PredictionMachine.com is a site devoted to using collections of statistics to simulate games thousands of times for the main purpose of selling you gambling information. Here’s how they describe it:
PredictionMachine.com strives to be the most accurate and trusted source for predicting sports outcomes straight-up and against-the-spread.
In sports wagering, money management is critical. Knowing the likelihood of success of any play (SU, ATS, O/U, or even Futures, Parlays, Teasers or other exotics) is of utmost importance – not just for deciding what to wager, but how much. By playing the game 50,000 times before it's actually played, all of our picks come with a specified level of confidence (no dimes, no stars – just the facts) and a Play Value Key and calculator to decipher what that means to you.
The main attraction of this site is what it calls “The Predictalator.” From the website:
The technology has the ability to account for all of the statistical interactions of the players (playing or not playing/injured), coaches, officials, fans (homefield advantage) and weather in each game. The Predictalator plays every game 50,000 times before it's actually played. This provides PredictionMachine.com the ability to assign probabilities to the likelihood of just about any outcome occurring in any event as well as to project individual statistics and more including straight-up, against-the-spread and over/under pick percentages for each game.
Anyway, the Predictalator revealed its picks for the 2016 NFL season and after running the numbers, it doesn’t look good for the Detroit Lions. The site has the Lions finishing with a 6-10 record, good for last in the NFC North and 27th overall in the NFL.
That’s fine. Many models and analysts alike are predicting bad things for the Lions this year and they have good reasons to believe such things. But it’s when the Predictalator goes beyond the act of pure game predictions that it finds itself way out of its league.
The site tries to take its predictions a step further by ranking each of the team’s units. The Lions are calculated to rank 24th on both offense and defense. But here’s where things start making no sense. The Predictalator says the Lions will finish with the 29th ranked passing offense.
Let’s stop there for a moment.
Since Matthew Stafford has been at quarterback, the Detroit Lions have ranked in the bottom four in any major passing statistic (yards, yards per attempt, passer rating, completion percentage) just once: in Stafford’s rookie year in 2009. According to Football Outsiders, the Lions’ passing offense finished 17th or higher in DVOA in every single season since 2009, finishing in the top 10 in both 2011 and 2012. Not only has this been, at the very least, an average passing offense, it has occasionally been one of the best in the league.
Yet this prediction model has the Lions’ passing offense falling off a cliff in 2016. Why? We’ll never know for sure with their methods hidden, but one can only assume the loss of Calvin Johnson has something to do with it. Nevermind the Lions replaced his contributions with two receivers who combined for 1,605 yards last year.
While predictive models advertise themselves as unbiased and scientific, those models are only as good as the stats people feed them. Time and time again, predictive models — especially those used for gambling purposes — have shown themselves to be unreliable. Like any predictions, human or computer based, you have to take them with a grain of salt, because stats are best used when they are analyzing past results, not predicting future ones.