forums

Phil Steele's 2019 Surprise Teams

Advertisement
X275 said:
I'm curious as to what information exactly goes into this computer model? Like how can someone just put random information in a computer and come back with any semblance of an accurate prediction? You couldn't possibly predict variables like who's going to be the starting QB, how they are going to perform in a new offense, what the offensive line is going to be, etc.

Not attacking you btw, just kind of confused as to how people come up with these kind of things and pass it off as somewhat legitimate
Click to expand...

If there’s one thing there’s plenty of in sports, it’s metrics to use for a computer model. How accurate those models are, I have no idea, but there’s plenty of data to feed into em.
 
1mg of Epi said:
If there’s one thing there’s plenty of in sports, it’s metrics to use for a computer model. How accurate those models are, I have no idea, but there’s plenty of data to feed into em.
Click to expand...
Plenty of preductive computer models fail with their datasets also. Just look at the difference between hand timed and automated 40 yd numbers. Its the subtle biases that ***** up models.
 
JD08 said:
Plenty of preductive computer models fail with their datasets also. Just look at the difference between hand timed and automated 40 yd numbers. Its the subtle biases that ***** up models.
Click to expand...
Yea.. I’m not big on stats to begin with, there’s so many variables that go into each one that without any context they can be meaningless.

****... half this board was convinced D’no had a top 10 offense for a while.
 
1mg of Epi said:
Yea.. I’m not big on stats to begin with, there’s so many variables that go into each one that without any context they can be meaningless.

****... half this board was convinced D’no had a top 10 offense for a while.
Click to expand...
Stats are fine to a point, but people put far too much faith in predictive analytics. They somehow think that a computer did it so it's devoid of any subjectivity. In practice, there's often even more bias because they can hide behind the "data doesn't lie" defense.
 
Advertisement
1mg of Epi said:
If there’s one thing there’s plenty of in sports, it’s metrics to use for a computer model. How accurate those models are, I have no idea, but there’s plenty of data to feed into em.
Click to expand...
Yea that's what gets me, but then again, these people aren't required to be right, so it's really just a reason to talk about it, which is exactly what we're doing
 
X275 said:
Yea that's what gets me, but then again, these people aren't required to be right, so it's really just a reason to talk about it, which is exactly what we're doing
Click to expand...
Yep. It get clicks. ****, one of the metrics they use may very well be “fan base most likely to click on article”.... and there’s your top 10 most likely to surprise.
 
JD08 said:
Stats are fine to a point, but people put far too much faith in predictive analytics. They somehow think that a computer did it so it's devoid of any subjectivity. In practice, there's often even more bias because they can hide behind the "data doesn't lie" defense.
Click to expand...

Pro football focus is the worst. They use straight up numbers to rate a player, ignoring who that player lined up against.
 
1mg of Epi said:
Pro football focus is the worst. They use straight up numbers to rate a player, ignoring who that player lined up against.
Click to expand...
That's where displaying multiple metrics becomes relevant. So you can see the difference between say yards after catch overall versus YAC against teams with a winning record. That helps give the accurate picture instead of cherry picked stats to further a narrative.
 
Advertisement
JD08 said:
That's where displaying multiple metrics becomes relevant. So you can see the difference between say yards after catch overall versus YAC against teams with a winning record. That helps give the accurate picture instead of cherry picked stats to further a narrative.
Click to expand...

The models are not perfect, but they're way better than the human prognosticators. (Evidence: Vegas, MLB front offices, NBA front offices, etc.) The benchmark to compare the models too isn't perfection, it's the next best alternative - which in this case is the "eye test".

Of course, models should be open to criticism, otherwise they never improve. The guys building models are constantly re-evaluating and tweaking them. But fans usually aren't educated enough to come up with valid criticisms of models. They paint in broad strokes, criticizing the entire concept of analytics, as opposed to presenting specific criticisms (e.g. overestimating home field advantage, not adjusting quickly enough to the most recent data points, and so forth).

It's always funny to me when fans dismiss the entire concept of analytics whenever a specific model or evaluation turns out to be wrong - but when more subjective analysis is wrong, as it often is, you don't hear fans say "we shouldn't be using humans to evaluate sports!"
 
JD08 said:
That's where displaying multiple metrics becomes relevant. So you can see the difference between say yards after catch overall versus YAC against teams with a winning record. That helps give the accurate picture instead of cherry picked stats to further a narrative.
Click to expand...
P5 teams with a winning record
 
Advertisement
Canedog said:
I know, right. Steele right doe. If Miami beats Florida we in four one hellova ride this Fall.
Click to expand...

I feel like we've been through this. After we beat Oklahoma and Ohio State we thought no one else on the schedule was close. Then it turned out those teams were over rated and we lost 4 or 5 games.
 
X275 said:
I'm curious as to what information exactly goes into this computer model? Like how can someone just put random information in a computer and come back with any semblance of an accurate prediction? You couldn't possibly predict variables like who's going to be the starting QB, how they are going to perform in a new offense, what the offensive line is going to be, etc.

Not attacking you btw, just kind of confused as to how people come up with these kind of things and pass it off as somewhat legitimate
Click to expand...

I can’t speak to Steele’s models but I know how I do it is by regressing the data towards the mean using past years data while weighting the most recent year more heavily.

I’ll use averages for previous teams and then use generally available data about newcomers to get a starting spot. Then I regress based on recruiting ranking or performance in HS (versus peers) and then complete projections.

Do something similar for data available for coaching staffs.

It’s a difficult process, and the accuracy for individual players is hit-and-miss, but you can get a pretty good gauge of team performance with this method.

Vegas does a similar process in win projection modeling as well.
 
Advertisement
Advertisement
Back
Top