Discuss football with over 60,000 fans. Free Membership. Join now!

 FAQFAQ  RegisterRegister   ProfileProfile   Log in to check your private messagesLog in to check your private messages   Log inLog in 

FootballsFuture.com Forum Index
FootballsFuture.com Home

Vikings by Numbers: Advanced Statistical Analysis

 
Post new topic   Reply to topic    FootballsFuture.com Forum Index -> Minnesota Vikings
View previous topic :: View next topic  
Author Message
Krauser


Joined: 20 Apr 2013
Posts: 1836
PostPosted: Sun May 12, 2013 5:02 pm    Post subject: Vikings by Numbers: Advanced Statistical Analysis Reply with quote

Several websites publish advanced NFL statistics and analysis: Football Outsiders and Pro Football Focus are probably the best known, but there are many others, including Advanced NFL Stats, Football Perspective, and Smart Football.

Some of them take published numbers (like passing stats) and analyze them to provide more detail. Volume stats like passing yardage and TD passes -- which accumulate with more attempts -- are ignored in favor of efficiency (yards per attempt, TDs per attempt, etc). The best known simple examples are yards per carry for RBs and yards per catch for receivers. Similar numbers exist for sacks allowed per pass attempt (for OLs), yards given up per coverage snap (for CBs) and catches per targeted pass (for WRs). In general, these numbers are quite a bit more revealing than the raw numbers.

Even more sophisticated analysis is available. Multivariable efficiency stats aim to pull several numbers together to give an overall index of performance. The standard NFL passer rating is the best known example, as is ESPN's QB rating. Other stats are simple enough but aren't included in the usual data from the NFL, so they have to be compiled by staff reviewing play-by-play on tape and writing down which CB was in coverage for that completed pass, who missed a tackle on that running play, or a block on that sack, etc.

The most complicated stats start by developing a model based on past performance (several years of NFL games, analyzed play by play) to predict what kind of result on a given play correlates with success (how many yards rushing on 1st and 10 do you need to gain to make that play successful?, etc). Then the standard NFL statistics are broken down according to situation (down and distance, etc), to show the difference in value of a pass that gains 12 yards on 3rd and 10, vs one that gains 12 yards on 3rd and 15. The best example of those model-based analyses are Football Outsiders' Defense-adjusted Yards Above Replacement (DYAR) and Defense-adjusted Value Over Average (DYOA). If you're up for it, here's FO's explanation of their methods: http://www.footballoutsiders.com/info/methods

These are also the kinds of numbers that lead to debates about whether NFL coaches punt too much, or how to best manage the clock in the last couple of minutes of a half.

I think we could use a thread to collect and discuss these "advanced" stats, as they apply to the Vikings. They can offer a clearer picture of how players are performing, what the coaches tendencies are, etc. Statistics are always open to debate, but they can be a useful starting point.
Back to top
View user's profile Send private message
Krauser


Joined: 20 Apr 2013
Posts: 1836
PostPosted: Sun May 12, 2013 6:00 pm    Post subject: Reply with quote

Since I mentioned Football Outsiders' DVOA, let's start with that. Here's their final DVOA stats from the 2012 regular season: http://www.footballoutsiders.com/dvoa-ratings/2012/final-2012-dvoa-ratings

The Vikings were:
--14th in the league, 2% better than average (#1 Seattle was 38% better than average, #32 Kansas City was 40% worse than average)
--17th in the league when you take strength of schedule into account ("weighted DVOA", which calculates opponent strength based on DVOA not wins and losses)
--15th in the league on offense, just barely above average
--21st in the league on defense, 3% worse than average (better defenses produce negative percentage scores, the opposite of offenses)
--5th in the league on special teams

In the second table, we learn they were:
--expected to win 8.8 games last year based on their play-by-play statistical performance and that of their opponents ("Pythagorean wins")
--the 3rd most consistent team in the league in terms of week-to-week putting up similar overall numbers ("Variance")

The Variance score is the only major surprise there for me. I guess the decline of the pass defense after Cook got hurt was matched by the improvement of the rush offense once Peterson got back to full speed, so a couple of changes mostly cancelled each other out. Contrast that to San Francisco, 31st in the league in Variance (2nd most inconsistent team), who got killed by the Giants and Seahawks but also had several dominant wins.

...

I think some people here would be surprised that the defense was worse than the offense compared to the rest of the league. You can click through the DVOA archives to see how the numbers changed throughout the year.

The Vikings' high-water mark for DVOA was after the dominant win over the Titans, in week 5: 7th in the league with 23.5% DVOA. The offense was 4.6% above average (12th), and the defense was -11.2% better than average (9th).

After the bad loss to Washington and the underwhelming win over the Cards, they were still 11th in the league, but their offense had fallen off a cliff: total DVOA of 10.4% (11th), offense -2.6% (negative is bad) (16th), defense -4.5% (negative is good) (12th). Special teams were ranked 2nd and holding the team up in the standings.

Chris Cook broke his arm the next week in the blowout loss to Tampa Bay, and after that the pass defense suffered greatly. By the end of the win over the Rams, the last game Cook missed, the defense had dropped 14 spots and was ranked 26th in the league, 7% worse than average. Once Cook came back, they rallied to finish 21st in the league, 3% worse than average.

Meanwhile the offense steadily improved after Harvin went out. After the week 9 loss in Seattle (Harvin's last game), the offense DVOA was -6.6% (negative is bad), 21st in the league -- a huge drop off from their position after the Titans game. That number improved every week afterward, though they were still below average (-2%) after the Texans game, and only broke barely into positive digits after the big win at home against the Packers.

...

So you can pull together this story for the Vikings in 2012:

Games 1-5 (up to the Titans win): Decent offense built around gadget plays to Harvin, even with Peterson limited. Decent defense against the run and pass. Good special teams. Record: 4-1

Games 6-8 (Redskins, Cards, Bucs): Wheels fall off the offense (Ponder forces some terrible INTs, coverage limits Harvin). Peterson is doing more but not at full speed yet. The whole defense plays terribly against RG3, does OK but not great against a terrible Cards offense, then Cook breaks his arm against the Bucs as we get blown out. Record: 1-2.

Games 9-12 (Seattle, first games against Packers and Bears): Harvin gets hurt. Peterson is amazing but Ponder and the passing game are terrible. Defense can't stop the pass, probably the worst numbers in the league over these weeks. They beat the Lions anyway. Record: 1-3

Games 13-14 (Bears, Rams): Peterson is still amazing. Ponder plays better, at least stops making mistakes. The defense is still weak but makes some big plays against Cutler. Record: 2-0.

Games 15-16 (Texans, Packers): Peterson is still amazing. Ponder plays pretty well. Chris Cook comes back and the defense plays great until Winfield is forced off with an injury at halftime against the Packers. Sherels has to step in and gets torched, but we hold on to win. Record: 2-0

...

Good signs:
-- Musgrave designed an effective offense for our personnel that worked well at the beginning of the year, including the big win against the Niners.
-- The offense improved toward the end of the year, even missing Harvin. A lot of that is thanks to Peterson, but Ponder played well the last few weeks as well.
-- Special teams were great all year

Bad signs:
-- The wheels fell off the Musgrave offense around the time Ponder started throwing INTs, even before Harvin got hurt, and didn't get put back together until the last few weeks.
-- Even with a record performance from AP over the last 8 weeks, the offense's overall efficiency improved only gradually
-- When Chris Cook was out, the defense was terrible.


Last edited by Krauser on Sun May 12, 2013 6:18 pm; edited 1 time in total
Back to top
View user's profile Send private message
Krauser


Joined: 20 Apr 2013
Posts: 1836
PostPosted: Sun May 12, 2013 6:17 pm    Post subject: Reply with quote

One more for now...

Chase Stuart at Football Perspectives has put together a stat called Game Scripts: http://www.footballperspective.com/game-scripts-the-best-teams-of-2012/

The idea is to calculate the average score (leading or trailing) for each team over the course of the year. A team that scores early and holds leads all game will have a positive score. A team that falls behind early and trails all game has a negative score. A team that falls behind early, rallies late, and wins a lot will still have a negative score despite a winning record, because they're usually trailing.

How did the Vikings do?

9th in the league at +2.9 points (leading by 2.9 points on average at any given moment all year).

Patriots were first, +7.7.
Seahawks +5.5
Broncos +4
Niners & Falcons +3.6
Bengals +3.5
Packers & Bears +3
Vikings +3

Meanwhile, the 1998 Vikings were +8.6, in the top 20 all time.

Another similar stat from the same guy: time spent leading, tied or trailing http://www.footballperspective.com/time-spent-leading-tied-or-trailing/

This number isn't weighted by the size of the lead, just measures the time with any lead or deficit.

How did the Vikings do?

2nd in the league for most time leading, 58.7%. Patriots 1st with 65%, Niners and Seahawks 54%.

7th in the league for least time trailing, 29%. Patriots 1st again with 17%. Seahawks 22%, Niners 25%.

32nd in the league for time spent tied, only 12%.

...Numbers probably reflect high success rate on offense on the first drive of the game.
Back to top
View user's profile Send private message
HighHopes


Moderator
Joined: 01 Jun 2008
Posts: 27095
PostPosted: Sun May 12, 2013 7:25 pm    Post subject: Reply with quote

^^Thats insane, we were leading for the 2nd most amount of time? Laughing
_________________
XBL/PSN: Chainedsniper
Back to top
View user's profile Send private message
Purplexing


Joined: 13 Jan 2008
Posts: 4443
Location: Outside Valhalla, looking in.
PostPosted: Sun May 12, 2013 9:37 pm    Post subject: Reply with quote

HighHopes wrote:
^^Thats insane, we were leading for the 2nd most amount of time? Laughing


Surprises due to simplicity of some stats leads to a yearning for more information about the stats.

Minnesota leading for a high percentage of games ( 'Game Scripts' ) might be explained by several things, which I guess would include, but are not limited to, running the ball effectively with the lead (e.g. AP), and getting an early lead by kicking FGs from longer distances than most other teams (e.g. Walsh).

One deficiency about 'Game Scripts' is that Baltimore doesn't appear in the top 10 NFL teams, yet they won the SB. So that stat is informative, but it doesn't individually point to success. That's where multi-variate analysis (M-VA) comes in.

The idea behind M-VA is to consider a number of predictive factors and calculate based on historical W-L data and those factors how much each factor contributes to the success of a team or player or coach. The factors with the highest weights should then get the most attention of coaches, to the extent coaches can influence those factors.

The ESPN QBR and the NFL passer ratings are not truly based on multi-variate analyses (M-VA), but rather, are multi-variate models that are calibrated based on the opinions of experienced sports analysts. Going off memory, the NFL passer rating has a maximum of 158.3, which is due to the size of weights applied to 4 separate factors. And the weights were a consensus pick of statisticians working for the NFL at the times the formula was 'assembled' (vs. calibrated). Please correct my recollection, if I am wrong on any of the details.

After I get time to read the details of these advanced stat analyses, I'll add more thoughts. But, I suspect that many analyses that are effective predictors will probably be correlated with the two keys to winning games; e.g. field position and turnovers.
Back to top
View user's profile Send private message
Bermuda Viking


Joined: 19 Mar 2005
Posts: 4912
Location: Paget, Bermuda
PostPosted: Sun May 12, 2013 9:55 pm    Post subject: Reply with quote

Krauser that was informative and entertaining at the same time. I loved the post and thank you for the effort!

So much of it made sense as to last year. Sure Baltimore won, but it's always the extra play beyond the statistics that wins each game. The Vikings were not supposed to win the games they did last year and this year the prediction is the same.

I really enjoyed your post.
_________________
I love the Vikings, win or lose.
Back to top
View user's profile Send private message
Klomp


Joined: 11 Aug 2011
Posts: 6608
PostPosted: Sun May 12, 2013 10:00 pm    Post subject: Reply with quote

Purplexing wrote:
One deficiency about 'Game Scripts' is that Baltimore doesn't appear in the top 10 NFL teams, yet they won the SB. So that stat is informative, but it doesn't individually point to success. That's where multi-variate analysis (M-VA) comes in.


That's because these stats were from only the regular season. I don't believe Baltimore was a top 10 regular season team.

There were 8 teams with a better regular season record than the Ravens, and 4 others who finished with an identical 10-6 record as them. But once they got to the playoffs, all bets were off.

The Baltimore Ravens were not the best team in the NFL last season. They were just the team that went on the best run in the playoffs.
Back to top
View user's profile Send private message
Purplexing


Joined: 13 Jan 2008
Posts: 4443
Location: Outside Valhalla, looking in.
PostPosted: Sun May 12, 2013 10:04 pm    Post subject: Reply with quote

Klomp wrote:
Purplexing wrote:
One deficiency about 'Game Scripts' is that Baltimore doesn't appear in the top 10 NFL teams, yet they won the SB. So that stat is informative, but it doesn't individually point to success. That's where multi-variate analysis (M-VA) comes in.


That's because these stats were from only the regular season. I don't believe Baltimore was a top 10 regular season team.

There were 8 teams with a better regular season record than the Ravens, and 4 others who finished with an identical 10-6 record as them. But once they got to the playoffs, all bets were off.

The Baltimore Ravens were not the best team in the NFL last season. They were just the team that went on the best run in the playoffs.


That is correct, which is derived from the regular season stats showing one set of leaders, and post season results showing another result; e.g. Balt vs. SF, with Baltimore winning.

And, one set of stats (e.g. regular season 'Game Scripts') is not sufficiently correlated to success in the regular season and post season. Thus, multi-variate analysis approaches are preferred, and multiple stats are analyzed in the articles listed by Krauser.
Back to top
View user's profile Send private message
Krauser


Joined: 20 Apr 2013
Posts: 1836
PostPosted: Mon May 13, 2013 1:02 am    Post subject: Reply with quote

The point I think is most interesting about Game Scripts is that the Vikings spent a lot of time leading, but most of the leads were narrow. For all his many faults, Ponder did a decent job not screwing up in games we were leading, with the single glaring exception of the loss at Lambeau. His other big mistakes happened in games we won anyway (Cards, Titans) or would've lost no matter how well he played (Seahawks).

The narrative about the season that I put together from DVOA is interesting too. I think the part of the year that made a lot of fans give up on Ponder was that stretch in the middle of the season starting with the Redskins game and extending to the loss in Chicago. I think most fans have underestimated how much worse the defense was in midseason than before Cook's injury -- they declined ever further than the offense, though of course the offense included the best rushing season in nearly 30 years.

...

A common and useful criticism of advanced stats is how poorly they do at predicting the future. Football Outsiders was started by Pats fans, and has always rated them highly (for good reason). It's been fun watching them try to hide their disappointment over the past few years as their top-rated teams have cratered in the playoffs against lower rated teams (the Giants in 2007 and 2011, the Ravens last year*). The best stats on a team's performance to date won't tell you how they'll play next week.

(*Some would include the Packers in 2010, as a wild card team that surprisingly won it all, but the FO DVOA stats actually had them at the top of the league even when their regular season record was only good enough for the #6 seed. So in that case, advanced stats were an accurate indicator of quality that became more obvious in the playoffs.)

...

Advanced stats came out of baseball (Bill James's sabermetrics, "Moneyball"). In baseball, there is much more data to analyze (pitch by pitch), with information much more specific to individual players (mostly pitcher vs hitter). They play 162 regular season games, the same players play most of the every game, there aren't many injuries. In football, outcomes depend almost entirely on team performance (making it hard to tell who to credit, the RB or the OL, the QB or the WR, etc), the plays are complex and often unpredictable, and injuries and other factors are so important that a team might hardly be the same team from a few weeks earlier (compare the Vikings in week 5 with week 10, minus Harvin and Cook).

So I don't think these stats are perfect, or even reliable. They're just interesting to consider and debate. I'll add to this thread from time to time when I come across something interesting.

Feel free to add to the collection in the meantime.
Back to top
View user's profile Send private message
CriminalMind


Joined: 25 Aug 2011
Posts: 4774
Location: Toronto, CA
PostPosted: Mon May 13, 2013 2:27 am    Post subject: Reply with quote

These stats have predictive power in a tube meaning if a team plays exactly like they have in the past in every given situation as before normilized for opponent, then the expected value trends to actualized results. The problem is teams rarely play exactly the same as is the past in the playoffs, when ppl are injured or players come back from injuries.

As u said, baseball would likely have the greatest predictive power cuz events are largely independent 1 vs 1. Thats how fantasy was started
_________________
Back to top
View user's profile Send private message
Purplexing


Joined: 13 Jan 2008
Posts: 4443
Location: Outside Valhalla, looking in.
PostPosted: Mon May 13, 2013 8:53 am    Post subject: Reply with quote

Krauser wrote:
The point I think is most interesting about Game Scripts is that the Vikings spent a lot of time leading, but most of the leads were narrow.


An alternative would be to weight the time leading with the point margin of the lead, or deficit, and recalculate the stat. It would increase the score of dominant teams that lead by large margins for long time periods. But I'm unsure if it would yield significant increases in predictive power.

Krauser wrote:

The narrative about the season that I put together from DVOA is interesting too. I think the part of the year that made a lot of fans give up on Ponder was that stretch in the middle of the season starting with the Redskins game and extending to the loss in Chicago. I think most fans have underestimated how much worse the defense was in midseason than before Cook's injury -- they declined ever further than the offense, though of course the offense included the best rushing season in nearly 30 years.


I've previously shown that Ponder's 5 worst games in 2012, per passer rating, were against the top 4 NFL pass defenses (Chicago twice), per passer rating against for the entire 2012 regular season.

Your point about losses also considers the defense's performance.

...
Krauser wrote:

A common and useful criticism of advanced stats is how poorly they do at predicting the future. ..... The best stats on a team's performance to date won't tell you how they'll play next week.


^ Theory of changing conditions, or imperfect correlation, in a nutshell.

Example: flip a coin twice. Because the probability of a heads or tails is 50%, the first flip should indicate the opposite for the second flip, not the same result. But over the course of hundreds of 2-flip trials, 50% is only observed in 50% of the 2-flip trials. In 25% of 2-flip trials, two heads will be observed, and in the other 25%, two tails will be observed.


Krauser wrote:

So I don't think these stats are perfect, or even reliable. They're just interesting to consider and debate.


Good summary point. But we should strive to find the best unbiased stats to discuss, rather than blindly discuss one stat that supports one theory, as did the author of the article you quoted on Ponder.

At the risk of beating a dead horseradish, I repeat my prior claim that the best stats to predict outcomes in football games are turnovers and field position. But stats collected and published on those two factors are incomplete because they do not have sufficient details and are only two factors in a myriad of factors that determine wins and losses. The key word is 'published'. I believe teams may collect stats internally that help them better determine the best coaching and game-plan strategies to win games.
Back to top
View user's profile Send private message
Krauser


Joined: 20 Apr 2013
Posts: 1836
PostPosted: Mon May 13, 2013 9:19 am    Post subject: Reply with quote

Quote:
An alternative would be to weight the time leading with the point margin of the lead, or deficit, and recalculate the stat. It would increase the score of dominant teams that lead by large margins for long time periods. But I'm unsure if it would yield significant increases in predictive power.


That's exactly what the original Game Scripts stat does. +2.9 for the Vikings means they led by an average of 2.9 points across the entire season (second by second, weighted by the value of the lead / deficit). By that measure, the Pats and Seahawks were clearly the 2 best teams in the league, the Vikings were 9th. Couple that with them leading for the 2nd most amount of time (ignoring the value of that lead), and you can see they tended to have narrow leads most of the time.

Quote:
At the risk of beating a dead horseradish, I repeat my prior claim that the best stats to predict outcomes in football games are turnovers and field position


IIRC the major factors of the multivariate analysis that Football Outsiders use are plays that lead to first downs (gaining at least 7 or 1st and 10, gaining at least 4 on 2nd and 5, converting 3rd and short and 3rd and long) plus avoiding turnovers. If you keep making first downs without turning over the ball, you're going to score and you're opponent will be hard pressed to keep up. Their models have rewarded the Brady/Manning/Rodgers style of highly efficient passing interspersed with highly efficient running plays (Pats RBs, or someone like LeSean McCoy, who tend to have fewer rushes but a good YPC average, and heavy production in the passing game) over vertical passing offenses (like Baltimore) that tend to be boom or bust.

They don't even particularly like long running plays, IIRC you don't get any more value for running 70 yards as 40 (sorry, AP) -- the idea is that your successful plays should be consistent and replicable. Their favorite RBs will gain 6, 5, 8, 4, 11, 6, and 7 yards on consecutive rushes (usually spaced apart by multiple passes). Peterson before last year tended to be more like 4, 9, 3, 0, 4, -1, 27. Even if the YPC is comparable, the higher number of "stuffs" (as they call rushes for no gain or negative yards) leads to stalled drives, which in turn limits the first downs you need to shift field position and score points. So Adrian's never been as highly rated by FO as you might think, and even in 2012 was often 2nd or 3rd RB in their weekly stats despite rushing for >200, surpassed by a some guy who went 13 carries for 94 yards, plus 6 catches for 55. Over the year, he did finish first among RBs (going from memory here), but it was hardly the blowout you'd expect.
Back to top
View user's profile Send private message
EvilNuff


Joined: 10 Feb 2007
Posts: 845
PostPosted: Wed May 15, 2013 12:25 am    Post subject: Reply with quote

Krauser wrote:
...the idea is that your successful plays should be consistent and replicable. Their favorite RBs will gain 6, 5, 8, 4, 11, 6, and 7 yards on consecutive rushes (usually spaced apart by multiple passes). ...


This is exactly why I disliked Barry Sanders as RB. Amazing fantastic runner, but he gave Detroit no consistency on offense and no ability to set up 2nd/3rd and shorts. This has also been a concern of mine about Peterson to be honest. I think if we had a top passing attack in this league AP would be less amazing than he looks given our relatively weak pass game.
Back to top
View user's profile Send private message
Display posts from previous:   

Post new topic   Reply to topic    FootballsFuture.com Forum Index -> Minnesota Vikings All times are GMT - 4 Hours
Page 1 of 1

 
Jump to:  
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum




Powered by phpBB © 2001, 2005 phpBB Group