System Check: Should You Care About Divisional Matchups?

If you’re like me, then you’re constantly surprised when people mention that an NFL game is divisional. People tend to say really nebulous things, like “it’s a divisional game, so the teams are more focused on winning,” or “the crowd will be a big factor,” as if teams didn’t give 100% in non-divisional games, or Seahawks fans were quiet against the Packers. It’s kind of hard to tease out the implicit betting advice from these kinds of statements. If both teams are “more focused on winning,” does that mean the game will be close? Does that mean the underdog is more likely to cover? If the crowd will be a “big factor,” does that mean the home team has a bigger advantage than usual? If the teams “know each other,” does that mean there will be less scoring? If these things are true, shouldn’t it already be built into the lines? How can this possibly be actionable betting advice?

Hypothesis 1: Underdogs cover the spread more often in divisional games.

In this article I will evaluate several hypotheses like this one. I will separate every regular season game from 2013 through 2017 seasons. In this case, those groups are divisional and non-divisional games, and throw out all the games that ended with a push, or where neither team was favored. There are 466 games in the divisional group, and 780 games in the non-divisional group. Next, I will calculate the rate of success in each group. In this case, “success” is when the underdog covers the spread. This happened 49.14% of the time (229/466) in divisional matchups, and 49.49% of the time (386/466).

Next, I will use Student’s t-test to determine if the difference between the two groups in statistically significant. In our data, we saw underdogs cover slightly more often in non-divisional matchups (49.49%-49.14% = 0.35%). It might be the case, though, that the true probability is the same for both groups, and the difference is due to random chance. The statement that the true probability is the same in both groups is our null hypothesis. We calculate the p-value, which gives the probability that in a world where the null hypothesis is true, our data would still show a difference at least as big as the one we observed. If the p-value is less than 5%, we reject the null hypothesis and declare that there is a real difference in the probability of the underdog covering in divisional and non-divisional games. In this example, our p-value comes out to 19.23%. It’s tempting to think that’s low, but from a statistical perspective, that’s not nearly low enough to reject the null hypothesis.

Verdict: Hypothesis 1 is false.

Even if we did find a stylistically significant pattern, that doesn’t necessarily mean we have a profitable betting angle. Suppose we found that underdogs covered 52% of the time in divisional games. That doesn’t mean it’s profitable to bet the underdog in every divisional game, because that 2% gets absorbed by the house edge. When betting against the spread, you typically get a line like -110. This means you don’t double your money if you win: you would need to bet $110 to profit $100 on a win. If you take a bet with a -110 line and have a 52% chance of winning, you will profit 90.9% of your bet 52% of the time and lose 100% of your bet 48% of the time, so on average your expected value is 52%*.909 – 48%*1 = -0.7%

Hypothesis 2: Underdogs playing a divisional game at home are more likely to cover the spread than underdogs in all other games.

Group 1: Divisional games where the home team is an underdog: 163 games. Success rate: 48.47% (79/163)

Group 2: Non-divisional games, and divisional games where the home team is favored: 1083 games. Success rate: 48.47% (536/1083)

P-value: Effectively 0. (1.49*10^-12)

Expected value of betting the favorite in Group 1: -1.62% at -110, 0.61% at -105.

Verdict: Hypothesis 2 is true but useless.

While the difference between the two groups is small (1.02%), it is statistically significant. The trend is in the opposite direction than expected: favorite covers more often, and that is more pronounced for divisional games where the favorite is on the road. However, the probability of visiting favorites covering the spread in divisional games (51.54%) is low enough that you expect to lose 1.62% of your bet on average when betting at -110. If you can get lines of -105, your edge is positive (0.61%), but so small that it’s probably not worth your time.

Hypothesis 3: Underdogs playing divisional games where the spread is at least 10 points are more likely to cover than underdogs in all other games.

Group 1: Divisional games where the spread is at least 10 points: 59 games. Success rate: 54.24% (32/59)

Group 2: Non-divisional games, and divisional games where the spread is less than to points: 1401 games. Success rate: 49.11% (583/1187)

P-value: 0.0416.

Expected value of betting the underdog in group 1: 3.54% at -110, 5.89% at -105

Verdict: Hypothesis 3 is true.

Even though the sample of divisional games where a team is favored by 10 points is small, the underdog is almost 5% more likely to cover in those games, and the trend is statistically significant. With a 54% chance of success, betting the underdog is a winning bet. However, with an edge of just 3.5% on a type of game that only happens about once a week, it would take a long time to place enough bets for your return to stabilize.

Hypothesis 4: Divisional games are more likely to hit the under than non-divisional games.

Games thrown out: games where the total is pushed. That is, the final total score of the game is equal to the Vegas total.

Group 1: Divisional games: 477 games. Success rate: 54.09% (258/477)

Group 2: Non-divisional games: 792 games. Success rate: 48.36% (383/792)

P-value: 0.169.

Verdict: Hypothesis 4 is false.

The gap here is substantial (4.69%), and the success rate of Group 1 is high enough that it would make a profitable bet, except the trend is not statistically significant, so I can’t recommend betting the under on every divisional game. And interesting thing about this is that divisional games are lower scoring. Divisional games have an average final total of 43.8 points, compared to 46.2 for non-divisional (p = .003). There is a trend here, but bookmakers know this and build it into the lines.

Hypothesis 5: Divisional games where one team is favored by at least 10 points are more likely to hit the under than other games.

Group 1: Divisional games with a 10+ point favorite: 61 games. Success rate: 65.57% (40/61)

Group 2: Non-divisional games, and divisional games where the spread is less than 10 points: 1208 games. Success rate: 49.75% (601/1208)

P-value: 0.029.

Expected value of betting group 1: 25.18% at -110, 28.02% at -105.

Verdict: Hypothesis 5 is true and useful. Bet the under on divisional games where one team is favored by at least 10 points.

This one is extremely surprising and extremely exciting. In the last 5 years, divisional games where a team is favored by at least 10 points have hit the under 65% of the time. That gives us a substantial edge of 25%. Even though the sample of divisional games with a 10+ point favorite is small, the success rate gap of over 15% is big enough for this trend to be statistically significant. Even more interesting is the fact that this only holds for divisional games: in the same timeframe, non-divisional games with a 10+ point favorite only hit the under 49.4% of the time (39/79). I don’t have a theory I like for why this is the case. You could argue that heavy divisional favorites do a better job than non-divisional favorites of shutting down the opposing offense because of familiarity, but if that were the case, I’d also expect those favorites to cover more often, which they don’t.

Of course, if this bet becomes widespread practice and Vegas sees a lot of action on the under for these games, the lines will adjust and erase this edge. Until then, I’m happy to make this bet. Since every game is divisional in week 17, there are five such games this weekend: OAK-KC (52.5), NYJ-NE (44.5), CIN-PIT (45.5), ARI-SEA (38.5) and SF-LAR (48.5).

Jacob Herlin is a data analyst at BettingPros and FantasyPros. You can find him on Twitter.