A Month's Rent in Nebraska

“Chick-fil-a is some good shit! Put some of them waffle fries on the sandwich with cheese & ketchup. Soooo good, make you wanna slap yo momma!”

-Channing Crowder

Sorry for the delay but here are the results from week 3 and doubly sorry for the font issues.  To recap, I tried to create a model to predict NFL winners against the spread.  Wojay and Kyle Kelly have been helping me to create data and last week we unveiled the first model.  Also we made believe that we were starting with $1600 dollars with which to bet ($100 per game).  Kyle Kelly calculated how much money we would be with at the end of the week if we bet according to model.  For those who are just interested in how the model did, amazingly (and surprisingly) enough we ended up winning $373.33.  Below is a table that displays the results.  Me, Wojay and a Random Nebraskan also picked against the spread.  The bolded column is the model prediction and highlighted cells are the predictions that the model got correct.

Game	Spread	Predicted Difference	Model Prediction	Mike_Z	Wojay	Random Nebraskan	Actual Result
Patriots AT Bills	Pats by 8.5	Bills by 0.15518	Bills	Patriots	Bills	Patriots	Bills
Jaguars AT Panthers	Panthers by 3.5	Panthers by 8.58	Panthers	Jaguars	Panthers	Jaguars	Panthers
Broncos AT Titans	Titans by 6.5	Broncos by .037	Broncos	Broncos	Titans	Titans	Broncos
Giants AT Eagles	Eagles by 9	Eagles by 6.27	Giants	Eagles	Giants	Eagles	Giants
Texans AT Saints	Saints by 4	Texans by 1.35	Texans	Saints	Saints	Saints	Saints
Lions AT Vikings	Lions by 3.5	Lions by 17.88	Lions	Lions	Lions	Lions	Vikings
Dolphins AT Browns	Browns by 2.5	Browns by 6.29	Browns	Dolphins	Browns	Dolphins	Dolphins
49ers AT Bengals	Bengals by 3	Bengals by 1.45	49ers	49ers	Bengals	49ers	49ers
Jets AT Raiders	Jets by 3	Raiders by 3.06	Raiders	Jets	Jets	Raiders	Raiders
Ravens AT Rams	Ravens by 4	Ravens by 9.74	Ravens	Ravens	Ravens	Ravens	Ravens
Chiefs AT Chargers	Chargers by 14.5	Chargers by 15.03	Chargers	Chiefs	Chargers	Chargers	Chiefs
Falcons AT Buccaneers	Bucs by 1.5	Falcons by 3.38	Falcons	Falcons	Falcons	Falcons	Buccaneers
Cardinals AT Seahawks	Cardinals by 3.5	Cardinals by 13.04	Cardinals	Cardinals	Cardinals	Cardinals	Seahawks
Packers AT Bears	Packers by 3.5	Packers by 17.39	Packers	Packers	Packers	Packers	Packers
Steelers AT Colts	Steelers by 10.5	Steelers by 7.88	Colts	Steelers	Colts	Steelers	Colts
Redskins AT Dallas	Cowboys by 3	Cowboys by 1.12	Redskins	Cowboys	Redskins	Redskins	Redskins

Below is a table that shows how each of us fared:

	Percent Right	Season Results
Model Prediction	0.625	0.625
Mike_Z	0.4375	0.4375
Wojay	0.5	0.5
Random Nebraskan	0.4375	0.4375

As you can see, the model was correct on almost 63% of the games.  If the model can hold this percentage, I will be beyond excited.  On the other hand, you other three experts all lost money.  Of course those reading this shouldn’t be surprised at me and Wojay’s results.  As for the random Nebraskan, you have to remember this is a college football state.

As I said above, Kyle Kelly interpreted the money lines and came out with how much money the model won.

	Money	Total
Starting		1600
End Week 3	373.33	1973

Anyways, the predictions for week 4 will be posted either Friday or Saturday.  If the model can keep up this performance, we may be on to something.

Sometimes Art is Shit Thrown on a Wall

“I couldn't find London on a map if they didn't have the names of the countries. I swear to God. I don't know what nothing is. I know Italy looks like a boot. I learned that. I know London Fletcher. We did a football camp together. So I know him. That's the closest thing I know to London. He's black, so I'm sure he's not from London. I'm sure that's a coincidental name.”

            -Channing Crowder

So, it’s finally here:  A model to predict weekly lines.  For those outside the know:  This is an over/under bet on whether a team will win by a specific score.  For example, the Patriot’s line is 8.5 points.  That is if you pick the Patriots you think they will win by at least 9.

Any Thoughts?

Due to popular demand (OK fine 2 people asked if I could try this), I decided to attempt this.  And it is a bitch.  I’m still working out how to best model this, but I want to emphasize this is a working model.  That is, I think this model sucks and can be improved.

Anyways, on to what I modeled.  Data were collected by me and two volunteers, Wojay and Kyle Kelly (to whom I’m very grateful).  Numerous variables were recorded and, after modifications and the use of composite variable, plugged in to a regression modeling processes with the idea of maximizing R_A².  Due to time constraints only offensive variables were used.  The following six variables accomplished this goal:

1.      Rushing attempts

2.      Pass Completion Percentage

3.      Rushing Touchdowns

4.      Passing Touchdowns

5.      Turn-overs

6.      Spread

A regression model was obtained that predicted the amount of points will score.  Estimates were made by taking the difference of the two teams playing each other.  I want to emphasize that injuries and strength of the opponent’s defense were not modeled, but will be in future updates.  For those interested, the model statistics are included at the end.  Below, is the spread offered by sportsbook.com on the Thursday before the game, the difference in points scored between the two teams, which team is predicted to beat the line by the model, my prediction (model free), Wojay’s prediction, and a random Nebraskan’s prediction.  Further, I included a smart-ass reason for my prediction.

·         Patriots at Bills

Spread	Predicted Difference	Model Prediction	Mike_Z	Wojay	Random Nebraskan
Pats by 8.5	Bills by 0.15518	Bills	Patriots	Bills	Patriots

Mike Z Thoughts:  I fully believe the Patriots will win this game.  In fact, I think the Bills suck.

·         Jaguars at Panthers

Spread	Predicted Difference	Model Prediction	Mike_Z	Wojay	Random Nebraskan
Panthers by 3.5	Panthers by 8.58	Panthers	Jaguars	Panthers	Jaguars

Mike Z Thoughts:  The Jags suck and Cam Newton moves the ball.

·         Broncos at Titans

Spread	Predicted Difference	Model Prediction	Mike_Z	Wojay	Random Nebraskan
Titans by 6.5	Broncos by .037	Broncos	Broncos	Titans	Titans

Mike Z Thoughts:  Winning by a touchdown is a lot to ask for a team with octogenarian at QB.

·         Giants at Eagles

Spread	Predicted Difference	Model Prediction	Mike_Z	Wojay	Random Nebraskan
Eagles by 9	Eagles by 6.27	Giants	Eagles	Giants	Eagles

Mike Z Thoughts:  The fact that Tom Coughlin tried to sign me at CB, makes me feel as if the Giants cannot stop Maclin and Jackson.  Note:  If I do sign with Giants disregard the previous thought.

·         Texans at Saints

Spread	Predicted Difference	Model Prediction	Mike_Z	Wojay	Random Nebraskan
Saints by 4	Texans by 1.35	Texans	Saints	Saints	Saints

Mike Z Thoughts:  The Dolphins have never beaten the Texans.  Therefore, fuck the Texans.

·         Lions at Vikings

Spread	Predicted Difference	Model Prediction	Mike_Z	Wojay	Random Nebraskan
Lions by 3.5	Lions by 17.88	Lions	Lions	Lions	Lions

Mike Z Thoughts:  The Vikings have no talent besides Adrian Peterson.  The Lions have Suh.  I’m more terrified of Suh than AP.

·         Dolphins at Browns

Spread	Predicted Difference	Model Prediction	Mike_Z	Wojay	Random Nebraskan
Browns by 2.5	Browns by 6.29	Browns	Dolphins	Browns	Dolphins

Mike Z Thoughts:  Let’s just move along…

·         49ers at Bengals

Spread	Predicted Difference	Model Prediction	Mike_Z	Wojay	Random Nebraskan
Bengals by 3	Bengals by 1.45	49ers	49ers	Bengals	49ers

Mike Z Thoughts:  Never trust a red-head…especially at QB.

·         Jets at Raiders

Spread	Predicted Difference	Model Prediction	Mike_Z	Wojay	Random Nebraskan
Jets by 3	Raiders by 3.06	Raiders	Jets	Jets	Raiders

Mike Z Thoughts:  Jason Campbell does still start for the Raiders right?

·         Ravens at Rams

Spread	Predicted Difference	Model Prediction	Mike_Z	Wojay	Random Nebraskan
Ravens by 4	Ravens by 9.74	Ravens	Ravens	Ravens	Ravens

Mike Z Thoughts:  As a Dolphins fan, I’ve seen how hard it is to win without a real receiver.  Sorry Sam Bradford.

·         Chiefs at Chargers

Spread	Predicted Difference	Model Prediction	Mike_Z	Wojay	Random Nebraskan
Chargers by 14.5	Chargers by 15.03	Chargers	Chiefs	Chargers	Chargers

Mike Z Thoughts:  It’s really hard to win by more than 3 scores.

·         Falcons at Buccaneers

Spread	Predicted Difference	Model Prediction	Mike_Z	Wojay	Random Nebraskan
Bucs by 1.5	Falcons by 3.38	Falcons	Falcons	Falcons	Falcons

Mike Z Thoughts:  Ummm, The Bucs just aren’t good.

·         Cardinals at Seahawks

Spread	Predicted Difference	Model Prediction	Mike_Z	Wojay	Random Nebraskan
Cardinals by 3.5	Cardinals by 13.04	Cardinals	Cardinal	Cardinal	Cardinals

Mike Z Thoughts:  Honestly, who cares about this game.  If you have a vested interested in this game, then you missed the purpose of the American Dream.

·         Packers at Bears

Spread	Predicted Difference	Model Prediction	Mike_Z	Wojay	Random Nebraskan
Packers by 3.5	Packers by 17.39	Packers	Packers	Packers	Packers

Mike Z Thoughts:  “I Don’t Always Throw Interceptions…But When I Do I Prefer To Throw Them In The Redzone.”

            -Jay “The Most Interesting Man on the Planet” Cutler

·         Steelers at Colts

Spread	Predicted Difference	Model Prediction	Mike_Z	Wojay	Random Nebraskan
Steelers by 10.5	Steelers by 7.88	Colts	Steelers	Colts	Steelers

Mike Z Thoughts:  The Colts just tried to sign me as a QB.  I’d be an upgrade

·         Redskins at Cowboys

Spread	Predicted Difference	Model Prediction	Mike_Z	Wojay	Random Nebraskan
Cowboys by 3	Cowboys by 1.12	Redskins	Cowboy	Redskin	Redskins

Mike Z Thoughts:  I picked the Cowboys based on the sole perception that they have been less obnoxious than Redskins fans…recently.

Stats Section

Analysis of Variance

                                            Sum of           Mean

        Source                   DF        Squares         Square    F Value    Pr > F

        Model                     6     5222.24760      870.37460      47.99    <.0001

        Error                    57     1033.73678       18.13573

        Corrected Total          63     6255.98438

                     Root MSE              4.25861    R-Square     0.8348

                     Dependent Mean       23.48438    Adj R-Sq     0.8174

                     Coeff Var            18.13379

As can be seen the R_A² is .82, which is very high.  However, because we are using two weeks of data, we are quite limited.  Further, each observation is nested within a team which causes dependence issues.  That is, points scored are nested within teams.  To get around this, I plugged team averages into the regression formula.  A more sophisticated model is necessary to combat the nesting issue, but I didn’t have time to model this.  An unfortunate result of this, is that a few of the difference scores are either very high or low.  Hopefully, by modeling the opponent’s defense these extreme scores will be moderated.

Parameter Estimates

                                         Parameter       Standard

       Variable     Label        DF       Estimate          Error    t Value    Pr > |t|

       Intercept    Intercept     1        1.88452        4.32534       0.44      0.6647

       OR_Att       OR_Att        1        0.13531        0.07692       1.76      0.0839

       TO                         1       -0.73234        0.42615      -1.72      0.0911

       P_Comp                     1        9.54637        6.69066       1.43      0.1591

       Spread       Spread        1       -0.23835        0.09401      -2.54      0.0140

       OR_TDs       OR_TDs        1        4.26551        0.81393       5.24      <.0001

       OP_TDs       OP_TDs        1        5.87377        0.53473      10.98      <.0001

This table shows how each variable effected the prediction.  This is shown in the parameter estimate column.  For instance, if we consider the turn-over variable (TO), we have the following interpretation:  For every turn-over, we expect the amount of points scored to decrease by .732.  Note:  Pass completion percentage (P_Comp), is not interpretable unless you divide its estimate by 100.  The spread estimate may also be difficult to interpret.