AVT and you (1 Viewer)

[bagger.]

as you are not doing the work to understand why player x will or will not have more receptions.

Sorry but no matter how much research you do or how many theories you run your stats thru, you still can't predict who will or will not have more receptions

You can't predict anything with certainty, but FF is about playing odds.I'll give you 2/1 odds that Steve Smith has more receptions per game than Keary Colbert...do you want some of that action?

You're using an extreme example which isn't fair IMO because the casual fan could also have predicted Smith to have more catches. So you're not proving that your(bagger's, whatever) research reaped any reward.How about Fitz or Boldin, who will have more receptions?

That and the fact to get any action, the appropriate odds would have to be more like 7 or 8 to 1 which would reflect Smith's odds against injury which is the only way that the Colbert bettor would win.

I didn't say the Colbert bettor would win, simply that the unresearching fan would also know Smith would catch more

knowing that he will catch more does not better your draft. knowing how much more does.

 

[bagger.]

I'll just repeat what GregR said, but in a slightly different way.

Assume we have Peyton Manning ranked as our top QB -- i.e., even though we know he doesn't have a 100% chance of finishing the year as the best fantasy QB, we think he's more likely to finish #1 than anyone else is.

Let A = the number of fantasy points we expect Manning to score if he finishes the year as the top fantasy QB.

Let B = the number of fantasy points we expect Manning to score if he does not finish the year as the top fantasy QB.

Let C = the probability that Manning will finish the year as the top fantasy QB.

Here is how AVT would do projections:

Manning's projected fantasy points = A

Here is a much more accurate projection:

Manning's projected fantasy points = A * C + B * (1 - C)

Since A > B and 0 < C < 1, note that A > A * C + B * (1 - C).

Therefore, your projection for your highest-rated QB should be less than your projection for the typical QB1 (generally speaking). So going back to my post above, if the typical QB1 scores X points, it is (usually) goofy to project your highest-rated QB to score more than X, but it (usually) correct to project your highest-rated QB to score less than X (reflecting a non-zero chance that your highest-rated QB will not finish as the QB1).

While this is great in theory, the practical application of not standing by your projections makes them useless, especially if you espouse the use of VBD.This method will get a hodge-podge of mathematic %s of whether they will be QB1, 2 or 3 or worse and any point differential you see from these won't be worth anything.

 

[LHUCKS]

I'm roped at work... will get back to this Sunday to explain the flawed logic.

 

[GregR_2]

This may seem like nitpicking... and I figure bagger knows what I'm about to say, but some of what he said could be misleading to someone new to the hobby. So I'm going to do it anyway.

That is the typical fallback of people who project stats for players without any regard to constraints. Your projections have to be constrained in a way that is consistent with league history. This means players on the same team have to be within team constraints. This means that players across the league cannot as a group all be higher than what players have been in years past.

Unless you have a good, specific reason for doing otherwise. Such as someone projecting passing and receiving higher going into 2004's point of emphasis on pass interference.Though to nitpick myself, my good specific reason is itself historical production when that same rule change went into effect. But that is still a great segue to my next nitpick

There will only be so many 1,000 yard rushers in 2006. It is easy to justify why double the amount backs that typically rush for 1,000 yards could do it, but you'll be wrong overpaying for marginal backs letting value at other positions go.

This is the typical trap that people fall into with the RB "landgrab" that occurs in the first three rounds. People continually get burned on that next breakout player because they are hoping that RB X will hit the 1,000 yard rushing projection, even though they have already ranked more than the historical amount of 1,000 yard rushers already.

You have to go with historical numbers that reflect what it is you are trying to project. In my example I gave previously, the "correct" historical numbers to have used would have been recent passing #'s, modified up by the same amount as the pass interference rule change has produced in previous seasons.In this case, my nitpick is on the wording used. The number of historical 1,000 yard rushers is heavily influenced by injuries. Unless you feel you can adequately predict which RBs will get injured, you probably aren't projecting injuries and so shouldn't use historical figures including injuries as your bounds.

So if you are going to use AVT to look at rushing yardage, and you just use final season results, you probably are hurting yourself by underestimating the lower RBs more than you are helping yourself. A better measure would be per game stats with some criteria of games as the starter/primary back to eliminate short-term flukes.

If you don't overproject you see the diminishing return of that marginal back in the 3rd round and all of a sudden there are some players at another position that stand out as a clearly better alternative.

I also want to point back up to what MT said earlier as his point #3. If you are going to build into your numbers the uncertainty that you have in your projection, it isn't a bad thing to be under historical numbers.That is something I struggle with. It doesn't intuitively seem right to me to lessen the value of the top players while effectively increasing the value of the middling players (since top players can only go down, while middling players have a chance of vastly outperforming what you set for them). But when I think through it logically, I don't know that it is wrong to do it that way. If I think Peyton's most likely point total is 300 and Delhomme is 200... but the true EV (expected value) is more like Peyton 260 and Delhomme 220... if I ignore that I am not building into my value comparison my true feelings on the range over which both players may end up scoring.

It's probably fine that you do it either way, as VBD should just be 1 tool in your decision making. If you project Peyton at 300 pts, you still need to factor in your uncertainty in that number in some other fashion when making your decisions. While it is nice to capture it in your projection for quicker use, it also may add enough extra work to making your projections that many may prefer to just gut that part out.

 

[GregR_2]

...

While this is great in theory, the practical application of not standing by your projections makes them useless, especially if you espouse the use of VBD.This method will get a hodge-podge of mathematic %s of whether they will be QB1, 2 or 3 or worse and any point differential you see from these won't be worth anything.

I kind of dealt with this in my reply I was typing while you posted this.The point differential will indeed be worth something. And I'd imagine most experienced FF people take into account what MT is saying, in some manner. It may be that you had Steve Smith projected at 800 yards because of uncertainty about his previous injury, but recognized he was a lot more likely to go way over that than he was to go under. The question is, did you then bump him up to 950, or did you keep him at 800 and then say, "I'll remember he has more upside when it comes time to compare him to other guys with similar projections".

Either way, you should be including this stuff in your decision. Now whether you put it in the number or include it in a qualitative sense is up to you. MT is talking about if you include it in your projection (and there is nothing wrong with that), then you should expect to have numbers that don't aline with AVT numbers as well.

 

[bagger.]

I agree with GregR's points about league rule changes or other league-wide dynamics that effect everyone, and should be thought of. My point was more going to the fact of people over-projecting players without checks or balances.

I agree that when making projections you should incorporate risk (whether that is injury risk or risk of a player not proving he has been a success for multiple years). However, once you incorporate that risk into your projections your end-rankings need to come close to tracking AVT if you want to have point dropoffs mean anything with intra and inter positional rankings.

Bottom line, the goal of projections is not to get a weighted average of expectations of a player. The goal is to have projections that when it is your turn to pick in a draft, what is on your sheet of paper most accurately reflects the points projected to be scored by available players, and the amount of points that will dropoff at each position in between your current draft pick and your next draft pick.

What Maurile talks about is something I think about in my initial projections, but you can't end there. At the end of the day you have to make a stand and say player X is my #1 RB and as such will score xxx points more than the next RB.

There is obviously a range of values where that player will reasonably end up at. But you have to pick a number that you feel best reflects the expected outcome of his performance, and how that compares to other players at that position and the historical performance at players at that position.

 

[GregR_2]

...

Bottom line, the goal of projections is not to get a weighted average of expectations of a player. The goal is to have projections that when it is your turn to pick in a draft, what is on your sheet of paper most accurately reflects the points projected to be scored by available players, and the amount of points that will dropoff at each position in between your current draft pick and your next draft pick.

What Maurile talks about is something I think about in my initial projections, but you can't end there. At the end of the day you have to make a stand and say player X is my #1 RB and as such will score xxx points more than the next RB.

...

You're saying you want his projection to be the points you project him to score. But isn't the overall goal of VBD though to come up with a number that represents his value? And while the weighted projection is further from points, isn't it closer to actual value?For example, maybe you have Curtis Martin as the #9 RB by what you think is most likely to happen, and have DD as the #8, ahead by 10 FP. But you realize CuMar could again be in the top 3 or 4 in the league, while you don't think DD could realistically make it that high.

If the weighted projection shows Martin's expected numbers to be higher than DD's expected numbers, doesn't that tell you that over the range of what you considered the possible outcomes (and how likely they are), that Martin is the more valuable pick?

I'll add, I'm not sold that weighted projections are the way to go, if only because of the fact they may significantly increase the amount of effort of doing projections. But I'm not convinced yet that they don't have merit as a way of quantifying the risk that you feel about the player and your projections for him, which plays into his final value.

 

[Bri]

as you are not doing the work to understand why player x will or will not have more receptions.

Sorry but no matter how much research you do or how many theories you run your stats thru, you still can't predict who will or will not have more receptions

You can't predict anything with certainty, but FF is about playing odds.I'll give you 2/1 odds that Steve Smith has more receptions per game than Keary Colbert...do you want some of that action?

You're using an extreme example which isn't fair IMO because the casual fan could also have predicted Smith to have more catches. So you're not proving that your(bagger's, whatever) research reaped any reward.How about Fitz or Boldin, who will have more receptions?

That and the fact to get any action, the appropriate odds would have to be more like 7 or 8 to 1 which would reflect Smith's odds against injury which is the only way that the Colbert bettor would win.

I didn't say the Colbert bettor would win, simply that the unresearching fan would also know Smith would catch more

knowing that he will catch more does not better your draft. knowing how much more does.

you don't know thatedit to clarify, you don't know exaclty how many more passes Smith will catch.

To attempt to put an end to this-

You could research and find out that QB1 targets player X 70% of the time and he has 150 catches three years in a row while player Y only gets 5% and only has 10 catches per year the last three years.(don't check my math, I didn't bother;-) ) That does NOT mean that player X will catch 70% in the 4th year and player Y will catch 10% again in the 4th year.

There are countless examples of WRs struggling and improving. If you charted Harrison's first 5 year targets they'd probably be stair-like. Walker was "nothing" and then suddenly a gem.

A youngster developping is too large a variable for you to definitively state an exact number of catches or receptions a WR will make. I can see "ranges" as good predictions but exact? cmon

 

[bagger.]

You're saying you want his projection to be the points you project him to score. But isn't the overall goal of VBD though to come up with a number that represents his value? And while the weighted projection is further from points, isn't it closer to actual value?

Not to me. To me the overall goal of VBD is to assess the point differential between positions and pick the position that outscores the "baseline" by a larger amount. This requires a projection as to what his actual points scored will be. In Joe's typical scenario it is knowing what players score what (which makes VBD perfect for known results, but as results become more unclear it rapidly deteriorates).

For example, maybe you have Curtis Martin as the #9 RB by what you think is most likely to happen, and have DD as the #8, ahead by 10 FP. But you realize CuMar could again be in the top 3 or 4 in the league, while you don't think DD could realistically make it that high.

If the weighted projection shows Martin's expected numbers to be higher than DD's expected numbers, doesn't that tell you that over the range of what you considered the possible outcomes (and how likely they are), that Martin is the more valuable pick?

I guess when I project stats I incorporate that into my projections. In that scenario I would have Martin ahead of DD since his upside is greater and that upside needs to be quantified in my projections.

I'll add, I'm not sold that weighted projections are the way to go, if only because of the fact they may significantly increase the amount of effort of doing projections. But I'm not convinced yet that they don't have merit as a way of quantifying the risk that you feel about the player and your projections for him, which plays into his final value.

I do weighted projections in that I weight past years stats and then adjust from there. I even think the way Maurile approaches it is another good and different way to do it to further help refine projections. However, to ignore AVT on its face is foolish because history tells you that QB1 will score x. If the player you have as QB1 is projected to score x-40 and the standard deviation for QB1 is 20, your projection is off and needs to be corrected.

 

[GregR_2]

Let me give an example to try to illustrate why the QB being that low may be ok.

Let's say we have a league with 5 QBs, to make it manageable. To make it even more manageable, let's take some of the ambiguity of predicting out of it and just deal with the fact they have a variance. Each QB scores one amount 80% of the time, and has a great year 20% of the time and scores more. Their scores are:

Bob: 200 points 80% of the time / 250 points 20% of the time.

Al: 190 / 250

Joe 180 / 240

David 160 / 230

Tom 150 / 230

With that distribution, about 1/3 of the years all 5 QBs will have an average year (which means Bob would QB1). The other 2/3 of the time, at least one QB will go off and be QB1. After doing a little math, one can find that the QB1 will average somewhere around 226.

Now if you were asked to predict "What will the QB1 score" next year regardless of who is that QB1, you should say 226 which is the AVT value. Because that is what the average QB1 will score.

However if you were to do projections for each individual QB, you would get the most accurate results either going with 200 for Bob (if you want the number most likely to show up) or 210 (if you want a weighted average). Even though Bob will be the QB1 nearly half of the time, the AVT average for the QB1 is not the number he is most realistically going to score. It's higher than what he is most likely to score.

That's the difference between predicting a specific player, and predicting what the top QB will score. You may have every reason to think that Peyton will score around 330 fantasy points, as he's had 3 of 6 years right in that range, with one clunker and one amazing season, and another that wasn't far off. Just because you don't think anyone else will score more than that isn't a reason then to think Peyton will have an above average year and jump him up to the 370 points that the QB1 normally scores. You're more accurate to predict Peyton near what he may actually score, than you are to say he will be the one guy who has an atypical season just because he was furthest up on your list when you ranked them by your realistic numbers.

 

[Dreamer]

It's a shame we lost all the yellow board discussion. We talked this into the ground.

 

[radballs.]

It's a shame we lost all the yellow board discussion. We talked this into the ground.

Yeah, but this is an important enough topic to be addressed each year. I'd much rather have more draft strategy threads than news reporting threads in the Shark Pool. These threads can turn into a beating a dead horse kind of thing. But, any lessons or new data that can be garnered from strategy threads are incredibly valuable.

 

[Chase Stuart]

Projections is a zero sum game. If Chad's ranking goes down, someone else's ranking goes up which means that person will be expected to score more points.

I'm surprised no one caught this.

 

[Dreamer]

It's a shame we lost all the yellow board discussion. We talked this into the ground.

Yeah, but this is an important enough topic to be addressed each year. I'd much rather have more draft strategy threads than news reporting threads in the Shark Pool. These threads can turn into a beating a dead horse kind of thing. But, any lessons or new data that can be garnered from strategy threads are incredibly valuable.

Absolutely. Threads like these are what make this board great.