There is a lot of talk about the large point spread in the Patriots/Texans game. New England is a 15.5-point favorite over Houston tonight, tied for the second largest spread ever in a non-Super Bowl playoff game behind only Minnesota/Arizona in 1998.  The over/under is 44.5, which means the projected final score is 30-14.5.

Let’s say the Texans pull off the upset. Are they more likely to do so in a low-scoring game, or in a shootout?  If Houston wins 14-13, they will have come in under their projected points total by 0.5, but held New England to 17 fewer points than expected.  If the Texans win 31-30, they would have exceeded their projected points total by 16.5 points, while holding New England to exactly the number of expected points.

So, which result is more likely? My intuition says a low-scoring game, but what do the numbers say? There have been 24 games since 1985 where a team won despite being an underdog of at least 14 points.  As it turns out, intuition is correct: on average, these underdogs exceeded their projected points for total by 7.8 points, but held their opponents to 13.3 fewer points than expected.

In fact, only once — a week 17 upset win by Detroit — did a 14-point (or worse) underdog win despite allowing more points than projected or over 24 points. Perhaps the best model for a Texans win today would be this JaMarcus Russell-led upset over the Eagles in 2009, when Oakland won 13-9.