterations) for quality of teammates and opposition. For our purposes, we will only cover the defensive half of RAPM. It’s an incredibly interesting stat, as its approach is completely different than other “all encompassing” statistics such as PER and Win Shares/48 (Click here to read more about adjusted plus-minus and what it does).

A simple method to see how steals correlate to overall defensive value is to look at a regression between steals per 48 minutes and defensive RAPM for every qualifying player on ESPN.

So is there a meaningful correlation? At first glance, no:


There is a very, very slight positive correlation between these two values, but with an R squared value of .036, it is clearly negligible and not meaningful at all. Initially, it appears that the quantity of steals has no correlation, and therefore no impact, on the defensive value a player adds.

However, discounting steals as having value in projecting a player would be silly if we ignore the importance of position. Guards, forwards, and centers play completely different defensive positions, so expecting them to approach steals in the same way seems a little silly. In fact, by immediately splitting all the players into two categories (PG, SG, SF) and (PF, C) we are able to shed light on the positions where steals add the most defensive value.



We see here that there is a decently strong correlation between steals and DRAPM for the backcourt positions and small forward while centers and power forwards have virtually no statistical correlation between steals and DRAPM. This makes very good intuitive sense. When we think of good defenders as bigs, steals really aren’t the thing that we picture. It makes sense that if steals were important to a player’s defensive value, it would be more apparent for guards and wings.

Our analysis doesn’t have to end here; looking at what the data indicates, we might want to test out the idea that steals correlate stronger with defensive value at point guard than at other positions. In isolating PG, SG, and SF, however, we get a mildly surprising result:




It appears that of these three positions, steals correlate with DRAPM for small forwards more than they do with DRAPM for the guard positions, and it also carries the strongest coefficient at small forward. For every additional steal small forwards get, they tend to contribute 1.35 more points per 100 possessions on average on defense. This flies in the face of our initial reactions when we think of which players get the most value from steals.

When we think of steals positively affecting defense, we think of players like Ricky Rubio and Patrick Beverley, pesky smaller guys that play relentless, aggressive D. Having steals correlate with good defense the most at small forward instead could be the result of a couple of things. In general, SF tends to attract the best athletes. Better athletes tend to be better defenders in general, and it would also make sense that being a better athlete can lead to the ability to get more steals.

Additionally, if we take the correlation as also carrying some causation, small forwards tend to have the most freedom to roam on defense and therefore have more opportunities to make a difference on defense through steals, resulting in steals being more important. Still, it is interesting that the most notable “steals are really important” arguments are about guards, not small forwards.

Through the positive correlations that we have found throughout, I think it is safe to say that steals in no way correlate with worse defense, or over-gambling to the point of hurting on defense, especially in the case of guards and small forwards. Of course, we also have to careful in interpreting the correlation.

In no way have we proven that a player getting additional steals directly causes his team’s defense overall to perform better; rather, we have found that players playing the 1-3 spots that tend to get more steals also tend to positively contribute more overall on the defensive end. Drawing the conclusion that there is causation here can be intuitively done, but it is by no means proven.

However, I think that the most important thing to take from these findings is that any analysis of steals as an indicator of defensive performance must come with a consideration of position. We have shown that the correlation between steals and defense is drastically different when looking at different positions, so it appears problematic to examine the impact of steals without taking that context into account.

Data courtesy of
Image courtesy of CBS Sports

Xavier Weisenreder
Georgetown University Class of 2016

Follow Xavier on Twitter: @BeMoreChillNext
Follow GSABR on Twitter: @GtownSports
Like GSABR on Facebook


One thought on “Steals, Regularized Adjusted Plus-Minus, and Ignoring Position

  1. “steals correlate with DRAPM for small forwards more than they do with DRAPM for the guard positions, and it also carries the strongest coefficient at small forward.”
    Those two mean the same thing. You’re saying basically the same thing twice.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s