FanPost

Predicting the SEC West: Warning Math

Before getting into the details, if you hate computer rankings, Elo rankings, etc. then move on. This analysis assumes that an algorithmic approach to evaluating teams is fairly effective. Comparing teams is hard for a computer. It is also hard for a human. I've used the Sagarin ratings because they are widely available and have an ok reputation in the sports analytics industry. Typically generating a point spread via Sagarin will be within a point of the opening Vegas line.

Now, to state the obvious, this division is historically good. All seven teams are in the top 35 of all three versions of the Sagarin rankings. The average ranking for the division is higher than all but 13 schools and four of those 13 are in the SEC west. Oklahoma, Oregon, UCLA,Georgia, Baylor, FSU, Stanford, Michigan State and Notre Dame are the only teams that rate as better than the average SEC West team.

Everyone but Arkansas is in the Ap top 15 but Arkansas is no slouch. The difference between Alabama, the top rated team in the division, and Arkansas is about 17 points at a neutral site. Sure that is a big gap but that is more parity than any division other than the ACC Costal division (where everyone is mediocre), the Big 12 (if you look at all 10 teams), and the PAC-12 North (the second best division in football). There are no easy weekends.

The Rankings

Team

Predictor

Alabama

93.66

A&M

90.25

Auburn

90.27

Ole Miss

85.31

LSU

87.89

Arkansas

78.03

MSU

80.09

HFA

3.54

Turning these into point spreads is pretty simple. I'm only going to convert games that will actually be played; i.e. we don't care about the spread if we played at Bama, So here are the future spreads (Home team in rows, away in columns):

Alabama

A&M

Auburn

Ole Miss

LSU

Ark

MSU

SEC-E#1

SEC-E#2

Alabama

-6.95

-6.93

-17.11

-11.22

A&M

-8.48

-5.9

-15.76

-14

Auburn

-3.56

-5.92

-15.78

-10.41

3.22

Ole Miss

4.81

1.42

-8.76

-9.95

LSU

2.23

-6.12

-2.09

-20.44

Ark

12.09

3.74

6.32

8.38

5.3

MSU

6.62

6.64

-5.6

-5.56

-15.28

As you can see Predictor really, really likes Alabama, and to a lesser degree Auburn and A&M. I think it might be a little too in love with Alabama. It is worth noting that Alabama benefits from playing A&M and Auburn at home. If those were road games they would be even money.

You might be asking yourself, "How the hell do you have a spread of -13.67?" Vegas will round it off and charge you extra vig. For our purposes these extra decimals are fine since we are just going to map the spreads to win percentages. Games that have been played have already been marked 0 or 100% for the losing/winning home team.

Alabama

A&M

Auburn

Ole Miss

LSU

Arkansas

MSU

SEC-E#1

SEC-E#2

Alabama

73.64%

73.58%

92.61%

100.00%

84.00%

A&M

77.79%

70.52%

100.00%

100.00%

88.79%

Auburn

62.86%

70.58%

100.00%

82.33%

38.32%

Ole Miss

32.94%

44.77%

78.49%

100.00%

81.32%

LSU

41.83%

71.19%

0.00%

0.00%

57.66%

95.35%

Arkansas

14.35%

36.52%

28.21%

22.47%

31.36%

MSU

27.32%

27.26%

69.59%

69.46%

90.54%

Now that we have win probability for each game the season can be simulated. Monte Carlo simulation is a fancy way to approximate something that is hard to compute exactly. Since there are 27 division games there are 134217728 possible combinations of results, not counting games vs the SEC East. Even knowing two results (MSU over LSU and A&M over Ark) there are still over 25 million ways the season could play out. The way it works is a random number between 0 and 1 is drawn for each game. If that number is less than the home team win probability we call that a win for the home team. Otherwise we call it a win for the away team. We simulate all the games, see who the division winner was, record the result then repeat as many times as we want. The whole thing takes a few minutes on a cheap computer to simulate the season a million times. There is a little error when you do this but gets small enough, fast enough, that nobody worries about the error (at most a tenth of a percent in this case).

So here are the results of 1000000 simulations with the Predictor spreads:

Team

Div Win %

Average wins

Alabama

42.22%

6.35

A&M

22.01%

5.73

Auburn

20.46%

4.99

Ole Miss

7.02%

4.52

LSU

4.46%

3.97

Arkansas

0.28%

1.72

MSU

3.55%

4.13

Not as promising as I would like but there are a few things to keep in mind.

  1. The Predictor model really, really likes Alabama. Sagarin's other ratings have them a few points worse. I thought about running both but this is long enough.

  2. Because we have played the weakest opposition of any team in the division our rating carries more uncertainty. We haven't played a team that is a good measuring stick. We will know may more after this weekend.

  3. We have a good cross division matchup. Auburn playing Georgia as opposed to anyone else in the SEC East is killing their chances. If we had to host Georgia or even USC instead of Tennessee it would cut our mean wins down closer to 4 and knock another percent or two off our division chances.

  4. Things suddenly look WAY better if we win this weekend.

Team

Div win %

Mean wins

Alabama

26.37%

5.68

A&M

25.23%

5.73

Auburn

22.87%

5

Ole Miss

15.15%

5.19

LSU

5.52%

3.97

Arkansas

0.35%

1.72

MSU

4.50%

4.13

We would actually be higher since we have evidence than we are better than we currently think (I believe we will win but the point spread wants to see proof). There would also be some fluctuations based on other teams performances. MSU would probably jump up to around 10% with a win over the Aggies and LSU would drop down just above Arkansas with a loss. If we want to get to Atlanta, this weekend and next weekend in College Station are the most critical games. Show up and be loud as hell. We need every tiny edge.

Honestly, if you told me after the Bama game last year 7% chance of winning the SEC West at this point I would have found that hard to believe. This team is way ahead of schedule in what could be the hardest division in the history. If we go 3-3 in division games we are light years from where we were when Freeze arrived.

If you want more detail about why Monte Carlo isn't terrible, why a generic HFA value is ok, or other sketchy mathematical tools of the sports analytics trade talk to me in the Grove (I'll be mobile since our tent was wiped out by ESPN) and I'll talk your ear off. I'll be the guy in the three piece suit who looks like he hasn't seen the sun since last weekend.

This post is a Red Cup Rebellion FanPost. Please don't sue us.