## The predictive value of overtime margin

### by Ken Pomeroy on Wednesday, January 29, 2014

Overtime games can produce strange results. On January 15, Mississippi beat LSU by 14 after going to an extra period when the Tigers’ Anthony Hickey missed a 15-foot shot at the end of regulation. If Hickey makes the shot, one might view LSU as two points better than the Rebels on that night. Instead, we got five more minutes of data and it turns out that LSU was 14 points *worse*. In a way, it defies logic.

If you had to pick one, would it be that the Rebels were 14 points better than the Tigers that night or were they zero points better? Some would say you don’t really have to choose because the Rebels won the game, so end of discussion. Those people probably aren’t interested in my work. But hypothetically, if you care about using that game for your own evaluation purposes, you have to wonder at some point if that 14-point win should be treated differently than a game where the teams were tied with five minutes to go and Mississippi won by 14 in regulation.

In order to examine this, I used the same technique I’ve used in the past to examine the meaning of scoring margin and one-point home wins. I took all cases since 2003 where two teams played each other twice. If one of those games went to overtime, I recorded the result in the other contest. For instance, last season Towson beat William & Mary by 13 in double overtime and then lost to the Tribe by seven in the rematch. That is viewed as some sort of truth of what a 13-point overtime win means in real life. Of course a 13-point overtime win at home doesn’t really mean that team is seven points worse on the road. That’s why I’m looking at 11 years worth of data.

Even over 11 years, one is left wanting for more data. There were just four cases of a home team winning an overtime game by 13 points and then playing that opponent again. And Towson’s subsequent seven-point loss was actually the best performance by the overtime winner in the other game. But there are many more examples of overtime margins in the single digits. Enough that we can compare the findings from this exercise to a similar approach using non-overtime games to shed some light on the importance of overtime scoring margin.

The results of this approach are summarized in the plot below.

The plotted points represent the average road rematch margin for the home margin given. For instance, in the 89 cases of a team losing an overtime game at home by one point, the rematch resulted in an average margin of -3.6 points. (Don’t take “rematch” literally here. The order of the games doesn’t matter. If two teams played an overtime game in the second of the pair of games, I treated the first game as the rematch.)

The lines represent the linear trend of those points, weighted by the number of observations, so the goofy points in the double-digits of the overtime cases don’t affect the regression much.

Because the slope of the OT regression is flatter than the regulation regression, the result is that overtime margin of victory clearly has meaning, but it’s not quite as useful as regulation margin from a predictive standpoint. Each additional point of home regulation margin increases the expectation on road margin by 0.36 points. In overtime, each additional point of home margin increases the expectation on road margin by just 0.27 points. It takes about a 14-point home regulation win for us to expect that the rematch would more often than not go to the road team. In the overtime cases, it takes about 19-point win to reach that point.

It should go without saying that in the real world, you wouldn’t want to simply use the last time two teams met to make your best guess on what the outcome will be the next time those teams play. It was done here because this is about the simplest approach I can think of to investigate the use of overtime performance from a predictive standpoint.

The results are consistent with what I think most people would expect. In some cases, overtime margin is just as useful as regulation margin, but in other cases, particularly where key players may have fouled out, it figures to be less useful. When all of those cases get averaged together, you get the outcome shown here. Overtime margin isn’t entirely indicative of the difference between two teams, but it’s a lot better than assuming those teams were completely even on that night.