0 |
Subject: A new statistical look at TNSP production
Posted by: Mighty Geeks
- [49915418] Wed, Oct 24, 2001, 18:11
I have attempted to analyze the statistical performance of opposing defensive teams and apply that analysis to the average scores of several of the players I am looking at adding to my team. Like any mathematical approach to the game, the flaw is that nothing that ever happens is "average" on any given week. Nonetheless, I think it is useful to look at opponents when selecting players to add to your teams.
The actual method I used is fairly easy to understand, but difficult to explain, so please try to bear with me. First, I used a statistical source to capture the following stats for each defense, passing yards allowed, passing TDs, rushing yards allowed, and rushing TDs. I also computed NFL averages for these same stats. I could do the same for WR and TE yards and TDs, but I'm not yet sure that it is worth the effort. Then I took these average numbers and figured out how many TNSPs that would be worth per game. For instance, the average Defense gives up 232 passing yards and 1.4 passing TDs. This means an average net of 316 TNSPs. This formula does not take into account fumbles or interceptions. I think I like that, but I'm not sure yet. I then look at a particular Defense, such as Arizona and determine that they give up an average of 268 yards and 1.8 TDs. This results in an average TNSP of 376. I then convert this into a ratio of 1.19 and compute that any QB lucky enough to face the Arizona D will get 1.19 points for every point he usually gets. I then multiply the average TNSP/game for each QB by this factor and arrive at a predicted TNSP for each QB.
The next thing I do is perform the exact same analysis for the next three weeks to find players who have a nice cushy schedule coming up. In the QB realm, an example of someone looking good is McNabb, who faces Oakland, Arizona and Minnesota for the next 3 weeks, with multiples of 1.10, 1.19 and 1.05 respectively.
Here is a quick look at top 10 predicted QB scores for the next week using this method:
McNabb 392.4 Garcia 353.6 Warner 330.1 Culpepper 322.6 Manning 296.8 Brooks 277.5 Flutie 275.8 Batch 265.8 Johnson 257.2 Fiedler 255.5
As a number cruncher from way back, I can assure you that these results will not come to pass, but they are interesting to look at sometimes. The more interesting stat for me is the three week outlook and I will try to paste my table for that as a reply. It might come out in HTML or it might not. I'll see what I can do.
Let me know if you find any of this interesting. |
1 | Mighty Geeks
ID: 49915418 Wed, Oct 24, 2001, 18:18
|
Name | | TSNP | TSNP/Game | Week 7 | Week 8 | Week 9 | 3 Week Tot | McNabb, Donovan | Philadelphia | 1792 | 358.4 | 392.4 | 426.5 | 376.5 | 1195.4 | Garcia, Jeff | San Francisco | 1881 | 376.2 | 353.6 | 477.4 | 347.6 | 1178.6 | Manning, Peyton | Indianapolis | 1724 | 344.8 | 296.8 | 388.4 | 324.1 | 1009.3 | Brooks, Aaron | New Orleans | 1401 | 280.2 | 277.5 | 297.9 | 287.3 | 862.8 | Griese, Brian | Denver | 1655 | 275.8 | 245.3 | 302.0 | 261.0 | 808.2 | Gannon, Rich | Oakland | 1352 | 270.4 | 205.4 | 269.5 | 309.8 | 784.7 | Johnson, Brad | Tampa Bay | 1224 | 244.8 | 257.2 | 199.1 | 310.6 | 766.9 | Batch, Charlie | Detroit | 1053 | 263.3 | 265.8 | 270.0 | 229.1 | 764.9 | Fiedler, Jay | Miami | 1115 | 223 | 255.5 | 297.1 | 211.7 | 764.3 | Flutie, Doug | San Diego | 1469 | 244.8 | 275.8 | 210.7 | 244.0 | 730.5 | Plummer, Jake | Arizona | 1256 | 251.2 | 252.8 | 190.8 | 239.3 | 682.9 | Brady, Tom | New England | 1008 | 201.6 | 201.0 | 245.6 | 227.1 | 673.7 | Brunell, Mark | Jacksonville | 1019 | 203.8 | 174.8 | 248.3 | 205.7 | 628.8 | Green, Trent | Kansas City | 1242 | 207 | 196.5 | 195.9 | 220.1 | 612.5 | Testaverde, Vinny | New York Jets | 1147 | 191.2 | 254.7 | 176.7 | 164.6 | 596.0 | McNair, Steve | Tennessee | 945 | 236.3 | 190.7 | 192.9 | 202.6 | 586.3 | Collins, Kerry | New York Giants | 1118 | 186.3 | 175.7 | 187.5 | 221.7 | 584.8 | Weinke, Chris | Carolina | 1148 | 191.3 | 203.4 | 179.8 | 189.5 | 572.7 | Johnson, Rob | Buffalo | 944 | 188.8 | 178.6 | 179.2 | 167.9 | 525.8 | Grbac, Elvis | Baltimore | 1077 | 179.5 | 146.6 | 144.8 | 218.7 | 510.1 | Miller, Jim | Chicago | 755 | 188.8 | 193.6 | 147.0 | 153.5 | 494.1 | Stewart, Kordell | Pittsburgh | 770 | 154 | 187.6 | 132.1 | 119.9 | 439.6 |
|
2 | biliruben Sustainer
ID: 3502218 Wed, Oct 24, 2001, 19:18
|
Nice job, MG. Your method seems reasonable.
|
3 | 56ers
ID: 23820916 Wed, Oct 24, 2001, 19:33
|
This is very interesting. I like when statistical approaches are applied to seemingly random events. I feel the most value of this method will be when applied to the WR position since its the most random. A suggestion would be to divide the projected points by price so we can find bargains.
|
4 | saber34
ID: 53972114 Wed, Oct 24, 2001, 19:34
|
I've always wanted to do this kind of analysis for at least QB's and RB's but never got anything automated. Good show....!!
|
5 | Tu Papi
ID: 409241522 Wed, Oct 24, 2001, 21:45
|
I don't care if it comes to pass or not. You, sir, are a stud.
|
6 | POWWOW
ID: 495562121 Wed, Oct 24, 2001, 21:49
|
What about Kurt Warner? Mighty Geeks
|
7 | Roar Donor
ID: 27927417 Wed, Oct 24, 2001, 22:09
|
This is great. Culpepper is missing, too. I'm guessing the bye weeks have something to do with the missing QBs. Maybe a 3-week projected points/game would allow for comparing all QBs, including those with the byes.
|
8 | Raz
ID: 257382817 Wed, Oct 24, 2001, 22:25
|
First of all, great job MG.
Roar and POWWOW - I'm sure Culpepper and Warner are missing because of their bye weeks. But there would be no point in ignoring bye weeks, as the projected 3 week point totals should tell you who to buy for a decent "long term" hold. They shouldn't lead you into buying someone who won't play one of those weeks.
|
9 | LuckyBruin
ID: 4631651 Thu, Oct 25, 2001, 01:15
|
Well, according to this, it's a good thing I bought McNabb this week.
Luck to all....
|
10 | Guru
ID: 330592710 Thu, Oct 25, 2001, 08:52
|
Good effort A little friendly constructive criticism:
You are calculating a ratio for each defense vs. the league average. But you are multilying this ratio times the actual average of the QB. However, after only 6 weeks, it is likely that many QB's won't have faced average defenses.
As a relatively "simple" adjustment, you could adjust each QB's historical average to compensate for the strength of his actual opposition. For example, McNabb has faced StL, Sea, Dal, Ari, and NYG. What is the average of those five team defenses? If they are better than average, then his projected production could probably be bumped up a little. And vice versa.
It seems like you already have the data. You'd just have to figure out a way to link it up with historical schedule info.
On the other hand, perhaps this getting too refined for a tool which would still be admittedly crude.
Good job. Thanks for sharing.
|
11 | dgrooves
ID: 557332117 Thu, Oct 25, 2001, 13:37
|
Something interesting to note:
I have seen the data you are using elsewhere. I thought the yardage numbers we pretty high, especially for the RBs, so I did a little investigating. In compiling those stats, any yards generated by any QB who played that defense are included. For example, Carolina played Atlanta Week 2. The stats above include the passing and rushing yards of Chandler as well as the passing and rushing yards of Vick.
As for the runningbacks, the numbers would need to be scaled down. Those statistics say that Washington is giving up 202 yards per game to the runningback. To expect that the runningback who plays Washington will put up 202 total yards is unrealistic. These 202 yards per game are the average rushing and recieving yards per game of any RB that lines up against the Redskins.
None of this has any great effect on your numbers because they should be used as a guide, not as exact predictions. However, with this information, some of the TSNP total would be higher because rushing yards for QBs are included with their passing yards (rushing yards are worth twice as much). The TSNPs for the RB (if you post them) should be lower because of the reasons detailed above.
All in all, this is a great analysis. The true value in this analysis, though, lies in the relative rankings of one defense against another.
|
12 | Mighty Geeks
ID: 49915418 Thu, Oct 25, 2001, 13:38
|
PowWow & Roar: As others surmised, I dropped people who had a bye in the next three weeks. The key folks that I dropped were:
Warner 330.1 bye 475.9 Culpepper 322.6 bye 281.5 Farve bye 290.0 313.2 Chandler bye 218.9 247.8
(I assume no one was seriously considering Couch, Kitna or Banks).
Guru:
You have made an interesting point. I considered working that into my calculations and just got distracted by another project before I could do that. I think that this will not be much of a factor once the season is further along, but I will experiment with that next time the network drops and I don't have anything else to do!
56-ers: I have always been dubious about the value of this analysis with regard to WRs because of the big difference between cornerbacks. Some defenses do a great job on the #1 receiver and a bad job on the #2 receiver. I run it for WR as well, but I'm not sure I think much of the output. It does seem to do very well with TEs, because some teams have slow linebackers or run-first linebackers and give up a lot of yards and TDs to TEs. I've used it in regular FF for TEs for a couple of years.
Just for the record the predicted top 5 WRs for the next 3 weeks are: Troy Brown Terrell Owens Rod Smith Marvin Harrison Keyshawn Johnson
The top 5 RBs for the next 3 weeks are: Edge R Williams P Holmes C Martin Tomlinson
If there is interest, I'd be happy to paste the tables up here.
|
13 | dgrooves
ID: 557332117 Thu, Oct 25, 2001, 13:41
|
Correction:
"These 202 yards per game are the average rushing and recieving yards per game of any RB that lines up against the Redskins."
should say
These 202 yards per game are the average rushing and recieving yards per game of all RBs that line up against the Redskins.
|
14 | Mighty Geeks
ID: 49915418 Thu, Oct 25, 2001, 13:50
|
dgrooves: You are correct in that the source data is a bit murky, especially as regards running QBs. But I think that the way I use the numbers eliminates the problem you see at RB. In your Washington example, where they give up an average of 202 rushing yards per game, that gets compared to the league average of 136 yards and I just compute that playing Washington is worth a 20% increase in a RBs usual production. The fact that the baseline number is an individual's point production keeps the numbers honest.
However, I completely agree that this is only a very ROUGH guide. It can't take into account key injuries or momentum changes or a thousand myriad variables that make fantasy football interesting. Its just one more thing to think about. It also exposes that a lot of things that everyone knows aren't always true. For instance, the vaunted Tampa Bay D gives up more rushing yards than the average D and the completely awful Dallas D gives up fewer.
|
15 | dgrooves
ID: 557332117 Thu, Oct 25, 2001, 13:59
|
Mighty Geeks, I completely agree (thats why I said the "true value in this analysis, though, lies in the relative rankings of one defense against another"). I understand that you werent using their raw numbers, I just wanted to say that those numbers were "murky" (nice word choice).
|
16 | Roar Donor
ID: 412141319 Thu, Oct 25, 2001, 14:10
|
Raz, Having numbers for Warner, Culpepper, and others with upcoming byes are nice for people that already own those players and are wondering if they should keep them or trade them.
|
17 | Mighty Geeks
ID: 49915418 Thu, Oct 25, 2001, 15:02
|
Guru: Your suggestion proved to be pretty interesting. I've not double checked my code (this started as an exercise to get better at Perl), but it appears that it doesn't change the relative position of the QBs much, with one notable exception. Aaron Brooks was projected to be the #4 QB and your suggestion drops him to #9 over the next three weeks. Over his first five games, he faced two teams that are terrible against the pass (Carolina & Atlanta), two that are fairly bad (Minnesota & Buffalo) and one that is average (NYG). The net result is that his adjusted TNSP/game goes from 280.2 to 251.7 -- quite an adjustment. The biggest gainer was Kerry Collins, who went from 17th to 13th -- not enough to really matter. I'm convinced I should add that to the whole process now.
|
18 | Mighty Geeks
ID: 49915418 Thu, Oct 25, 2001, 15:55
|
Name | Team | TSNP | TSNP/Game | Adj TSNP | Week 7 | Week 8 | Week 9 | 3 Week Tot | Williams, Ricky | New Orleans | 1553 | 310.6 | 319.4 | 317.4 | 478.1 | 307.2 | 1102.7 | Holmes, Priest | Kansas City | 1648 | 274.7 | 282.5 | 370.7 | 187.1 | 422.8 | 980.6 | James, Edgerrin | Indianapolis | 1622 | 324.4 | 288.0 | 328.4 | 330.2 | 313.7 | 972.3 | Martin, Curtis | New York Jets | 1774 | 295.7 | 267.1 | 313.0 | 236.4 | 304.5 | 853.9 | Alexander, Shaun | Seattle | 1132 | 226.4 | 233.8 | 254.7 | 354.5 | 195.1 | 804.3 | Barber, Tiki | New York Giants | 858 | 214.5 | 201.1 | 304.8 | 199.8 | 261.2 | 765.8 | Tomlinson, LaDainian | San Diego | 1573 | 262.2 | 235.0 | 269.4 | 267.9 | 206.6 | 743.9 | Smith, Lamar | Miami | 1227 | 245.4 | 224.0 | 174.1 | 262.5 | 293.9 | 730.5 | Bettis, Jerome | Pittsburgh | 1304 | 260.8 | 241.1 | 267.2 | 145.9 | 258.0 | 671.1 | Stewart, James | Detroit | 1028 | 205.6 | 208.0 | 227.9 | 200.1 | 222.6 | 650.6 | Pittman, Michael | Arizona | 899 | 224.8 | 239.8 | 238.3 | 236.8 | 145.1 | 620.2 | Dunn, Warrick | Tampa Bay | 781 | 195.3 | 210.7 | 210.7 | 159.7 | 225.5 | 596.0 | Dayne, Ron | New York Giants | 942 | 157 | 147.2 | 223.1 | 146.2 | 191.2 | 560.5 | Mitchell, Brian | Philadelphia | 813 | 162.6 | 172.1 | 143.6 | 223.6 | 172.1 | 539.3 | Henry, Travis | Buffalo | 817 | 163.4 | 152.4 | 101.0 | 200.0 | 173.8 | 474.7 | Centers, Larry | Buffalo | 787 | 157.4 | 146.8 | 97.2 | 192.6 | 167.4 | 457.2 | Allen, Terry | Baltimore | 879 | 146.5 | 156.5 | 162.4 | 109.6 | 173.4 | 445.5 | Smith, Emmitt | Dallas | 788 | 157.6 | 155.8 | 202.4 | 94.3 | 141.9 | 438.6 | George, Eddie | Tennessee | 900 | 180 | 183.9 | 128.9 | 190.9 | 111.3 | 431.1 | Smith, Antowain | New England | 990 | 165 | 142.1 | 124.9 | 129.4 | 162.9 | 417.2 | Anderson, Mike | Denver | 801 | 133.5 | 157.4 | 179.4 | 131.3 | 104.3 | 415.0 | Hearst, Garrison | San Francisco | 825 | 165 | 152.7 | 107.9 | 163.4 | 135.2 | 406.5 | Garner, Charlie | Oakland | 762 | 152.4 | 144.9 | 143.0 | 127.3 | 112.6 | 382.9 |
|
19 | Mighty Geeks
ID: 49915418 Thu, Oct 25, 2001, 16:00
|
The Guru's suggestions really shook up the RB situation. According to his schedule, that James guy is overrated. I'm going to run out and sell him ASAP.
|
20 | kid westside
ID: 58082021 Thu, Oct 25, 2001, 18:56
|
Great info! is there a way u could do a WR chart, and also add the Rbs/Wrs that have byes.
|
21 | CanEHdian Pride
ID: 426351415 Thu, Oct 25, 2001, 19:02
|
That is great info! I was mulling over whether to drop CuMar for Lamar Smith this week or after this week and this table confirms that after this week the gap between the 2 players should be almost converged as Smith next 3 weeks would include the Jets as the third game and Martin heads into Miami. That should tip the scales into Martin's favor for the 3 weeks stretch after this week. It's nice to see something physical that jives with your mental process.
Great work Mighty Geeks!
|
22 | dgrooves
ID: 557332117 Thu, Oct 25, 2001, 19:12
|
CP, how exactly are NO, KC, and Mia better matchups than Car, Ind, and NYJ?
|
23 | CanEHdian Pride
ID: 426351415 Thu, Oct 25, 2001, 19:37
|
* "tip the scales into Smith's favor"
Their rankings in the chart above should converge once (in Smith's situation) Seattle is exchanged for the Jets and (in CuMar's situation) Carolina is exchanged with Miami.
Just showing that this charts can give you an idea of how far apart 2 players should be over an stretch allowing you to extrapolate by switching the matchups from week to week.
My bad on the typo...nice catch.
|
24 | bookie
ID: 279182210 Fri, Oct 26, 2001, 11:32
|
Mr Geek, You are the man.... thanks for the great analysis tool... I hope you plan on continuing to share this information, it will prove most useful going forward...
|
25 | Khahan
ID: 12432113 Fri, Oct 26, 2001, 11:59
|
Man, I wished my brain worked like this. Thanks MG. I'm sure this took a good bit of time, thanks for the effort and the thought to share it with us.
|
26 | Mighty Geeks
ID: 49915418 Fri, Oct 26, 2001, 15:39
|
Thanks for all the nice words. It took quite a bit of time to automate the process of collecting the data, but updating it week to week isn't that bad, so I should be able to post it going forward when the rest of my life doesn't get in the way. I want to emphasize that my real world experience with this kind of data is that it only provides general information -- the real world never fits so nicely into sophisticated statistical models and is likely to diverge vastly from one as simplistic as this.
I'll post the WR chart as another reply. Next week, I'll try to figure out how to get all the style stuff out of the tables (that is why the page takes *SO* long to load). I cheated and pasted the table into Word and it adds a lot of stupid crap to each cell!
You may also notice that the Guru's suggestion changed the WR chart quite a bit as well.
|
27 | Mighty Geeks
ID: 42771721 Fri, Oct 26, 2001, 16:49
|
I lied about pasting the WR chart. My PC locked up and then my wife talked me into leaving work early, so I never did paste it up. I will post a new message next week once I get the results from this week's games (Tues - Weds).
|
| Rate this thread: | If you wish, you may rate this thread on scale of 1-5. Ratings should indicate how valuable or interesting you believe this thread would be to other users of this forum. A '5' means that this thread is a 'must read'. A '1' means that this is a complete waste of time. If you have previously rated this thread, rating it again will delete your previous rating. If you do not want to rate this thread, but want to see how others have rated it, then click the button without entering a rating, or else click here. |
|
|
Post a reply to this message: (But first, how about checking out this sponsor?)
|
|