Justin Kubatko

Archives

by Justin Kubatko | permalink | trackback |

“Good teams win the close games.” That’s what the conventional wisdom would have us believe. But is that true? Do the good teams win the close games? To answer this question, I looked at how each team performed in close games versus how they performed overall. I chose to define a close game as a game that is decided by 5 or fewer points. If in fact it is true that the good teams win the close games, there should be a strong positive correlation between a team’s overall winning percentage and their winning percentage in close games.

Using all BAA/NBA regular season games from the 1946-47 season through the 2004-05 season, I calculated each team’s overall winning percentage and their winning percentage in close games. The correlation between the two winning percentages was 0.628 — a positive correlation, but not particularly strong. To put this another way, about 39 percent of the variability in overall winning percentage can be explained by its linear relationship with winning percentage in close games. A team’s winning percentage in close games is not a particularly reliable predictor of their overall winning percentage. Why would this be true? Because in a close game, luck plays a much larger role. The 2003-04 Detroit Pistons — the NBA champions — went just 12-17 in close games, while the team with the third-worst record in NBA history, the 1992-93 Dallas Mavericks, went 6-3 in close games. Did the Mavericks just “know how to win,” or is this result mainly a matter of luck?

Let’s look at how past NBA champions have fared in overall winning percentage, winning percentage in close games, and winning percentage in blowouts. Of the 59 champions, 30 finished first in overall winning percentage, and 17 finished second. Let me put this in a table:

 Overall Win % Rank Freq Percent 1 30 50.85 2 17 28.81 3 4 6.78 4 4 6.78 5 1 1.69 6 1 1.69 8 1 1.69 10 1 1.69

About 80 percent of NBA champions have finished with the first- or second-best record in the league. If the conventional wisdom is true, we would expect roughly 80 percent of NBA champions to finish first or second in the league in winning percentage in close games. However, that’s far from what actually happens:

 Close Win % Rank Freq Percent 1 12 20.34 2 10 16.95 3 8 13.56 4 5 8.47 5 8 13.56 6 1 1.69 7 1 1.69 8 1 1.69 9 2 3.39 10 1 1.69 11 1 1.69 12 3 5.08 13 1 1.69 14 1 1.69 16 1 1.69 17 1 1.69 20 1 1.69 22 1 1.69

Only 37 percent of NBA champions finish first or second in the league in winning percentage in close games, while 27 percent finish out of the top 5. Somewhat remarkably (at least if you believe the conventional wisdom), 11 of the 59 NBA champions (18.64 percent) have finished with a sub-.500 record in close games. Finally, only 7 NBA champions (11.86 percent) have had a winning percentage in close games that was better than their overall winning percentage.

If good teams don’t necessarily win the close games, then what do they do? Simple: They blow people out. I’ll define a blowout to be any game where the point differential is 15 points or more. The correlation between a team’s overall winning percentage and their winning percentage in blowouts is 0.899. That is, about 81 percent of the variability in overall winning percentage can be explained by its linear relationship with winning percentage in blowouts. I mentioned above that the 2003-04 Pistons were just 12-17 in close games, but they were 18-1 in blowouts. Meanwhile, the 1992-93 Mavericks — who had a better winning percentage than the Pistons in close games — were 0-44 in blowouts.

If we look at past NBA champions once again, we find that they are much more likely to finish near the top of the league in winning percentage in blowouts than winning percentage in close games:

 Blowout Win % Rank Freq Percent 1 25 42.37 2 13 22.03 3 9 15.25 4 2 3.39 5 5 8.47 6 1 1.69 7 2 3.39 10 1 1.69 12 1 1.69

About 65 percent of NBA champions have finished either first or second in the league in winning percentage in blowouts. (Recall that only 37 percent finished in the top two in winning percentage in close games.) No NBA champion has ever finished with a sub-.500 record in blowouts.

To further illustrate the disparity between looking at winning percentage in close games and winning percentage in blowouts, here are the top five teams from the 2004-05 season in winning percentage in close games:

 Close Games Team W-L Win % WAS 20-10 .667 ORL 17-9 .654 DEN 12-7 .632 NJN 12-7 .632 PHO 10-6 .625

Note that only one of the four conference finalists — the Phoenix Suns — made the list above. Detroit (12-9), Miami (12-10), and San Antonio (10-10) were 14th, 15th, and 17th, respectively, in this category. Now here are the top five in winning percentage in blowouts:

 Blowouts Team W-L Win % SAS 30-4 .882 MIA 18-3 .857 PHO 20-4 .833 DAL 20-5 .800 HOU 18-5 .783

The top three teams were all conference finalists (the Detroit Pistons finished in a tie for 10th in this category). If you were asked to name the five best teams in the NBA last season, the second list would be much closer to your answer than the first.

Let me wrap this up by summarizing the two key points of the article:

1. A team’s winning percentage in close games is not a particularly reliable predictor of their overall winning percentage.
2. A team’s winning percentage in blowouts is a much better predictor of their overall winning percentage than their winning percentage in close games.

The statement “good teams win the close games” does tend to be true, but a much stronger statement is “good teams win the blowouts.”

Published on Thursday, January 5th, 2006 at 3:47 pm

37 Responses

John R

Oof, I just wrote quite a long objection to your definition of close game and how its applicable here and I lost it because I mistyped the authorization code.

In brief, I think your definition of close game is flawed. There are numerous ways to arrive at a game decided by 5 or less points and not all of them have anything to do with the game being close or in contest down the stretch. What will be ultimately interesting is finding out why some teams, having good or bad overall records, win the close games. For that, we need a definition that describes a close game as one that is being contested near the end of the game and not actually at the finish, that is to say that either team has a reasonable chance to win within the last 2 minutes, or so, of the 4th quarter. To me, a game decided by 11 when one team was up by 5 with 1 minute to go is much more an example of a team winning a close game than a team down by 11 with 10 seconds to go and hitting 2 uncontested garbage time 3’s to cut it to your 5 points. Admittedly, defining close game in this way makes the research more difficult, but it is REALLY what someone means when they say good teams win close games. The final score of the game doesn’t always tell the story…see Clippers vs Spurs from 12/13/05. The Spurs were two unlikely Duncan free-throws, then a missed Kaman layup at the buzzer away from losing in regulation, but pulled it together to win by 8 in OT. An EXTREMELY close game, but one that would not show up in your stats at all, and one the Spurs did just enough to get the win.

Yes, I’m a Clippers season ticket holder. Yes, the injuries have been crushing.

I don’t think it is all that reasonable to draw conclusions about performance in close games based on final score alone.

JonL

Good article, but perhaps an area for further study is overtime games. Are good teams less likely to play OT games, are they more likely to win close OT games or games with a margin greater than 5, etc.

Kevin Pelton

Certainly games can go in and out of a five-game margin without really reflecting their true “closeness,” but I don’t think this happens frequently enough to get worked up about, especially when you consider that it’s not just difficult to do but impossible for anything beyond three or four years back.

Roland Beech did use this method last year for SI.com, and while he imbued it with a fair amount of significance, you might not want to when you see who his poster child was for poor clutch play.

John R

“I don’t think this happens frequently enough to get worked up about.”

Seems like something than can be proven or disproven instead of just speculated on, but that still leaves out games that are close but just not precisely 5 points close or closer.

My larger point is that focusing on a 5 point spread really addresses the question: do good teams win games decided by 5 points or less? Instead of: do good teams win close games, when the “conventional wisdom” is defining close as meaning still in contest near the end of regulation? I am not questioning any of the data or conclusions from it, only what has actually been addressed here.

John R

Not to beat a (possibly) dead horse, but just look at the nationally televised IND/GS tonight. One can’t possibly argue that game isn’t what people mean when they say “close game” (it was 2 points with 1:46 to play), but the final spread (10?) was more than 5, so it wouldn’t even be considered in the analysis. But its perfect (anecdotal) evidence of the conventional wisdom, a team without either of its all-stars (O’Neal, Artest) finds a way to win thus establishing that it is a “good” team.

I’m not trying to be difficult here, I just think the new basketball math is capable of better. Mr. Kubatko, Mr. Pelton, as a big fan I challenge you to go back to the drawing board on this one.

Justin Kubatko

John R, quite honestly I don’t have the time nor the data to do the study you are suggesting. If you want to do the research, I’m sure we could put it on this site. However, your idea is not foolproof. For example, how would you define a close game? Wherever you decide to draw the line, someone will find a game that they think is a close game that doesn’t meet your definition. You can’t win.

All that said, I really don’t think it would make much of a difference, and I think my conclusions would remain essentially the same.

Kevin Broom

I agree with Justin and KevinP here. In addition to being good use of the available data, Justin’s research echoes what we already know about scoring differential.

Andrew

It’s a very good article, but I think John’s point is still valid. I understand how much more complicated it would make the analysis, and I’m not asking someone else to do it. But certainly there is no harm in pointing out the limitations of this approach?

Garbage time and cruch time affect how teams play, and one of these two scenarios usually makes a significant apperance in most games.

In garbage time the leading team usually clears their bench, generally lets their guard down, and waits out the clock. This can give their opponent a chance to narrow the blowout into what looks like a closer game.

In cruch time, if one team is behind and running out of time, they abandon their standard efficient offense and they gamble. They intentionally foul and give their opponents some easy points and then they rush and throw up quick threes. This can easily backfire and turn a 3 point game into a 7 point game — and make a close game look like it wasn’t competitive at all.

KnickerBlogger

As per a 2 point game turning into a 10 point one, isn’t that just the same as a team turning a 20 point deficit into a 10 point one? Isn’t being able to knock down free throws at the end of a game or trading fouls for 3 pointers a part of winning?

How else can you define a close game? A team could be losing by 15 at the half, tie it in the 3rd, and lose by 10. Is it a close game? A team could be up by 2 with 4 to go, but if they lose by 10, was it a close game?

By not using the final score, you’re entering a slippery slope.

KnickerBlogger

ALSO - for all those that “lose” a post due to the wrong authorization code:

I’ve found that if you just hit the back button on your browser, the post is still there. At that point, I just hit referesh, and a new authorization code comes up, with your comment still intact.

(speaking of which my word this time was “mikan”)

TK

82 games definition of clutch time is a reasonable definition of close to me: “4th quarter or overtime, less than 5 minutes left,neither team ahead by more than 5 points”

I’d be far more interested in the team records in close games in the playoffs than regular season.

Tim Duncan, the poster child, has a relatively weak clutchtime offensive game especially for a superstar but still gives you the other stuff and he is surrounded by an overall well above average clutchtime cast (including several of the very best in the league clutch shooters like Horry) who cover for his weakness.

Andrew

“As per a 2 point game turning into a 10 point one, isn’t that just the same as a team turning a 20 point deficit into a 10 point one? Isn’t being able to knock down free throws at the end of a game or trading fouls for 3 pointers a part of winning?”

Yes. The only point is that these games won’t show up in this analysis. Teams who win close games in this fashion won’t get credited for it. And teams who lose close games in this fashion won’t be penalized. And equally important, games which weren’t truly close will slip in and be counted.

“By not using the final score, you’re entering a slippery slope.”

Absolutely. I don’t have a better solution. I was just saying that John’s main criticism had some validity.

TK

I would assume there is a very strong correlation between close game winning percentage in the playoffs and overall playoff record and that the champion is very likely to be #1 0r #2 on close game winning percentage. I am not sure it really means anything beyond the obvious, but it would be somewhat in contrast to the main conclusion drawn from looking at the regular season.

John R

It is not in fact a slippery slope. It is about picking a single worthwhile and meaningful definition. The score at the end of the game is not a worthwhile definition to either counter the “conventional wisdom”, because it has little to do with what the conventional wisom is talking about, nor to garner further potential useful information, such as how do the teams that win close games more often do it.

A more useful definition for close games would include finding the average length of a possession during the last 2 minutes of a game and coming to a determination about teams having some reasonable chance of winning based on historical data concerning deficits overcome. Though still not perfect, this is far less arbitrary than comparing the final score. My example from last night still stands. 2 point spread with 1:46. This is what the conventional wisdom means when it says close games, but because the final score was more than 10, it is not even considered in your analysis. I’d hazzard to guess more NBA games fall into my version of close games than yours and would therefore increase the data set and be much more meaningful in terms of reflecting how many close games an NBA team is in over the course of the season.

“As per a 2 point game turning into a 10 point one, isn’t that just the same as a team turning a 20 point deficit into a 10 point one?”

Based on your statements, it isn’t enough information. If these things happen in the first quarter, they are functionally similar, but if they happen in the last minute, they are quite different and exactly what should be examined if the conventional wisdom is to be challenged. When people say good teams win close games they mean good teams turn 2 points into 10 when the game is on the line.

“All that said, I really don’t think it would make much of a difference, and I think my conclusions would remain essentially the same.”
“I agree with Justin and KevinP here. In addition to being good use of the available data, Justin’s research echoes what we already know about scoring differential. ”

And so the new “conventional wisdom” is formed…

TK

But, to back up for a second, this was a very good article, examining conventional wisdom more closely and opening a topic for conversation.

I would be interested in the correlation of close game winning from regular season to playoff. I assume against better teams blowout level performances from a team could end up more closely matched and pull a larger share of games in the close category, maybe the close game win percentage of the champions move to a common level (at least 60%?). It would be interesting to see the close game win percentages of regular season and playoffs for recent year top contenders and see the movement or trends vs. how they finish up.

Maybe in addition to closegame win percentage it might be valuable to know the raw number of close game losses (winnable near the end games not won) and the number of blowouts experienced (games blown so badly and early they weren’t recoverable). It might be that there is a stronger correlation between these statistics and overall success than close game win percentage.

Justin Kubatko

John R wrote:

And so the new “conventional wisdom” is formed…

Hardly. Did you even read my conclusions?

The statement “good teams win the close games” does tend to be true, but a much stronger statement is “good teams win the blowouts.”

I stand by that statement.

TK

“A more useful definition for close games would include finding the average length of a possession during the last 2 minutes of a game and coming to a determination about teams having some reasonable chance of winning based on historical data concerning deficits overcome. ”

I think that is a solid suggestion for collecting additional useful information.

Justin’s study is a very good base but there is room to discover even more.

Kevin Pelton

It’s easy to sit on the sidelines and criticize others, but what work are you willing to put in to address these criticisms, JohnR?

I would love to read a more detailed study.

Chris

I found this to be a very informative article. I don’t have access to the links now but I do remember reading very similar articles on both Football Outsiders and Hardball Times about how good teams blow out their opponents and winning close games isn’t always an indicator of a good team. So I find it interesting that this occurs in pro basketball, football and baseball.

Kevin Pelton

Here’s the Aaron Schatz NFL offering.

JonL

I just compiled the OT numbers for the 04-05 season, but I’m not going to go into detail because it doesn’t look like there’s any strong correlation with winning %(If one can make such correlations anyway based on small sample sizes). Most teams are around .500, except for a few teams (the Spurs won all three of their OT games, the Hawks and Rapters were horrible in OT games, etc.)

John R

You are right, it is quite easy to criticize. In fact its what I do professionally. But that doesn’t make me incorrect in any way. As a rule I find the writers on this site very good, but absolutely allergic to criticism, even when it is fair like mine has been.

Stand by your statements all you want, that doesn’t change the fact that what you proved has nothing to do with conventional wisdom and everything to do with the new math’s belief system. That belief system is well on the way to becoming new conventional wisdom, despite insufficiently addressing the old.

To answer a question with a question, did you even read the whole of my posts or did you just take out one line so that I could be misquoted out of context? I’m not here to be difficult. I feel I’ve gone out of my way to use non-confrontational and respectful language. I’m just pointing out a flaw in a single definition. But if I was wrong you could put up better reasoning than the 2 pseudo personal attacks that are the posts at 1:26 and 1:31.

Justin Kubatko

John R wrote:

I’m just pointing out a flaw in a single definition.

Okay, let’s get focused. What is your specific definition of a close game?

TK

I know you asked John R… but since I think of the topic as being open for general discussion, here is my own off the cuff answer (using the process for selecting a standard suggest by JohnR): Close would be somewhere between a historically documented 40% chance to still win with 5 minutes to go to at least a 30% chance with 2 minutes to go. I’d tend toward the former over the latter as the definition but expressed a range that I think is reasonable. Less than these odds with that amount of time remaining doesn’t seem close enough to me to call close.

TK

I went looking for a database that might make it easy to view the data on the raw number of close games and blowouts lost, or the playoff data. Yahoo stats is set up to provide it at 3 and 10 point breaks but those categories wouldnt work for me.

Did you find a good database to use for this type query or is it possible to do an advanced query like this at your site? If it was highly manual, I respect the work that went into it.

TK

The value of looking at the flipside, loss side of the equation can be found with the 2003-04 Pistons. Of 28 losses, 17 were still close, 10 in the middle and only one blowout. That is the distribution of a team that is in most games and I believe partially offsets their below .500 record in close games.

In line with the article’s conclusion but an extension on it, perhaps losing close games shouldnt hurt your reputation that much, maybe delivering blowouts is a bigger deal - and I would add avoiding blowouts as well.

Not all good teams have this quality losses by margin profile, some get blownout more frequently and could have fewer close losses and might look better because of it on a percentage comparison. The more views, the more information, the better the understanding.

Coupling a pretty strong blowout frequency and the ability to almost totally avoid blowouts was a good combination for the 2003-04 Pistons.

John R

While I am rather good at criticizing, I am not by any means a statistical whiz. Any reasonable approximation that takes into account the fact that a game is being contested as opposed to the final score would seem to be a better fit to me. I acknowledge that this will be a (somewhat) arbitrary selection, though no more arbitrary than selecting a final score differential, so I would recommend selecting something that a statistical whiz would be able to cull from historical data.

Some possibilities:
1) Determining a league average for possessions used for the last 2, 3, 4, or 5 minutes and only include games that were not impossible to lose based on the league average points per possession times possesions remaining at that chosen time. This would seem to be the most inclusive case of a game still being contested, or close, that I can think of. Alternately, to be even more inclusive, select the maximum possessions ever used over the given time period or the league average times a small multiplyer.
2) Similar to TK’s method, determine the historical likelyhood of a team making a comeback at a given time and select a reasonable percentile chance and say that games within that spread at that time are games being contested. For example featuring entirely made up stats, 30% of teams that were behind by 10 with 2 minutes to go were able to come back and win would dictate games spread by 10 points or less at that time would be “close”. Multiple time points could be chosen requiring possibly an and or or test for inclusion. 90% chance to win with 5 minutes left and 30% chance to win with 2 minutes left, etc.

Again, I admit that any selection will be arbitrary. But more than any sport, the final score in basketball can be very misleading. I think that almost any reasonable definition of “close game” that really considers whether a game is actually being contested down the stretch would be better.

TK

This is a limited look but I found that the 03-04 Pistons were 11-3 in blowouts in the playoffs and 2-2 in close games. They lost a little ground in 04-05 on both going only 8-5 for blowouts and 1-2 in close games.

The 03-04 Spurs were 5-2 in blowouts, 2-3 close. In 04-05 not much different percentagewise 9-3 in blowouts 2-3 close.
Meeting the Lakers with the 2 allstars was the biggest difference in terms of playoff outcomes for those years for the Spurs.

In terms of number of games in each category, blowouts were much more common than close games. That meant suggest that blowout percentages are more important than close games. But this is just two teams, two years and not presented as a historically meaningful trend. That would take further checking if someone wants to. Time for the weekend.

TK

Correction: Spurs in close games 03-04 should have listed as 1-1.

TK

I should acknowledge that avoiding blowouts is recognized in the article’s second conclusion. I didnt give full notice to that part of it initially but ended up focused on it myself then found it again in your writing.

My quick mini study of the playoff for two team, two years is completely consistent with both of your conclusions.

And to reality check myself:

“I would assume there is a very strong correlation between close game winning percentage in the playoffs and overall playoff record and that the champion is very likely to be #1 0r #2 on close game winning percentage. ” TK (earlier)

The facts suggest otherwise. 03-04 champs were 2-2 in close games as were the 04-05 champs. Not bad, but probably not best, and not really distinguishing.

Pat

I think John R has a valid point, and while altering the method may not change the overall conclusions, it’s still good to do so.

I think a quick ‘n’ dirty way to approximate a “close” game would be to look at games where the scores at the end of some combinations of the quarters was less than five. For example, a “close game” could be defined as one in which the point spread was 5 points or less at the end of both the first and third quarters, or at halftime and the end of regulation. It’d require a bit more work than what you’ve done presently, but I’d imagine this sort of data is more readily available than the scores at some intermediate point like two minutes left in the game.

Kevin Broom

I think John makes some worthwhile suggestions, BUT obtaining the data for this seems a daunting task. Roland at 82games may be able to generate what we want over the past few years through his pbp database. Perhaps some smart programmer could do the same further back.

I’m intrigued by this comment: “But more than any sport, the final score in basketball can be very misleading.”

I wonder whether this is true. It does sometimes happen that a hotly contested game ends up with a larger scoring margin than might be expected (and vice versa — a blowout ends up closer), but points differential is an extremely effective measure of relative team strength. A team with a scoring differential of +10 per game better than a team with a +2 differential.

I agree that Justin’s study isn’t “perfect.” But, I still think it’s an excellent use of the available data. I’d like to see someone take on the next step — doing a game-state analysis akin to Roland’s definition of “close”.

Mike Goodman

I’d think the difference between a team’s W-L record and their ‘pythagorean W-L record’ would be a pretty good proxy for ‘winning close games’.

Unless game-by-game Pt-Diff is very non-normally distributed, a good team wins the majority of their blowouts and a lesser % of their close games.

Justin Kubatko

I’m willing to do another study, but I don’t have the time to collect the data. Several of you have made very good comments, and if you can provide the data then I will write another article.

Andrew

“I’m intrigued by this comment: “But more than any sport, the final score in basketball can be very misleading.”

I wonder whether this is true. It does sometimes happen that a hotly contested game ends up with a larger scoring margin than might be expected (and vice versa — a blowout ends up closer), but points differential is an extremely effective measure of relative team strength. A team with a scoring differential of +10 per game better than a team with a +2 differential.”

I think the comment is true if you look at any individual game — garbage time and cruch time strategies will warp the final score. More than other sports, basketball teams will gamble with inefficient strategies when the clock starts to run out.

However, these forces apply to all teams in most games throughout the year, so everything averages out with a large data set. This makes point differential reliable… and leads well into Mike Goodman’s suggestion.

Alan Reifman

It’s interesting to speculate on how the findings might (or might not) transfer to the college level. A couple things about the college game — the shorter distance of the three-point line and one-and-one free-throw situations — could lead to greater volatility at the ends of games in comparison to the NBA. This presumably would weaken the connection at the college level between overall winning percentage and winning percentage in close games.

Rob

Lead changes and ties internal to a game could also be an excellent way to determine “close”

Leave a Reply

Check Spelling
Activate Spell Check while Typing