That’s the assumption in an article I saw today relating the upcoming Miami Heat team to the 2007 New England Patriots that went undefeated until losing the Super Bowl to the Giants. The claim is that the Patriots went so hard through the beginning of the season that they essentially wore themselves out for the second half, culminating in their Super Bowl upset. On top of that, the article quotes another blog that said that the Giants didn’t beat the best team of all time; that team stopped existing after week 11. They beat a good, but not best of all time, Patriots squad. As evidence of their decline, it is noted that the Patriots won by 25 points a game in weeks 1 through 11, whereas they only won by 8 and change the rest of the way.
If all you had to look at were points scored, it’s understandable to think that they slowed down or wore down or whatever. But thanks to my models, I have a couple other measures I can look at. I can look at the expected win probability for the Pats in each of their games from week 2 on, and how many points they should have won by. I can also look at their opponents. And I think these paint a different picture.
Let’s start with the opponents. In weeks 2 to 11, when the Pats were running up the score on everyone, their opponents had an average final win percentage of 52.7%, or were about 8.5-7.5. In the second half of the season, including playoffs, that number is 49.6%, or 8-8. This isn’t a significant difference. How about what the model thought of their chances? The Patriots had an average win probability of 82% in their first half games, and 80% in the second (again, not significant). Looking at the expected point difference, the Pats were supposed to win by 14 points in the first half of their games and 10 in the second half, again not a significant difference with only 9 games in each half. So by all of these measures, the Patriots’ first and second half of the season were roughly equal. Yet in the second half, they actually won by far fewer points than they had been in the first half. So you might be able to argue that they slowed down.
However, there’s one other thing to take into account. As I said, the Patriots were supposed to win by about 14 points on average in their first 9 games, and by 10 in their second 9. Yet they actually won by 25 and 8. This is where you need to make a decision about model errors. If you think the model describes teams and how many points they will win by well on average, as I do, then any differences are essentially random chance. So in their first 9 games, the Patriots just happened to do really well, scoring 11 more points than the model thought they would. In the second half they did just as the model expected them to, occasionally winning by more or less than expected but only a point off on average.
Most people don’t like the idea of random chance; however, it does play a role. In weeks 2 through 11, the Pats had 1 turnover return for a touchdown (fumble or interception), 9 (of 13) fumble recoveries, and 1 kick return for a touchdown. In weeks 12 through 21, those numbers are 2, 4 of 14, and 0. So they broke even on touchdown conversions from unexpected sources (kick returns and turnovers turned into points are unreliable at best), but they recovered far fewer fumbles in the second half of the season. I don’t know if it’s enough to explain an unexpected 11 points per game; it would depend a lot on field position and the like. But I’m sure it contributed, and there are other game events that you can’t really predict that could lead to the unexpected scoring.
Alternatively, you could think that the model unpredicted the Patriots’ scoring for other reasons. The basketball-reference blog thinks, for example, that the Pats were trying harder in the first half and burnt themselves out. In that case you would think that in the second half they would have lower predicted scores or winning probabilities, but that isn’t the case. The opponents were just as strong, including playoff teams, but the Pats should still have won 80% of their games (and they actually won more). As far as I can tell, the Patriots were the same team the whole way through the season, they just managed to score a bunch of unexpected points in the first half of the season. Vegas agrees; the Patriots covered every week in the first 10 games except for against the Colts and only once the rest of the way.
Overall, again I think this comes down to your opinion of random chance. As best as I can tell, the Patriots did as well as could be expected, and equally well, in both halves of the season. They just happened to score a lot more points in the first half, probably helped in part by recovering an unexpectedly high number of fumbles in the first nine games (and fewer in the last nine). But I’m sure some people will believe that they were trying harder and got tired down the stretch, or other teams caught on to what they were doing. As for how this affects how to think of the Heat, I think the answer is not at all. As long as they don’t overplay the starters and risk actual fatigue or injury, I’m sure the Heat will do just fine in the playoffs after trying to win in the regular season.