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!
These are just rounding issues. I've got the QBs projected for 325.9 completions and 24.2 passing TDs, and the receivers collectively with 325.9 receptions and 24.2 receiving TDs.Dirty Weasel said:
You're correct here. I'll get back to you on this one, but my run-pass ratio is too low. Thanks for pointing it out.
Part of this is because my run-pass ratio is too low. But part is also because Tomlinson has averaged 15.9 games played per season, which is higher than his expectation for this year. If we expect him to play in 14.5 games (the average for a starting RB is about 13), and multiply his career-average 403 touches by 14.5/15.9 we get 368 touches.
This will be changed when I increase the run-pass ratio. But the Chargers do have better WRs this year than they've had in a very long time; and Gates may not be as big a factor as he has been in recent years. So I do expect the WRs to do better as a group than they have in previous seasons.Thanks for your comments.
His SOS was easy in 2006 and hard in 2007. It looks easy again in 2008. I've got his YPA at 7.1 right now.I do have the Chargers' run-pass ratio too low right now, as Dirty Weasel pointed out.Ministry of Pain said:
Chambers has never competed for Targerts with Gates, Tomlinson, and V.Jax before. (Until last year, that is, when he averaged 3.5 catches per game as a Charger. More in the playoffs, but that was when Gates was out.) He was always the clear #1 receiving option in Miami. It's not so clear in San Diego, where the balls will be distributed around to more receivers.
Good catch. This is a mistake in my Excel spreadsheet. Will be corrected.Musesboy said:
I've got each Redskin RB averaging at least 4 yards per carry, so you may be looking at Dodds' projections?Michael J Fox said:
Again, you may be looking at the wrong projections (I've got Duckett projected for 85 carries, the second-most of any staffer), but I do think both Julius Jones and Maurice Morris as better overall runners than Duckett, with Duckett being mostly the short yardage and goal line back. (I do have Duckett with more rushing TDs than any other Seahawk.)
This is helpful. Thanks.Andy Herron said:
weird. i pulled it from the same link. thx for the correction, your projections look reasonable.
Rivers had a hard schedule in 2006. It was about half a FP/G tougher than average.Rivers had an almost identically difficult schedule in 2007, which was about half a FP/G tougher than average. Rivers also averaged slightly fewer FP/G in '07 than '06.
even with D Rhodes coming back to take some of the carries? I think they want to get Addai a littel more rest so he doesn't get hurt again. not much, but I doubt he will do more than last year. sorry for jacking the thread.
This was actually a global error on my part.When I start doing projections, the first thing I do is project total plays and run/pass ratios for each team. My first cut at doing so uses a formula derived using regression analysis based on the previous year's stats.The formula for total plays is = 575 + 11 * [last year's yards per rush] + 0.0946 * [last year's passing yards] - 46.2 * [yards per pass attempt] + 707 * [last year's rush attempts divided by last year's total plays]The formula for run percentage is = 0.133 + 0.000159 * [last year's pass attempts] - 0.0706 * [last year's win percentage] + 0.611 * [last year's rush attempts divided by last year's total plays]That's just the starting point, which is generally a more accurate starting point than just using last year's total plays and run-pass ratios.Then I make adjustments based on coaching changes and personnel changes, etc.But my starting points were off because I was using 2006 win percentages instead of 2007. Forgot to update them.The changes aren't major. But for the Chargers, it increases my initial projection from a run percentage of 46% to 47%. I've further moved it up to 48% based on the fact that their SOS looks relatively easy to me this year.(The Chargers had a run percentage of 50.7% last year, compared to a league-wide average of 45.1%. They should be expected to regress toward the mean a bit. I think 48% seems about right.)So my updated projections for the Chargers team-wide passing / receiving totals are:312 completions out of 504 pass attempts, 3551 passing yards, 23 passing TDs, 14 interceptions.The higher run-pass ratio also bumps Tomlinson up to:304 carries, 1401 yards, 15 TDs / 54 receptions, 402 yards, 2 TDs.I've also updated the run-pass ratios of all other teams based on 2007 win percentages.
I may bump this up a bit. It's worth looking at.Some relevant info:The Colts scored 19 rushing touchdowns last year, well above the league-wide average of 12. (They were tied with San Diego for second most rushing TDs in the league, behind only Minnesota.) I expect them to be above average again, but with some regression to the mean. Maybe in the 15-16 range, with the RBs getting 13-14 of those.I've currently got their RBs collectively scoring 13 rushing touchdowns this year. In the past three years, Colt RBs have scored 16, 13, and 18 rushing TDs. So I'm on the low end there.Right now I've got Addai scoring 9 of the Colt RBs' 13 rushing touchdowns. I could see giving him 10 of 14 if I bump the RB total to 14. Or giving him 11 of 14 if he steals one from Rhodes (since I think Addai is a better goal line back than Rhodes).Addai also had 3 receiving TDs last year, but I don't see that happening again. His 3 TDs on 41 receptions means he scored on 7.3% of his catches -- more than double the average of about 2.8% for RBs. My projection for him is that he'll score on 3.6% of his receptions -- which translates into 1.48 touchdowns on 41 catches, but that gets rounded down to 1 touchdown reception as displayed on the site.Thanks for your comments.
Gary Barnidge TE 6'6" 243 Louisville Pick Analysis:Barnidge had excellent college production - 108 receptions, 17 TDs and almost 14 yards per catch. He gets in and out of breaks well and has the ability to change direction. He caught passes from Brian Brohm at Louisville and it's a little surprising he fell this far.Based on your and Andy's comments, I've bumped Felix Jones to 26 receptions for 204 yards and a TD.I've also bumped the RB rush attempts up to 397 (and QB pass attempts down to 531). Both small changes, but that gives them a run percentage of 45% compared to last year's 44%.
I'm planning to write an article for FBG on this.Very briefly . . .gheemony said:
That would be a great way to do projections, and I'm aiming for the same results as that method, but I don't have any simulations set up. When distributing rushes and targets to players, I do take into account that there's less than a 100% chance that the starter will be healthy all season.
OK, maybe not.
For his career, Boldin has scored 29 TDs on 413 receptions, or 7.0%. The NFL average for WRs is about 8.1%. With five season' worth of production to go on, I don't think it's a fluke that he's truly below average in the TD department. (I don't really mean that Boldin himself is below average around the end zone; I just mean that the team tends to go to either Fitzgerald or the TEs rather than to Boldin when they are in the red zone. Leonard Pope and Ben Patrick combined for 7 TDs last year, and Jerame Tuman typically scores on a high percentage of his catches as well.) Moreover, I see Leinart getting more playing time this year, and I don't think he's as good as Warner, especially in the TD department. So I've reduced the TD numbers for all ARI receivers this season compared to what they'd be if I thought Warner would get the majority of the snaps again.Verbal Kint said:
Really? You might be right, but I think Dunn really started to show his age last year, and at this point is not as good a runner as Graham. Bennett is harder to get a read on. He played pretty well last year, but in the games in which both Graham and Bennett got carries (i.e., when they were both healthy), Graham's carries outnumbered Bennett's 120 to 25.
Interesting. I think Stevens is a better receiver than Troupe, but I'm not sure how much I trust him to stay out of trouble. I'll add some production for Stevens, though.Thanks for your comments.
Wow. ETA:I'm planning to write an article for FBG on this.Very briefly . . .gheemony said:
The first thing I do for each team is project total number of offensive plays (not including sacks), and run-pass ratio. My first cut at this uses the previous year's stats fed into formulas derived using regression analysis over the last 10 seasons' worth of data for each team. (Actually, I don't remember if I used 10 years or 15 or some other number . . . I did the regression analysis a few years ago.)
Then I tweak those numbers based on any coaching or personnel changes, or sometimes just gut feel.
Once I've got projected total plays and run-pass ratios for each team, that gives me the total number of pass attempts (and targets) and rush attempts for each team.
The next thing I do is divide the pass attempts, targets, and rush attempts among various positions. For now, I always give QBs 100% of the pass attempts, and 0% of the targets, and I give TEs 0% of the rush attempts. Rushes are distributed to QBs, RBs, FBs, and WRs, and targets are distributed to RBs, FBs, WRs, and TEs. I start with the percentages for each team from the previous year, and then again adjust them based on any coaching or personnel changes.
Then, within each position, I distribute the pass attempts, targets, and rush attempts to individual players.
My default for QBs when the starter is set is to give 15/16 of the QB pass attempts and QB rush attempts to the starter, and 1/16 to the backup. Where the starter's job security isn't great (or where there hasn't been a starter named yet), those numbers will change. (They'll also change where one QB is more of a runner than another. Like I have Vince Young with 15/16 of the pass attempts, but 15.5/16 of the rush attempts, while Kerry Collins has 1/16 and 0.5/16, respectively.)
This is the part -- distributing passes, targets, and rushes to individual players within their positions -- where I don't rely much on data from previous years. In some cases, where little has changed among WRs, for example, I'll keep the ratios of targets pretty close to what they were last year. But in most cases, things change from year to year as younger players start to take on more of a role in the offense, or guys return from injury, or whatever. So this part is largely subjective, based on media reports of how coaches plan to use their players this year and my overall feel for each team situation based on everything I've seen.
So to take Tomlinson as an example -- I've got the Chargers down for 969 total offensive plays (not including sacks); I've got 48% of those plays being runs; I've got 87.2% of those runs going to RBs; and I've got 75% of the RB runs going to Tomlinson. So Tomlinson gets 969 * .48 * .872 * .75 = 304 rushing attempts.
By the same token, 52% of the Chargers' plays will be passes, and I've got 23.1% of the targets going to RBs, and 65% of those going to Tomlinson. So Tomlinson gets 969 * .52 * .231 * .75 = 76 targets.
Once I've got rush attempts and targets for each players, I multiply them by my individual player projections for yards per rush, touchdowns per rush, receptions per target, yards per reception, touchdowns per reception, etc.
When I do individual projections for those things (as well as completion percentage, yards per attempt, touchdowns per attempt, and interceptions per attempt), I rely heavily on Bayesian inference analysis. This is the part that will take up the bulk of my article. To use yards per carry as an example, I've divided the universe of running backs into groups based on long-term YPC. For example, 0.4% of NFL RBs average more than 5.05 yards per carry over 350+ carries, 0.8% average between 4.95 and 5.05, 1.6% average between 4.85 and 4.95, 2.4% average between 4.75 and 4.85 . . . yadda yadda . . . 8.5% average between 4.35 and 4.45, 10.5% average between 4.25 and 4.35, 13% average between 4.15 and 4.25, 11.3% average between 4.05 and 4.15, 9.7% average between 3.95 and 4.05 . . . yadda yadda . . . 0.8% average between 3.25 and 3.35, and 0.4% average fewer than 3.25 yards per carry.
So I take that distribution as the prior probability for a given running back before he has his first carry.
Once he has, say, 39 carries for 130 yards (3.33 YPC), I can recalculate the percentage chance that he's in each of those groups (between 4.75 and 4.85, between 4.65 and 4.75, between 4.55 and 4.45, and so on). His observed YPC is 3.33. So the chance that he's truly a 4.45-4.55 YPC RB can be calculated based on how many standard deviations 3.33 is from 4.5 over his 39 carries. When we add up the probabilities that he's in each individual group, we can get his expected true YPC. (As it turns out, 39 carries is too small a sample size to be all that significant. After just 39 carries at a 3.33 average, his expected true YPC is at about 4.03 -- just a bit below the 4.16 NFL average for RBs. But once he has 390 caries at a 3.33 average, his expected true YPC drops to 3.61.)
I go through a similar analysis for each player for completions per pass attempt, yards per pass attempt, touchdowns per pass attempt, interceptions per pass attempt, touchdowns per rush attempt, receptions per target, yards per reception, and touchdowns per reception.
As input, I will often use a player's career stats. But in the case of older players, I will tend to use just the last few years' worth of stats (to avoid using his younger, faster days). For players who've switched teams, I will sometimes not use stats from old teams if I don't think they are representative of his current situation. (I've excluded Randy Moss's Raider years, for example.) And so on. So there's still a subjective component to it. But after playing around with a number of different ways of doing individual projections for stuff like yards per carry, I really like the Bayesian method I've got set up. I've found it to give very realistic projections, and (unlike the regression analysis I'd played around with before) I love the way it takes into account sample size. Peyton Manning, for example, has had sky-high stats that would get regressed back to the mean too much using any regression analysis that tries to apply to both experienced and inexperienced QBs. But since he's been so good for so long, Bayesian inference analysis doesn't regress him back that much -- he really is that good, as he's demonstrated consistently.
So anyway . . . after I have all the individual player projections done, I add them all up to compare total receiving stats to total passing stats. Where there's a divergence, I have them meet in the middle. So if a team's QBs are projected to complete 61% of their passes, while the receivers collectively are projected to catch 59% of their targets, I'll split the difference to make completions equal receptions. Then do the same with yards and touchdowns.
Then I add up all stats for all teams and make sure they fit in with league-wide historical norms, and make any adjustments I need to there.
Then I post a thread in the shark pool asking for corrections.
That would be a great way to do projections, and I'm aiming for the same results as that method, but I don't have any simulations set up. When distributing rushes and targets to players, I do take into account that there's less than a 100% chance that the starter will be healthy all season.
Great post MT. It's probably not all that surprising that I use a very similar approach to projections. I'll post on that subject later when I have more time to give it a worthwhile post and read for all of you.
Thanks for the great post and detail.Have you thought about running your regression analysis before each season, instead of relying on the equations you derived a few years ago? Don't know if it would change your projections of offensive plays very much, but with all of the changes that occur in coaches, coordinators, players, etc., over time, it might help to update the regression analysis on a regular basis. Also, if you use fixed time period for the analysis (i.e., most recent 10 years), then older observations that might reflect very different circumstances for the team would be replaced by more recent observations. I'm sure your tweaking is meant to account for some of this change, but updating the regression equations might be helpful.
I'm planning to write an article for FBG on this.Very briefly . . .gheemony said:
The first thing I do for each team is project total number of offensive plays (not including sacks), and run-pass ratio. My first cut at this uses the previous year's stats fed into formulas derived using regression analysis over the last 10 seasons' worth of data for each team. (Actually, I don't remember if I used 10 years or 15 or some other number . . . I did the regression analysis a few years ago.)
Then I tweak those numbers based on any coaching or personnel changes, or sometimes just gut feel.
Once I've got projected total plays and run-pass ratios for each team, that gives me the total number of pass attempts (and targets) and rush attempts for each team.
The next thing I do is divide the pass attempts, targets, and rush attempts among various positions. For now, I always give QBs 100% of the pass attempts, and 0% of the targets, and I give TEs 0% of the rush attempts. Rushes are distributed to QBs, RBs, FBs, and WRs, and targets are distributed to RBs, FBs, WRs, and TEs. I start with the percentages for each team from the previous year, and then again adjust them based on any coaching or personnel changes.
Then, within each position, I distribute the pass attempts, targets, and rush attempts to individual players.
My default for QBs when the starter is set is to give 15/16 of the QB pass attempts and QB rush attempts to the starter, and 1/16 to the backup. Where the starter's job security isn't great (or where there hasn't been a starter named yet), those numbers will change. (They'll also change where one QB is more of a runner than another. Like I have Vince Young with 15/16 of the pass attempts, but 15.5/16 of the rush attempts, while Kerry Collins has 1/16 and 0.5/16, respectively.)
This is the part -- distributing passes, targets, and rushes to individual players within their positions -- where I don't rely much on data from previous years. In some cases, where little has changed among WRs, for example, I'll keep the ratios of targets pretty close to what they were last year. But in most cases, things change from year to year as younger players start to take on more of a role in the offense, or guys return from injury, or whatever. So this part is largely subjective, based on media reports regarding how coaches plan to use their players this year and my overall feel for each team situation based on everything I've seen.
So to take Tomlinson as an example -- I've got the Chargers down for 969 total offensive plays (not including sacks); I've got 48% of those plays being runs; I've got 87.2% of those runs going to RBs (note: fullbacks are separate from RBs); and I've got 75% of the RB runs going to Tomlinson. So Tomlinson gets 969 * .48 * .872 * .75 = 304 rushing attempts.
By the same token, 52% of the Chargers' plays will be passes, and I've got 23.1% of the targets going to RBs, and 65% of those going to Tomlinson. So Tomlinson gets 969 * .52 * .231 * .75 = 76 targets.
Once I've got rush attempts and targets for each player, I multiply them by my individual player projections for yards per rush, touchdowns per rush, receptions per target, yards per reception, touchdowns per reception, etc.
When I do individual projections for those things (as well as completion percentage, yards per attempt, touchdowns per attempt, and interceptions per attempt), I rely heavily on Bayesian inference analysis. This is the part that will take up the bulk of my article. To use yards per carry as an example, I've divided the universe of running backs into groups based on long-term YPC. For example, 0.4% of NFL RBs average more than 5.05 yards per carry over 350+ carries, 0.8% average between 4.95 and 5.05, 1.6% average between 4.85 and 4.95, 2.4% average between 4.75 and 4.85 . . . yadda yadda . . . 8.5% average between 4.35 and 4.45, 10.5% average between 4.25 and 4.35, 13% average between 4.15 and 4.25, 11.3% average between 4.05 and 4.15, 9.7% average between 3.95 and 4.05 . . . yadda yadda . . . 0.8% average between 3.25 and 3.35, and 0.4% average fewer than 3.25 yards per carry.
So I take that distribution as the prior probabilities for a given running back before he has his first carry. (In the future, I may use different priors based on draft position.)
Once he has, say, 39 carries for 130 yards (3.33 YPC), I can recalculate the percentage chance that he's in each of those groups (between 4.75 and 4.85, between 4.65 and 4.75, between 4.55 and 4.45, and so on). His observed YPC is 3.33. So the chance that he's truly a 4.45-4.55 YPC RB can be calculated based on how many standard deviations 3.33 is from 4.5 over his 39 carries. When we add up the probabilities that he's in each individual group, we can get his expected true YPC. (As it turns out, 39 carries is too small a sample size to be all that significant. After just 39 carries at a 3.33 average, his expected true YPC is at about 4.03 -- just a bit below the 4.16 NFL average for RBs. But once he has 390 caries at a 3.33 average, his expected true YPC drops to 3.61.)
I go through a similar analysis for each player for completions per pass attempt, yards per pass attempt, touchdowns per pass attempt, interceptions per pass attempt, touchdowns per rush attempt, receptions per target, yards per reception, and touchdowns per reception.
As input, I will often use a player's career stats. But in the case of older players, I will tend to use just the last few years' worth of stats (to avoid using his younger, faster days). For players who've switched teams, I will sometimes not use stats from old teams if I don't think they are representative of his current situation. (I've excluded Randy Moss's Raider years, for example.) And so on. So there's still a subjective component to it. But after playing around with a number of different ways of doing individual projections for stuff like yards per carry, I really like the Bayesian method I've got set up. I've found it to give very realistic projections, and (unlike the regression analysis I'd played around with before) I love the way it takes into account sample size. Peyton Manning, for example, has had sky-high stats that would get regressed back to the mean too much using any regression analysis that tries to apply to both experienced and inexperienced QBs. But since he's been so good for so long, Bayesian inference analysis doesn't regress him back that much -- he really is that good, as he's demonstrated consistently.
So anyway . . . after I have all the individual player projections done, I add them all up to compare total receiving stats to total passing stats. Where there's a divergence, I have them meet in the middle. So if a team's QBs are projected to complete 61% of their passes, while the receivers collectively are projected to catch 59% of their targets, I'll split the difference to make completions equal receptions. Then do the same with yards and touchdowns.
Then I add up all stats for all teams and make sure they fit in with league-wide historical norms, and make any adjustments I need to there.
Then I post a thread in the shark pool asking for corrections.
That would be a great way to do projections, and I'm aiming for the same results as that method, but I don't have any simulations set up. When distributing rushes and targets to players, I do take into account that there's less than a 100% chance that the starter will be healthy all season.
Fabulous post. Thank you. This definitely increases my confidence level in your projections and I'll read them very closely. I only wish I knew as much about statistics as you did so I could try this at home.
Thanks for taking the trouble to outline your methods, that was a great read
I have one question that may be related:Don't underestimate the complexity of Grudens O. It's Troupes 1st year in the system. Also, I read something about at OTAs that Troupe did something wrong, and Gruden screamed something indicating he wanted Stevens back.Apparently, Bennet is dumb as a rock so now that he has had a little while to learn the system it should help. (he was acquired pretty late last year) Gruden also loves speed.The longer EG stays away from the team, the higher you bump the other RBs.Also, Dunn "showing his age" may have had something to do with the Falcons completely sucking. It isn't unheard of for a player to hang it up if they don't think the team is going to be competitive. (see Moss in Oakland) Not saying Dunn is that kind of guy, but Atl was different last year.Maurile Tremblay said:
Honestly, during the season I don't spend a lot of effort on my own fantasy teams. I play in about 10 leagues. I love doing the drafts. But once the season starts, I'm doing the Roundtable feature here as well as the upgrade/downgrade feature and some other stuff . . . and pretty much all my FF energies are spent on FBG rather than my own leagues.So when someone offers me a trade I don't spend a lot of time evaluating it. I just accept it if it obviously makes my team better and reject it otherwise.Musesboy said:
I didn't think it was a bad question.
Very good question. (I'd been away for a few days. Sorry.)The answer, I'm afraid, is a poor one. I failed to account for how much the KC offensive line and overall offense had regressed over the past few years. I had LJ pegged, therefore (based on his career effectiveness), at too high a YPC and touchdowns-per-rush. And when I gave him 72%-75% of the KC RB carries (which is probably about right for a projection), LJ kept ranking up there with Tomlinson and ADP. The correct solution would have been to re-think what kind of YPC and touchdowns-per-rush figures were realistic for Johnson in the recently suckified KC offense, but my harebrained solution was: "That can't be right. I better give him fewer carries so his totals will be more realistic."I've re-worked all the KC projections now (to be updated on the site shortly). Thanks for calling me out on this.AnonymousBob said:I didn't think it was a bad question.
That's what we're here for.Very good question. (I'd been away for a few days. Sorry.)The answer, I'm afraid, is a poor one. I failed to account for how much the KC offensive line and overall offense had regressed over the past few years. I had LJ pegged, therefore (based on his career effectiveness), at too high a YPC and touchdowns-per-rush. And when I gave him 72%-75% of the KC RB carries (which is probably about right for a projection), LJ kept ranking up there with Tomlinson and ADP. The correct solution would have been to re-think what kind of YPC and touchdowns-per-rush figures were realistic for Johnson in the recently suckified KC offense, but my harebrained solution was: "That can't be right. I better give him fewer carries so his totals will be more realistic."I've re-worked all the KC projections now (to be updated on the site shortly). Thanks for calling me out on this.I didn't think it was a bad question.
