Late Game Timeouts: If the Game is Tied, Let Them Play

By John Ezekowitz

Last week I began my analysis of intentionally fouling while up three at the end of college basketball games with the story of Tom Izzo’s split second decision not to call timeout with his team down in a 2nd Round NCAA Tournament game earlier this year. Izzo was rewarded with a game-winning 3-pointer, prompting Basketball Prospectus’s John Gasaway to wonder about the effectiveness of timeouts at the end of close basketball games. I took up his question and now feel prepared to offer some first pass answers.

In this first of a series, I will use my dataset of every close college basketball game from 2009-2010 to analyze whether timeouts effective when you hold the ball and the score is tied near the end of the game.First, a word on my methodology and assumptions. I catalogued a possession if it was the last possession in a period (end of the 2nd half or an overtime) where the team could tie or take the lead (ie: from being down three to being tied). Because both offensive and defensive strategies vary so greatly depending on the score, I’ve decided to analyze timeout effectiveness as four separate problems (when down three, down two, etc, etc). I think this is the best way to control for strategy. If anyone has suggestions on how to better integrate the data, I’d love to hear them. Finally, I included timeouts called by both teams, making the assumption that the value of a timeout does not change based on which team called it. The vast majority of timeouts were called by the team with the ball.

In the case of teams with the ball when the score is tied, the data clearly show that it is more effective not to call timeout. In my 2009-2010 dataset, 452 teams fit the above criteria. 235 of those teams called timeout, 217 did not. Of the teams that called timeout, only 35.7 percent scored on the subsequent possession. Teams that did not call timeout scored 53.0 percent of the time. A simple two sample t-test with unequal variances shows that this difference is strongly statistically significant (p=0.0002). A logistic regression with timeouts as the independent variable and whether the team scored as the dependent variable showed that calling a timeout was a significant predictor of successfully scoring (p<0.001) and that teams that did not call timeout were twice as likely to score as teams that did.

This clearly shows that teams that do not call timeout score more often, but do they score more points? To assess this aspect, I looked at a team’s points per possession when they held the ball and the score was tied. Teams that called timeout scored an average of 0.773 points per possession whereas teams that did not call timeout scored an average of 1.06 PPP. Another hypothesis test showed that this difference, too, was statistically significant (p=0.022). (Stats note: since PPP was not normally distributed, I used the Wilcoxon Rank-sum test to see if that difference was statistically significant). Thus teams that do not call timeout not only score more often, but also score more points on their possessions than teams that do.

This analysis raises interesting questions about the value of a timeout. In this situation, the team with the ball will generally adopt whatever strategy they feel will best allow them to score, since a score this late in the game will likely win it for them (teams that scored in this sample won 69 percent of the time). I think it is safe to assume that both the coach and the players for the team with the ball have a good understanding of what sort of play will best allow them to score. The defense, on the other hand, may not be in the same situation. The coaches who have scouted the team may have some idea of what play their opponent will run, but the players may not. In economics, this would be known as asymmetric information.

Calling a timeout gives the coaches of the defending team a chance to balance this asymmetry by explaining to their players what the other team may run (or simply by imparting advice about defense). Thus I would hypothesize that in this situation, the value of a timeout is greater for the defending team.

Finally, I’d like to point out again that this is my first pass at these questions. There are obviously other factors to control for, such as the makeup and strengths of the team with the ball and the team defending (coaches may feel more comfortable letting teams with strong guard play go without a timeout, leading to selection bias). Any thoughts on further analysis would be greatly appreciated.

I hope you will continue to read for my analysis of timeout effectiveness in other close game situations. I think the results will surprise you.

About the author


View all posts


  • 1. It was a 2nd-round game, not Sweet 16, when Michigan State beat Maryland on Lucious’ 3.

    2 . Does your dataset provide the ability to cut situations by time remaining? My hypothesis is that a disproportionate share of the final possession timeouts are taken in situations where scoring is unlikely (i.e., with <3 seconds remaining). Would be interested in seeing the result cut by 5 second segments or simply removing timeouts in the final 3 seconds.

    • Andrew,

      Unfortunately after having gone thru all the best available p-b-p data, I am not confident in the timestamps for quite a few of the games and thus did not log the time remaining. I would say anecdotally that your comment may have some merit, but in general most of the “tied” possessions occurred with enough time on the clock to advance the ball up the floor comfortably for a last shot.

      Additionally, as you will see from the future posts, not calling a timeout is certainly not always the dominant strategy. I really think there is something to the game being tied situation that provides advantages for the defense in calling a timeout.

  • for the questions your are asking, effect sizes tell us much more than do probability levels . . . what was the R Squared in your regression analysis?

    • I should have included this in the post. Its a logistic regression so we can only get a pseudo-R^2 (and the way that is calculated varies slightly based on which statistical package you use). I used Stata and the Pseudo-R^2 was .226; not fantastic, but certainly there is something there. This is why I want to introduce more variables about team make-up.

  • Question – were timeouts remaining included as a data point? Curious if there was a difference when last possession involved no timeout because of a lack of timeouts rather than strategy. Teams/players may often be more assertive knowing there is no timeout to call.

  • I have an assignment for a graduate level course in which I need to duplicate findings of an economic study. Is there anyway I could get the data set for this study? I would be very appreciative. This was a really cool topic that was enjoyable to read.

  • Do you think timeout or not, that the actual variable may be if they set up a half-court set and ended up with a bad shot at the buzzard (more likely with a timeout called) or if they ran their natural flow of offense and may have took their “best shot” opposed to the “last shot”?

Leave a Reply

Your email address will not be published. Required fields are marked *