This is a sample guest message. Register a free account today to become a member! Once signed in, you'll be able to participate on this site by adding your own topics and posts, as well as connect with other members through your own private inbox!
The Giants allowed 94% as many FPs to opposing RBs as the league average defense did. Tomlinson scored 45.3 FPs against them. The Eagles allowed 104% as many FPs to opposing RBs as the league average did. Tomlinson scored 3.3 FPs against them. That's random.I'm not sure we're using the same definition of random, though. If Tomlinson's median expected yardage total against the Saints is 100 and against the Steelers its 60, that doesn't eliminate any piece of randomness.
Thanks Chase... exactly the explanation I was looking for... I was assuming that it would be the case (player X usually exceeeding his average performance against a particular team) - but if there hasn't been any proof of it... then I rest my case...
There was a good discussion on this about Jamal Lewis against Cleveland, and how he bombed on opening day 2004, afte rushing for 500 yards against them in 2003.I'm out.
In regards to Chad, hasn't that S/CB...Baxter owned him since he was in Baltimore?
Well like I said above, I want Dodds to say which one and you guys like this approach. I'm OK with him being wrong/human if it happens to be incorrect
I guess I just prefer the "out on a limb" type predictions.I don't believe you answerred my Q about what San Fran has done to stop this trend of allowing 100 yard games to these WRs. Not to belittle but I think it's a valid point amidst this discussion.
Huh? Who said anything about 3+ TD games? I would not expect anybody to be able to predict that. Nor would I expect Dodds to try to predict when Tomlinson rushes for 200+ yards. Those are statistical outlyers. Tomlinson rushing for 100 yards or Peyton Manning passing for 300 yards are not.BTW, I agree with your previous statement that receiving yards are less predictable. I don't have a problem with him not projecting any receivers for over 100 yards.
If you & I were sitting in a bar, I'd feel pretty confident picking one WR over another to get 100 yds (or a TD, or herpes, or whatever). However, if I were putting out projections for customers every week, I think I'd do it the way DD does. I used to feel differently.
Here we goI respectfully disagree with you and Chase on this one... and here is why:
Last year - there was, on average:
3.8 QBs per week with 300yds game
8.1 RBs per week with 100yds game
9.8 WBs per week with 100yds game
I understand completely that predicting who these 21 players will be is extremely difficult (if not impossible to do)... but to say that we should base our projection on "if that game was played 50 times - Larry Fitzgerald would have a gaussian distribution around 85 yards" - then I predict 85 yards for him is totally wrong because this game will only be played once! (there is no Normal distribution here... it's a "one time event")
No sane person would ever predict a 10/204/2 game to a WR, and I understand that - but these will happen, just ask Plaxico Burress last year...
We know that, on average - every week 10 WRs will get at least a 100yds game... we know that these outlier games will happen... we know that we don't know who will produce them...
But to say, since I don't know who will get them... I don't predict any is totally wrong in my opinion...
To draw an analogy... we know that there are 21 gifts under the Christmas tree... we know that some kids were nicer this year than others (i.e. Boldin getting 8 - 100yds game in '05 and Fitz getting 7)... we know that Santa Claus will give these gifts on sunday... we just don't know who was nice enough to get one - so we say that none will get any?...
My 2 cents...
You're up to 4 cents now....
Like I mentioned in another post here:I understand this thinking (i.e. [how FBG is doing projections]) - and I can see that this is the only way to have "safe" projections.By "safe" projection I mean - FBG is saying to their subscribers "If you start Anquan Boldin - you should expect him to have between 63 and 97 yards depending on the matchup".---Of course, I'm not dishing the way FBG is doing the projections - far from it... the "other way" you mention in your reply would have a far larger error factor than the way Dodds do it...I was just surprised that the "safe" projections wouldn't account for more players having a 300/100yds game - since we know that 21 will usually reach that plateau...---This is like offering someone a slim chance for a million bucks, once a year... or offering them a chance to have 50K often in the upcoming weeks... the sane choice (way to do predictions) is trivial...
Fair enough - point understood... and I agree with that...So, who's getting herpes on week1?
More importantly, a lot more RBs than WRs average near (say, within 20 yards of) 100 yards/game.
What can I say... I'm financially inclined... got change for me?
Let me give you an analogy. In real life I'm an actuary. One job of an actuary is to project what the losses will be when we write a set of insurance policies. Now, suppose we write 1,000 policies and we determine that, on average, 10% of the drivers (or 100 of them) will get into an accident next year, and we determine that we need to charge $8,000 per accident, or a total of $800,000. Thus, on average for all the policies, we need to charge $800 each.Our "projection" for your number of accidents is 0.1. What we do not do is try to guess which 10% of the drivers will have accidents, charge them each $8,000, and give everyone else their policy for free. If we can predict that some drivers have a higher chance of an accident (for example, people with bad prior driving records,) and others have a lower chance of an accident (for example, people with good prior driving records,) we should reflect that, but we're not trying to guess.You might criticize this approach by saying "you know 100 people will get into an accident, but you're not charging anyone for a full accident." The trick is that while we know approximately 100 people will get into an accident, for any particular driver there's no way to know what will happen. We have to work off an average.That's just what FBG is doing with its projections.
I strongly disagree with this.Weekly projections should only concern that week. You don't name your starters for the season but for the week. You win or lose for their points for that week. Week, week, week, see what's the point in rolling with season projections and dividing them? Since when does a player get 76.8 yards each week? You know he never does but instead it's 102 and 51 and 76 and such.See, I understand the use of those projections to form your own rankings and draft day thoughts but we're done with that. We're in season(except for some last minute folks of course)You bring up the NFL average 209 yards per game passing average. What does that matter? You won't be happy if your QB only gets those average numbers unless they're loaded with TDs. Further, I can't recall the Cards ever coming anywhere near that low passing output last year, heck or any Denny Green team for that matter. I just don't see how this stat applies at all.Back to per game average for WR. You know Fitz is gonna have some 100 yard games and it's probably safe to say it's against lesser Ds. Well then why not predict it?Trick Q-You pickup a player off the WW. Umm Travis Wilson. Braylon has gone down and is out for the year and Trais is starting. You projected Wilson to have 3-400 yards on the year. Well it's week 2. You're not going to roll with those 3-400 and divide them by 14 games are you? Well then why do it above?We're in week to week mode and I disagree
I'm not suggesting there is no randomness invloved just that I think it is less than you think. I guess where we disagree is how much of the variance from week to week for the top players at each position is based on randomness and how much is based on predictable factors. I would love to see a study done on this.
This is all irrelevant. If I roll enough six-sided dice, I'd expect at least one to land on six. But my projection for each individual die will still be 3.5.
It doesn't matter that it's a one-time event. Using probabilities is still the correct approach. Ask any competent gambler.The projections for each player are essentially OVER/UNDERs. Each week, out of the 16 game played (in non-bye weeks), what's the average total score in the highest-scoring game? What's the average highest OVER/UNDER set by Vegas each week? If the answer to the second question is substantially lower than the answer to the first question (and it is), does that mean Vegas is setting its OVER/UNDERs incorrectly? (It's not.)
Ron Mexico? Mark Chmura?Fair enough - point understood... and I agree with that...So, who's getting herpes on week1?
This is also a good analogy.
2=alot?
is this a joke?Did you miss the season stats and averages and yearly projections and SOS ratio which involves the year too?
apparently I editted/added while you replied
Believe me - I understand this point - I worked at an actuary firm for 4 years (calculating pension plans solvability ratios most of the time)... prior to now developing mathematical models insuring us that we pick the most valuable stocks/bonds "most of the time"...Enough of my job - isn't saturday?
... from your analogy - I don't expect FBG to try and match their prediction to the total number of receiving yards on week1 in the NFL - but to try and figure out (if possible - and I don't mean it is) which WR will outperform their averages and post a 100yd game... the exact contrary of the insurance policies example to make sure the insurance company will make money (even though they have no idea who will have the accidents)I want FBG to make educated guess in saying who has a better chance of having an accident this weekend because he's riding a motorcycle without a helmet and talking to a cell phone (i.e. playing against the weakest pass defense)...
And 7 WRs
The difference is that Scarpati has a lot less information on each individual driver than Dodds does on each individual starter in the NFL. If Scarpati were to analyze each individual drivers history week by week for the past year and where and how much they are going to be driving in the coming week, Then adjust the amount each driver is charged on a week by week basis, I think the amount each driver is charged over the course of a year would more accurately reflect his risk.Edited to add:Certainly this approach would not be cost effective for Scarpati's company because they only care about the overall revenue. It doesn't matter to their bottom line if they are off on each individual driver, its the averages that matter.Thats not the case for FF though. We need to make starting linup decisions on a week by week basis. Weekly variance has a much bigger impact on our results.
We don't project the likelihood that a particular player will have a 100 yard game, although it wouldn't be a bad thing to project since some leagues award bonuses for reaching 100 yards.I don't know what the exact formula would be, but it would probably take projected yardage as the input. So a projection of 100 yards might imply a 50.3% chance of rushing for 100+ yards, a projection of 90 yards might imply a 42% chance, a projection of 80 yards might imply a 36% chance, and so on. (I'm making these numbers up.)
I don't think it is irrelevant... All I'm saying is that I would think the projections for Larry Fitzgerald against the 49ers pass defense would be more than his usual 70-80 yards per game (the 3.5 expected result)... since this week he is rolling a die that has 4 faces with 6 - one with a 4 - one with a 1...I know the projections are done this way - to a certain degree... don't think I'm saying the OVER/UNDER is not a good way to set "safe" projections (i.e. those that are the closest to the actual results most of the time)...
Sure, but that is irrelevant to the principle. The principle is that optimal projections will not look like actual results if there is any element of uncertainty at all. So in Scarpati's case, he will not pick out a majority of the drivers and project them to have zero accidencts, even though at the end of the year a majority of drivers will indeed have had zero accidents. It's the same in fantasy football. There is a difference only in degree.
This is understood Maurile - and I would never say it is not the right way to do it (the OVER/UNDER analogy is the best way to say this)...The fact of the matter is that FBG week1 projections are there to say to your opponent "if we played that FF weekend a 1000 times - I would have beaten you on average by 8.3 points"
This is exactly what we do.
The strength of the opposing defense definitely factors into the weekly projections, and since there's a correlation between past performance by an offensive player against an opponent and past defensive strength of that opponent, and there's also a correlation between past defensive strength and current defensive strength for a given team, and there's also a correlation between current defensive strength and current projection for an offensive player facing that defense . . . you can expect there to be a correlation between past performance by an offensive player against a particular defense and the current projection for the player against that same defense. But it's probably not as strong as a lot of people might guess.
It matters because if his company is better at doing projections than competing insurers, Scarpati's company will be able to charge lower premiums for lower-risk drivers, and thus get more business (while pricing out the higher-risk drivers that its competitors do not realize are so risky).Government regulation may screw that up, but where rates are freely set by the market it would be the case.
Fair enoughI think, what we are missing is the distribution of yardages that gives the results (the underlying percentages)...For example, suppose that DeShaun Foster and Chris Brown are both projected to gain 65 yards this week (pure speculative example)...What would be interesting is finding out that the 65 yards expected value for Foster comes from (5% chance of having 200yds - 5% for 150 - 15% for 120... down to 5% for 10 yards)... while Brown's 65 yards value comes from (50% 70 yards - 50% 60 yards) - extreme example obviously!This would help guys swinging at the fences with the last roster spot... or playing it safe - depending on your matchup...Obvisouly these percentages would come from the SOS, expected number of touches and other variables that are needed to make projections...
Excuse my 0 post count, but I have to throw up a case against this. While 300 yard games and 100 yard games are slighlty uncommon, to call them "rare" is a disservice to the elite QBs and WRs and RBs that do it week in and week out. Last year, Anquan Boldin and Larry Fitzgerald combined to have 15 games of 100+ receiving yards in 29 total games played by them. Better than 50%. Fitzgerald had 15 catches for 231 yards and 1 TD in 2 games vs. San Francisco while Boldin had 19 catches for 272 yards and 2 TDs in those same two games. And I'll be damned, in Kurt Warner's 1 start vs. San Francisco last year, he threw for 354 yards. I think it's safe to say at this point that Arizona has had success throwing the football up and down the field against the 49ers, and they simply didn't do enough in the offseason to make me think otherwise. I personally think the Cardinals offense is going to run a clinic on the 49ers threw the air and with the help of Edge. But I think the running game is going to need to take a few weeks to gel together, so they will rely on Kurt a bit more in the first few weeks of the season.
Yes, that would be cool (not only for the weekly projections, but for the yearly projections as well). This is exactly why FF owners who are well informed about individual players and teams and have a good draft strategy will still outperform (on average) FF owners who print out a cheatsheet at the last second to take to their draft -- even if the cheatsheet is excellent. A cheatsheet or a simple set of projections can never include all the information you just mentioned, so a cheatsheet made by an expert is never a perfect substitute for being generally well informed yourself.
My point is that, to them, annual revenue is what matters. Week to week variance for individual drivers is irrelevant. In FF, the opposite is true. If FBG's goal is to be able to add up all the weekly projections and and see how close they come to the actual yearly totals then the approach they are taking is correct. In reality though, yearly totals are irrelevant (once the draft is over). Accurately predicting week to week variance has a much greater impact on our end results so it is worth it to make the effort to try to predict that variance as accurately as possible.
I don't see a difference there at all. We're past the draft stage, so from here on out, a fantasy owner must do his projections for a certain week. Similarly, an insurance company will do its projections for a given year. A week is shorter than a year, but that's not really a difference. An insurance company cares about its results over 15 years just like a fantasy owner cares about his results over 15 weeks -- but in both cases it's the smaller units of time that matter when doing the individual projections.Week-to-week variance among drivers is irrelevant, but so is quarter-to-quarter variance among players. But that's okay because nobody does weekly projections for drivers or quarterly projections for players. Year-to-year variance among drivers, however, is exactly analogous to week-to-week variance among football players. In both cases, that's what matters quite a bit, and that's what people are trying -- imperfectly -- to project.
I don't concern myself with yearly stuff at all when I do the weekly projections. I attempt to predict how the game will be played. Rushing attempts, ypc, etc. I am WAY more concerned with predicting SOLID game data.From that data, I fit the players to that game data projection. I believe all other methods will introduce HUGE bias. I also believe once you predict the outcomes of all the games that the TOTAL data should match historic norms. Not how many 300 yard games, but total passing yards for the week. Total passing TDs for the week, etc. The data needs to reflect what we know about the league. That's how you end up with a stable platform of which to build individual numbers from.Let's say I run my simulation and Team X predicts to have 108 rushing yards and 0.9 TDs. After you give some to the QB who scrambles for about 7 yards a game, the role playing change of pace back who gets 13 yards, there are only 88 yards left for the #1 RB. But of course he has a chance to get 100 yards. He is going to touch the ball 22 times in this example. If he averages 5 ypc instead 4 ypc he ends up 110 yards. But I don't want to just make up that 5 ypc. I wanted it grounded in the actual matchup of the simulation. Hence in this example he had 4 ypc and a 88 yard prediction.As a whole, are projections appear much more conservatively. But what you are really seeing is a better model that more accurately predicts all players with NO BIAS. I let the math tell me how many runners project to have 100 yards, etc. I don't get caught up in trying to splash whole numbers around like 320 yards and 3 TDs. I know those models are severely flawed because the numbers as a whole don't match up to league norms. By definition they have to be flawed. If you project the league as a whole to throw for 15% more yards then you are starting with a 15% error and can only be worse from there.
First of all, thanks for explaining the process - this is greatly appreciated...We can clearly see that a "top-down" approach is used:1. Average number of rushing yards / receiving yards / TDs on a normal NFL weekend;2. Incorporation of all the variables (injuries / SOS / roles / weather / etc.);3. Simulated games to average the results;Which is contrary to alot of other sites way of doing it - the "bottom-up" way... i.e. Looking at players one at a time (usually spending too much time on stars) and best-guessing the outcome - which, summing it up, will project an unusual weekend in the NFL...A good example of this was that I saw at a site once, where no IDP guys had a projected sack (probably since the expectations were under 0.5 and they didn't use decimals)... no sack on a given weekend in the NFL? The OL coaches would certainly throw a party...The Vegas "OVER/UNDER" analogy drawn my Maurile is matching perfectly the way you are doing your projections - and we all know "on average" (i.e. most of the time) their odds are the best you can find...From this, I reiterate that these are "safe" projections (surely the best way to go - I would do the same if I had to do so!)... i.e. that Monte-Carlo simulation (i.e. running the system often with the variables implemented and averaging the results) will say that... "if Anquan Boldin played against the 49ers defense on the first weekend of '06 for 10 or 100 or 1000 times - you would expect him to get, on average, 83 yards" (speculative example).What I would be interested in (and I mentioned it to Maurile in a prior post) is the distribution of these results (I know it's not possible to offer this to subscribers - I'm just saying that it's what's interesting)... Anquand Boldin is not projected to gain 83 yards against the 49ers because in the 10 simulations he had: 80 / 81 / 78 / 86 / etc. yards... but sometimes 163 / 43 / 83 / 81 / 144 / 63 / etc. ... and this is what the educated FFer is looking for (in my opinion - even though we can draw most of it ourselves)... I'm I going with Mason (safe bet) as my WR3 since his 10 simulation games range was in the 70-80 yards or I'm I going with Jennings (swinging for the fences) because his simulated range was from 10-150 yards?... Even though they have the same projected yards this week...---This is a cool debate - thanks for the input everyone...
First of all, thanks for explaining the process - this is greatly appreciated...We can clearly see that a "top-down" approach is used:1. Average number of rushing yards / receiving yards / TDs on a normal NFL weekend;2. Incorporation of all the variables (injuries / SOS / roles / weather / etc.);3. Simulated games to average the results;Which is contrary to alot of other sites way of doing it - the "bottom-up" way... i.e. Looking at players one at a time (usually spending too much time on stars) and best-guessing the outcome - which, summing it up, will project an unusual weekend in the NFL...A good example of this was that I saw at a site once, where no IDP guys had a projected sack (probably since the expectations were under 0.5 and they didn't use decimals)... no sack on a given weekend in the NFL? The OL coaches would certainly throw a party...The Vegas "OVER/UNDER" analogy drawn my Maurile is matching perfectly the way you are doing your projections - and we all know "on average" (i.e. most of the time) their odds are the best you can find...From this, I reiterate that these are "safe" projections (surely the best way to go - I would do the same if I had to do so!)... i.e. that Monte-Carlo simulation (i.e. running the system often with the variables implemented and averaging the results) will say that... "if Anquan Boldin played against the 49ers defense on the first weekend of '06 for 10 or 100 or 1000 times - you would expect him to get, on average, 83 yards" (speculative example).What I would be interested in (and I mentioned it to Maurile in a prior post) is the distribution of these results (I know it's not possible to offer this to subscribers - I'm just saying that it's what's interesting)... Anquand Boldin is not projected to gain 83 yards against the 49ers because in the 10 simulations he had: 80 / 81 / 78 / 86 / etc. yards... but sometimes 163 / 43 / 83 / 81 / 144 / 63 / etc. ... and this is what the educated FFer is looking for (in my opinion - even though we can draw most of it ourselves)... I'm I going with Mason (safe bet) as my WR3 since his 10 simulation games range was in the 70-80 yards or I'm I going with Jennings (swinging for the fences) because his simulated range was from 10-150 yards?... Even though they have the same projected yards this week...---This is a cool debate - thanks for the input everyone...
It is a very interesting discussion. With next week being a longer week, I will see if I can publish a from the gut piece that looks at certain situations for explosive break-throughs (ie teams that will be looking to attack a soft corner, teams that have mashed this run defense lately, etc). I will still predict the player at 94 yards and 0.7 TDs, but it will be a situation where people can see he has the potential for a HUGE game.
First of all, thanks for explaining the process - this is greatly appreciated...
We can clearly see that a "top-down" approach is used:
1. Average number of rushing yards / receiving yards / TDs on a normal NFL weekend;
2. Incorporation of all the variables (injuries / SOS / roles / weather / etc.);
3. Simulated games to average the results;
Which is contrary to alot of other sites way of doing it - the "bottom-up" way... i.e. Looking at players one at a time (usually spending too much time on stars) and best-guessing the outcome - which, summing it up, will project an unusual weekend in the NFL...
A good example of this was that I saw at a site once, where no IDP guys had a projected sack (probably since the expectations were under 0.5 and they didn't use decimals)... no sack on a given weekend in the NFL? The OL coaches would certainly throw a party...
The Vegas "OVER/UNDER" analogy drawn my Maurile is matching perfectly the way you are doing your projections - and we all know "on average" (i.e. most of the time) their odds are the best you can find...
From this, I reiterate that these are "safe" projections (surely the best way to go - I would do the same if I had to do so!)... i.e. that Monte-Carlo simulation (i.e. running the system often with the variables implemented and averaging the results) will say that... "if Anquan Boldin played against the 49ers defense on the first weekend of '06 for 10 or 100 or 1000 times - you would expect him to get, on average, 83 yards" (speculative example).
What I would be interested in (and I mentioned it to Maurile in a prior post) is the distribution of these results (I know it's not possible to offer this to subscribers - I'm just saying that it's what's interesting)... Anquand Boldin is not projected to gain 83 yards against the 49ers because in the 10 simulations he had: 80 / 81 / 78 / 86 / etc. yards... but sometimes 163 / 43 / 83 / 81 / 144 / 63 / etc. ... and this is what the educated FFer is looking for (in my opinion - even though we can draw most of it ourselves)... I'm I going with Mason (safe bet) as my WR3 since his 10 simulation games range was in the 70-80 yards or I'm I going with Jennings (swinging for the fences) because his simulated range was from 10-150 yards?... Even though they have the same projected yards this week...
---
This is a cool debate - thanks for the input everyone...
Exactly what the FBG subscriber is looking for... This will not affect the overall projections (nor should it)... but just added comments saying "Watson is projected to 63 yards - but in two of the 10 simulations, since the [insert second week opponent team name] don't have a LB who can run with him or a Safety that can tackle him, he had 112 and 131 yards"...
I don't think anyone took it that way... in fact, it started a (too unfamiliar) civilized debate... along with Dodds's explanation of his though process - very much appreciated
I don't think there's as big of a difference between high yardage games and multiple TD games as you might think.Which occurred more frequently in 2005:LT exceeding 140 yards rushingLT scoring 3 or more TDsThe answer: 3 TDs (3 to 2)Just saying, all of it is variable.
