VBD and FBG Projections (1 Viewer)

[Chase Stuart]

Your argument only works if you believe that fantasy drafts/auctions are "efficient" in the economic sense, which I personally think is untrue.

That's the point I was going to end up with, but I see we can skip the middle step. I believe fantasy drafts are efficient in the economic sense. I think the ADP market for money leagues is really good, and difficult to consistently beat a significant percentage of the time.

Let's look at another domain; do you think that the broad population is efficient at a game like blackjack? The values and percentages are much more clear in blackjack than in fantasy football, and there's a lot of money at stake, but I think it's very likely that if you were looking at the overall population, that people inefficiently invest their money at the blackjack table.Similarly, I think the broad population is very unlikely to be efficient at choosing values for players in drafts. Some people may be very good at it, but I can tell you that my main league isn't efficient top to bottom, and the Anarchy leagues certainly aren't, even if you leave out DonnyT.

I think the Staff vs. MB auction had some very good drafters, but I don't think the pricing was efficient; I don't think people's values were adjusted sufficiently for being able to start 4 WRs in a PPR system.

Which team is better, assuming you started RB-WR-RB? Having WR15 and WR20 instead of WR20 and WR29 as your starters is a nice upgrade, and you'll also have WR29 instead of WR36 as your backup WR. But, your top backup RB is now RB40 instead of RB24.

...

So it looks like RB24 over WR15 is probably a bad move. But it's close. I don't think it's a slam dunk either way, but if you go take two RBs and one WR in the first three rounds, I'd probably pull the trigger on a WR in the fourth. If you have only one RB through three rounds, you better grab that 2nd RB in round four. So as always, it depends on who will be available later in your draft, and who you've taken so far. And if each of your RBs miss 3 games instead of 2, then you will gain 40 points from having a better third RB, and then the two scenarios will be perfectly equal.

RB40, though, isn't an abstract concept that will give you a certain number of points, it's an actual player whose value is mapped to a probability function, not to a single number. Right now, it so happens that RB40 is Kevin Jones by FBG projections, RB24 is Thomas Jones, WR15 is Andre Johnson, WR20 is Deion Branch, WR29 is Vincent Jackson, and WR36 is Donte Stallworth.Personally, I would much rather have K.Jones (as top backup), A.Johnson, Branch and Jackson than T.Jones, Branch, Jackson and Stallworth. K.Jones could easily outscore T.Jones this year; he's RB40 because of his risk profile, not because T.Jones is a better runner or in a better situation.

Another way that RB40 is not just an abstract concept is that your backup RB could be a handcuff. If your RB1 is Steven Jackson, Brian Leonard's value to you is greater than RB49 (where he's currently projected), because Leonard is likely to produce more points when you need them than other RBs ranked in the same area. I'd rather have Kevin Jones than Brian Leonard, even if I had Steven Jackson at RB1, but I'd rather have better players at other positions and a handcuff for my stud, than crappy starters and better bench players.

We're not going to get anywhere if you're going to argue that RB40 could easily outscore RB24. Obviously you're supposed to prefer RB24 to RB40, otherwise what's the point of the argument? If I was to tell you that I think Stallworth could have another huge year and be more productive than Andre Johnson, that doesn't help us determine whether to go RB24 or WR15. (In practice, it might if you happened to think one player was really undervalued, but that's not a theoretical point. If, for example, you're convinced that Matt Leinart is going to lead all QBs in scoring, you wouldn't take Peyton Manning even if he tell to you in say, the 4th round. But that doesn't make Peyton Manning a bad pick in the fourth round, generally.)

 

[gheemony]

How do you come to that conclusion? You can't look at ADP to determine value.

Why not?

ADP measures expected cost, not expected value. VBD is a tool to examine where the expected cost of a player (where you'd have to draft him or what you'd have to pay at auction) exceeds the expected value of that player. If you try to map VBD directly onto ADP, you lose the reason to use VBD in the first place.

Let's say you had 120 VBD believers do 10 drafts, 12 drafters each.Would the ADP from those 10 drafts not tell you a great deal about the value of each player?

It would tell you a lot about the perceived value of each player. It also would tell you a lot about what VBD baselines people are using. In the staff vs. MB auction, it was clear to me that I was using different DD parameters than the majority of the drafters. This was particularly notable in the mid-range of the RB and WR curves, where RBs were consistently going for $5 more than my projections, and WRs were consistently going for $5 under my projections.

What ADP from those drafts will not tell you is, is it better to use Joe's Secret Formula, Worst Starter, or something else as a baseline? Joe's Secret Formula attempts to map onto ADP, which I think makes it a circular definition that is of limited usefulness.

The specific question you asked is how much more valuable is RB24 to have on your team than WR36. I think looking at ADP answers this, just like looking at ADP would answer the question of how much more valuable QB1 is than PK1, and RB10 than DT10. While everyone knows draft strategy/VBD comes out of ADP (you wait on players with low ADPs even if YOU value them highly), people often forget that the direction arrow goes both way. Draft strategy/VBD informs ADP, too.

How much more valuable, if any, do you think RB24 is than WR36?

Your argument only works if you believe that fantasy drafts/auctions are "efficient" in the economic sense, which I personally think is untrue. I think every draft and auction has inefficiencies, and in many cases those inefficiencies are systematic--for example, in leagues with scoring which favors WRs or TEs, RBs are severely overvalued by drafters. See the Anarchy leagues where WRs and TEs routinely get passed over for RBs who are projected to score 100 fewer points and 50+ fewer VBD points. If you move your baseline for the Anarchy leagues so that Ronnie Brown has more VBD points than Jeremy Shockey, because that's what ADP suggests, you are hiding the fact that Jeremy Shockey is more likely to help you win that league, which is the true definition of value.Obviously, the value of RB24 vs. WR36 depends on the league setup. RB23-25 are Ahman Green, Thomas Jones, and Jamal Lewis by Dodds' projections; WR35-37 are Jennings, Stallworth, and Curry. I think it's clear that in typical scoring leagues, the first group is more valuable than the second group, so there is some merit to having a baseline other than "worst starter" in a 1/2/3 league. But Joe's Secret Formula places those RBs as having more value than players like Burress, Santana Moss, Randy Moss, Hines Ward, Chris Chambers, and I'm not at all sure that the value is still there. Not to mention RB vs. QB value, where Ahman Green in Houston is valued above Brees and Bulger.

Finally! A voice of reason! I always felt this about Joe's formula, but couldn't explain why. This is it. It's the trap of trying to adjust baselines to make the VBD values match expected cost. Well done.

 

[gheemony]

I think the baselines set for "Maurile's Auction Method" are the best available in DD. His article gives a very good explanation of why he sets the baselines where he does. And there is a good discussion about expected fantasy starts from a player. I didn't have time this year to incorporate all of Maurile's concepts, but I certainly will for next year.

 

[Big Pink Monkey]

I think the baselines set for "Maurile's Auction Method" are the best available in DD. His article gives a very good explanation of why he sets the baselines where he does. And there is a good discussion about expected fantasy starts from a player. I didn't have time this year to incorporate all of Maurile's concepts, but I certainly will for next year.

Gotta link?BTW, why doesn't FBG have a search feature for their articles?

 

[CalBear]

We're not going to get anywhere if you're going to argue that RB40 could easily outscore RB24. Obviously you're supposed to prefer RB24 to RB40, otherwise what's the point of the argument? If I was to tell you that I think Stallworth could have another huge year and be more productive than Andre Johnson, that doesn't help us determine whether to go RB24 or WR15. (In practice, it might if you happened to think one player was really undervalued, but that's not a theoretical point. If, for example, you're convinced that Matt Leinart is going to lead all QBs in scoring, you wouldn't take Peyton Manning even if he tell to you in say, the 4th round. But that doesn't make Peyton Manning a bad pick in the fourth round, generally.)

Right now, FBG projects T.Jones to score 163 points, and K.Jones to score 95.8 points. My argument is, using those single numbers hides a lot of interesting information. In particular, it hides the probability curve. Here's a hand-wave at the probability curves for T and K Jones:

T.Jones:

10% 200+

20% 180-200

40% 140-180

20% 100-140

10% 0-100

K.Jones:

10% 200+

10% 180-200

20% 140-180

30% 100-140

10% 75-100

20% 0-75

The projections, which provide just a single number, don't really capture the fact that the shape of the probability curve for these two backs is quite different, in a way that increases the value of K.Jones relative to T.Jones. To put it another way, the further you go down the list of projections, the less useful the projections become as single numbers; other factors become more relevant in deciding between players. (Take a look at the last 50 or so picks in the Staff vs. MB auction; they have almost no relationship to anyone's projections).

Part of the upshot of this is that I don't think it's correct to take K.Jones, who is projected to score 96 points on the year,

and project him to score 6 points per game when one of your top two RBs is out. I would say that scoring an even 6 points per game is among the least likely possibilities for K.Jones this year; zero and 10 points are both more likely than 6 points.

 

[Chase Stuart]

Part of the upshot of this is that I don't think it's correct to take K.Jones, who is projected to score 96 points on the year, and project him to score 6 points per game when one of your top two RBs is out. I would say that scoring an even 6 points per game is among the least likely possibilities for K.Jones this year; zero and 10 points are both more likely than 6 points.

It doesn't matter, if his expected value is 6ppg. That's the number we should use for this exercise (unless you're arguing that you'd start RB4 ahead of Jones if Jones was known to be injured. That's a reasonable argument, but I think it unnecessarily complicates an already difficult issue).I've given my thoughts on comparing WR15 to RB24. What are yours?

 

[CalBear]

Part of the upshot of this is that I don't think it's correct to take K.Jones, who is projected to score 96 points on the year,

and project him to score 6 points per game when one of your top two RBs is out. I would say that scoring an even 6 points per game is among the least likely possibilities for K.Jones this year; zero and 10 points are both more likely than 6 points.

It doesn't matter, if his expected value is 6ppg. That's the number we should use for this exercise (unless you're arguing that you'd start RB4 ahead of Jones if Jones was known to be injured. That's a reasonable argument, but I think it unnecessarily complicates an already difficult issue).I've given my thoughts on comparing WR15 to RB24. What are yours?

What I'm arguing is that the exercise of plugging in the expected value for a backup, based on an expected number of games missed, is not very likely to result in interesting information. I understand the statistical perspective you're coming from, I just don't think the numbers are solid enough to be the basis of a real statistical analysis. Garbage in garbage out principle.Let me exaggerate to make my point. Let's say you're trying to decide between two RBs whose probability curves look like this:

200 points, 50%

1-199 points, 0%

0 points, 50%

101+ points, 0%

100 points, 100%

0-99 points, 0%

These two backs have identical statistical expected value, but their value is not at all similar. Their value is also not encapsulated in their projections; which one you would prefer depends on your style and on the rest of the players on your roster. To look at a 150-point back and say "oh, he's going to score 50 more points than those two" is really not correct.

When I look at WR15 (currently A.Johnson by FBG projections), and specifically at WR15-17 (Johnson, Burress, and S.Moss), I see the end of a tier. WR18-20 are Calvin Johnson, Braylon Edwards, and Deion Branch, who I think have significantly more questions than the WRs ahead of them. I think the difference between A.Johnson and C.Johnson is larger than the difference in their projections, because C.Johnson comes with significantly more risk.

Similarly, I see Jones as near the end of a tier (completed by the two RBs after him, J.Lewis and Lynch), but I see the difference between Jones and some of the backs in the next tier (such as D.Foster, J.Jones, and F.Taylor) to be less than the difference in their projections, because the risk associated with those backs is already priced into the projections (and in many cases, to too large an extent).

I'm pretty clear that I would rather have A.Johnson in my starting lineup than T.Jones as my RB3, so my DD baselines are not going to be set so that RB24 is more valuable than WR15.

 

[Big Pink Monkey]

Right now, FBG projects T.Jones to score 163 points, and K.Jones to score 95.8 points. My argument is, using those single numbers hides a lot of interesting information. In particular, it hides the probability curve.

This is essentially what I do, but I compact the curve down to an aggregate expectation. I started doing this after learning about poker. If I have JJ and someone puts me all in, I can think about the likelihood of certain hands to make my decision:10% bluff

50% QQ-AA

40% AK

The aggregate expectation is what I'm interested in because I probably can't put him on a single hand. Similarly, a single projection is unlikely to be correct. Different scenarios have different reasonable expectations attached. If Kevin Jones comes back to start, he could be okay. If he doesn't, he's worthless. Or he could come back a little and share time with Tatum Bell.

I'm not much of a math guy, so could you explain the difference between an aggregate expectation and a probability curve, if there is one?

 

[CalBear]

I'm not much of a math guy, so could you explain the difference between an aggregate expectation and a probability curve, if there is one?

A probability curve is just a way to visualize the different possibilities for an event. If you look at Peyton Manning, his probability curve will be pretty tight: low probability for 400+ points, low probability for 250- points, high probability for 300-350 points. If you look at, say, McNabb, his curve will be broader and flatter: possibly similar to Manning in probability for 400+ points, moderate probability for 250-350 points, some probability of 0-250 points due to injury. In non-statistical terms, his upside is similar to Manning's, but his situation isn't as reliable and he is more of an injury risk; a probability curve would help you look at the two QBs and compare the range of possibilities (as opposed to just comparing Manning projected at 333 points and McNabb projected at 278).

 

[Big Pink Monkey]

A probability curve is just a way to visualize the different possibilities for an event. If you look at Peyton Manning, his probability curve will be pretty tight: low probability for 400+ points, low probability for 250- points, high probability for 300-350 points. If you look at, say, McNabb, his curve will be broader and flatter: possibly similar to Manning in probability for 400+ points, moderate probability for 250-350 points, some probability of 0-250 points due to injury. In non-statistical terms, his upside is similar to Manning's, but his situation isn't as reliable and he is more of an injury risk; a probability curve would help you look at the two QBs and compare the range of possibilities (as opposed to just comparing Manning projected at 333 points and McNabb projected at 278).

In practical terms, I think we're doing the same thing. I have a draft board that tells me certain things, but I'm keeping those scenarios in the back of my head. I look at the way the draft is going and figure out who is appropriate for my team beyond the simple numbers. However, I still try not to take anyone whose expectation is below the expectation for the slot I'm in.

 

[Art_VanDalay]

This is the type of analysis that keeps people coming back to FBG for superior info. Great job!

 

[gheemony]

So what's best way to incorporate probability curves/distributions into projections/VBD? I'm looking for a better way than simply using your gut. Rather, instead of saying "I believe A. Johnson's and C. Johnson's expected values are the same but I'm going to draft A. Johnson instead becase I perceive A. Johnson's distribution of values to be less risky." Can we quantify it?

As an example, I like the FootballOutsider's projections because they include Risk Ratings. They rate the probability that a player will NOT meet their projection. So I can weigh two players with similar projections and easily identify which one has the least risk.

Another example is BaseballProspectus (it may be Baseball HQ). With their projections, they include a % likelihood that the player will outperform their prior year's numbers. That's really helpful.

 

[gheemony]

I think the baselines set for "Maurile's Auction Method" are the best available in DD. His article gives a very good explanation of why he sets the baselines where he does. And there is a good discussion about expected fantasy starts from a player. I didn't have time this year to incorporate all of Maurile's concepts, but I certainly will for next year.

Gotta link?BTW, why doesn't FBG have a search feature for their articles?

I found it in DraftDominator. There is article link in the Help menu (I think). Not sure if the article is posted. You should PM Maurile.

 

[Big Pink Monkey]

So what's best way to incorporate probability curves/distributions into projections/VBD? I'm looking for a better way than simply using your gut. Rather, instead of saying "I believe A. Johnson's and C. Johnson's expected values are the same but I'm going to draft A. Johnson instead becase I perceive A. Johnson's distribution of values to be less risky." Can we quantify it?

That's why I do an aggregate expectation. The risk is built into the projection and the value number. Other than that, I just make a judgment call.If I have one guy that will score 300 or 0 and another that will definitely score 150, I'm not just picking based on a perception of value. I'm picking based on how that risk profile fits in with my team. It is quantified, but it's difficult to include that much information in a draft sheet.

Also, there aren't a lot of players that can have a clearly defined risk since injuries are the most common factor in missing expectations if your projections are any good. Unsettled starting positions and the like are the most common ones you can plan for and those get more clear as the offseason progresses. Even when working with probabilities, you're still making a best guess.

 

[Chase Stuart]

Part of the upshot of this is that I don't think it's correct to take K.Jones, who is projected to score 96 points on the year,

and project him to score 6 points per game when one of your top two RBs is out. I would say that scoring an even 6 points per game is among the least likely possibilities for K.Jones this year; zero and 10 points are both more likely than 6 points.

It doesn't matter, if his expected value is 6ppg. That's the number we should use for this exercise (unless you're arguing that you'd start RB4 ahead of Jones if Jones was known to be injured. That's a reasonable argument, but I think it unnecessarily complicates an already difficult issue).I've given my thoughts on comparing WR15 to RB24. What are yours?

What I'm arguing is that the exercise of plugging in the expected value for a backup, based on an expected number of games missed, is not very likely to result in interesting information. I understand the statistical perspective you're coming from, I just don't think the numbers are solid enough to be the basis of a real statistical analysis. Garbage in garbage out principle.Let me exaggerate to make my point. Let's say you're trying to decide between two RBs whose probability curves look like this:

200 points, 50%

1-199 points, 0%

0 points, 50%

101+ points, 0%

100 points, 100%

0-99 points, 0%

These two backs have identical statistical expected value, but their value is not at all similar. Their value is also not encapsulated in their projections; which one you would prefer depends on your style and on the rest of the players on your roster. To look at a 150-point back and say "oh, he's going to score 50 more points than those two" is really not correct.

When I look at WR15 (currently A.Johnson by FBG projections), and specifically at WR15-17 (Johnson, Burress, and S.Moss), I see the end of a tier. WR18-20 are Calvin Johnson, Braylon Edwards, and Deion Branch, who I think have significantly more questions than the WRs ahead of them. I think the difference between A.Johnson and C.Johnson is larger than the difference in their projections, because C.Johnson comes with significantly more risk.

Similarly, I see Jones as near the end of a tier (completed by the two RBs after him, J.Lewis and Lynch), but I see the difference between Jones and some of the backs in the next tier (such as D.Foster, J.Jones, and F.Taylor) to be less than the difference in their projections, because the risk associated with those backs is already priced into the projections (and in many cases, to too large an extent).

I'm pretty clear that I would rather have A.Johnson in my starting lineup than T.Jones as my RB3, so my DD baselines are not going to be set so that RB24 is more valuable than WR15.

You're still not arguing draft theory. Do you see why?You're argument is no different than saying Antonio Gates with the 30th pick is a bad pick, because you think Vernon Davis will finish the season as the top fantasy TE, and can be had at least 30 picks later. That's not an argument that taking TE1 in the third round is bad draft theory, which is what prompted this debate.

Arguing either is fine, but I thought we were discussing draft theory.

 

[CalBear]

You're still not arguing draft theory. Do you see why?

You're argument is no different than saying Antonio Gates with the 30th pick is a bad pick, because you think Vernon Davis will finish the season as the top fantasy TE, and can be had at least 30 picks later. That's not an argument that taking TE1 in the third round is bad draft theory, which is what prompted this debate.

Arguing either is fine, but I thought we were discussing draft theory.

I am using an illustrative example; I'm bringing up Kevin Jones not because he's personally important, but because he illustrates my point, which is:The usefulness of projections in draft theory is inversely proportional to the number of points projected.

That is, projections are a good tool if you're deciding between Clinton Portis and Steve Smith, but they're not a good tool if you're deciding between Ladell Betts and Greg Jennings. The further down you go, the less you can rely on projections to provide meaningful differentiating data; other factors become more important.

To return specifically to the question of baselines and VBD: Personally, I don't think there is a single correct baseline, and in fact I look at mine every year and tweak them every year based on the actual player population. But I'm pretty confident that trying to map VBD to ADP is going to give you sub-optimal results.