Do NFL Teams with the Lead Play Worse in the Second Half?

Anyone who bets NFL halftimes on a regular basis knows that teams with a big lead will be favored by slightly less than if the game is close or tied. In fact we can observe this effect in almost any sport, for varying reasons. It is worth understanding why teams give back their lead in football, as it has implications for both live and halftime betting, as well as our understanding of the sport.

There are really three possible sources of "lead disadvantage" that come up in other sports, and hence are likely to exist in the NFL as well. The first is that the refs tend to favor the team that is down in the game. Even if they don't necessarily favor the losing team on purpose, they will fail to make calls against them, with the subconscious idea of not kicking a man when they are down. This is HUGE in basketball and makes a difference in hockey as well.

The second source is that teams that are up in the game play worse, or at least "worse" when it comes to covering a point spread. Teams with the lead focus on running out the clock, which leads them to score fewer points than they otherwise would. Some of this "worse" play is offset by the other team, who in desperation to get back in the game, will also employ sub-optimal (from a point spread perspective) strategies in an effort to force turnovers or score quick points/goals.

The final source comes from the final possession of the game. Across sports, the team with the lead will usually just run out the clock if the game is not close, while the team that is behind will always try to score, even if it has no influence on the final outcome of the game. The exception is hockey and perhaps soccer, where this works in the other direction due to empty net goals.

To understand the source of the halftime lead disadvantage in football, we first need to find out how large it is. We will find this by building a regression model predicting the average scoring margin of victory in the second half based on the first-half point spread, the first half scoring margin, and which team received the opening kickoff.


From the model above, we can see that after accounting for the strength of the teams and which team started out with the ball in the first half, we would subtract 10% of the home team's first half lead to find the final average margin of victory. So for example, if the home team led by 10 at the half, we'd chop off one point from our final second half score difference, after accounting for the full game point spread and who started with the ball. This 10% represents the intrinsic disadvantage of having the lead in the NFL.

To explain this 10%, we start by checking for the impact of ref bias. We use a statistic known as Expected Points Added to measure the impact of penalties. Briefly, EPA measures the difference in the number of points we'd expect each team to score at a given down, distance, and yards from the end zone, before and after each play. To determine the impact of a penalty we take the difference of points the offense would be expected to score before and after the penalty. We exclude all plays in the final two minutes of the game from our analysis, as these tend to include kneel downs and other worthless plays. We will look at the final two minutes in a separate analysis. Using the same data set as we did for scoring margin, we model the change in expected points from penalties:

We can see that there is a slight impact from ref bias, with a team expected to lose 1.39% of their first half lead due to the difference in penalty calls in the second half. This impact is statistically significant, although it will never add up to more than a couple tenths of a point per game. It is perhaps somewhat surprising that points added from penalties has no correlation to the point spread, as perhaps better teams could be expected to draw more penalties, particularly since we have included penalties such as pass interference calls. But the data suggests otherwise.

We now move on to see whether teams actually play better or worse due to having a lead. We will once again use EPA to measure this, using the difference in EPA per play of the two teams. Once again, the final two minutes are excluded. We measure the impact of the lead in three areas: rushing EPA per play differential, passing EPA per play differential, and EPA per play differential across all plays. We start by looking at rushing EPA per play:

The above coefficients, which measure expected points added per rushing play, seem very small. For each point the home team is favored, we'd expect them to gain 0.010 EPA more per rush, offset by -0.0043 EPA for each point of their first half lead. Not huge, but keep in mind there are around 25 running plays in a half across both teams. A team with a 10 point lead would generate over a full point less due to this lead effect in the running game.

The above model measures the difference in home EPA per play and away EPA per play, meaning the impact is working both ways. Given that their opponent is behind, the home team expects their opponent to throw, so the opponent's running game works better when they do run, and vice versa. We move on to passing EPA:

The effect of the lead is more pronounced in the passing game. Because passing is more important than rushing as there are more passing plays in a game on average, plus passing is better than rushing meaning good teams pass more, we see a higher correlation between the point spread and pass EPA, at 0.020 vs. 0.010 for rushing. The score differential is also five times more influential, with the opposite sign - teams that started with the lead average 0.02 EPA more per point of their first half lead. With around 30 pass plays in an average half, this suggests that a 10 point half time lead can have a 10*0.02*30 = 6 point impact on passing EPA in the second half, across both teams.

Part of the increased success in the passing game no doubt has to do with game theory and team strategies, as teams with the lead expect the pass and defend it more effectively, and vice versa. But there may also be a much smaller effect that comes from using the first half lead in our above model. While one half of a game is almost useless in terms of sample size, a team that outscores their opponent in the first half is probably playing a little better, on average, than would be expected by the point spread alone. This stronger performance level flows through into the second half passing statistics. Because scoring points and being better at football than the other team is much more correlated to the passing game than the running game, we might see more of an impact of the first half when we try to predict passing performance, as opposed to rushing. In any case, the predictive value of the first half, compared to the full game point spread, is probably very slight.

The team with the lead performs much better in the passing game, and slightly worse in the running game. Hence, we may be surprised by our model of EPA across all plays:

We find that overall, the first half lead has no significant impact on team EPA per play in the second half. Teams simply play no better or worse, regardless of the first half outcome. This is a little bit of a surprise - on rushing plays, the team with the lead only suffered an EPA loss of 0.004 points per point of first half lead, while on passing plays, the team with the lead gained 0.02 points of EPA per point of first half lead. Most teams throw over half the time, so we'd expect to see a net benefit from this, since the passing benefit is much larger than the loss in the run game.

The reason lies in play selection by teams with the lead, and the fact that overall, passing is almost always better from an EPA perspective. Modeling the difference in pass plays as a percentage of all plays, we find that teams throw the ball 1.4% less than their opponent in the second half per each point of first half lead (this relationship breaks down for large first half leads but is reasonable in most games). In modern football, passing tends to be about 0.2 EPA per play better than rushing, and the change in play mix roughly offsets the advantage teams with the lead have when they do throw the ball.

We now move on to the final effect, that of the last possession. We measure the last possession by focusing on scoring differential in the final two minutes of the game. We would expect the team that is behind at the half to outscore their opponents over this period as they often effectively get an extra possession here, particularly after we take the point spread into account.

As expected, we do see that the team who trailed at the half has a massive advantage at the end of the game, with teams losing 7.9% of their first half lead during this time period, after accounting for the difference in team strength. Also notable is that despite making up just 2/60 or 3.3% of the game, the models suggest that the final two minute scoring margin can be best predicted by using 12% of the point spread. When one considers that many end-game situations wind up being kneel-downs or garbage time, the true influence of skill in competitive games is likely even greater. There are more plays per minute at the end of the game, but also, more of these plays tend to be high-leverage passing plays, which amplify the difference in skill between teams. Good teams ALL either throw the ball well or defend the pass well - running sucks, and the market knows this meaning if a team isn't good in the passing game, they won't be favored.

In summary, we have found that the intrinsic disadvantage of having the first half lead is 10.0% of the lead. Our models suggest that 1.4% of the lead is lost due to ref bias, team per-play effectiveness makes no net impact, and 7.9% is lost due to end-game possession effects. The other 0.7% is likely just statistical variance. As bettors, ref bias and end-game effects are something we need to account for, but are unlikely to present any obvious betting angles, besides the rare case of a coach that likes to try and score with a big lead in the last two minutes of the game - rare in the NFL, but more common nowadays in college football (ugh).

The difference in team effectiveness does however represent a possible second half betting angle. Passing is more effective with the lead, and studies have shown that it is usually wrong not only from a scoring differential perspective but also from a games won perspective to run out the clock until the very end of the game. Coaches with a history of staying aggressive in the second half with the lead may be solid second-half bets, while those who go into a run-heavy shell are easy fades.