We've reached the first playoff weekend in the NFL and I'd have to say it's kind of boring on a whole when looking at any advantage based on the location of the game. Gone are the days of Florida teams like Tampa Bay or Miami that can beat Philadelphia or New England respectively at home, but then lose to Chicago by 38 points and Minnesota by 24 points respectively on their next road match. Although none of these teams reach traditional statistical significance (p<0.05) when looking for a difference in their performance home vs. away, there are some interesting trends to observe in these data.
For the past several years I've been keeping track of how many points on average a team scores (offense) versus how many points they give up (defense) over each game of the NFL season, noting whether those games were played either home or away for each team. [Beats playing softball as a hobby]. By using exactly 16 games worth of scores collected from this season and last to find the average, I tabulate predicted scores each week based on the average of what a team's offense normally scores versus what the opposing team's defense normally allows at that point in the season. For example, later today Houston's offense, which is scoring on average 25.13 points per game, is playing Indianapolis' defense, which on average allows the other team to score 21.50 (graph below). I would expect Houston to score around 23 points based on the average of these two numbers, and Indianapolis scoring about the same based on the opposite comparison.
By calculating the score based on only season averages, all but one of the games this weekend are predicted to be within one point. As you might have guessed, there are many factors that can affect the final score, with one of those being where the game is played. To try to calculate this location affect, I've used the predicted score based on a 16 game average as my standard for that week and anything scored above and below is recorded as a number. Staying with the previous example, if Houston scores 27 points today, their offense would have scored 3.69 points above what they were expected, while the Indianapolis defense would have allowed 3.69 points more than expected. We can separate these above/below metrics for each game by location (home or away), and then compare the distributions using a simple t-test.
Given below are the NFL teams (separated into sides) playing this weekend ranked by the probability of observing this location effect by random chance using a t-test (last column). The average amount of points an offense has scored or a defense has allowed over the last 17 weeks is given in blue, followed by the average amount of points over/under from this metric that particular team has scored/allowed at home and away games. I also include the median over/under for all games this season. Chicago and Philadelphia sides appear in 3 out of the top 5 slots in this ranking.
The team-side with the lowest p-value based on location effect is the Chicago defense, that on average allows 17.69 points per game for the season, but holds it's opponent to 5.6 points less than this metric when playing at home. I've attempted to account for this location effect in my predicted score by multiplying it by the p-value and then adding this to the predicted score. So for the Chicago defense, I take -5.6 (home effect) times 0.05294 (t-test p-value) and add it to 17.69 (allowed season average) to come up with 12.38 predicted allowed points after adjusting for location. Finding the mean of opposing offense/defense season averages with the location adjustment will give the predicted adjusted scores for each team. As you can see below, two of the games this weekend switched predicted winner after accounting for this location effect.
I'll try to write more on the Chicago/Philadelphia game tomorrow but I'm struggling as is to get this out before the first playoff game. I'm sure this post is riddled with all kinds of grammatical errors but I'm posting as is to get it in before Houston/Indianapolis.
-Michael Edwards, Bioinfo Solutions LLC
* Microsoft Excel for Mac 2011 and JMP Pro ver.14 were used to collect, analyze and display NFL data.