Everything you know about betting the NFL bye week is wrong
By HARVARD SPORTS ANALYSIS COLLECTIVE
There is a huge difference between “statistical significance” and “practical significance”.
Statistical significance is what most researchers report after a study, and usually comes in the form of a p-value. This value lets us know the probability that a result occurred by chance. If this value is low, usually below 0.05, you can reject the null hypothesis, whatever it may be.
However, statistical significance does not mean practical significance. Practical significance simply refers to how useful the results of your study are. If you are testing whether two things are equal, and they are in actuality slightly different, if you increase your observations enough that difference will become statistical significant. But if that difference is so small that it doesn’t really matter, then your study hasn’t shown any real practical significance. It is also possible to have a study that shows some promise of practical significance, but is nowhere near statistical significance.
When doing research for this article, we ran into both these cases.
We started out by looking at whether teams had letdowns the week before or after their bye week, thanks to a suggestion from twitter. We thought that might be something with teams coming off bye weeks – they would have gotten extra rest, which would affect their performance, and perhaps the betting lines were overestimating (or underestimating) that effect.
Looking at the data though, the lines have been pretty straight on. Teams coming off bye weeks have gone 386-376 ATS, remarkably close to 50 percent. And in terms of totals, after a bye week teams hit the Over 372 out of 775 times, which, coming in at 48 percent, isn’t close to statistical (or practical) significance.
After finding no real effect, at least in terms of ATS or Over/Under, we thought there might be a difference depending on what week a team had its bye. Maybe if they had it later on, when they were more tired, the extra rest would help them more than if they had a bye week early on.
Below, we have plotted the record of teams coming off a bye week both ATS and Over/Under, split up by which week they had a bye. As you can see, there is no clear trend, and the increased variability at the end is most likely attributable to smaller sample size.
We then looked at how teams did the week before they had a bye. In terms of ATS, there was no difference at all: teams compiled a record of 381-388 ATS. In terms of totals, however, it was more interesting.
Teams playing the week before a bye have gone Over in 415 out of 775 games, for a rate of 53.5 percent. With this many observations, this result was nearly significant, with a p-value of 0.0523.
However, the effect size (only 3.5 percent above a 50-50 bet) is so small that it may not have much practical significance. Assuming you bet $105 to win $100 on all these games, you would have made $3,700. This isn’t a terrible return – with 775 bets, that’s a return of 4.5 percent per bet.
However, this is over a vast many games – if you started betting on these types of games now, because the expected winning percentage is so close to 50 percent it would be likely that you would lose a bunch of those bets, and probable that you could experience a losing streak early on, sapping all your capital.
So for this reason, although betting on the Over in games where one team has a bye the next week is very close to statistical significance, I don’t feel like it has much practical significance.
Plotting the same chart as before, we can see that the success of the Over bet is due to a stretch late in season when the success rate spikes up. This is further reason for not betting the Over in these games now. We will be looking at that period again once it gets closer to see if it is significant by itself.
The other thing we wanted to look at was regarding the big news of this week: the firing of Joe Philbin. Simply put, we wanted to look at how teams that fired their coach midseason did the week after.
Using this list, and adding to it the firing of Dennis Allen last year and Gary Kubiak the year before, I made a database of all 24 such occurrences since 2000. Looking at how these teams did, only nine out of those 24 teams managed to cover the spread in their next game.
This is not that close to statistical significance because of the small sample size, but with an estimated winning percentage of 37.5 percent ATS it may be of practical significance.
If you had bet $105 to win $100 against all these teams, you have won $555, for a return of 22 percent. However, with such a small sample size this may just be fluke and this trend may not continue, so you will have to decide how to balance the statistical significance of this trend (or lack thereof) with the practical significance.
The good news is you’ll have a week to consider this, as the Dolphins have a bye week this weekend.