FanPost

A Perfectly Wrong Preseason Prediction is Still a Useful Preseason Prediction

Last season I remember CollegeFootballPoll.com predicting the Yellow Jackets would go 12-0 in 2015. This stuck in my memory for two reasons:

  1. They really bought into the hype train following the 11-3 campaign of 2014, 12-0 seemed unrealistic to me
  2. Their predictions for 2014 and before were actually very poor
As we all know and want to forget, 2015 ended up 3-9, a solid 9 wins off the CollegeFootballPoll.com prediction. Looking back at their predictions, we get the following:
Untitled.0.png
A least squares fit with an R2 of 0.76 is pretty darn good for college football, with only 2009 as an outlier (around 0.90 without 2009). As far as the slope goes:
  • Slope = 1.0 = perfect prediction
  • Slope = 0.0 = no correlation at all (meaningless prediction)
  • Slope = -1.0 = perfectly wrong
So CollegeFootballPoll.com predictions with [y=-0.726x + 12.64] are practically perfectly wrong. Which is still useful! If we know the website is consistently wrong, we can safely bet on the opposite of its prediction. Like talking to your pathologically lying ex.

I was eagerly awaiting their next prediction, and they are unsurprisingly predicting just 2 measly wins for 2016. That's not a bad thing, if you ask statistics:
Untitled2.0.png
Based on their perfectly wrong predictions of the past, a 2-10 prediction for 2016 most likely means an 11-1 season. The 95% confidence interval of the regression is +/- 2.2 wins, so if their wrong-ness holds we can expect something between 9 and 13 regular seasons wins (yes, we get an extra regular season win, compensation for the "vacated" 2009 ACC title and the NCAA can suck it).

Do I believe we will win 11 games? Not really, but I sure hope we will. Regardless, I enjoyed looking at CollegeFootballPoll.com hilariously wrong predictions.

FanPosts do not necessarily reflect the opinions of the FTRS writing staff, and are spotlighted purely to create discussion.