Tuesday, November 17, 2009

Using Stats to Make a Roster Decision: An NFL Example


A group of local ultimate "think tank" members (as we're dubbed) have been geeking out over a very cool site called Pro Football Focus.

Looking at the best and worst players at each position this year, one of the thinkers asked where Cleveland Browns QB Brady Quinn would fall in the ratings of quarterbacks.

It's very true that Quarterbacks rely on many teammates for success. Receivers much catch, people must block, and people need to run the plays as directed in order to get success. However, QBs remain the most important people on the field and have greater control over the outcome of the game than any other player.

Sounds like a main offensive Handler's role in ultimate, no?

I went to the pro football focus site and looked for methodology... and didn't get much in order to answer the Quinn question. Looking for the Wages of Wins QB ratings, I ended up on a new site that builds on the wages of wins football measure (unlike the basketball wages of wins, the NFL model has been criticized).


I took their formula (the model looks legit) and came up with the following 2009 wins produced of Quinn, Derek Anderson and Peyton Manning

Advanced NFL Stats= Brady Quinn versus Derek Anderson

QB Wins Added
Brady Quinn -1.542092593
Derek Anderson -3.963324675
Peyton Manning +3.300733894

Formula: QB Wins Added = (Comp% * 0.18) - (Int/Att * 50.5) - (Sack Yds/Att * 1.57) - 8

-Brady Quinn will lose 1.5 games more than the average quarterback based on his current performance
-Derek Anderson will lose 3.96 games more than the average
-Brady should continue to start, as he is the best of two below average options
-Peyton Manning is over 3 wins better than the average quarterback.

How it Applies to Ultimate?

I wish it did apply more. Right now, the best we have for stats is a clipboard or iphone used during the game, trying to capture everything under time pressure. It's easy to miss stuff and accurately measure what's really happening.

If you are doing stats already, and have data on your teams, please send it my way!

I will re-ask people to consider the following
-video tape their teams games
-break their games down play by play and stat by stat
-Work together with other teams to build a data set of a particular division to determine the average performance of players in a particular division/level

If that was done, we could move forward with a formula to determine whether players were average or below average on defense and offense.

Coordinating teams might be a problem, and one team breaking down video for all other teams is too much of a workload. However, teams like DoG have been using data for years to make better decisions about who's contributing.

It's time for teams to build on this. Don't wait for other teams to do it first and gain the advantage.


march said...

There's a new iPhone/Touch app called UltiStats that should hopefully make stat taking during games much easier. Unfortunately it came out after the Canadian season was over but I'm looking forward to playing around with it. The developer (Chad Boulay) added a feature that allows you to email the stat info to yourself so hopefully this will be the start of good stats in ultimate. Something which I feel is missing from the sport.

moses said...

The UPA's score reporter site needs a spot to upload those UltiStats stat files so that they can be associated with each game and the database of stats can start getting built up. They also want to come up with a standard for what stats get recorded during a game. Ultimate Canada - maybe you guys take the lead on this?

One downside to ultistats as I see it right now: I'm pretty sure it only records stats for one team, so each team would have to be responsible for their own. But kudos to the developer for writing the program for such a popular platform...Hopefully he'll make some money for his efforts.

Maxmillien said...

Another suggestion would be to look at ESPN's PER stat. They attempt to quantify how a player does on a per minute basis so that stats are not inflated based on playing time. It can show how a player who plays less is more valuable than someone who may play more but cause more turnovers. It's used for basketball but can work for ultimate too.

Daniel said...

Before the ultimate community can really start to use stats to determine the value of players, we need to figure out which stats are valuable towards making that kind of determination.

For instance the most popular stat recorded seems to be turnovers. What does that tell you unless you differentiate between throwing turns and receiving turns or consider turnovers in conjunction with a completion percentage? Can a player with 5 turnovers be better in a game than a player with only 1? I think so but we don't use stats in a way that can back that up.

Here are a few that I think would be relevant:
For Offense
- drops
- throw-aways
- throw-aways as a result of drops by receiver
- touches
- completion percentage (all throws)
- completion percentage (discounting throws dropped by receivers)
- number of points played in a game
For Defense
- blocks
- times thrown-at
- number of points played in a game

Who has the time to keep track of all of that?

Druski said...

I downloaded Ultistats and am looking forward to a chance to use it in a tournament setting. With the sheer quantity of data it is able to record (essentially recording every 'touch', the outcome of every pass, and proper attribution of turnovers), I suspect the data can be manipulated to be very useful - including in the ways Daniel outlines above, or most of them anyway.

Not looking as forward to seeing what this will imply for my *own* stats... going to the NFL example I wonder how many games I cost my team per year :P

However, as the old maxim goes: that which can be measured, can be improved. So, there is hope for me yet...

Sport Management Steven said...

Thanks for the excellent responses.

March, I got to see that ulitstats app on the weekend. It looked good compared to the previous palm version and seemed easy to do, but had the very limitation Moses points out. (one team)

Moses- Definitely something for UC to look into leading. At the very minimum we should be standardizing the reporting of scores so those outside our tournaments can follow from home.

MAX- NBA PER is a good start to the idea or concept. Hollinger (Mr PER) has built a good model, but it has been criticized heavily for the weights he puts on scoring. Inefficient scorers still look pretty good.

Dan- No one has time to record the important stats on site (past posts have listed key stats). It's best to record your games (set the camera up one end zone and share camera work). Split the games up amongst your team and take two hours to break down game film at your leisure. As your tournaments and games pile up you will see some very interesting team and player trends. Giving various players the chance to decode game 'tape' might give them a different idea of how they play.

One key: standard rules breaking down plays. What is a drop, what constituents a throw away, each decoder has to be consistent.

Druski said...

BTW the idea of a Canadian score reporter (and whatever augmentations people suggest) is a fairly popular idea, judging by suggestions at the Ultimate Canada Conference this weekend. I won't say it's a bad idea, though people should recognise the amount of work that has already gone into creating/maintaining the system the UPA already has, and factor that into any stand-alone Canadian suggestions. If anyone has any workable solutions though, I think this would be great - if you have an idea (and ideally some infrastructure it could work on) I would suggest submitting it (including an action plan) to Ultimate Canada.

Final point: Ideas + action = progress... Ideas without action = rhetoric. Let's see some intiative from those that would improve things!