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!
Tell me how you would test that Chris Johnson's decline last year was due to regression to the mean as opposed to holding out in training camp and poor offensive line play.We both agree there was a decline. That's not the point. The point is, "What caused it?"
OK, so after week 1 in 2011, Brady was on pace for 8,272 yards and 64 TDs. Obviously that's not a sustainable pace; so do you bench him in week 2?If you did, you missed out on 423/3. After week 2 his pace was down to 7,520 yards and 56 TDs. Is it now time to bench him?If you did, you missed out on 387/4. After week 3 his pace was down to 7,077 yards and 59 TDs. Is it now time to bench him?And so on. In fact, Brady's worst game all season turned out to be 198/2 against Pittsburgh, and while you probably wouldn't have benched him for that, it's predictable that he might not do as well against one of the league's best defenses, regardless of what his previous 6 weeks had been.Brady was just as "due for a clunker" after week 1 as he was after week 3; being "due" doesn't have any predictive value.
Regarding the bold, seriously???? Yeah, your only choices are to either a) delusionally believe Brady will get 8200 yards and 64 TDs, b) bench Brady. There is no in-between. No scenario where you expect Brady to be a top-3 QB the rest of the way....only either Brady gets 8200 yards, or you bench him. Perhaps I misread the bold.
the reason doesnt matter.in projections, once you start veering into the far end of the bell curve and getting x standard deviations away from a player's "baseline", it is exceedingly likely he will not hit those numbers again.i dont need to placate you with some made up reason that at the end of the day nobody really knows based on new coaching, o line or whatever. i just know that a reasonable expectation is to drop production off a record year to some degree.
No, I actually had Brady and never benched him at all last year. It's not a question of benching a stud because you think he's due for a clunker. It's a question of knowing that one is coming and trying to figure out when it's going to happen. Again, it's not "this guy is due for a clunker because he's had too many good games in a row". It's "this guy's on pace for 8,000 yards which is next to impossible so let's try to figure out what factors along the way will bring this guy back down to earth."
Or don't even bother figuring out the reasons. Just know that reality will take hold, and favorable variance cannot last forever. The reasons, in some cases, are irrelevant.
You can only create a true baseline if you can control for all variables. That's absolutely impossible. There's no point in pretending that it is since personnel, schedule, scheme and dozens of other factors change every year. And with no baseline, regression to the mean has little meaning.So yeah, you'd better be able to analyze the variables and explain why you think a player's output will drop/rise in order to figure out if last year was based on luck and thus likely unrepeatable or if the variables caused a change in a player's baseline.
The reasons are all that matter. If a decline happened 100 percent of the time after a great season then I might agree with you. But it doesn't. We need to be able to identify the reasons for the potential decline so we know if it is going to happen or not.
Are you suggesting that Brady was more likely to have a clunker against Pittsburgh because he had a good start to the season? That if he'd had three games to start the season where he went for 200 yards with 2 INTs, he'd be more likely to have a good game against Pittsburgh?
No, he was more likely to have a clunker against Pittsburgh because Pittsburgh was a good team. But after seeing his early pace of 8,000 yards, I knew there'd have to be few down weeks for him for various reasons. It's simple math if you expected his pace to be around 4500-5200 yards.
Regression to the mean is a fancy way for someone to say:I know DeMarco Murray averaged 5.49 ypc last year, but take away that outlier game vs Stl in which he had a 91 yard run, and he only averaged 4.65 ypc. Since you probably can't count on him breaking a 91 yard run every year, in 2012 he's more likely to be closer to 4.65 ypc than 5.49 ypc.
His having a few down weeks has nothing to do with his early pace. He was going to have a few down weeks because that's generally what happens. If someone averages 250 yds/game he doesn't get 250 every game, some games he gets 300 some he gets 200.
This is exactly what I'm saying.
No, it's exactly the opposite of what you're saying. What you're saying is that it's "simple math" that if you expected his pace to be around 4500-5200 yards, he would have down weeks if he'd started above that pace. The math is just as "simple" that if he'd started below that pace, he would have "up" weeks later--that he would be more likely to have a good game against Pittsburgh if he'd started off poorly. You may not understand that that's what you're saying, but it is.Reasonable predictions for Brady's performance against Pittsburgh can only be adjusted upwards based on good early-season play. Maybe you don't adjust them, but you certainly don't adjust them downwards.This is exactly what I'm saying.
I dont understand what you're arguing with me about here. If you have a guy that everybody reasonably expects to get between 4500-5200 yards and he starts on a pace for 8,000, it's going to be pretty freakin' likely that he has some down weeks to get back to a realistic end of year total. How else is his pace going to reduce? That's all I'm saying. It has nothing to do with Pittsburgh lol, except for the fact that since Pittsburgh has a great defense I'd put more money on them bringing him back down to earth than anyone else. But they are two separate things. I'd put just as much money on Pitt giving Brady a hard time regardless of how he had started the season.No, it's exactly the opposite of what you're saying. What you're saying is that it's "simple math" that if you expected his pace to be around 4500-5200 yards, he would have down weeks if he'd started above that pace. The math is just as "simple" that if he'd started below that pace, he would have "up" weeks later--that he would be more likely to have a good game against Pittsburgh if he'd started off poorly. You may not understand that that's what you're saying, but it is.Reasonable predictions for Brady's performance against Pittsburgh can only be adjusted upwards based on good early-season play. Maybe you don't adjust them, but you certainly don't adjust them downwards.This is exactly what I'm saying.
You're arguing with a straw man; no one thought Brady was going to throw for 8000 yards.Let me put it this way. Let's say you had done game-by-game projections for Brady in 2011, and had predicted that in week 4 against OAK he'd throw for 250 yards and 1 TD. After seeing him put up enormous numbers in weeks 1-3, would you adjust your week 4 projections up or down? What about if he stunk up the joint in weeks 1-3?
Regression to the mean isn't a cause. It's an effect. Sometimes it's an effect of losing an offensive tackle.
Up.His enormous numbers in weeks 1-3 are evidence that he's better than I'd previously thought (or is in a better situation or whatever). And if he's better than I'd previously thought, I should adjust my future projections up.But — and here's where regression to previous expectations come in — my new week four projections for him will still be lower than his average from weeks one to three.
No we couldn't. We can't even determine whether adding a skilled WR2 PREDICTABLY results in a loss of fantasy points to the WR1 because WR2 cannibalized touches or results in an increase because WR2 draws more coverage. Sometimes it hurts, sometimes it helps. We don't even agree on when the WR2 added to the formula is skilled or not! Look no further than the debate on Titus Young.We can't come up with a conclusive answer on whether the addition of a proven short-yardage RB is a net gain or loss for the lead back. Could be a gain if the SYB converts some short downs to keep some drives alive that otherwise would have stalled. Could be loss if that SYB vultures some goal-line carries that would have been TD's for the lead back. Which will it be?
Bingo. Which is ironic considering how Kutta was accusing people of not understanding what regression really means.
He's likely to have some down weeks because everybody is likely to have some down weeks.But if he starts out on a pace for 8,000 yards, that actually makes him less likely to have some down weeks than if he'd started out on a pace for 2,000 yards. (Holding constant what a "down week" means.)
There's really no difference in 1 and 2. They essentially say the same thing. Luck is nothing more than an inability to calculate an outcome of those numerous variables you state in #1. And just because you got the right answer doesn't mean you actually know how to do the math, though I suspect many on these boards feel it does. And given our penchant for remembering our hits better than we remember our misses, I'd say most of us think we have a good hold on how to apply these variables.But think critically and honestly about it for a minute. How can you say a player's weight training or focus on route running made THE difference with any confidence? How many WR's work out at Fitz's camp every year? Don't they all work on their route running? So do they all improve statistically the next year? If not, why not? When does it result in a better year and when does it not? To put it in terms of your post, how do you know whether a WR benefited from poor DB play or caused poor DB play by improving his own game? Did the CB slip because he used poor technique, because the field was a mess, or because the WR sold him an amazing double move? Do you really break down enough film on enough fantasy relevant players to know enough to affect your rankings and projections?You implicitly recognize this problem in your next paragraph when you talk about how changing teams and how surrounding players can affect a player's stats. The same problem inherent in the application of statistical principles like regression to an individual case is present when trying to apply things like "worked on route running" and "added 15 pounds of muscle". Have you tracked in any disciplined manner how often a player did those things and did better, the same or worse the next year? And be honest, you're getting this key "information" from reporters and bloggers who are getting it from people who may have agendas or biases in telling it. We see this every year. Player X added weight this off-season. But player Y added weight last year but felt it slowed him down so he's cutting weight this year. I've been on these boards for years and it's the same thing every year. OMG, I'm bumping player A because he's added some muscle this year to get stronger. I'm also bumping player B because he cut some weight to get faster. Like we know enough about player A's or B's individual physiology to know whether either strategy will be a good one. But I can guarantee come next year we'll do the same because we'll remember that time we heard about WR Jones going to Fitz's camp and how he had that good year. But we aren't so quick to remember that time that player WR Smith went to Fitz's camp and didn't do any better the next year. That one just kinda fades into obscurity and if we do remember it, we shrug it off as, get this, just bad luck.If we were really disciplined and tracked our hits and misses and what our thinking was on these players, I'm guessing we'd find that we miss as often as we hit when we focus in on things like "worked on his route running this off-season" or "added 10 pounds of muscle to be able to fight off the jam at the line".
I'd be curious to see the the formula you use to predictably determine all these means players will regress to.btw, you people don't have to quote the entire ####### thread when you reply to each post.
I don't use a formula. In fact, I do less in the way of projections or number crunching than I ever have. I've been playing since 1996. I think tiers is the best strategy because variance wipes out differences in ranking rather quickly. I just get amused when someone thinks they have a system figured out. As I stated previously, I suspect most of them reinforce their belief by up-playing their hits and downplaying their misses. I know I am guilty of that and am constantly on guard to keep it from biasing me. What I have seen is that player health probably affects things in as cleanly and drastically a manner as anything else, but unfortunately for us the coaches and players guard and distort that information so even it is suspect unless it's an IR type injury.
Well that is why you do the team projections which is the real frame work. Of course what really happens is not going to be the same as you project. But if your using past performance for the team and there have not been major coaching or personnel changes, you should be able to predict pretty well the play distribution based on past performance. If there have been major changes, for example in Cleveland right now, I was suggesting that perhaps looking at Childress offense with the Vikings being a good guideline for their play distribution now with Richards. I expect their offense to be very different than past seasons.I remember talking about the passing offense getting a big boost back when the rule changes were announced, I forget what year that was but I think it was the same season Ricky Williams retired(before he came back). Anyhow we speculated what the impact of the rule might be and there was talk about the TE position possibly getting more targets and passing offense increasing across the board. So I considered a 5-10% boost in targets/receiving yards for each team during the seasons following that for that reason. I think the main effect of this may be realized now so I would not suggest giving a 5% boost to every teams passing yards in 2012 but I would give 2011 passing stats greater weight than 2010 or 2009. We might see total yards fall back some seasons but I see the overall trend still being more likely to increase than decrease moving forward.
Most people in this thread have defined RTTM to mean "the affect of random luck" and how it affects a player's stats in the following year. So that is what I mean by my statement above.I suppose losing an offensive tackle during the year could fall under that definition, but if you read the entire thread you will see that I mentioned Brees losing Carl Nicks in the off season as a reason why Brees will not repeat his numbers from last year. I do not think that qualifies as "random luck."Would you say that if Brady lost Welker, Gronk, and Hernandez in the off season and his stats declined that following year that Brady was regressing to the mean?
See my response to MT.And please show me where I accused people of not understanding what regression means.
Let me ask you something.Brady and Stafford both had career years throwing the ball last year. Who do you think is more likely to repeat that performance and why?
The people who forget the bold don't really understand what the concept is. Absolutely it can work both positively and negatively....'Instinctive said:
Absolutely agree. It's silly to use a mean other than an inviduals mean. But situation doesn't change a players mean. By definition, his "mean" should be how he would perform in an average situation, with an average team, against an average schedule with an average coach, ect. His mean should be a starting point for projections, not an ending point. If a WR has a career of 700 receptions and a YPC of 12, and this year he moves from Cleveland to New England, then yes, you should expect his YPC to go up.
Again agree with this. A lot gets hidden when you look at simple stats. I pointed out DeMarco Murray earlier, who had a 170 carries last year, and averaged something like 5.5 ypc. But a single huge game was responsible for some much of that stat. In fact, his 1 91-yard run added over 0.5 yards to his season YPC average. He's a prime eexample of where it's useful to look at the actual data. It's not fair to totally discredit his 25/253 performance against the Rams, but I think it's dangerous to project him forward with a 5.5 ypc.
This is where I think a lot of people get confused. You kind of are and aren't regressing Gronkowski. In terms of Gronkowski's ability you are not regressing him. He's still averaging right around his YPC and TD%. But you are regressing his Targets, you're basically saying that he got twice as many targets last year and that's not what he'll get in an typical year (regardless of Lloyd as you said that Brady had a career year and won't repeat his 611 attempts).
They all must regress is a misunderstanding of regression to the mean. It's a probability expectation.
Yes So I was correct on TD's but wrong on yards. But you forgot to quote my other proposed bet:
My point still stands. I think we use "regression to the mean" too much in this hobby. The "luck factor" is just not as strong as some would like to make it. Use regression to the mean for your long-term roulette wheel predictions, but for FF, let's continue to look for the reasons stats go up and down, and not just chalk them up to regression.
Well put, but I would take it farther. I think the term is often misused. A regression equation is a model based on data. It can be used to make predictions, but the prerequisites are seldom met (and often ignored) in discussions like this.
Saying at least one QB will throw for 5K yards isn't about regression to the mean. Regression to the mean is fundamentally about small sample sizes, not "luck" as you put it. You knew that those players wouldn't get 100 TDs again. Why is that? Because you knew the only reason they averaged 100 TDs over 64 games is because for four players, 64 games is a small sample size, and they weren't really 100TD/64 game players.
That's true to a certain extent. But really I knew it because there were specific reasons why I thought it wouldn't happen again. Things change year to year. Players' situations change, coaches come and go, etc.Chase Stuart said:
It's not useless, but I would agree it's not accurate, either. Regression to the mean works best with a large group of players and doesn't work in absolutes on individual players.
I don't understand why it is so hard for some folks to grasp this concept. Regression to the mean is just another tool we can utilize to project stats. I guess if you don't want to utilize it, for whatever reason, don't. I understand the concept and it factors into my projections. It's certainly not the lead factor, but the information is nonetheless useful.
It's not that people don't understand the concept. It's that the concept is misused quite often by people who think they understand it.
