A Splash for Awareness

Recently my Facebook news feed has become flooded with videos of my friends splashing themselves with ice cold water to raise awareness for ALS. 

Voices have sprung up poo-pooing on it saying that it is silly viral fad and doing no good. Raising awareness is usually a term everyone uses with little understanding of what we are being made aware of. So I asked- are people becoming more aware? 

What better place to look than to our modern source of knowledge and awareness - Google! 

I looked up the worldwide Google  trends in the last 90 days related to ALS. With no surprise my crude assessment shows a big spike since the start of the viral fad. Conclusion - people are reading about it. 

It is a big jump to say people are being made more aware but at least they are reading up on it. Plus as a bonus ALS Association has raised a lot of money

 "The organization’s national office has received $5.5 million for Lou Gehrig’s disease research since July 29, compared to $32,000 in the same period last year." 

(Source:http://time.com/3111699/ice-bucket-challenge-als-donations/ )

So why not people of the internet- Keep splashing! 

Olympic Special - Men's Hockey Simulation Results

Living in Canada it is hard not to get the olympic bug and no bigger hype than that of Hockey. So I decided to try running a simulation for the tournaments final 4 teams (Canada, US, Sweden, Finland).

It is obviously a low number of games and the inconsistency of the model is through the roof. I was even lucky the whole model converged. I thought of adding other tournament or recent friendly games but those always have different players and seem irrelevant in this case. I will update the post later this week for the medal games.

Bottom line - even less than the NFL simulations take this for a grain of salt. Still fun to do anyway.

Simulation Results

Note: This is their probability of being the best team in the tournament. 

Gold: Sweden 44% (USA-26%, Canada-14%, Finland-4%)

Silver: USA 29%

Bronze: Canada 15%

PS- I hope the model is wrong. Go Canada Go!

From Data to the Super Bowl: Super Bowl Predictions

The Super Bowl game, we made it!

In review - Both probabilities correctly predicted last weeks conference championship games. The original super bowl prediction was rather close missing the correct prediction of having Denver in the final by one game (pesky bengals!)

We have an interesting super bowl ahead of us with the best defense playing the best offense. Historically in four super bowls where the top defense played the top offense the defensive team wins 3 out of 4 times (75%).  This supports the old Mantra....Defense Wins Championships. This was questioned by a recent analysis by the Advanced NFL Stats team that found that elite offenses in most cases will out pace elite defenses. This analysis took into account total performance and is not specific to playoffs. Either way should be a game for the history books.

 
 For my analysis the total 2013-2014 network of games including playoffs was taken into account.  The strength of this analysis is not only does it take into account the scores and the margins but also who it was against (and how those teams performed against other teams). All currently reported rankings of offense and defense are based on crude performance of yards and thus may over/under estimate true rankings. This analysis did not have weighting on any single game (NOTE: when this was done for playoff games the conclusion was the same).

Both teams did not play each other this season so all estimates are based on indirect comparisons from the network. See Network explanation here. So this week I will put the 95% probability range along with the estimate probability. Without any further adieu the results:

Super Bowl Championship Probability

DEN VS SEA ------------------------------------> SEA Win 78% chance (95% CrI = 57%-91%)


CrI -  Credible Interval. This interpretation of the 95% is that this is the certainity of our estimate. As you can see it all falls above the 50% giving us a higher certainty of a SEA win.

From the Data to the super Bowl: Championship Probabilities

We are almost there! Three more games. Here are this weeks probabilities for the conference championship games. 

As a reminder it is important to note that although we have a probability, that probability has a level of certainty. I have included the distributions for each game's probability to depict how the estimate is derived (Direct and Indirect data - see prior posts). Direct comparisons come from when these  teams played each other and the other data comes from the network of other teams they played. As you can see in both games there is a slight discordance between the direct and indirect evidence. Season performance against similar teams (network based) highly benefits both DEN and SEA which both had better records than their relative opponents and  won by larger margins. In contrast, the games played against each other were either split (SF/SEA) or in favor of NE. This highly affects the level of certainty around our estimate. Should be another exciting weekend of football in what has been a great playoffs thus far!



Game 1: NE @ DEN --------------------------> DEN win - 59.8%

Game 2: SF @ SEA ----------------------------> SEA win - 64.7%

From the Data to the Super Bowl: Playoff chances

So the wildcard weekend was just as the name says...WILD! Tough weekend for predictions and many of the brackets thrown in the trash (including my own). I was not going to adjust the prediction but for the fans I will post something for this week.

Based on some feedback rather then communicating certainty I am going to post probability of victory.  Certainty is based on not only my estimate but the  95% credibility region around that estimate. It is important to note that although we have a probability, that probability may have its own certainty around it as an estimate. Here are the probabilities of victory  for the 4 playoff games this week.



Game 1: NO @ SEA --------------------------> SEA win - 70.6%

Game 2: IND @NE ----------------------------> IND win - 51.2%

Game 3: SF @ CAR ---------------------------> SF win - 57.8%

Game 4: SD @ DEN ---------------------------> TIE  - 50%


Bonus- Looking Under the hood:  SD @ DEN further analysis (for Danny and Muno)



As discussed in prior entries the analysis being conducted is placing all the games into a network and running a simulation of the ways the games could possible happen based on prior games. The analysis then takes into account not only if the teams have played each other (direct comparison) but how they performed against teams they both played (indirect comparisons) and the teams those teams played etc.. It then weighs those indirect comparisons less then if they played each other directly. The analysis weighs direct comparisons most heavily. Comparisons further down the network are weighed less.

So lets take the SD @ DEN game for example. They split both their games against each other by an almost equal margin, additional the comparisons closest in the network (within the division) there records are only off by one game 5-1 (DEN) and 4-2 (SD) (with similar performances). So their direct and indirect comparisons are very close. When graphed (for you visual people- See Below) we see that the estimates of their direct and indirect both mirror each other right down the middle of the effect estimate (the chance of winning)- The middle of the curves or the highest density its right at 0.  Putting our probability right in the middle at 50%! Should be interesting to see how this one plays out.

Another great weekend of football ahead!

Graphic representation of the direct and indirect estimates for the SD @ DEN Game.

From the Data to the Super Bowl - Playoff Predictions

So here we are....what everyone has been waiting for....(dramatic pause).... The Playoffs. The data has been compiled and the lessons learned.  Still a work in progress.
 


Playoff Predictions

   
Wild Card Games

Game 1: KC (5) @ IND (4) -------->    IND (4) Win (Low Certainty)

Game 2: SD (6) @ CIN (3) -------->    CIN (3) Win (Medium Certainty)

Game 3: SF (5) @ GB (4) --------->     SF (5) Win (Very High Certainty)

Game 4: NO (6) @ PHI (3) -------->    PHI (3)  Win (Very Low Certainty)


Divisional Round

Game 1: IND (4) @ DEN (1) ------>   DEN (1) Win (Medium Certainty)

Game 2: CIN (3) @ NE (2) -------->   CIN (3) Win (High Certainty)

Game 3: SF (5) @ SEA (1) -------->   SEA (1) Win (Medium Certainty)

Game 4: PHI (3) @ CAR (2) ------>   CAR (2) Win (Very Low Certainty)


Championship Games 

Game 1: CIN (3) @ DEN (1) ------->  CIN (3) Win (Low Certainty)

Game 2: CAR (2) @ SEA (1) ------->  SEA (1) Win (High Certainty)


Super Bowl 

CIN @ SEA ---------------------------->  SEA Wins (Medium Certainty)


Super Bowl Champions: Seattle


Comments:

Thanks for all the feedback and lots of great brainstorming on this latest NFL experiment. Based on the feedback there will be two major changes to work on in the off-season for next season's analysis. 

1. Taking into account home field advantage
2. A moving average to better predict the power rankings and take into account recent performance for adjustment on the head-to-head predictions.


Happy New Year!

M.