Thursday, November 26, 2015

WEEK TWELVE NFL PREDICTIONS

Below please find the 2015 week twelve NFL predictions based on the model we built last year. This model uses a learning algorithm fed by data from the last eleven weeks. To date, success rates this season are as follows: Week 2, 50%: Week 3, 62.5%: Week 4, 60%: Week 5, 64%; Week 6, 57%; Week 7, 50%; Week 8, 71%; Week 9, 53.8%; Week 10, 42.7%; Week 11, 57.1%.

This week the crowd disagrees with the model on four games, where the model predicts PHI over DET, ATL over MIN, TB over IND, and BAL over CLE. I'm sympathetic with the crowd on the TB @ IND game, and I never know what to think about CLE, though it does seem to me that this should be an easy win for BAL. I'm not sure what the crowd is thinking here. My intuitions say that TEN will defeat OAK, even though both the model and the crowd say otherwise.

So here goes - (The % indicated after each game represents the crowd-sourced prediction from nfl.com as of this afternoon. The two digit decimal numbers following team names should be interpreted relative to each other for comparison to determine, in part, how close the game will be.):

  • PHI @ DET
    • PHI (.46) over DET (.45) - Close Game
    • Crowd Disagrees @ 60%
  • CAR @ DAL
    • CAR (.64) over DAL (.48)
    • Crowd Agrees @ 73% 
  • CHI @ GB
    • GB (.57) over CHI (.49)
    • Crowd Agrees @ 95%
  • NO @ HOU
    • HOU (.56) over NO (.44)
    • Crowd Agrees @ 63%
  • STL @ CIN
    • CIN (.55) over STL (.46)
    • Crowd Agrees @ 93%
  • MIN @ ATL
    • ATL (.51) over MIN (.50)  - Close Game
    • Crowd Disagrees @ 65%
  • NYG @ WAS
    • NYG (.52) over WAS (.44)
    • Crowd Agrees @ 78%
  • TB @ IND
    • TB (.54) over IND (.49)
    • Crowd Disagrees @ 56%
  • BUF @ KC
    • KC (.66) over BUF (.49)
    • Crowd Agrees @ 67%
  • OAK @ TEN
    • OAK (.45) over TEN (.39)
    • Crowd Agrees @ 85%
  • SD @ JAX
    • JAX (.49) over SD (.36)
    • Crowd Agrees @ 77%
  • MIA @ NYJ
    • NYJ (.49) over MIA (.42)
    • Crowd Agrees @ 79%
  • ARI @ SF
    • ARI (.57) over SF (.34)
    • Crowd Agrees @ 96%
  • PIT @ SEA
    • SEA (.60) over PIT (.57) - Close Game
    • Crowd Agrees @ 54%
  • NE @ DEN
    • NE (.64) over DEN (.52)
    • Crowd Agrees @ 73%
  • BAL @ CLE
    • BAL (.49) over CLE (.33)
    • Crowd Disagrees @ 59%

Games to watch: PHI @ DET, MIN @ ATL and PIT @ SEA.

Thursday, November 19, 2015

WEEK ELEVEN NFL RESULTS

Below please find the 2015 week eleven NFL prediction results based on the model we built last year. This model uses a learning algorithm fed by data from the last ten weeks. To date, success rates this season are as follows: Week 2, 50%: Week 3, 62.5%: Week 4, 60%: Week 5, 64%; Week 6, 57%; Week 7, 50%; Week 8, 71%; Week 9, 53.8%; Week 10, 42.7%; and this week 57.1%. 

Last year, the Week 11 predictions were at 42%, dropping from 77% the week before.

This week the crowd did slightly better than the model at 64%. Details appear below.

The average power ratings took an interesting turn after the last two weeks. KC is now rated first after a four game winning streak, followed by CAR (2), NE (3), SEA (4), PIT (5), ARI (6), GB (7), HOU (8), CIN (9) and TB (10). 

The % indicated after each game represents the crowd-sourced prediction from nfl.com as of last Thursday afternoon. The two digit decimal numbers following team names should be interpreted relative to each other for comparison to determine, in part, how close the game will be.

  • TEN @ JAX
    • JAX (.46) over TEN (.38)
    • JAX 19 over TEN 13 - Prediction Correct
    • Crowd Correct @ 77%
  • OAK @ DET
    • OAK (.45) over DET (.40)
    • DET 18 over OAK 13 - Prediction Incorrect
    • Crowd Incorrect @ 79% 
  • IND @ ATL
    • ATL (.52) over IND (.48)
    • IND 24 over ATL 21 - Prediction Incorrect
    • Crowd Correct @ 81%
  • STL @ BAL
    • BAL (.47) over STL (.46) - Close Game
    • BAL 16 over STL 13 - Prediction Correct
    • Crowd Incorrect @ 73%
  • TB @ PHI
    • PHI (.53) over TB (.49)
    • TB 45 over PHI 17 - Prediction Incorrect
    • Crowd Incorrect @ 73%
  • DEN @ CHI
    • CHI (.52) over DEN (.51) - Close Game
    • DEN 17 over CHI 15 - Prediction Incorrect
    • Crowd Correct @ 61%
  • NYJ @ HOU
    • HOU (.56) over NYJ (.51)
    • HOU 24 over NTJ 17 - Prediction Correct
    • Crowd Incorrect @ 71%
  • WAS @ CAR
    • CAR (.62) over WAS (.49)
    • CAR 44 over WAS 16 - Prediction Correct
    • Crowd Correct @ 92%
  • DAL @ MIA
    • MIA (.44) over DAL (.42)
    • DAL 24 over MIA 14 - Prediction Incorrect
    • Crowd Incorrect @ 59%
  • KC @ SD
    • KC (.59) over SD (.45)
    • KC 33 over SD 3 - Prediction Correct
    • Crowd Correct @ 74%
  • GB @ MIN
    • MIN (.59) over GB (.51) - Close Game
    • GB 30 over MIN 13 - Prediction Incorrect
    • Crowd Correct @ 51%
  • SF @ SEA
    • SEA (.56) over SF (.36)
    • SAE 29 over SF 13 - Prediction Correct
    • Crowd Correct @ 94%
  • CIN @ ARI
    • ARI (.58) over CIN (.57) - Close Game
    • ARI 34 over CIN - Prediction Correct
    • Crowd Correct @ 70%
  • BUF @ NE
    • NE (.66) over BUF (.53)
    • NE 20 over BUF 13 - Prediction Correct
    • Crowd Correct @ 88%

WEEK ELEVEN NFL PREDICTIONS

Below please find the 2015 week eleven NFL predictions based on the model we built last year. This model uses a learning algorithm fed by data from the last ten weeks. To date, success rates this season are as follows: Week 2, 50%: Week 3, 62.5%: Week 4, 60%: Week 5, 64%; Week 6, 57%; Week 7, 50%; Week 8, 71%; Week 9, 53.8%; Week 10, 42.7%.

My intuitions are a bit uncomfortable with the CHI over DEN prediction. But they were last week, when the model predicted KC over DEN. So, maybe the model is detecting something I'm not. Also, I'll be surprised if the Texans really do defeat the Jets, but after their victory over the Bengals last week, it's really hard to say.

So here goes - (The % indicated after each game represents the crowd-sourced prediction from nfl.com as of this afternoon. The two digit decimal numbers following team names should be interpreted relative to each other for comparison to determine, in part, how close the game will be.):

  • TEN @ JAX
    • JAX (.46) over TEN (.38)
    • Crowd Agrees @ 77%
  • OAK @ DET
    • OAK (.45) over DET (.40)
    • Crowd Agrees @ 79% 
  • IND @ ATL
    • ATL (.52) over IND (.48)
    • Crowd Agrees @ 81%
  • STL @ BAL
    • BAL (.47) over STL (.46) - Close Game
    • Crowd Disagrees @ 73%
  • TB @ PHI
    • PHI (.53) over TB (.49)
    • Crowd Agrees @ 73%
  • DEN @ CHI
    • CHI (.52) over DEN (.51) - Close Game
    • Crowd Disagrees @ 61%
  • NYJ @ HOU
    • HOU (.56) over NYJ (.51)
    • Crowd Disagrees @ 71%
  • WAS @ CAR
    • CAR (.62) over WAS (.49)
    • Crowd Agrees @ 92%
  • DAL @ MIA
    • MIA (.44) over DAL (.42)
    • Crowd Agrees @ 59%
  • KC @ SD
    • KC (.59) over SD (.45)
    • Crowd Agrees @ 74%
  • GB @ MIN
    • MIN (.59) over GB (.51) - Close Game
    • Crowd Disagrees @ 51%
  • SF @ SEA
    • SEA (.56) over SF (.36)
    • Crowd Agrees @ 94%
  • CIN @ ARI
    • ARI (.58) over CIN (.57) - Close Game
    • Crowd Agrees @ 70%
  • BUF @ NE
    • NE (.66) over BUF (.53)
    • Crowd Agrees @ 88%

Games to watch: GB @ MIN, CIN @ ARI, DEN @ CHI, and STL @ BAL.

POST WEEK TEN NFL STATS


Model 1 (M1) is the procedure from last year and is based on normal averaging. So far, this has been the most accurate model. M2 is based on moving averages, and M3 is based on weighted percentages (thereby allowing games later in the season to count more than earlier games.) AVE is, of course, the average of the three. AVE POWER is the sum of M1, M2 and M3.

Sunday, November 15, 2015

THE SOCIAL-NETWORK ILLUSION

The MIT Technology Review recently posted an interesting demonstration about how ideas and beliefs in social networks can look more prominent than they actually are. See The Social-Network Illusion That Tricks Your Mind.



The explanation is rather straightforward: "... Lerman and co have discovered a related paradox, which they call the majority illusion. This is the phenomenon in which an individual can observe a behavior or attribute in most of his or her friends, even though it is rare in the network as a whole.

They illustrate this illusion with a theoretical example: a set of 14 nodes linked up to form a small world network, just like a real social network (see picture above). They then color three of these nodes and count how many of the remaining nodes link to them in a single step.

Two versions of this setup are shown above. In the left-hand example, the uncolored nodes see more than half of their neighbors as colored. In the right-hand example, this is not true for any of the uncolored nodes.

But here’s the thing: the structure of the network is the same in both cases. The only thing that changes is the nodes that are colored.

This is the majority illusion—the local impression that a specific attribute is common when the global truth is entirely different."

There are a lot of important lessons here that may be extendable to the extreme polarization in the American politic, among other things, especially since the illusion may also be exaggerated by the specific friends we keep and their role in influencing our own representations of reality on social media.

For those who are interested in such things, I'm considering offering a seminar on Networks and Network Analysis next Fall. This will be a nice companion to the Complex Systems seminar I will be teaching next semester.

Saturday, November 14, 2015

WEEK TEN NFL RESULTS

Below please find the 2015 week ten NFL predictions based on the model we built last year. This model uses a learning algorithm fed by data from the last nine weeks. To date, success rates this season are as follows: Week 2, 50%: Week 3, 62.5%: Week 4, 60%: Week 5, 64%; Week 6, 57%; Week 7, 50%; Week 8, 71%; Week 9, 53.8%; and this week 42.7%.

This is the first week to drop below 50%, and a few weeks like this are to be expected. Note that last year there were weeks at 31%, 38% and 42%. But this particular week was unusual on other fronts as well. The model was correct on 6 of the 14 games, but the crowd was correct on only 4 of them with a tie on a fifth, the DAL @ TB game. Furthermore, there was only one game in which the crowd was correct and the model incorrect, and this was the ARI @ SEA game.

Please keep in mind that we have yet to roll out any predictions based on the feature detection drive analysis network we are currently building. We are moving quickly along on that, though it's still difficult to say whether predictions for this year will happen this semester. At the latest, we will be re-predicting the games from this Fall next semester using new methods which are obviously needed.

(The % indicated after each game represents the crowd-sourced prediction from nfl.com as of this afternoon. The two digit decimal numbers following team names should be interpreted relative to each other for comparison to determine, in part, how close the game will be.)

  • BUF @ NYJ
    • BUF (.53) over NYJ (.53) - Very Close Game
    • BUF 22 over NYJ 17 - Prediction Correct
    • Crowd Incorrect @ 72%
  • DET @ GB
    • GB (.52) over DET (.35) - Easy Win for GB
    • DET 18 over GB 16 - Prediction Incorrect
    • Crowd Incorrect @ 98% 
  • DAL @ TB
    • DAL (.44) over TB (.44) - Very Close Game
    • TB 10 over DAL 6 - Prediction Incorrect
    • Crowd Split 50%-50%
  • CAR @ TEN
    • CAR (.58) over TEN (.42) - Easy Win for CAR
    • CAR 27 over TEN 10 - Prediction Correct
    • Crowd Correct @ 94%
  • CHI @ STL 
    • STL (.55) over CHI (.44) - Easy Win for STL
    • CHI 37 over STL 13 - Prediction Incorrect
    • Crowd Incorrect @ 87%
  • NO @ WAS
    • NO (.48) over WAS (.41)
    • WAS 47 over NO 14 - Prediction Incorrect
    • Crowd Incorrect @ 84%
  • MIA @ PHI
    • PHI (.55) over MIA (.41) - Easy Win for PHI
    • MIA 20 over PHI 19 - Prediction Incorrect
    • Crowd Incorrect @ 78%
  • CLE @ PIT
    • PIT (.50) over CLE (.36) - Easy Win for PIT
    • PIT 30 over CLE 9 - Prediction Correct
    • Crowd Correct @ 91%
  • JAX @ BAL 
    • BAL (.48) over JAX (.44)
    • JAX 22 over BAL 20 - Prediction Incorrect
    • Crowd Incorrect @ 72%
  • MIN @ OAK
    • MIN (.54) over OAK (.51)
    • MIN 30 over OAK 14 - Prediction Correct
    • Crowd Incorrect @ 51%
  • NE @ NYG
    • NE (.70) over NYG (.53) - Easy Win for NE
    • NE 27 over NYG 26 - Prediction Correct
    • Crowd Correct @ 83%
  • KC @ DEN
    • KC (.58) over DEN (.58) - Very Close Game
    • KC 29 over DEN 13 - Prediction Correct
    • Crowd Incorrect @ 93%
  • ARI @ SEA
    • SEA (.60) over ARI (.60) - Very Close Game
    • ARI 39 over SEA 32 - Prediction Incorrect
    • Crowd Correct @ 68%
  • HOU @ CIN
    • CIN (.65) over HOU (.55) - Easy Win for CIN
    • HOU 10 over CIN 6 - Prediction Incorrect
    • Crowd Incorrect @ 96%

Thursday, November 12, 2015

WEEK TEN NFL PREDICTIONS

Below please find the 2015 week ten NFL predictions based on the model we built last year. This model uses a learning algorithm fed by data from the last nine weeks. To date, success rates this season are as follows: Week 2, 50%: Week 3, 62.5%: Week 4, 60%: Week 5, 64%; Week 6, 57%; Week 7, 50%; Week 8, 71%; Week 9, 53.8%.

Note that this week the model agreed with the crowd in 10 of the 14 games as opposed to 10 of the 13 from last week. On the very close prediction of the DAL @ TB game the crowd split 50/50.

Contrary to the numbers, my intuitions strongly disagree with the predictions below on the KC @ DEN game. I'm not sure how the model could predict KC unless there's something wrong with the math. As it stands, the game is predicted to be very close. Maybe I'll be surprised and be correct on this one.

So here goes - (The % indicated after each game represents the crowd-sourced prediction from nfl.com as of this afternoon. The two digit decimal numbers following team names should be interpreted relative to each other for comparison to determine, in part, how close the game will be.):

  • BUF @ NYJ
    • BUF (.53) over NYJ (.53) - Very Close Game
    • Crowd Disagrees @ 72%
  • DET @ GB
    • GB (.52) over DET (.35) - Easy Win for GB
    • Crowd Agrees @ 98% 
  • DAL @ TB
    • DAL (.44) over TB (.44) - Very Close Game
    • Crowd Split 50%-50%
  • CAR @ TEN
    • CAR (.58) over TEN (.42) - Easy Win for CAR
    • Crowd Agrees @ 94%
  • CHI @ STL 
    • STL (.55) over CHI (.44) - Easy Win for STL
    • Crowd Agrees @ 87%
  • NO @ WAS
    • NO (.48) over WAS (.41)
    • Crowd Agrees @ 84%
  • MIA @ PHI
    • PHI (.55) over MIA (.41) - Easy Win for PHI
    • Crowd Agrees @ 78%
  • CLE @ PIT
    • PIT (.50) over CLE (.36) - Easy Win for PIT
    • Crowd Agrees @ 91%
  • JAX @ BAL 
    • BAL (.48) over JAX (.44)
    • Crowd Agrees @ 72%
  • MIN @ OAK
    • MIN (.54) over OAK (.51)
    • Crowd Disagrees @ 51%
  • NE @ NYG
    • NE (.70) over NYG (.53) - Easy Win for NE
    • Crowd Agrees @ 83%
  • KC @ DEN
    • KC (.58) over DEN (.58) - Very Close Game
    • Crowd Disagrees @ 93%
  • ARI @ SEA
    • SEA (.60) over ARI (.60) - Very Close Game
    • Crowd Disagrees @ 68%
  • HOU @ CIN
    • CIN (.65) over HOU (.55) - Easy Win for CIN
    • Crowd Agrees @ 96%

Games to watch: BUF @ NYJ, DAL @ TB, KC @ DEN, and ARI @ SEA.