Dr. Krzyzewski or: How I Learned to Stop Worrying and Love the Stall

[throatybeard]

I hear you but really, how many previous tourney games have slipped away from us after we've gone into the stall? I remember the 1986 championship game and the 1998 Elite Eight game.

Not the 1998 Kentucky game. We were up 71-54, but with too long left to run the stall. In fact, we were taking some quick threes.

Then they murder us on their next five possessions, getting three points on four of the possessions and four on the other one to cut it to 71-70. It was a blitzkrieg the likes of which you probably never saw before and probably will never see again. It makes JWill and Reggie Miller look like guys screwing around on their back yard hoop. (After all, Reggie only hit two threes).

Too, I thought the basic deal with the 1986 Louisville game was that Dawkins and Henderson just couldn't hit a bull in the butt with a banjo down the stretch. But there was a 45-second shot clock back then, making a stall more attractive. It's been 25 years since I saw that game (my parents didn't own a VCR untill 1987), so I can't speak too definitively to it.

Anyway, the best thing is that now we have a thread in which to merge all future horse-beatings of this topic.

[dyedwab]

...so here goes :-) This is a very interesting thread, btw.

My take is that there are three basic question about "stall ball" and how they apply to the Michigan game.

2. Should we run "stall ball"?
3. How should we run "stall ball"?
4. Did we run "stall ball" effectively?

To the first question, I think the numbers are pretty clear that "stall ball" is a fairly effective strategy - the math works. I grant that there are potential issue with our players losing their aggressive mentality, but, it generally seems to work. I admit that I was actually happy to see us go into a stall against Michigan because they had fought back from the 15 point lead, it seemed that they were making every shot, and they seemed to be dictating pace. Now, we didn't run it that effectively, but we still won...

To the second question, I have always thought they we started our offense too late during the shot clock. I have always thought that we bring the shot clock down too low before we initiate offense - I've preferred that it start at 13-14 secs, not the 10 that we normally due. However, with Nolan's ability to turn the corner on nearly every defender and his array of shots in the lane, I've become less worried about. (And with Kyrie able to do it also?). Against Michigan, I'm not sure that it would have mattered when we started our offense because...

...how did we execute at the end of the game? Not all that well. Instead of having Kyrie or Nolan taking his man one-on-one off the dribble to initiate, they dribbled into a trap - with three men guarding him. And it befuddled us... And, on the other side of the court, we didn't do as good a job stopping MI as we needed too - thus the game was closer than we would have liked.

To sum up, it seems that our failure to execute at the end of the game is leading to questions about strategic decisions that, at least to me, seemed pretty sound.

[TampaDuke]

The whole point of the stall is the math. No strategy gives 100% success. If you've watched thousands of games, like you say, and in 100 of them the stall was employed, how many resulted in a loss? Probably only a couple. Sure those stand out in our memories, but if our success rate with any strategy is 98% (and I'm not saying it is; obviously I just made those numbers up), then it's a pretty good strategy.

Now, if we'd lost 30 of those 100 games, then it doesn't sound like such a good strategy. But the number of games we've lost after employing a stall is a lot closer to 2 per 100 than it is to 30 per 100, and that's why the mathematical equation is a better judge than a one game sample.

Stall ball certainly isn't pretty and I think Kedsy brings up good points about % chance of winning in stall vs. not, but I think that it does give the team a better chance of winning if the team executes poorly down the stretch and if the team executes well, it doesn't take long for the game to be out of reach.

I understand these points, but would you agree that employing stall ball wouldn't be the proper tactic, mathematically, if a different tactic resulted in greater odds of winning?

The fact that stall ball might lead to wins in, say, 90% of our games doesn't by itself demonstrate it's the best tactic.

As several others have hit on, what's missing from much of this analysis is a comparison with non-stall tactics and the expected percentage of winning when employing those tactics.

[Kedsy]

I understand these points, but would you agree that employing stall ball wouldn't be the proper tactic, mathematically, if a different tactic resulted in greater odds of winning?

I think the analysis is probably a bit more complicated than that, but I'd be willing to entertain the idea. Unfortunately, I don't have those stats. If you do, please share; don't be shy.

[Jderf]

I would love to have an entire season of "almost losses." That would make us undefeated.

It would also put a large portion of our fan base into psychiatric institutions.

[juise]

I would love to have an entire season of "almost losses." That would make us undefeated.

This sounds like UNC's last 6 weeks (save the Duke ACCT game).

[Dr. Tina]

I love the name of this thread! Gave me a good laugh...that's all...

[TampaDuke]

I think the analysis is probably a bit more complicated than that, but I'd be willing to entertain the idea. Unfortunately, I don't have those stats. If you do, please share; don't be shy.

I don't have the stats, and I agree that the analysis would be quite complicated. Not necessarily meaning to single you out here, I'm just pointing out that a review of the percentage success rate of stall ball doesn't necessarily answer the question of whether it's the best tactic amongst several admittedly good options.

I'm not a stats guy, but I guess if someone where to do analysis, my suggestion for a starting point would be a comparison of def and off efficiency of each team while using stall ball versus employing typical non-stall tactics, while somehow also factoring in the number of expected possessions each tactic would permit both teams. That's just my initial thought, however. Personally, I'd also like to see home versus non-home forums factored in as well (stall ball never really bothers me at Cameron, but on a hostile floor I tend to think momentum, particularly when playing as the favorite in a tournament format, is vastly underrated).

[Newton_14]

Awesome thread with great discussion and cool title. This is why I love DBR. Good stuff and I commend all on both sides of the debate.

I have to confess that I love the stall and fully believe it is the best option when you have the personnel to run it properly. I have to also confess that in every "big" game, if we get a double digit lead at any time in the 2nd Half, I immediately start watching the clock and praying we can make it to the under 8 Minute Timeout and still have a double digit lead.

My "magic" number is 13 points. If we can make the under 8 TO with at least a 13 point lead I always feel very confident we will win. Especially when we have a PG that is effective at getting into the lane and finishing. We have Nolan who is great at this. If Kyrie had not gotten hurt he would have been the ultimate.

I am sure K has the math calculated to a science with the margin of the lead/time left, but he varies the strategy game by game based on variables such as foul trouble, fatigue, opponent, how the game has gone, etc. So he starts the delay at different times, almost always sometime after the under 8 TO. I have seen him start it right out of the timeout, or sometime between there and the 3 minute mark, depending on the variables.

He also will sometimes "ease" into the delay, meaning he will deliberately slow the pace after the under 8 TO, but not make it obvious, and not use the entire 35 seconds. Then at some point he will go into the full delay with Nolan holding it out top in the corner.

I do agree with "InSpades" on the idea of waiting longer to inbounds after made shots to burn even more time. Not sure why we do not do that. I also agree with Royal08, in that I have watched many games where a team had a good working lead with less than 6 minutes and gave it all away by not working the clock and taking quick shots.

I did not get upset with the delay against Michigan. Like Matt pointed out we lost much of the lead before we went to the delay. I was upset with the lineups K went with. I thought he made a mistake by going small to adjust to Michigan vs going big to force them to adjust to us. I know Mason had a bad game, but most of his mistakes were on offense. Same for Miles.

I thought the line up of 4 guards with Kyle at the 5 was a bad idea for 3 reasons:
1. We had rarely deployed that lineup with or without Kyrie, and had not deployed it at all since prior to Christmas except for a couple of times with less than 2 minutes and the opponent fouling on purpose. That put the guys in different roles than normal on offense and I feel it confused them.
2. We just got Kyrie back. So we have a player that has not been on the floor since December, and that too made it tough on the other 4 guys as they are not used to having him out there, and they are in a rarely used lineup configuration anyway.
3. The 1-3-1 zone was extremely difficult and our smallish lineup was probably the worst lineup for us to attack that zone. We needed one of our "2 Bigs lineup with Kyle at the 3" to give us our best chance against that zone. (My Opinion)

The lineup of 3 guards + Kyle and Ryan was a little better, but still a lineup we have rarely used and one of the guards was Kyrie, which again made it more difficult on the other 4 players.

I am convinced that K went to those 2 lineups for defensive purposes exclusively, but the fact that Michigan was scoring anyway seemed to negate any benefit of having that lineup out there. Maybe K also used those two lineups to give us five 3 point shooters against the zone, but the problem there was we did not move the ball fast enough to get one of those shooters a wide open look.

I rarely question K on strategy. There is a reason he has won 900 games. Sunday was the rare occasion where I questioned who he had on the floor, and like someone else mentioned, there is nothing wrong with questioning the King on occasion, and in the case of K, I think he would agree with that.


That's my take on both the delay and the strategy down the stretch Sunday. At the end of the day, I am simply glad the end result was win number 900. I get to start watching the clock in the 2nd half at least one more time this season, praying for that 13 point Duke lead at the under 8 TO!

[InSpades]

If we wanted we could truly do some math and analyze this properly. Assume that their points scored and our points scored is a distribution (the sum of binomial trials?). Figure out the variances and when things overlap with the other team making up the lead differential.

We would need to make some assumptions about possession times for each team in each scenario (stall vs. non-stall). As well as points per possession in each scenario. It wouldn't be *that* difficult and I think we could come up w/ some reasonable conclusions (such as going to stall in a certain scenario gives us a 97% chance to win, whereas not going to stall gives us a 95% chance to win).

The flaw in the math I saw before is that it didn't really take into account variances. I mean, pretty much any strategy we employ is going to win us these games if things go averagely. It's the times that things go badly that we need to worry about and aren't really reflected well without looking at full distributions. I think we will all agree that stall ball decreases variance but also puts us closer to losing. While playing normally will increase variance but hopefully keep us further ahead on average.

Someone want to volunteer to do the math? . This may be a project for the off-season!

[Newton_14]

If we wanted we could truly do some math and analyze this properly. Assume that their points scored and our points scored is a distribution (the sum of binomial trials?). Figure out the variances and when things overlap with the other team making up the lead differential.

We would need to make some assumptions about possession times for each team in each scenario (stall vs. non-stall). As well as points per possession in each scenario. It wouldn't be *that* difficult and I think we could come up w/ some reasonable conclusions (such as going to stall in a certain scenario gives us a 97% chance to win, whereas not going to stall gives us a 95% chance to win).

The flaw in the math I saw before is that it didn't really take into account variances. I mean, pretty much any strategy we employ is going to win us these games if things go averagely. It's the times that things go badly that we need to worry about and aren't really reflected well without looking at full distributions. I think we will all agree that stall ball decreases variance but also puts us closer to losing. While playing normally will increase variance but hopefully keep us further ahead on average.

Someone want to volunteer to do the math? . This may be a project for the off-season!

Excellent idea for a summer project and discussion. Surely there are a few "Math Experts" amongst our regular DBR Members...

[Richard Berg]

The fact that stall ball might lead to wins in, say, 90% of our games doesn't by itself demonstrate it's the best tactic.

You're right. But just for completeness, a different strategy with a higher win rate in "equivalent" game scenarios isn't necessarily better, either. Variance/skew in the results matters. The best choice could be radically different in the regular season vs a single-elimination tournament.

[MarkD83]

Here is a crude attempt at the math. Lets take the up 13 with 8 minutes left scenario. (Please check my math if you have a chance.)

This is 480 seconds and lets assume Duke takes 30 seconds off the clock and the other team takes 10 seconds to do something (score, miss etc.). This is 40 seconds for a possession each which means 12 possessions per team in 480 seconds.

If we divide the point differential into 11 bins separated by 0.2 points per possession ranging from Duke scores 1 point more per possession to Duke scores 1 point less per possession. If all things go according to plan Duke wins in all scenarios (-12 with a 13 point lead is the worst case).

However, on the last few possessions of the game the other team could have the ball and be within 2 or 3, depending upon how Duke scores. For example, lets take the almost unbelievable scenario where Duke hits 2 freethrows on each possession and the other team hits three pointers each time down...

Duke up 13; 3 ptr by other team Duke up 10; 2 free throws Duke up 12; etc...with ~2:00 left the game is a 1 possession game and the other team will change their strategy because the game is wide open once again. So...what is the point differential to make the game a 1 possession game with ~ 1 minute left.

If Duke hits 2 free throws and the other team scores a three pointer two out of three possessions and a two the other possession (a differential of 0.67) Duke would have a three point lead and the ball with 40 seconds left. Lets count this as a possible OT game (Duke misses and the opponent hits a three at the buzzer to tie the game).

So the entire distribution goes from +1 to -1 and -0.67 is the cut off for a tie, so 83.33% of the distribution is a tie or win and 16.67% chance of a loss.

Hummmmm...I think I would like Duke to be leading by 16 with 8 minutes left.

[sporthenry]

Of course, it is difficult to criticize K. But I have to say that saying the effectiveness of stall ball when we are up 10 is 90% effective is misleading when I'm sure playing non stall ball is very similar. I think the most important part of the situation is being up 10 with x minutes left.

For getting to 900 wins, stall ball is obviously very effective but in a one and done elimination style, statistics have a funny way of not always working. So forgive me if I'm a bit hesitant to get away from what got us that lead. This game might have been a bit different b/c we really started stall ball when they switched to the zone (which they probably should have played more but I guess they feared Duke's offense). And we really didn't get many opportunities to run as KI isn't back to his full speed so I get the pulling the ball out especially vs a zone where there is absolutely no pressure. I also believe it was started when Kyle was saddled with foul trouble. And KI never saw the 1-3-1 and Michigan's offense was fluky and I could understand wanting to cut their possessions.

However, the original poster mentioned Duke's offense with the likes of Jay Williams and Boozer and how they didn't need to run stall ball however, I would claim that with a full strength KI, this Duke's offense is its strength. Using kenpom, our offensive efficiency obviously fell once KI left comparing our 4 major non conference games with KI to the conference games. But Duke put up top 5th best offensive efficiency games against MSU, Butler, and K-state when MSU and K-state are the 29 and 30th most efficient defenses in the country. So if KI can get his speed and quickness back, I would say that stall ball shouldn't happen until Thornton is checking in as that was our best offense all year.

[devildownunder]

Two problems with stall ball.

1) The way we run it, the strategy too often amounts ZERO offense and horrendous shot on every possession. From a purely strategic standpoint, this is my problem with it. The only year I can remember in which we were consistently good at executing an offense via stall ball was 2004, Chris Duhon's last year. Senior-year Duhon knew how to work the point in a 4-corners-style offense (which is what this is, and boy does it leave a bad taste in my mouth saying that), so we frequently got great shots out of it. Of course, that didn't stop us from losing to Maryland while running it in the ACC tournament final.

2) But, I admit that my biggest problem with stall ball is that it just feels like such a cheap, unmanly, "loser's mentality" approach to winning (there. I said it!). I mean, really, the basic idea is, let's try to work out a mathematical formula that allows us to stop playing anything that looks like basketball and win the game by simply standing around, milking the clock for as many as 10 minutes. I get angry and frustrated just thinking about it.

It takes a lot for me to get to this point with a strategy because I am, typically, a VERY pragmatic guy when it comes to sports. I want to win. And I'm not typically for style points or anything like that. Heck, even with stall ball, if I knew it were going to work, I'd probably hold my nose and accept it. But it's not infallible. The original post suggested that K has been trying to boil it down to an invincible formula. Well, if that's the case, K has yet to find that formula -- if it even exists. We've lost games with stall ball (most painfully for me the 2002 sweet 16 to Indiana). We've blown leads and fallen behind with stall ball. We've won games with stall ball. And we've done all those same things by continuing to run our offense. K's winning percentage has been outstanding over the years, whether he's used this stomach-turning (to me anyway) approach or not. So (and here I will stress that I know it's K's program, he has no idea who I am, nor does he care, and he'll do what he wants. Doesn't mean I can't have my opinion.) I would much prefer he went at it in a matter that made me feel proud to put on one of my Duke basketball T-shirts. No matter what the outcome, when we go to stall ball, I'm ashamed to admit I'm a Duke fan. It's hard for me to even look other fans in the face after we trotted out that stuff.

[devildownunder]

Not the 1998 Kentucky game. We were up 71-54, but with too long left to run the stall. In fact, we were taking some quick threes.

Then they murder us on their next five possessions, getting three points on four of the possessions and four on the other one to cut it to 71-70. It was a blitzkrieg the likes of which you probably never saw before and probably will never see again. It makes JWill and Reggie Miller look like guys screwing around on their back yard hoop. (After all, Reggie only hit two threes).

IIRC, the biggest problem we had with Kentucky 1998 was that we had a very young team and K had to keep calling timeouts to try to build and maintain that lead. He made a lot of adjustments and did a lot of soothing. Unfortunately, that left us with zero timeouts for the last 8 to 10 minutes of the game. When Kentucky started making its decisive run, only the TV timeouts stood in their way, and that wasn't enough to stem the tide. We most certainly didn't run stall ball. In fact, we didn't seem to have much of any plan during the critical stretch of that game because we became rattled. Chalk that one up to youth. That was the Battier class's freshman year. Super-talented group. There were better days ahead.

[trinity92]

I really do think that Kyle having 4 fouls played a role in our going to the stall on Sunday. (And I don't care what K said afterward - I have eyes - that was stallball.) With Kyle defending Michigan's "5", I think he had a greater chance of picking up his 5th foul than he would normally.

Matches points out something that very few have acknowledged here-- Kyle's fourth foul made limiting the number of times Kyle had to defend and risk fouling out more important than limiting Michigan's offensive possessions in general. To me, it was the best reason to have gone to the stall.

Having said that, we were playing against the 1-3-1, which takes time to pick apart even if you're running your normal offense-- if our goal was to shorten the game, Michigan did us a favor. We could have initiated our offense normally with instructions from the coaches to be methodical and had a much better chance at a good shot. In addition, Michigan would have had to defend rather than being able to rest and be fresher on offense. The zone we were facing was designed to protect the middle from our superior size, not to create turnovers on the perimeter. I don't think there was a big downside to running our offense trying to crack it.

Although I've struggled with going to 4 corners in the past, I've gotten to the point where I see us up and starting to stall and think "well the game's over now-- good win." I just don't think the stall was employed effectively against Michigan. I mean, we won, but I'm not prepared to say we won because of how we ran the stall-- we may have won in spite of it.

[InSpades]

Here is a crude attempt at the math. Lets take the up 13 with 8 minutes left scenario. (Please check my math if you have a chance.)

This is 480 seconds and lets assume Duke takes 30 seconds off the clock and the other team takes 10 seconds to do something (score, miss etc.). This is 40 seconds for a possession each which means 12 possessions per team in 480 seconds.

If we divide the point differential into 11 bins separated by 0.2 points per possession ranging from Duke scores 1 point more per possession to Duke scores 1 point less per possession. If all things go according to plan Duke wins in all scenarios (-12 with a 13 point lead is the worst case).

However, on the last few possessions of the game the other team could have the ball and be within 2 or 3, depending upon how Duke scores. For example, lets take the almost unbelievable scenario where Duke hits 2 freethrows on each possession and the other team hits three pointers each time down...

Duke up 13; 3 ptr by other team Duke up 10; 2 free throws Duke up 12; etc...with ~2:00 left the game is a 1 possession game and the other team will change their strategy because the game is wide open once again. So...what is the point differential to make the game a 1 possession game with ~ 1 minute left.

If Duke hits 2 free throws and the other team scores a three pointer two out of three possessions and a two the other possession (a differential of 0.67) Duke would have a three point lead and the ball with 40 seconds left. Lets count this as a possible OT game (Duke misses and the opponent hits a three at the buzzer to tie the game).

So the entire distribution goes from +1 to -1 and -0.67 is the cut off for a tie, so 83.33% of the distribution is a tie or win and 16.67% chance of a loss.

Hummmmm...I think I would like Duke to be leading by 16 with 8 minutes left.

My general idea of how the math would work would be something like...

Michigan has 12 possessions left. They score 2 points on a possession with a probability of 55%. So you could create a full distribution of the probability of Michigan scoring from 0-24 points for the rest of the game (obviously we could add in 3s as well).

Duke while playing the stall scores 2 points on a possession with probability of 40%. So you could create Duke's full distribution of scoring from 0-24 points.

You could also combine the distributions and look at Michigan - Duke's score.

On average Duke would give up about 4 points of their lead over 12 possessions at these scoring rates. Not a surprise, Duke is expected to win even scoring less than Michigan. However when things vary from the expected Michigan has a % chance to win.

Once you had the distributions you could have results like... Michigan scores 18 points 5% of the time... Duke scores 8 or less points 30% of the time so in that scenario Duke loses 1.5% of the time. If you sum up all of these you would get Duke's probability of losing the game.

The goal would be to decide if the decreased number of possessions makes up for Duke's decreased chances of scoring. This would obviously vary heavily with time and score (as well as your scoring assumptions). There may be too many variables to have it be very meaningful but I'd like to see the results.

Edit: It may be easier to do the math using a normal approximation for the scoring. Though I'm not sure if 12 trials is enough to make a normal approximation entirely justified. Probably good enough though. In that case you could say that Michigan's score would be approximately normal with a mean of 12*.55 and a variance of 12*.55*.45*(4). I think that's the appropriate variance formula (the 4 coming from each basket being worth 2 points). So then Duke's score would be normal with mean of 12*.4 and variance of 12*.4*.6*4. You can subtract them to create a new normal approximation, though at the moment calculating the variance of that is slipping my mind. Regardless... the result is a normal with mean of say 6 and a variance of some number and you can just look up on a normal distribution table the probability of that being less than 0 (which equates to the probability of michigan winning).

[Matches]

1) The way we run it, the strategy too often amounts ZERO offense and horrendous shot on every possession. From a purely strategic standpoint, this is my problem with it. The only year I can remember in which we were consistently good at executing an offense via stall ball was 2004, Chris Duhon's last year. Senior-year Duhon knew how to work the point in a 4-corners-style offense (which is what this is, and boy does it leave a bad taste in my mouth saying that), so we frequently got great shots out of it. Of course, that didn't stop us from losing to Maryland while running it in the ACC tournament final.

Oh I dunno, I thought we were pretty good with it last season. Scheyer was a big part of that IMO - he just seemed to have a knack for making a play at the end of the shot clock.

The key IMO to running stall ball well is to have guys who can get into the lane at will and create offense quickly. Kyrie at full strength should be a wizard at it - he's just not at full strength. Nolan also has the potential to be really good at it, though sometimes his decisionmaking is suspect. If Kyrie gets back to or near full speed, I think you'll see much better execution of stallball than we did last Sunday.

2) But, I admit that my biggest problem with stall ball is that it just feels like such a cheap, unmanly, "loser's mentality" approach to winning (there. I said it!). I mean, really, the basic idea is, let's try to work out a mathematical formula that allows us to stop playing anything that looks like basketball and win the game by simply standing around, milking the clock for as many as 10 minutes. I get angry and frustrated just thinking about it.

It does feel like LawyerBall sometimes, I'll concede. I hated Four Corners with a passion. Probably my acceptance of stallball has something to do with the fact that it's being run by guys wearing darker blue...

[Udaman]

Watching the Michigan game I was mad at the stall ball as well...then I heard Coach K's comments, and then went back and watched the end again. He was right. We were not in the "stall" offense. Why do I say this:

1) Coach K always gives a signal when he wants us to move to the stall. It's taking his index finger pointing it straight up and then twirling it in a circle. He never did that.

2) With the stall, we almost always wait until there are under 10 seconds left on the shot clock to start our play. Against Michigan, we did so several times with 15 plus seconds left.

3) With the stall, if a guard has it and is not being guarded, he'll just sit there and dribble. We only did this once. If you watch it again, you'll see we were passing the ball back and forth, and looked kind of timid, and then when there were 20 second or so left, we would hold it a while, and then pass it in with 13 seconds or so.

4) On two possessions we took fast shots - Nolan took a terrible 3 that he missed. Irving drove and made a bank shot but charged.

We weren't stalling.

That said - I agree that Coach K should have called a timeout to settle us down.