Does the ESPN FPI really mean anything?

#1
BearcatVol

BearcatVol

Well-Known Member
Joined
Jan 22, 2014
Messages
139
Likes
132
#1
How much faith do you have in the ESPN FPI after 2 Weeks into the 2016 season? If these rankings and projections truly represent the SEC then the rankings of the East and West are very lopsided. I'm not so sure.

ESPN Football Power Index 2016 –Week 2

Rank School FPI PROJ W-L

SEC East
6 Tennessee (21.4) 10.3-2.4
18 Florida (14.8) 8.3-3.9
31 Georgia (10.5) 7.6-4.5
42 Missouri (6.9) 6.2 5.8
49 Vanderbilt (5.5) 5.1-6.9
54 South Carolina (4.8) 5.9-6.1
87 Kentucky (-3.9) 3.2-8.8

SEC West
2 Alabama (25.8) 10.4-2.1
7 LSU (20.3) 8.1-4.1
11 Texas A&M (18.5) 8.6-3.5
12 Ole Miss (17.2) 7.5-4.6
13 Auburn (16.8) 7.5-4.6
34 Arkansas (10.1) 6.7-5.3
39 Mississippi State (7.8) 5.5-6.5

ESPN Football Power Index - 2016 - ESPN
 
  • Like
Reactions: 1 person
#4
#4
Naw...just another way to keep them and college football in the forefront of sports discussions (water cooler material). Final poll is what counts...:)

GO VOLS!
 
#7
#7
ESPN stats are nothing more than made up algorithms that were created in an attempt to become apart of the NFL. Loosely.

Look at the QBR. There was a "Quarterback Rating" statistic for years and nobody really had any issues with it. Then when ESPN decides to center their show around the NFL for 10 months out of the year, out comes this statistic from nowhere that they made up. They practically become an All-NBA station during the other 2 months and created another stat out of thin air (its name escapes me). Now it's time for college football. With the new playoff system, it would be very beneficial for them to create some sort of new ranking system that will keep fans on their toes....even though ESPN's opinion doesn't matter, as the playoff committee has the only say. The FPI.

Bottom line, ESPN's stats are garbage attempts at marketing.
 
  • Like
Reactions: 1 person
#8
#8
Didn't espn's FPI say we were going to blow out northwestern and iowa when the polls said we wouldn't? I'd say it's more accurate at then end of the year when the sample size is larger.
 
  • Like
Reactions: 1 person
#9
#9
There's not a lot of performance difference between prediction models for future events. ESPN FPI finished 12th out of 68 models tracked in the prediction tracker last season but is off to a slow start this year, presently sitting at 51 out of 59 models being tracked. Last year the top model correctly predicted the winner in FBS games 78.631% of the time. ESPN FPI correctly predicted the winner 77.721% of the time. This year, after only 2 weeks, the top model is at 83.908% while ESPN FPI is at 75.862%. This gap will possibly close somewhat as the season progresses.

ThePredictionTracker NCAA RESULTS

The best predictor is probably Vegas - except for evenly matched teams - then nobody really has much of a clue.

Prior to the 2013 season, Phil Steele analyzed 20,882 FBS games to see what the predictive power of the Vegas lines were at identifying the winner given the amount of points a team was favored by. For example, if a team was favored by more than 31 points, that team won 98.99% of the times (of the 20,882 games 596 games had a team favored by more than 31 points).

more than 31 points 98.99%
24.5-31 96.03%
17.5-24 93.24%
14.5-17 86.84%
10.5-14 79.01%
7.5-10 73.61%
3.5-7 65.06%
less than 3 52.22%

Phil Steele

Last I checked the line for our game this week was at 27.5 suggesting a 96.03% probability that WGWTFA.
 
  • Like
Reactions: 2 people
#11
#11
orangeluvr said:
All the science I don't understand,
It's just my job five days a week,......Rocket Man :blink:
Click to expand...

You mean Josh Dobbs is really Elton John who is actually Reginald Kenneth Dwight?

It's all starting to make sense...or it's not! :ermm::eek:hmy::crazy::blink::banghead2:
 
#13
#13
ChattaTNVol said:
There's not a lot of performance difference between prediction models for future events. ESPN FPI finished 12th out of 68 models tracked in the prediction tracker last season but is off to a slow start this year, presently sitting at 51 out of 59 models being tracked. Last year the top model correctly predicted the winner in FBS games 78.631% of the time. ESPN FPI correctly predicted the winner 77.721% of the time. This year, after only 2 weeks, the top model is at 83.908% while ESPN FPI is at 75.862%. This gap will possibly close somewhat as the season progresses.

ThePredictionTracker NCAA RESULTS

The best predictor is probably Vegas - except for evenly matched teams - then nobody really has much of a clue.

Prior to the 2013 season, Phil Steele analyzed 20,882 FBS games to see what the predictive power of the Vegas lines were at identifying the winner given the amount of points a team was favored by. For example, if a team was favored by more than 31 points, that team won 98.99% of the times (of the 20,882 games 596 games had a team favored by more than 31 points).

more than 31 points 98.99%
24.5-31 96.03%
17.5-24 93.24%
14.5-17 86.84%
10.5-14 79.01%
7.5-10 73.61%
3.5-7 65.06%
less than 3 52.22%

Phil Steele

Last I checked the line for our game this week was at 27.5 suggesting a 96.03% probability that WGWTFA.
Click to expand...

Interesting analysis, but it would be far more interesting to eliminate the easy picks and see how they really performed with games like UGA vs UT and don't have games like UGA vs Podunk U to pad the W/L stat.
 
#14
#14
AM64 said:
Interesting analysis, but it would be far more interesting to eliminate the easy picks and see how they really performed with games like UGA vs UT and don't have games like UGA vs Podunk U to pad the W/L stat.
Click to expand...

I think the Vegas analysis is the only thing I've seen where anyone has done that with straight up matches. Against the spread most prediction models are going to be close to 50%. It is impossible to predict the future with much certainty. Obviously, if teams are extremely mismatched you have a good edge straight up but all things being equal there's not a tremendous edge with evenly matched teams. For example, Freak in his weekly pick contests always looks for teams that are more evenly matched thus it eliminates or at least minimizes the edge someone more informed might have over the truly ignorant. :)
 
#15
#15
AM64 said:
Interesting analysis, but it would be far more interesting to eliminate the easy picks and see how they really performed with games like UGA vs UT and don't have games like UGA vs Podunk U to pad the W/L stat.
Click to expand...

I don't think that works statistically as you just skewed the curve. Get your point. But not really fair to do that, mathematically speaking. You could take out a subset I guess though.
 
#16
#16
ESPN's FPI isn't perfect, but I find it and other good statistical models (such as Sagarin) to be much more "meaningful" than the AP and Coaches poll rankings. However, it is important to note that this early in the season, all models suffer from limited data. You can only glean so much from two games.

Perhaps the most interesting thing from ESPN's FPI is that it gives us almost as good of a chance to win the SEC (31.2%) as Alabama (34.6%). To some extent, this may be a function of Alabama playing in the SEC West and having a "more difficult road" to get to the SEC Championship Game, but still interesting.
 
  • Like
Reactions: 1 person
#18
#18
The FPI is much more reliable in predicting a games outcome then any other poll around...

But unless your using it for that purpose, it is pretty useless
 
#21
#21
after further thought, the FPI does mean something when there's a giant disparity in the chance to win....against the weakest opponents on the schedule, i'd wager that it's probably pretty accurate.

when that FPI chance to win is in that 40-60% range, i'd wager that it's more evenly split as a predictor of a winner.
 
New posts
You must log in or register to reply here.
Advertisement



Back
Top