Sep 18, 2021; Pullman, Washington, USA; USC Trojans helmets sit during a game against the Washington State Cougars in the first half at Gesa Field at Martin Stadium. Mandatory Credit: James Snook-USA TODAY Sports

Football Updates (08/06/15)

Generally, this annual article is a one-stop shop to learn about everything new and improved we are doing for the upcoming football season. This year, with the launch of Soccer and our Daily Fantasy Tools section as well as the Live ScoreCaster overhaul demanding their own dedicated columns, this should be quick and to the point about what we have done to modify (improve) the football simulation engines and some features that are new to football this year, yet have been implemented elsewhere.

Trend Machine and Mock Draft Machine Upgrades

Last year’s version of this article (Site Updates 8/19/14) primarily focused on the launch of the Trend Machine and the Fantasy Football Draft Machine as well as the site redesign. ┬áThe site redesign, particularly with its mobile-friendly focus, is going strong (and expect even more infographics and design upgrades as we recently brought on a tremendous full-time designer).

The Trend Machine, a searchable database of over 35 years and 35,000 straight-up, against-the-spread and over/under outcomes, remains one of our flagship products and is now FREE to everyone AND also includes college football, college basketball, the NBA, NHL and MLB.

And, the Fantasy Football Draft Machine, rebranded as the Mock Draft Machine, is available now as part of our preseason fantasy information in the NFL. The Mock Draft Machine (great for actual drafts as well) simulates through fully customizable Fantasy Football drafts to gauge best strategies and recommended picks for every selection and scenario. New this year is the ability to let the Predictalator take over and predict what others in the league will do (you can draft for everyone or just you). Also new, the Mock Draft Machine will evaluate and grade drafts as they happen to help you understand your place in real-time. It allows for uploading of other ranking systems as well.

Game Matchups (Individual Game Purchases)

With the addition of our Trend Machine and Daily Fantasy data as well as more robust information now available on every player, team and game, we unveiled our Game Matchup pages available for every game of every sport (including a free pick for each active sport). This hub for all things picks, analysis, data and trends for each game is available as part of the subscription to each sport, yet can also be accessed with an individual purchase (great if you have an extra interest in a game you are watching or miss out on the ability to utilize or buy a full week).

Among other items, these pages include spread and total picks on original, current consensus and cuztomizable lines as well as the ability to review in-game projections and evaluate halftime picks. Also included in the Game Breakdown are: Live ScoreCaster charts during the game, Trend Machine information on the teams and matchup, Power and Team Efficiency Rankings, Previous Matchups, Schedules/Results and more. Check out the Game Matchup – My Picks page to see every game.

Picks Calendar

Ultimate transparency and integrity are incredibly important to us (as you will see below). In addition to being able to review all pick performance for each sport in the ResultsFinder, we have launched a Picks Calendar that links to the actual picks article from any sport of any previous date.

Football Engine Modifications

The vast majority of you reading this section (or even this article) are probably not too concerned with or interested in the tweaks we put in place for the NFL’s new extra point rules for 2015, but know that we have taken the necessary steps to thoroughly research and handle all actual rule changes (the Cowboys, Bills, Bears and Ravens are each projected to essentially make all extra points, while the Saints and Buccaneers project to miss at least 5% of extra points).

On to what you likely want me to talk about – engine improvements. As it relates to what is seen on this site, I spend far more of my time reviewing and analyzing performance and conducting research to incorporate into our models than I do actually creating content. Furthermore, I generally spend most of that time reviewing best practices and research regarding how we handle and utilize data. I also obviously spend a lot of time building new, innovative products that users can benefit from, but performance remains the core priority for this business…

One likely noticeable and straight-forward engine tweak that we have made will better differentiate the difference between straight-up percentage and against-the-spread percentage in extreme cases in college games. We conducted thorough analysis in the offseason that graded the predictive value of our in-game Live ScoreCaster straight-up predictions relative to what actually happened at an R-squared of over 0.985 for NFL, NBA and MLB (LSC basically explains around 98.5% of what occurs in a professional sporting event). While still strong (roughly 0.80 – way stronger than most non-sports models I’m used to), college football and college basketball lagged behind the pro sports. While some of that may make sense, we used that information to put quite a bit of effort into more appropriately illustrating dominance in college sports.

The most important revelation regarding our football engines centers around the way in which we handle early season data. You will or have likely seen on the site that we promote 14 consecutive weeks during the NFL last season in which, picking every game, the Predictalator was .500 or better against-the-spread. That’s true. From Week 7 through the Conference Championships, the Predictalator went 86-50 (63% ATS) on playable picks. Great.

The problem, specifically with last year, is that the first six weeks were essentially the opposite with 33-48 (41% ATS) and just one week above .500. Not great.

The same stretches over previous seasons were, fortunately, never as bad at their worst and picking every NFL game has turned a healthy profit against-the-spread in four of five NFL seasons, including last year.

While that six week stretch was brutal for all who lived it, we learned some very interesting things along the way…

A year ago, just as we did today, we published projected scores for all 256 regular season NFL games in the upcoming season (the first time we have made this article available). Simply reviewing our original prediction for those games from the NFL Preview in comparison to the actual consensus closing spread for those actual weeks when they happened yields some amazing results:

  • Weeks 1-6: 53-39 (58% ATS)
  • Weeks 7-17: 76-89 (46% ATS)

There is clearer gradual degradation as the season goes along (Weeks 1-4 were 65% ATS while Weeks 14-17 were 45% ATS), but that puts the numbers simply in the same terms as what we discussed above. It’s also along the lines of what one would expect for a preseason analysis (something that starts strong and fades to essentially a coin flip by the end of the year). The problem is that it ultimately did not translate as well to what we published in the actual weeks. Obviously, in-week projections should be much stronger in later weeks, but this shines some optimism (through a trying situation) on early weeks and methodology.

I have long said that the absolute most difficult, and yet important thing for anyone in my position to do is to figure out how to weight (or trust) what has happened this season as compared to expectations leading up to the season. It’s a sliding scale and much more difficult early in the season. I have spent A LOT of my career dealing with this issue. However, never has the analysis been as clear as with what we saw while tracking this information this way thoroughly for the first time.

While I obviously cannot promise what any week or series of weeks in the future will do with respect to pick performance in the NFL, know that we have found a very real, tangible and positive means by which we can expect to improve early season performance (without disrupting the integrity of the process or performance later in the season or overreacting to small sample sizes).