# What happens during crunch time for the entire NBA

Click for a full-sized interactive version.

After my last post about the Houston Rockets crunch time woes, Bluemars from the forums asked if I could do a similar analysis for all NBA teams. That seemed like a pretty neat idea, and the above chart is the result. The chart shows all teams’ offensive efficiency plotted against their assist ratios during various crunch time moments, defined as being down by five points or less (or tied) with a certain number of minutes remaining. Basically, it’s an attempt to show which teams have heroball tendencies and what results those tendencies produce. For extra fun, the chart also shows each team’s win percentages during their crunch time minutes.

Technical information

On the x-axis is assists per 100 possessions. On the y-axis is points per 100 possessions. The grey lines show the averages of each measures for all teams.

You can, and should, use the slider in the upper left hand corner to change the time period of analysis. There are seven. The first is an entire game, which is how each team behaves on the two measures of interest throughout the course of an entire game. The next six are all crunch time moments. They show how each team behaves when they’re down by five points or less (or tied) starting from five minutes remaining to 30 seconds remaining in a game.

The size of each team’s circle indicates that team’s winning percentage during that time period. Larger circles indicate higher winning percentages. You can mouse over a circle to see the complete data for that team during any time period. So, for example, with four minutes remaining and down by five or less, the enormous black dot on the far right of the graph is the ethereal San Antonio Spurs. They have a ridiculous winning percentage of 63.6%, pass more, and score more than any other team in this situation. I hate them for being so good.

In order to track the progress of a specific team, you can either click on its circle or its name in the lower left hand corner to highlight its circle. That circle will remain highlighted as you adjust the time period using the slider.

Important limitations

• Because not all games are close, the sample size of the games (and possessions from those games) as one moves closer to the end of games becomes very small.
• Each average is an unweighted average. Each team’s performance is considered equally when calculating the average, regardless of how many crunch time possessions and games the teams played.

Observations

Heroball is the king of crunch time

Move the slider to the entire game time period. Look at where the average lines intersect. Now move the slider through each time period chronologically. Notice how the intersection of the average lines steadily moves downwards and to the left. That means as crunch time becomes, uhh, “crunchier,” teams are passing less and scoring less. The Houston Rockets are pretty representative of what the entire league seems to do. Highlight them and slide through the time periods. You’ll see that they start off reasonably well, but with between 4 and 3 minutes remaining the ugly specter of heroball emerges. By the time 30 seconds are remaining, they’re firmly at the bottom and on the left of the chart, meaning they don’t pass and don’t score.

Passing the ball is better than not passing the ball

This observation seems so plainly obvious I almost hate having to write it. But clearly it needs to be shouted from the rooftops because not all teams seem to get it, especially during crunch time. If you look at the general dispersion of the teams, you’ll notice that it trends to the right and upwards. In other words, the more assists a team has, the higher that team’s offensive rating tends to be. This isn’t an optical illusion. In fact, I wanted to know the exact value of this trend, so I plotted it.

Click for a full-sized interactive version

I took out the bells and whistles from the initial chart and just plotted each team as a blue dot. That red line you see is the trend line. It shows the general direction of where teams are plotted on the chart. As you flip through the time periods, you’ll see that offensive rating increases as assist ratio increases in every single period, including the entire game.

The lowest slope (the grade) of any trend line from any chart is 1.69. The highest is 3.05. In normal person speak, that means that an additional assist per 100 possessions (or playing in a way that might produce an additional assist) correlates to between 1.69 and 3.05 additional points per 100 possessions. PASS THE BALL.

Unless your name is Kevin Durant

The upper left quadrant is interesting. Teams in this quadrant need to have high offensive efficiency but low assist ratios. Not surprisingly, there are practically no teams in this quadrant in any of the charts, because, as we just discussed, passing the ball leads to scoring. While the Portland Trailblazers flirt with this quadrant a couple of times, the only team that is strongly in this quadrant in all time periods is the Oklahoma City Thunder. I also hate Kevin Durant.

Teams that pass more, win more

As you scroll through the time periods, take a look at the dispersion of circle sizes. Generally speaking, larger circles are to the right of the assist ratio average line, especially in the 3 and 2 minutes left charts (afterwards it probably becomes much harder to win, period). I don’t think this is a coincidence.

Crunch time teams I like

This doesn’t necessarily mean they win a lot of close games, but it means these teams tend to deviate less from a “normal” offensive strategy during crunch time (aka, they heroball less)–Golden State Warriors, Miami Heat, Atlanta Hawks, Memphis Grizzlies.

Also worth mentioning are the Toronto Raptors. They are the anti-Rockets. They are the only team I found that moves completely against the grain. Their normal offense is quite selfish and inefficient but they become more unselfish and efficient as crunch time progresses. Track their circle. It’s interesting.

About the author: Richard Li is an independent researcher and consultant. He likes numbers and pictures.

• Red94 says 2 YEARs ago New post: Crunch Time Revisited
By: Richard Li

[caption id="attachment_13731" align="aligncenter" width="300"> Click for a full-sized, interactive version[/caption>

Earlier this year, I took a look at the Houston Rockets performance during crunch time compared to that of other NBA teams. The results were ugly. With 30 seconds left in games in which the Houston Rockets were down by five or less, the Rockets offensive rating was 37 points per 100 possessions. That's just abysmal, but probably wasn't very surprising for Rockets fans who had become used to throwing things at their TVs during close games.

At the end of February, I grabbed updated crunch time data. Now I'm finally getting off my ass to do something with those data. The chart above visualizes data from both times when data were collected.

What's similar between the two time periods is that teams inexplicably still rely on heroball during crunch time (again, defined as late in games down by five points or less). I feel like coaches should have to write HEROBALL IS DUMB on the chalkboard Bart Simpson style. To reiterate, tomes of data all conclude that isolation is the least efficient offensive play, and common sense dictates that it becomes even less efficient if the opponent knows it's coming. But, if you look at the chart, you'll see that offensive ratings and assist ratios both decrease as games become tighter. I give up.

What's different is that the Rockets went from being one of the worst crunch time teams in the league to being a somewhat average one. Their offensive rating with 30 seconds left while down by five or less rose to 85 (obviously we're working with small sample sizes here). More importantly, at least for me, is that increase is coupled with an increase in assist ratio. That means the better offense is (probably) resulting from extra passes being made, and not simply because heroball shots that were previously rimming out are now falling. If the latter were the case, we would see the Rockets circle move straight up, indicating an increase in offense without any increase in assist ratio. Instead, we see it move up and to the right. Recent crunch time highlights of driving and kicking to open shooters (e.g., Harden to Beverley) come to mind.

In fact, that upper left quadrant is probably the most important one. Since assist ratio and offensive rating are closely intertwined (assists result in points by default), we would expect teams to trend upwards and to the right. The ones that truly heroball, and do so successfully, would be found in the upper left quadrant. These teams are offensively efficient but don't pass. The only team that is consistently in this quadrant, in both time periods, is the OKC Thunder. If you're going to heroball, you better be #35 on the Thunder, or else you're just dumb.

On a slightly more flexible note, the data also show that a certain amount of individual shot-making ability is helpful during crunch time. The second most interesting quadrant is the bottom right. Teams in this quadrant have high assist ratios, but low offensive ratings. Basically, they only score if there's an assist involved. Like the upper left quadrant, it's pretty hard to make it into this quadrant, since, as stated before, assists automatically lead the points. It's telling that the two teams who can't get out of this quadrant are the Chicago Bulls and Minnesota Timberwolves. They are two teams who pass well but have either poor shooters or players who aren't necessarily known for creating for themselves (yes, I put Kevin Love in this category). Someone like James Harden obviously can, and it looks like the Rockets as a team are learning how to utilize all his skills during crunch time, and not just his shooting.

• bluemars says 2 YEARs ago

These data are only for when the team is down by five or less, or tied. They do not represent when the team is winning. Those data are available, but I did not include them for the reasons you mentioned.

I tried the chart with change in assist ratio and offensive rating, but ultimately decided to go with the nominal values because they are easier to interpret. I added the dashed red average lines to try and show how teams move away from their original average as games near their end.

My fault for not reading more carefully!

• Richard Li says 2 YEARs ago

Thanks for taking my suggestion! One thing that this analysis doesn't take into account is the fact that you want to kill time at the very end of the game when you're up, even if it reduces your efficiency a little. I suspect this may contribute to the overall drop in assist ratio and explain the particularly precipitous drop in assist ratio of the Spurs in the last minute. Only including samples in which the team is losing would, of course, eliminate this bias.

With respect to the issue of colinearity between assist ratio and offensive efficiency, examining change in assist ratio versus change in offensive efficiency (during crunch time versus the rest of the game) would be a possible workaround.

These data are only for when the team is down by five or less, or tied. They do not represent when the team is winning. Those data are available, but I did not include them for the reasons you mentioned.

I tried the chart with change in assist ratio and offensive rating, but ultimately decided to go with the nominal values because they are easier to interpret. I added the dashed red average lines to try and show how teams move away from their original average as games near their end.

• thenit says 2 YEARs ago

According to NBA.com: USG...........Ast/game........%Ast.............FGA...........FGM...........3PA..............3PM......................
Harden.............................27.5%..........5.2...............34.6%...........16.4.............7.3.............6.4................2.1........................
Durant..............................30.3%..........4.9...............28.7%...........18.9.............9.4.............4.8................2.0........................
LeBron.............................29.2%..........6.6...............35.4%............15.8............9.3.............3.4................1.4........................
Who is playing "Hero-Ball"?? Harden has the less USG of the three.
What Harden needs to do is to cut down to four 3 point attempts per game.....May bring up Hardens FGM to level of Durant/LeBron per game.

Ypu are not getting the point it's the end of the game when the hero ball comes into play. USG doesn't really applied here
• bluemars says 2 YEARs ago

Thanks for taking my suggestion! One thing that this analysis doesn't take into account is the fact that you want to kill time at the very end of the game when you're up, even if it reduces your efficiency a little. I suspect this may contribute to the overall drop in assist ratio and explain the particularly precipitous drop in assist ratio of the Spurs in the last minute. Only including samples in which the team is losing would, of course, eliminate this bias.

With respect to the issue of colinearity between assist ratio and offensive efficiency, examining change in assist ratio versus change in offensive efficiency (during crunch time versus the rest of the game) would be a possible workaround.

• Jatman20 says 2 YEARs ago According to NBA.com: USG...........Ast/game........%Ast.............FGA...........FGM...........3PA..............3PM......................
Harden.............................27.5%..........5.2...............34.6%...........16.4.............7.3.............6.4................2.1........................
Durant..............................30.3%..........4.9...............28.7%...........18.9.............9.4.............4.8................2.0........................
LeBron.............................29.2%..........6.6...............35.4%............15.8............9.3.............3.4................1.4........................

Who is playing "Hero-Ball"?? Harden has the less USG of the three.

What Harden needs to do is to cut down to four 3 point attempts per game.....May bring up Hardens FGM to level of Durant/LeBron per game.
• Jatman20 says 2 YEARs ago Lets see if Durants numbers don't change now that Westbrook is out. I only feared Durant/Westbrook as the season started; with some help from Reggie Jackson and Ibaka (IMO Serg is the same as last year or regressed-w/o K-Mart I expected better numbers to start the season). Durant needs to score 40 points every night unless Serg picks up his game and Reggie becomes more consistent. Elite teams will beat the Thunder unless we see, "Hero-Ball"!! Welcome to our world--injuries to Asik, Beverley (2 times), Lin (back spasms), Parsons (back spasms), Greg Smith.......Hardens feet are terrible. Shooting starts with a good base. Harden walks out there like he is walking on egg-shells....prob lacks some sensation due to constant swelling in his feet. I attribute losses to no backup center (DMo hasn't cut it), Terrence (lacks experience)/Casspi Struggles vs bigger PFs, F Garcia has been a poor backup SG (offense has struggled), and PG play due to injury. Like AB but not good defensively. Normally not a problem because backup PG's tend to not play defense. Depth across the board hurts. Hard for starters to play 48 mins a game....Rockets score enough points. C. Lee and Bass may have helped. I wish the owner hadn't nixed that deal. Lee would have been the perfect PG.....true spot up shooter that can play defense. His DefRtg was great on the road at 6'5"........MCW (Phi) is a 6'6" PG.
• Sir Thursday says 2 YEARs ago

Perhaps if we could make a guess at the percentage of assisted shots that are made baskets we could neutralise the intrinsic correlation slightly? Naively, let's assume that assisted shots are made at the same rate as the team's overall FG% (almost certainly an oversimplification, but it's the easiest estimate to hand). Then if you divide the assist ratio by the FG%, you'll end up with a modified assist ratio (MAR?) that factors out the effect of missed shots wiping assist-worthy passes off the board.

ST

• Richard Li says 2 YEARs ago

I think there's one fly in the ointment of this data. Assists will tend to go up when scoring goes up, because unless someone scores there cannot be an assist at all.
Using SportVU data to track the number of passes compared to points would show a clearer correlation between ball movement and scoring.

100% true. I recognized that and almost included it in the limitations section, but decided not to because the piece is long enough as it is.

SportsVU doesn't track data at the team level, or team level data isn't available to the public if it does. So that's not an option. My ideal selfishness/heroball measure would be something like passes in the front court for every minute of frontcourt possession, both at player and team levels. But again, not an option.

Assist ratio is imperfect, but it's the best option available. There's certainly colinearity between assist ratio and offensive rating, for the exact reasons you just mentioned.

The important thing to recognize is that an extra assist per 100 possessions would have to replace a current possession that does not result in an assist. That possession could still result in points. So we would be replacing unguaranteed points with guaranteed points. In this situation, we would definitely expect an uptick in points per 100 possessions, but it shouldn't be a complete field goal's worth (2 to 3 points), maybe a fraction of a point or something.

However, the slope of trend line is, indeed, about a complete field goal's worth, or even a little bit higher. My interpretation of that is it's not just the assisted field goal itself on that single replacement possession that's increasing offensive ratio, but the style of play over several possessions that would result in an assist in one of those possessions. So teams are benefitting from the assisted field goal itself, but also the other possessions that don't necessarily result in an assist also have a higher likelihood of ending in points.

• Jeby says 2 YEARs ago

I think there's one fly in the ointment of this data. Assists will tend to go up when scoring goes up, because unless someone scores there cannot be an assist at all.
Using SportVU data to track the number of passes compared to points would show a clearer correlation between ball movement and scoring.

• TeamBall says 2 YEARs ago

Another fantastic piece of work, thanks! Its just so much more entertaining and fun to watch team ball vs. boring predictable ISO plays repeatedly, especially in the closing minutes of a tight game.

• Mario Peña says 2 YEARs ago I wonder how this measures out relative to the history of the NBA. I wonder if the culture of basketball changed with players like Jordan, Iverson, McGrady, Carter and Bryant and how they approached crunch time which in turn impacted younger generations perceptions of acceptable play at the end of the game.
• thejohnnygold says 2 YEARs ago

Thanks for doing this--it went pretty much as expected. This will only make future hero-ball-iso end of game situations more frustrating. :(