NFL Statheads Are Losing Their War With The Spread. Here's Why.SIn last week’s post, we showed that several of the larger statistically-inclined websites which offer NFL picks have had a rough go of it to start the season, at least as far as their picks against the spread. This was a surprise, because these results come on top of a string of several successful seasons in which each site consistently posted winning (>50%) results.

Today, we dive in to explore why the sites may have fallen behind the 8-ball.

First, we allocated picks from Football Outsiders, numberFire, Team Rankings, and Prediction Machine, four of the websites that we discussed last week. While there are several other competing analytics sites worth mentioning (including Statsational, CFBMatrix, and Massey-Peabody, who have had decent seasons), we stuck with these four because they had readily available for several seasons going back in time, and/or because they pick all NFL games each weekend.

Because each site puts out picks at different points in the week (Football Outsiders, for example, puts theirs out on Wednesday, and edits through till Sunday), a complete study would ensure the line and time of each site's pick was identical. In the future, we hope to contrast and include this information.

In any case, thus far in 2013, six NFL games have finished within half a point (on either side) of the closing line, and the average contest has finished 9.6 points from the spread (median distance, 8 points).

First, let’s see how often each site picks the home team to cover, and how often each picks the favorite to cover

Site

Home team (%)

Favorite (%)

Football Outsiders

52.1

43.6

numberFire

42.1

42.7

Prediction Machine

45.4

40.2

Team Rankings

29.7

33.3

These results aren’t totally surprising; the public often backs favorites and home teams, leaving skilled bettors to often choose underdogs or road teams. In fact, over at ESPN Insider, Dave Tuley touts a “dogs or pass” philosophy. Still, it’s noticeable that all four prediction sites tend to back underdogs, and most tend to favor the road team, at least through week 8.

Perhaps these numbers explain the poor performance through week 8, as home teams (55% ATS) and favorites (54% ATS) have cleaned up.

Next, we consider if the sites are similar with respect to their picks. The following cells indicate the percentage of picks in which the two sites (one on the row, the other on the column) have agreed on. Surprisingly, the websites appear to be using methods or algorithms which are independent of one another, with percentages around 50 indicating that whether or not each pair of sites agrees on a pick is more or less a coin flip.

Football Outsiders

numberFire

Prediction Machine

Team Rankings

Football Outsiders

X

58.7

55.3

47.9

numberFire

58.7

X

50.4

53.7

Prediction Machine

55.3

50.4

X

51.2

Team Rankings

47.9

53.7

51.2

X

Better yet, what if all four sites agree? Here are the frequencies, and cover percentages, based on the number of websites which backed the home team.

Number of sites picking the home team

4

3

2

1

0

Frequency (%)

21

25

49

22

4

Home record (ATS)

12-7-2

15-10

26-23

9-13

2-2

This suggests the sites are in complete harmony with respect to picking the road team, home picks are a healthy 12-7-2. Of course, those numbers are too small a sample to yield any statistically significant results.

Let’s now take a look at all the individual teams and see how the group as a whole has done for each. The table below show the total record against the spread for all four sites broken out for games involving that team. For instance, the four sites are 15-17-0 ATS when picking games involving the Chicago Bears while their picks are 17-11-0 ATS for the games with the Atlanta Falcons. It looks like these sites are having particular trouble with the Bears, Packers, Saints, 49ers and Titans. Most of the other records for each team are at or very near .500 with very few teams having a winning record. In fact, there are only 5 teams that these four sites in total have a winning record for so far this season: Falcons (17-11), Bills (18-14), Vikings (18-10), Patriots (18-14), and the Raiders (14-10-4). Is there any obvious common thread between these teams? Please leave serious and sarcastic suggestions in the comments.

TeamWLT
Arizona Cardinals15170
Atlanta Falcons17110
Baltimore Ravens14140
Buffalo Bills18140
Carolina Panthers16160
Chicago Bears7174
Cincinnati Bengals13154
Cleveland Browns12200
Dallas Cowboys15170
Denver Broncos12164
Detroit Lions14180
Green Bay Packers9190
Houston Texans13150
Indianapolis Colts13150
Jacksonville Jaguars15170
Kansas City Chiefs13190
Miami Dolphins13150
Minnesota Vikings18100
New England Patriots18140
New Orleans Saints8200
New York Giants15170
New York Jets16160
Oakland Raiders14104
Philadelphia Eagles15170
Pittsburgh Steelers13150
San Diego Chargers12124
San Francisco 49ers10220
Seattle Seahawks14180
St. Louis Rams13190
Tampa Bay Buccaneers16160
Tennessee Titans9154
Washington Redskins12160

Next let’s separate out each site and look at winning percentages. The graph below shows a summary of winning percentages (ATS) for each site. Each wedge represents a team and the larger the wedge the higher the winning percentage. Each group of four plots represents a site. The divisions are labeled and the conferences are represented by blue and red wedges for the NFC and AFC, respectively. So, for instance, Number Fire is doing a really good job picking games against the spread involving NFC North opponents whereas they are not doing so well against the AFC North.


NFL Statheads Are Losing Their War With The Spread. Here's Why.

Finally, if you're further interested, the results that were used to generate the above image can be found in the original version of this post. No site has a perfect record with any team this year, but Football Outsiders is getting killed by Chicago (0-6-1) and the Saints (0-7). Though they aren’t the only ones being scorched by the Saints. White NumberFire is 4-3 with the Saints, TeamRanking sand Prediction Machine are 1-6 and 3-4 respectively.

Cheers.

This post was written jointly by Michael Lopez of StatsByLopez (@StatsByLopez) and Gregory J. Matthews of Stats In The Wild (@StatsInTheWild).

Originally published on Stats In The Wild, republished here with permission.