The 'effect of being up X'

[J.E.]

I've dug a little into an effect that Matthew Goldman and Justin M. Rao first discovered and talked about in a presentation at SSAC'13. You can watch the video of the presentation here (the for this analysis relevant part starts at ~minute 16, but since it's an awesome presentation there's no harm in watching all of it) and you can find the full paper here

The part I dug further into is, specifically

We also find a strong motivating effect of losing—the trailing teams displays an overall boost in efficiency[..]

If the above quote holds true then we should see, e.g., higher PPP of an attacking unit that's down 10, compared to when the game is tied (assuming all players are average).

To try to quantify how large the effect is I put in additional binary variables into my RAPM framework, one variable for being up a certain amount of points before each possession started. Those variables run from 'down 57' to 'up 57'.

Thanks to the RAPM framework we're controlling for player quality and can then take a look at how much the offense (or opponent defense) is being influenced by being up X (down -X)

I find the same effect as Matthew Goldman and Justin M. Rao did, and, according my results, it's huge and most likely linear. The following chart displays the results



According to this, an away team that is down 10 points is expected to score ~3.5 points more PPP than if it were tied. Inversely, if they were up 10 they would be expected to score ~3.5 points less than if they were tied. Down 20 you're expected to score ~6.2 PPP more than 'normal' - that's like replacing an average offensive player with LeBron

Some notes:
- With adjustment built in, a 5 man unit that's down 10 is required to score more than average points to get an average rating in RAPM. This leads to another side effect: Assuming you have two teams A and B playing a third team C. Both Team A and Team B win by 15 against Team C. Team A jumps out to an early 15 point lead and holds it the entire game, while Team B is tied until the last quarter and outscores Team C by 15 in the last 5 minutes. With the adjustment Team A will get more 'credit' for their 15-point win
- It's probably a good idea to check whether giving less weight to those 'garbage time' possessions is also a good idea. That's something Mark Cuban is a fan of
- This analysis can't tell us whether a) defenses start to clamp down when down (1) and loosen up when up (2), or b) offenses start to fool around (become less efficient) when up (3) and give more effort when down (4). We don't know because the variables for (1)+(3) and (2)+(4) are identical in the regression
- You should really watch the above video as Matt provides some interesting hypotheses on why we observe this effect

[bbstats]

This is awesome!

Have you considered using this as another RAPM variable (if it's linear)? That would go a long way in helping adjust player values for garbage time.

[nbo2]

Awesome - this was my guess for the adjustment that Wayne Winston thought only he and Jeff Sagarin were taking into account a few years ago.

Will 2013-14 RAPM on stats-for-the-nba.appspot.com be updated to include this effect?

[AcrossTheCourt]

This is very cool. For a while, we've just been treating all possessions the same, so it's good to see this work done. I've always wondered about garbage time possessions. I know I've seen an old model adjusted with a "leverage index," which I think gives possessions during close games with little time remaining heavy weight and maybe the playoffs too.

By the way, do you use a variable or adjustment for having no rest (i.e. a game the day before for the home/away team)?

[mzappitello]

Is this on a unit by unit basis or just with all teams together?

The steepness of the fit of the data might be an interesting characteristic of each team, pairing, or unit.

[DSMok1]

By the way, do you use a variable or adjustment for having no rest (i.e. a game the day before for the home/away team)?

That would be a useful variable to add, I believe. My rest day analysis showed a pretty pronounced effect: (look most of the way down the page): http://godismyjudgeok.com/DStats/APBRme ... =2661.html

[boooeee]

Very interesting. Does anybody know if anyone has taken a more direct look at this? Meaning just looking at actual possession outcomes, after stratifying by scoring margin? I suppose there's some selection bias that could sneak in there though (bad teams tend to trail more often than not).

[colts18]

Does RAPM take into account garbage time? What about intentional fouls. J.E., how did you take into account intentional fouls? Did players get credit for them. It seems like it would make sense to ignore intentional fouls and the last minute of garbage time because some teams aren't playing hard defense.

[J.E.]

Have you considered using this as another RAPM variable (if it's linear)? That would go a long way in helping adjust player values for garbage time.

Yes. This is already a variable in my current RAPM calculation. I still need to do some more research to find out whether it's a good idea to give less weight to garbage time possessions

Is this on a unit by unit basis or just with all teams together?

That's everybody lumped together.

By the way, do you use a variable or adjustment for having no rest (i.e. a game the day before for the home/away team)?

I don't, but might in the future. Although I suspect the "days rest" effects are mostly going to cancel each other out

[DSMok1]

By the way, do you use a variable or adjustment for having no rest (i.e. a game the day before for the home/away team)?

I don't, but might in the future. Although I suspect the "days rest" effects are mostly going to cancel each other out

They would cancel out at the player level for the most part, over the course of the season (providing a balanced schedule in that regard)... but you would be able to calculate the exact values for that variable, because you have matchups on a game by game level.

[sethypooh21]

Simplest explanation is the team that's down is more likely to try and score quickly whereas the team that's leading will try and milk the clock. And while I think the causation is slightly messy, the observed results would mesh with Knarsu's observations about shot quality decreasing as the shot clock runs.

[xkonk]

Simplest explanation is the team that's down is more likely to try and score quickly whereas the team that's leading will try and milk the clock. And while I think the causation is slightly messy, the observed results would mesh with Knarsu's observations about shot quality decreasing as the shot clock runs.

That's my first guess as well, coupled with some regression to the mean. JE, could you run the same analysis but only looking at the first half of games? If the effect shrinks a decent amount, it would fit with the idea of leading teams simply running the clock.

[J.E.]

JE, could you run the same analysis but only looking at the first half of games?

Sure. I think ultimately this should really be run for quarters, with maybe the 4th quarter split up into even more parts. I would believe that a 20 point lead 6 minutes into the 2nd quarter has a different effect on the defense than a 20 point lead 90 seconds before the game ends

[J.E.]

Ran it for first half only, doesn't look very different to me

[xkonk]

Some of the patterns in the data look extremely similar - the bounces from -2 to 1, the dip at 10, etc. I can see the values aren't exactly the same, but is this maybe more similar than we should expect? I know that's a vague question, but I would have guessed that you'd see something a little different just from cutting the sample in half.