Saturday, June 20, 2009

A Study of Losses in the AP & Coaches Polls

What type of loss do the voters hate? What teams and conferences do they love? Last year it was non-conference schedules - this year it's the polls.

We’re going to be asking (and answering) two main questions here: first,
are there any constants in how far teams drop in the polls after a loss?
And second,
have the polls shown any biases for or against teams or conferences?

the Setup

As usual, this study only deals with the BCS era, the past 11 seasons from 1998-2008. All in all, there were 183 weeks of AP rankings and 184 weeks of Coaches rankings. (Yes, the coaches put out one more poll than the writers – I deal with that anomaly in the Methodology section.) At averages of 67 AP voters and 60 coaches per week, multiplied by the standard 325 votes per voter (25 “votes” for #1, 24 for #2, 23 for #3, etc.), that’s over 4 million AP votes and over 3½ million Coaches votes in the last 11 years. (Don’t worry, we’re not going to be dealing with all of them. Well, not really.) During those 183 and 184 weeks, there were 3,516 games played by AP Top 25 teams & 3,498 games played by Coaches Top 25 teams. Ranked AP teams lost 1,033 games and ranked Coaches teams lost 1,049 games, winning percentages of just over 70% for each. Not bad.

After dropping some games from each (again, check the Methodology section), we’re left with 1,028 AP games and 1,041 Coaches games to analyze and gather data from. So here’s our starting point:

0 - Base Numbers AP games avg. drop Co games avg. drop
Total Number of Top 25 Losses 1998-2008 1,028 -7.9 1,041 -7.7

In all those 1,028 AP losses, the average drop in the polls was -7.9 spots, while the average of the 1,041 CO games was -7.7 spots. The way we’re going to make sense of all these games is to divide these games into groups based on possibly relevant factors. Factors such as…

Factor #1: Upset - was the loss an upset? (upset = beaten by a lower-ranked team)

1 - Upset? AP avg drop Co avg drop
Number of Top 25 Upsets (U) 656 -9.1 671 -8.6
Number of Top 25 non-Upsets (noU) 372 -5.7 370 -5.9

Relevant? Yes, considerably. About 64% of the losses in the Top 25 were upsets, and teams that were upset dropped around -3 spots more in the polls than teams which were beaten by a higher ranked team. Let’s look at some others to get the feel of them.

Factor #2: Margin of Loss - does it matter how much you lose by?

2 - Margin of Loss AP avg drop Co avg drop
3 points or less 241 -7.0 244 -6.8
4 to 8 points 252 -7.2 258 -7.0
9 to 16 points 212 -7.5 215 -7.6
more than 17 points 323 -9.3 324 -8.9

Relevant? Maybe, at least in the Coaches poll. There's progression, but not much difference between a -1 point loss and a -15 point loss in the AP...

Factor #3: Month - does it matter when you lose?

3 - Month AP avg drop Co avg drop
Sept (+Aug) 252 -10.6 250 -10.3
Oct 317 -8.2 327 -7.9
Nov (+Dec) 329 -6.6 330 -6.6
Bowl 130 -4.9 134 -4.9

Relevant? Yup. Definitely for both polls.

Factor #4: Location - does it matter where you lose?

4 - Location AP avg drop Co avg drop
Home (vs) 325 -8.5 322 -8.1
Neutral (v) 187 -5.4 192 -5.4
Away (@) 516 -8.3 527 -8.2

Relevant? Doesn't look like it. Even though there were significantly more away game losses by Top 25 teams, they dropped the same average number of spots as home game losses. (But what's up with the neutral site games? 3 whole spots less?)

Factors #3 + #4: Month + Location

Month Loc. AP avg drop Co avg drop
Sept vs 91 -11.0 88 -10.3
Oct vs 124 -8.2 128 -7.9
Nov vs 110 -6.9 106 -6.6
Sept v 13 -12.1 12 -11.6
Oct v 14 -6.4 16 -7.1
Nov v 30 -4.4 30 -4.4
Bowl v 130 -4.9 134 -4.9
Sept @ 148 -10.3 150 -10.1
Oct @ 179 -8.4 183 -8.0
Nov @ 189 -6.8 194 -6.8

Ah, that's it. The vast majority of neutral site games are conference championships or bowl games, played really late in the season. A September neutral site game actually dropped teams an average of between -1 and -2 spots more than a September home or away loss. So this confirms that Month is relevant and Location isn't.

So that’s how this whole study works. Here are the other categories I looked at and whether or not they’re relevant:

the Results

Location - does it matter where you lose? No
Number of Losses – does it matter how many losses you have? No
Overtime – does it matter if the game was tied in regulation? No
Conf. Game – does it matter if the loss was a conference or non-conference game? No
Initial Rank – does it matter how highly you were ranked to start the season? No
Previous Year’s Wins – does it matter how good you were the year before? No
Previous 3 Year’s Wins – does it matter how good you were the last three years? No
Difference in Rank – does it matter how evenly you're matched with your opponent? Not really. (Yes, sort of, when it's a neutral site game, but no when it's a home or away game.)

Upset - does it matter if you were beat by a higher-ranked team? Yes
Margin of Loss (MoL) - does it matter how much you lost by? Yes
Month - does it matter when you lost? Yes
BCS Conf – does it matter if you’re in a BCS conference? Yes
Opp. BCS Conf – does it matter if the team you lost to is in a BCS conference? Yes
Rank – does it matter how highly you’re ranked? Yes
Opp. Rank – does it matter how highly your opponent was ranked? Yes
Opp. Previous Year’s Wins – does it matter how good your opponent was the year before? Yes
Opp. Previous 3 Year’s Wins – does it matter how good your opponent was the last three years? Yes
# of Weeks at Ranking – does it matter how long you’ve been ranked that high? Yes

(If you’re interested in all of these categories' breakdowns, click here. If you can think of any other statistically measurable category that you think might be relevant and correlate, just let me know and I’ll check it out.)

So now that we have a better idea of which categories are relevant, let’s go deeper and gauge just how much of a correlation there is for each one. We can do this by combining categories, just like we did with the earlier Month+Location. When we do so we see that:

In the AP Poll, the three most relevant factors are Upset, Month, and Rank.
In the Coaches Poll, the three most relevant are Upset, Month, and MoL (Margin of Loss).

So to answer our first question, yes, there are some patterns that we can follow to gauge (very generally) what a factors are involved in how far a team drops after a loss. We can say with certainty that, all other things being equal, getting upset will drop you more than losing to a higher ranked team; losing in September will drop you more than losing a bowl game; losing as the #23 team will drop you more than losing as the #3 team, etc. But here’s the thing – there is no magic formula that you can apply that will tell you exactly how many spots a team is going to drop after a loss. Sometimes a Top 5 teams that is upset by a TD in September drops -4 spots, sometimes they drop -11 spots. Sometimes a team ranked just outside the Top 10 that loses to a higher-ranked team in October by two TD’s drops -3 spots, sometimes they drop -10. That’s just the nature of a subjective ranking system.

Despite this, we can examine those above relevant factors to create a base of averages that we can use to compare and contrast how the polls see each team and answer our second question. Here’s the averages that we’ll be using:

pink=highest drop avg - blue=lowest drop avg

AP Poll_____________Coaches Poll
Upset Month Rank AP avg drop
noU Bowl #01-05 10 -2.3
noU Bowl #06-10 12 -3.3
noU Bowl #11-15 11 -3.6
noU Bowl #16-20 16 -4.5
noU Bowl #21-25 13 -4.5
noU Nov #01-05 4 -2.8
noU Nov #06-10 11 -4.0
noU Nov #11-15 28 -4.8
noU Nov #16-20 30 -5.1
noU Nov #21-25 39 -5.9
noU Oct #01-05 4 -4.3
noU Oct #06-10 22 -4.7
noU Oct #11-15 28 -6.5
noU Oct #16-20 33 -5.0
noU Oct #21-25 30 -7.5
noU Sept #01-05 4 -7.5
noU Sept #06-10 8 -5.5
noU Sept #11-15 16 -5.6
noU Sept #16-20 23 -8.0
noU Sept #21-25 30 -8.9
U Bowl #01-05 13 -3.8
U Bowl #06-10 14 -4.6
U Bowl #11-15 17 -5.5
U Bowl #16-20 13 -7.3
U Bowl #21-25 11 -9.0
U Nov #01-05 53 -5.3
U Nov #06-10 43 -7.1
U Nov #11-15 36 -7.6
U Nov #16-20 45 -7.7
U Nov #21-25 40 -9.9
U Oct #01-05 35 -6.7
U Oct #06-10 36 -7.8
U Oct #11-15 36 -9.1
U Oct #16-20 40 -10.1
U Oct #21-25 53 -12.6
U Sept #01-05 24 -7.6
U Sept #06-10 37 -10.0
U Sept #11-15 32 -11.4
U Sept #16-20 40 -12.4
U Sept #21-25 38 -17.0
Upset Month MoL Co avg drop
noU Bowl -3 or less 15 -3.1
noU Bowl -4 to -8 12 -3.2
noU Bowl -9 to -16 15 -3.5
noU Bowl -17 or more 22 -5.6
noU Nov -3 or less 13 -3.8
noU Nov -4 to -8 20 -2.8
noU Nov -9 to -16 25 -5.2
noU Nov -17 or more 52 -6.7
noU Oct -3 or less 18 -3.3
noU Oct -4 to -8 26 -4.7
noU Oct -9 to -16 30 -6.4
noU Oct -17 or more 42 -8.4
noU Sept -3 or less 14 -3.6
noU Sept -4 to -8 17 -5.8
noU Sept -9 to -16 13 -9.3
noU Sept -17 or more 36 -9.9
U Bowl -3 or less 18 -5.0
U Bowl -4 to -8 14 -5.1
U Bowl -9 to -16 20 -5.1
U Bowl -17 or more 18 -7.3
U Nov -3 or less 54 -6.6
U Nov -4 to -8 64 -6.8
U Nov -9 to -16 46 -7.6
U Nov -17 or more 56 -7.8
U Oct -3 or less 65 -8.6
U Oct -4 to -8 58 -8.2
U Oct -9 to -16 36 -8.8
U Oct -17 or more 52 -9.8
U Sept -3 or less 47 -9.3
U Sept -4 to -8 47 -11.0
U Sept -9 to -16 30 -12.1
U Sept -17 or more 46 -13.5

the Payoff

So how do we use those averages? By examining every game and comparing how many spots a team dropped to the average for that combination of factors. For instance, if you click on each of the school names below, you'll be taken to their school page where you'll see something like this:

Iowa State Losses while in the Top 25, 1998-2008
date AP rk CO rk vs/@ oAP rk oCO rk oConf Opponent MoL AP drop avg diff CO drop avg diff
10/19/02 #09 #13 @ #02 #02 Big12 Oklahoma 3-49 -8 -04.7 -03.3 -5 -08.4 03.4
10/26/02 #17 #18 @ #07 #07 Big12 Texas 10-21 -5 -05.0 00.0 -5 -06.4 01.4
11/09/02 #21 #22 @ #12 #12 Big12 Kansas St 7-58 -9 -05.9 -03.1 -8 -06.7 -01.3
10/01/05 #23 #26 @ NR #42 Big12 Nebraska 20-27 -9 -12.6 03.6 - - -
11/26/05 #26 #25 @ NR NR Big12 Kansas 21-24 - - - -8 -06.6 -01.4

Iowa State's first loss happened in October, they were ranked #6-10 (AP), but it wasn't an upset = the average for all teams with those factors was a -4.7 spot drop. The Cyclones dropped -8 spots, which was not only significantly more but was one of the biggest drops for a team with that combination of factors (signified by the red background). Game #4 counts for the AP since the Cyclones were ranked #23, but not for the Coaches since they were out of the Top 25 - vice versa for game #5.

Combining all those games, we get this:

Team AP L's +/- AP Avg leasts mosts Co L's +/- Co Avg leasts mosts AP-CO diff All L's All +/- Tot. Avg
Iowa St 4 -2.7 -0.7 0 1 4 2.1 0.5 0 0 4.8 8 -0.7 -0.1

the +/- is the sum of the "diff" columns from above (differences between the average drops and the amount the team actually dropped). The higher it is, the more the polls like that team, the lower (negative) it is, the more the polls dislike them. (Why? who knows - that a study for another time. And somebody else.) Leasts & mosts are the times when a team's dropped spots were the least or most of all losses by teams with that combination of factors. AP-CO diff is the difference between the AP and Coaches polls - the higher the number, the more the Coaches liked the team, the lower (negative) the number, the more the AP liked the team.

The most important numbers are the colored averages. Those represent the average number of poll spots more than or less than the average that each team dropped PER GAME. So looking at Iowa's 10 losses while ranked in the AP, for instance, the Hawkeyes dropped -2.5 spots more per loss than the average team did in the same circumstances.

Got all that? Great. Here you go - have fun arguing.

(click on the column headers to sort)

Losses while in the AP & Coaches Polls, 1998-2008
Team AP L's AP +/- AP diff Avg leasts mosts Co L's Co +/- Co diff Avg leasts mosts AP-CO diff All L's All +/- Tot. Avg
Boston Coll 15 1.8 0.1 2 2 13 5.1 0.4 3 2 3.3 28 6.9 0.2
Clemson 17 -18.8 -1.1 1 5 19 -4.4 -0.2 1 2 14.4 36 -23.2 -0.6
Florida St 32 -16.0 -0.5 5 4 32 2.0 0.1 6 5 18.0 64 -14.0 -0.2
Georgia Tech 22 17.3 0.8 4 3 23 17.2 0.7 4 1 -0.1 45 34.4 0.8
Maryland 10 -5.5 -0.6 2 3 10 2.2 0.2 1 1 7.8 20 -3.3 -0.2
Miami (FL) 18 -2.0 -0.1 4 4 17 13.3 0.8 2 0 15.4 35 11.3 0.3
N. Carolina 5 -4.0 -0.8 1 0 3 -6.6 -2.2 0 1 -2.6 8 -10.6 -1.3
NC State 10 20.3 2.0 2 0 11 -0.8 -0.1 1 1 -21.1 21 19.6 0.9
Virginia 16 10.3 0.6 3 1 17 -4.1 -0.2 1 2 -14.4 33 6.1 0.2
Virginia Tech 28 6.7 0.2 4 3 28 5.8 0.2 3 2 -1.0 56 12.5 0.2
Wake Forest 6 -4.4 -0.7 0 1 5 -8.8 -1.8 0 2 -4.4 11 -13.2 -1.2
Illinois 8 -2.0 -0.3 0 1 8 -3.2 -0.4 0 1 -1.2 16 -5.2 -0.3
Iowa 10 -24.5 -2.5 0 5 12 2.6 0.2 0 0 27.1 22 -22.0 -1.0
Michigan 30 0.1 0.0 2 5 31 -1.1 0.0 5 4 -1.2 61 -1.0 0.0
Michigan St 15 -10.5 -0.7 1 3 17 -14.1 -0.8 1 5 -3.5 32 -24.6 -0.8
Minnesota 14 4.4 0.3 2 2 14 6.8 0.5 3 1 2.3 28 11.2 0.4
Northwestern 8 -6.6 -0.8 1 4 10 -4.2 -0.4 0 2 2.4 18 -10.8 -0.6
Ohio St 27 -6.4 -0.2 4 5 25 20.0 0.8 5 2 26.4 52 13.6 0.3
Penn St 19 18.1 1.0 2 1 20 -3.1 -0.2 2 3 -21.2 39 15.0 0.4
Purdue 21 18.7 0.9 4 1 23 -13.6 -0.6 1 5 -32.2 44 5.1 0.1
Wisconsin 24 -14.3 -0.6 3 4 26 -24.2 -0.9 2 6 -10.0 50 -38.5 -0.8
Colorado 14 -25.4 -1.8 1 4 15 -22.2 -1.5 1 4 3.2 29 -47.6 -1.6
Iowa St 4 -2.7 -0.7 0 1 4 2.1 0.5 0 0 4.8 8 -0.7 -0.1
Kansas 4 3.8 0.9 1 1 4 7.8 2.0 2 0 4.1 8 11.6 1.5
Kansas St 18 -1.9 -0.1 3 4 16 7.2 0.4 2 2 9.1 34 5.2 0.2
Missouri 16 24.9 1.6 3 0 16 20.8 1.3 4 1 -4.1 32 45.7 1.4
Nebraska 21 -4.8 -0.2 2 3 21 10.7 0.5 3 3 15.5 42 5.9 0.1
Oklahoma 19 21.2 1.1 7 1 18 31.9 1.8 5 1 10.8 37 53.1 1.4
Oklahoma St 12 12.2 1.0 4 0 12 -3.7 -0.3 3 2 -15.9 24 8.5 0.4
Texas 25 -11.2 -0.4 4 4 24 7.2 0.3 2 0 18.4 49 -4.1 -0.1
Texas A&M 23 41.4 1.8 7 2 24 12.9 0.5 4 2 -28.5 47 54.3 1.2
Texas Tech 10 -6.0 -0.6 1 2 14 10.1 0.7 1 1 16.0 24 4.1 0.2
Cincinnati 4 0.0 0.0 0 1 4 2.4 0.6 0 0 2.3 8 2.4 0.3
Connecticut 3 -7.6 -2.5 0 1 3 -9.7 -3.2 0 1 -2.1 6 -17.4 -2.9
Louisville 11 -11.5 -1.0 1 3 11 -8.6 -0.8 0 1 2.9 22 -20.0 -0.9
Pittsburgh 12 -16.6 -1.4 2 3 11 -6.4 -0.6 1 2 10.1 23 -23.0 -1.0
Rutgers 5 -7.3 -1.5 1 1 4 3.7 0.9 1 0 11.0 9 -3.6 -0.4
S Florida 7 -11.1 -1.6 0 2 7 -20.9 -3.0 0 2 -9.8 14 -31.9 -2.3
Syracuse 6 -14.1 -2.3 1 4 7 -11.5 -1.6 0 1 2.6 13 -25.5 -2.0
West Virginia 15 -34.7 -2.3 1 7 18 -25.0 -1.4 1 3 9.6 33 -59.7 -1.8
Arizona 5 -6.3 -1.3 0 2 5 1.6 0.3 0 1 7.9 10 -4.6 -0.5
Arizona St 17 18.0 1.1 4 1 19 33.2 1.7 5 1 15.3 36 51.2 1.4
California 16 0.5 0.0 2 2 14 -6.5 -0.5 1 1 -7.0 30 -6.1 -0.2
Oregon 26 -1.4 -0.1 3 5 24 -22.2 -0.9 3 6 -20.8 50 -23.6 -0.5
Oregon St 6 -3.4 -0.6 1 1 5 7.7 1.5 2 0 11.1 11 4.4 0.4
Stanford 4 8.6 2.2 1 0 5 4.9 1.0 1 0 -3.7 9 13.5 1.5
UCLA 16 -18.5 -1.2 2 3 18 -15.2 -0.8 3 5 3.3 34 -33.7 -1.0
USC 15 12.2 0.8 2 0 15 17.9 1.2 2 1 5.8 30 30.1 1.0
Wash St 8 5.0 0.6 1 0 7 10.5 1.5 1 0 5.6 15 15.5 1.0
Washington 13 4.5 0.3 2 1 13 14.9 1.1 2 2 10.4 26 19.4 0.7
Alabama 18 18.4 1.0 3 1 16 6.4 0.4 5 2 -11.9 34 24.8 0.7
Arkansas 17 26.9 1.6 6 0 17 14.2 0.8 7 2 -12.7 34 41.0 1.2
Auburn 24 -23.6 -1.0 3 8 23 -37.1 -1.6 2 6 -13.4 47 -60.7 -1.3
Florida 34 -2.9 -0.1 3 3 34 23.3 0.7 7 2 26.2 68 20.4 0.3
Georgia 31 19.2 0.6 8 4 32 33.7 1.1 5 1 14.5 63 53.0 0.8
Kentucky 4 -4.4 -1.1 0 1 6 1.0 0.2 2 2 5.3 10 -3.4 -0.3
LSU 23 1.1 0.0 3 5 23 11.2 0.5 4 0 10.1 46 12.3 0.3
Mississippi 6 14.9 2.5 1 0 7 3.7 0.5 0 0 -11.2 13 18.6 1.4
Miss St 12 13.5 1.1 3 1 11 12.8 1.2 4 2 -0.7 23 26.3 1.1
S Carolina 17 6.5 0.4 2 3 14 -8.1 -0.6 1 2 -14.5 31 -1.6 -0.1
Tennessee 31 -12.0 -0.4 7 7 31 -8.3 -0.3 3 6 3.7 62 -20.3 -0.3
Vanderbilt 2 1.6 0.8 0 0 2 -3.0 -1.5 0 0 -4.6 4 -1.5 -0.4
Air Force 4 11.6 2.9 0 0 6 -2.8 -0.5 0 1 -14.4 10 8.8 0.9
Ball St 2 -3.4 -1.7 0 1 2 -4.9 -2.5 0 0 -1.5 4 -8.3 -2.1
Boise St 6 -2.1 -0.3 2 1 6 -10.7 -1.8 0 1 -8.6 12 -12.8 -1.1
Bowl Green 3 -2.7 -0.9 0 0 5 -9.4 -1.9 0 2 -6.7 8 -12.0 -1.5
BYU 8 1.3 0.2 1 1 10 2.8 0.3 0 0 1.5 18 4.1 0.2
Colorado St 7 14.6 2.1 1 0 8 -5.6 -0.7 1 2 -20.2 15 9.0 0.6
E Carolina 5 5.4 1.1 0 0 4 5.7 1.4 1 1 0.3 9 11.2 1.2
Fresno St 10 11.5 1.1 2 1 10 -23.7 -2.4 1 5 -35.2 20 -12.2 -0.6
Hawaii 2 0.3 0.1 1 1 3 1.1 0.4 0 0 0.8 5 1.3 0.3
LA Tech 1 0.9 0.9 0 0 0 0.0 0.0 0 0 -0.9 1 0.9 0.9
Marshall 2 -3.4 -1.7 0 1 3 -3.7 -1.2 0 0 -0.3 5 -7.1 -1.4
N Illinois 2 0.0 0.0 0 0 3 2.1 0.7 1 0 2.2 5 2.1 0.4
Notre Dame 24 -22.4 -0.9 4 7 21 -6.6 -0.3 2 2 15.7 45 -29.0 -0.6
S Miss 7 -9.1 -1.3 0 2 7 -12.0 -1.7 1 3 -3.0 14 -21.1 -1.5
TCU 10 -7.9 -0.8 0 1 9 -6.3 -0.7 1 1 1.6 19 -14.3 -0.8
Toledo 1 1.6 1.6 0 0 1 -7.8 -7.8 0 1 -9.4 2 -6.1 -3.1
Tulsa 1 -6.1 -6.1 0 1 1 -8.2 -8.2 0 1 -2.1 2 -14.3 -7.1
Utah 1 0.6 0.6 0 0 1 -0.3 -0.3 0 0 -0.9 2 0.4 0.2
UTEP 2 -1.2 -0.6 0 0 2 0.2 0.1 0 0 1.4 4 -0.9 -0.2
Wyoming 1 0.9 0.9 0 0 1 -2.2 -2.2 0 0 -3.1 2 -1.3 -0.7

If your school doesn't appear, it for one of two reasons. First, it could be that your team has never lost while ranked in the Top 25. Those schools are:

Memphis, Tulane, Navy, & Miami (OH).

Props. Sort of. (Tulane and Miami (OH) spent 11 and 6 weeks in the Top 25 without losing, but Navy and Memphis had all of 1 week in the Top 25.)

Second, your team doesn't appear because they have never been ranked in the Top 25 during the BCS era. My condolences. Those schools are:

Duke, Indiana, Baylor, Central Florida, Houston, Rice, SMU, UAB, Army, Western Kentucky, Akron, Buffalo, Central Michigan, Eastern Michigan, Kent St, Ohio, Temple, Western Michigan, New Mexico, San Diego St, UNLV, Arkansas St, Florida Atl, Florida Intl, LA-Lafayette, LA-Monroe, Middle TN St, North Texas, Troy, Idaho, Nevada, New Mexico St, San Jose St, & Utah St.

The good news is that if your team has ever beaten a Top 25 team, they get their own page & their name above in blue is hotlinked like the teams in the table.

Analysis

In looking at those figures in the table, it seems that some teams need to be sending the voters thank-you cards (Texas A&M, Oklahoma, Georgia, Arizona St...), and some should be sending the voters do-better letters (West Virginia, Auburn, Colorado...) And there are a whole lot of other interesting situations to be found with each individual team. For instance, even though Ohio State has been much maligned for the past three years, the polls don't show it - in each of their six losses, they dropped less spots than the average every time in the Coaches poll and half the time in the AP Poll. (And even when they dropped more than the average in the AP, it was less than one spot more.) And did you know that while Oklahoma has 33 wins over Top 25 teams, only 4 of them were over non-conference foes?

But the real meat of this study comes not necessarily in looking at individual teams but taking in the larger picture.

BCS vs non-BCS Teams

Sure, most non-BCS teams have a legitimate complaint that the polls never give them enough respect in the preseason rankings. But does that bias carry over once teams start losing? In the AP, that's a resounding no - 11 of the 19 non-BCS teams actually have a zero or positive average in the AP (meaning they dropped less spots than the average). That's a much better percentage of teams than the BCS conferences where only 29 of 63 teams have a positive average. Air Force and Colorado State are in the Top 5 of teams who get a break, dropping over 2 spots less than the average for comparable teams when they lose. So if anything, the AP voters make up for their preseaon apprehension of non-BCS teams by dropping them less than BCS teams when they inevitably lose.

But the Coaches poll is a different story. Only 5 of the 18 non-BCS teams have a positive average, meaning that the overwhelming majority of non-BCS teams drop further in the Coaches poll after a loss than the average team does. (It's true that in general, the Coaches drop teams further for a loss than the AP, but not by that much - it's a difference of 0.2 spots.) In fact, the Coaches dropped non-BCS teams more than the average in nearly 2/3's of their losses. Looking at it one final way, all of the BCS teams combine for the average 0.0 in the Coaches poll - non-BCS teams combine for a -1.6 average, losing a spot and a half more per game than BCS teams. Remind me again why keeping their ballots secret is a good thing?

Conference vs Conference

What about the conferences? Are there any that receive better (or worse) treatment from the polls? Here's the numbers...

Conf AP L's +/- AP Avg leasts mosts Co L's +/- Co Avg leasts mosts AP-CO diff All L's All +/- Tot. Avg
ACC 149 -2.7 0.0 22 22 149 11.5 0.1 16 17 14.2 298 8.8 0.0
Big10 176 -23.0 -0.1 19 31 186 -34.2 -0.2 19 29 -11.1 362 -57.2 -0.2
Big12 166 51.4 0.3 33 22 168 84.8 0.5 27 16 33.4 334 136.2 0.4
BigEast 88 -95.5 -1.1 11 25 89 -60.6 -0.7 9 11 34.9 177 -156.1 -0.9
Pac10 126 19.2 0.2 18 15 125 47.0 0.4 20 17 27.7 251 66.2 0.3
SEC 219 59.0 0.3 39 33 216 49.8 0.2 40 25 -9.2 435 108.8 0.3
CUSA 22 -13.2 -0.6 1 4 21 -21.7 -1.0 2 6 -8.4 43 -34.9 -0.8
MAC 10 -7.8 -0.8 0 2 14 -23.6 -1.7 1 3 -15.8 24 -31.4 -1.3
MtnWest 24 24.1 1.0 2 1 28 -13.4 -0.5 1 4 -37.5 52 10.8 0.2
WAC 23 10.0 0.4 5 4 24 -33.0 -1.4 2 6 -42.9 47 -23.0 -0.5

That pretty much confirms what we just found looking at BCS vs non-BCS conferences: non-BCS teams get the shaft in the Coaches Poll but the AP doesn't seem to distinguish between BCS & non-BCS. The MtnWest drops a whole spot less than the average in the AP, but they drop half a spot more with the Coaches. And check out the BigEast - they're not getting any love anywhere. Pac10 vs SEC? Pretty much equal when they lose, but the AP tends to prefer the SEC while the Coaches prefer the Pac10.

Year by Year Since 1997, both the number of Top 25 losses to account for (usually between 80-100 per year) and the average number of spots dropped fluctuated, bobbing up and down. But there's no correlation between the two - more losses in a year doesn't mean teams drop further for each one. But there is an unmistakable trend that both the Coaches and AP Polls are punishing teams more for a loss now than they did before the BCS. This tells us two things: first, that wherever the premium was before (whether it was winning or playing a tough schedule or defending home-field), has shifted to not losing. Second, this trend towards more spots dropped per loss could be because there's more quality choices (teams) for voters to choose from when filling out their ballots. But this doesn't prove that there's more parity in the game today - it proves that people BELIEVE that there's more parity in the game today.

____________________________________

So that's that. Again, you can click on each team's name in the table (and paragraph) above to see a breakdown of their individual losses while in the Top 25 & wins against the Top 25.

What's that? More? Oh, alright. These tables detail teams' number of weeks in the Polls & the number of votes received during the BCS era and a comparison of teams' performances in and against the Top 10 / Top 25. Both of these new stats pages can now be found on the left sidebar and will be updated weekly during the season.

If you've made it this far, thanks for reading.

4 comments:

C R X said...

Regarding neutral site losses dropping less in the polls than home or away losses, could it be that neutral site games tend to be between more evenly matched opponents?

Ed Gunther said...

Thanks for the comment, CRX - I got an e-mail that basically asked the same thing...

"I think I can explain the anomaly of losing on a neutral field.

The teams that play on a neutral field are usually big names, somewhat more highly ranked than the average team. They are often closely matched and often games between rivals.

I think you will find that losing to the favorite in such games falls into your category of "the lesser team losing to a more highly ranked team".

OTOH if the more highly ranked team loses I think you will find voters in both polls are willing to cut the losing team some slack precisely because it is a big, highly-publicized game played against a rival or a quality opponent.

I did not happen to notice whether you looked at the amount a ranked team drops when they lose to a bad team as compared to losing to a good (but lower ranked) team. I think you covered that angle with the result that the worse the winning team is, compared with the rank of the losing team, the further the losing team drops.

I suspect you may find that the difference in rank between teams playing in neutral stadiums is, on average, less than the average difference between any random pair of the teams in your study."


The short answer to the main question, whether or not neutral-site teams are more evenly matched, is that yes, they are. Almost two-thirds of the neutral site games are between teams within 10 spots of each other in the polls (whether it's #1 vs #10, or #14 vs #23). Interestingly, about half of the losses by Top 25 teams at neutral sites were upsets, while half were not - it's about even in that respect.

A side issue that we run into is that we've got a lot of categories here but few games - upset + location + difference in rank + month... it comes out to about 30 different combinations for 180-ish games. And because most of those games are during bowls and between evenly-matched teams, most of those 30 categories take 5 or less games into account to get their average. So the averages aren't as stable as I'd like them to be, but still.

Even working with these averages, there does seem to be a correlation between difference in rank at a neutral site + how far a team drops. But remember, it's not really a matter of if there's a correlation but how much of a correlation there is. There's still a greater correlation between how far a team drops + month, and how far a team drops + rank, and how far a team drops + margin of loss. (For instance, the Top 25 team that dropped the least in the AP after a neutral-site loss was a November upset by a team ranked 11-15 spots lower than them. If there was a greater correlation, we'd expect it to be a bowl loss, or NOT an upset, or a loss to a better opponent.)

I originally debated whether or not to add this factor to the main list, but I decided not to just because there's no real correlation between difference in rank + how far a team drops when it's not a neutral site game. But I think it's worth mentioning now.

Gabe said...

You mentioned that if there are any other correlations that might be relevant to mention them- how about previous week win/loss margin, as well as opponent.

I have a feeling that if a team is coming of a huge game ie playing the #1 team, their loss the next week may drop them less depending on how much worse (lower ranked) the team they're playing that week is.

So when if in week 2 of the upcoming season, USC is ranked #1 (although I pray they're not) when Ohio State plays them, I would guess that if Ohio State looses their game in week 3 against Toledo it would have a smaller effect than if they had played a team ranked #23 in week 2.

The other correlation I think would be cool to see is what effect the previous week's win/loss margin has on the next week's rankings.


I loved your article by the way.

4rx coupon codes said...

Hey you make a very good job with your analysis