AVBD: Solution to a problem of standard VBD (1 Viewer)

[Uncle Grandpa]

I'm sure I'm not the first to point this out, but I haven't seen it mentioned so here goes...

Background: Standard VBD is calculated as the difference between a player's fantasy scores and the fantasy score of a predetermined baseline player. For example, if I project the WR1 to score 375 fantasy points next year, he has 75 VBD above the WR6 (my arbitrary baseline for WRs) who I think will score 300 points.

Issue: While VBD calculates the difference between a player and his respective baseline, it completely disregards the difference between that player and the other players above the baseline. I believe this ignorance leads to sub-optimal decisions. Here is an example which highlights the issue:

2015 Projections

RB1 - 400 (100 VBD) WR1 - 375 (75 VBD)

RB2 - 375 (75) WR2 - 335 (35)

RB3 - 365 (65) WR3 - 325 (32)

RB4 - 350 (50) WR4 - 315 (15)

RB5 - 345 (45) WR5 - 305 (5)

RB6 - 300 (0) WR6 - 300 (0)

VBD Order:

RB1 - 100

WR1 - 75

RB2 - 75

RB3 - 65

RB4 - 50

RB5 - 45 .........

Standard VBD: Tells me to draft RB1 before WR1.

Simple enough, but there's a problem. While the RB 1 has the highest advantage over the baseline, it doesn't have the highest overall advantage against its positional field. If you add up the point advantage for the RB1 over all of the RBs, the total is 265 points. Meanwhile, if you add up the point advantage of the WR1 over all of the WRs, the total is 295 points. Drafting the WR1 first now gives me the largest advantage over the entire field, not just the baseline player.

The solution - Advanced VBD: Instead of simply calculating VBD as the difference between player X and his baseline, you add the differences between Player X and all of the players below him until reaching the baseline player. Then divide that total by the number of players in the entire position minus 1. RB1 would be calculated as (400-375 + 400-365 + 400-350 + 400-345 + 400-300) / (6 - 1)

After doing this for all players, the AVBD order now looks like this:

WR1 (59 AVBD)

RB1 (53)

RB2 (28)

RB3 (20)

WR2 (19).......

This is obviously a tad more complicated than standard VBD, but it's not difficult to create the formulas in Excel.

ETA: See post 23 for refinement of this method.

 

[LawFitz]

Love where you're going with this.

Now just need to find an adjuster for positional attrition as well.

 

[TDorBust]

Well if you want to go deeper it should also take into account what will come around to you at your next pick or predefined 10/12/16 picks later.

So if by your example you are drafting 1.01 in the league and you have a choice between WR1 and RB1. The next time your pick comes around logic might dictate that 10RBs and 4WRs might be off the board. Wouldn't the value be even more so in the RBs favor since you are still able to get a much higher VBD player?

Personally I think this is why tiers are so important. You want to hopefully always grab the last guy in each tier of players as you are getting optimum pick location since everyone in a tier in theory should roughly wash each other out.

In the end my statement only makes sense if you are trying to build a redraft and not dynasty team as in redraft you will slightly draft to position of need but in dynasty its always about BPA as you have no clue what you will really need in 3 years. In redraft I would assume you try to get the highest VBD of anyone in your starting roster if your looking at it just as a list to draft from.

I would also say this is the issue with doing a big board that is not done manually as only you can tell where players are normally taken.

Example:

Your pick 1.11 and 2.02

Rodgers, Lacy, Calvin, Forte are all on the board a packer fan sits behind you drafting, who do you take? Hint my friends hate the Bears and I hear them outright stating they would never draft a bear, this would never be taken into account in any big board or algorithm.

 

[CalBear]

That's a interesting point, but there are a few things to think about. The main thing is that the curve of pre-season projections maps poorly onto the curve of season results, and the mapping is particularly bad for players near and below the baseline. That is, we're better at projecting what WR1 will score than what WR36 will score. For example, FFToday's 2014 projections (I can't find FBG's) for the #36 ranked WR was 63/814/5 (111 points) and it turns out that the #36 WR (Vincent Jackson) scored 112 points, which is right on. But the #24 RB was projected for 263/1133/5 (160 points), and the actual #24 RB (Tre Mason) scored 121 points. So VBD projections based on projections for a baseline player pretty significantly understated the value of top RBs relative to top WRs in 2014, especially since the actual RB1, Demarco Murray, scored 304 points, which is 50 more than Lesean McCoy had been projected to score.

Now, this is also a problem for standard VBD, and one way I'm dealing with it is to use a point-based baseline instead of a ranking-based baseline. I can come up with a reasonable projection for what RB24 will score based on historical values, and that's a more solid baseline than whatever we're projecting for the guy we're ranking 24th out of RBs, who's in some RBBC where we're probably over-projecting both guys. Over the past 5 years, the #24 RB has scored 138, 148, 119, 130, and 121 points, and the average of those five (131) is a way better baseline than 160. The #36 WR very consistently scores around 110 points.

So, we're over-projecting not only the #24 RB, but probably every RB after the first few. In 2014, RBs 1-7* scored more than they'd been projected to score, and RBs 8-24 scored less than they'd been projected to score. So if you're evaluating RB7 by adding up VBD points based on projections for RB8-24, you may be aggregating the plus-side errors of all those projections. If you built out projections for each position based on historical production at each slot (RB1-24, WR1-36, QB1-12), you'd mitigate that issue.

It's worth noting that this is basically how dynamic auction values (DVBD) are calculated; it's the player's percentage of the remaining VBD points in the player pool.

* And of course, the guys we projected to finish in those slots are totally different than the guys who actually finished in those slots, but that's an issue with any ranking system.

 

[Biabreakable]

Great suggestion Uncle Grandpa.

Could you explain how to set up a formula for this purpose in excel?

 

[CalBear]

Great suggestion Uncle Grandpa.

Could you explain how to set up a formula for this purpose in excel?

Put the numbers for each position in a column (say, B1-B24). Then in the next column, C23=B23-B24, C22=B22-$B$24+C23, and copy up from C22 to C1.

 

[GregR]

Yes this is the issue that had people trying or using the Average Starter baseline as it samples the whole position. And this is pretty much a twist on Average Starter is how I'd phrase it.

It's an Average Starter baseline but for each player, rather than it being the average of the whole position, it's the average of the position beneath him.

Which is a good idea, it could avoid some of the negatives of pure Average Starter, like negative x-values that have to be normalized.

A good test would be to work up auction values with it. One thing I would want to make sure is whether you are going to get a decent value compared to players above that weren't "sampled" in the baseline. I'd think you'd be good as the higher player's x-value contributing to the pool should handle it. But it would be good to try with some wildly different positional curves to see how they come out.

 

[GregR]

Great suggestion Uncle Grandpa.

Could you explain how to set up a formula for this purpose in excel?

Put the numbers for each position in a column (say, B1-B24). Then in the next column, C23=B23-B24, C22=B22-$B$24+C23, and copy up from C22 to C1.

Or just use AVERAGE().

 

[Uncle Grandpa]

Now just need to find an adjuster for positional attrition as well.

Thanks. What do you mean by "positional attrition"?

 

[GregR]

The solution - Advanced VBD: Instead of simply calculating VBD as the difference between player X and his baseline, you add the differences between Player X and all of the players below him until reaching the baseline player. Then divide that total by the number of players in the entire position minus 1. RB1 would be calculated as (400-375 + 400-365 + 400-350 + 400-345 + 400-300) / (6 - 1)

I just noticed the bolded.

What is the reasoning for dividing by the entire position rather than by how many players contributed to the baseline?

 

[Uncle Grandpa]

Well if you want to go deeper it should also take into account what will come around to you at your next pick or predefined 10/12/16 picks later.

So if by your example you are drafting 1.01 in the league and you have a choice between WR1 and RB1. The next time your pick comes around logic might dictate that 10RBs and 4WRs might be off the board. Wouldn't the value be even more so in the RBs favor since you are still able to get a much higher VBD player?

Personally I think this is why tiers are so important. You want to hopefully always grab the last guy in each tier of players as you are getting optimum pick location since everyone in a tier in theory should roughly wash each other out.

In the end my statement only makes sense if you are trying to build a redraft and not dynasty team as in redraft you will slightly draft to position of need but in dynasty its always about BPA as you have no clue what you will really need in 3 years. In redraft I would assume you try to get the highest VBD of anyone in your starting roster if your looking at it just as a list to draft from.

I would also say this is the issue with doing a big board that is not done manually as only you can tell where players are normally taken.

Example:

Your pick 1.11 and 2.02

Rodgers, Lacy, Calvin, Forte are all on the board a packer fan sits behind you drafting, who do you take? Hint my friends hate the Bears and I hear them outright stating they would never draft a bear, this would never be taken into account in any big board or algorithm.

I've done this in the past using older methods of valuing players for redrafts. However, it feels like it would be an algorithm built aside from the VBD calculation. In other words, calculate the player values (AVBD in this case), then use some sort of optimizer that considers roster/lineup requirements, your draft picks, the draft picks of others, and the players remaining.

 

[Uncle Grandpa]

The solution - Advanced VBD: Instead of simply calculating VBD as the difference between player X and his baseline, you add the differences between Player X and all of the players below him until reaching the baseline player. Then divide that total by the number of players in the entire position minus 1. RB1 would be calculated as (400-375 + 400-365 + 400-350 + 400-345 + 400-300) / (6 - 1)

I just noticed the bolded.

What is the reasoning for dividing by the entire position rather than by how many players contributed to the baseline?

There is the number of players in the position , but the number of differences between Player X and the others is n-1. The result is the average of the differences, or average advantage.

ETA: I reread your question and should note that the original should say, "then divide that total by the number of the players in the entire position (included in the baseline calcs) minus 1.

 

[ShaHBucks]

I draft based my gut

 

[ldizzle]

Doesn't football guys draft dominator do this? I seriously don't know.

 

[moleculo]

 

[GregR]

The solution - Advanced VBD: Instead of simply calculating VBD as the difference between player X and his baseline, you add the differences between Player X and all of the players below him until reaching the baseline player. Then divide that total by the number of players in the entire position minus 1. RB1 would be calculated as (400-375 + 400-365 + 400-350 + 400-345 + 400-300) / (6 - 1)

I just noticed the bolded.

What is the reasoning for dividing by the entire position rather than by how many players contributed to the baseline?

There is the number of players in the position , but the number of differences between Player X and the others is n-1. The result is the average of the differences, or average advantage.

ETA: I reread your question and should note that the original should say, "then divide that total by the number of the players in the entire position (included in the baseline calcs) minus 1.

Ok, I think we're on the same page, though the example had a result that followed the wording.

So for RB2... there are 4 RBs below him and the sum of differences is 140. You showed his final value as 28, or 140/5. So it should be 35, or 140/ (4 RBs below him that contributed differences)?

 

[GregR]

Doesn't football guys draft dominator do this? I seriously don't know.

Draft dominator will let you set 1 baseline player or value that gets used with every player at that position. You have control over that part, but it is the same for every player at the position.

What he is doing, is calculating a different baseline value for every player in the position, based on the average of just the players under him. That would take unique coding to that system to put in it.

Maurile's method, if I remember correctly, does some unique stuff for each position, but it is related to number of games likely to start, not related to changing the baseline value.

 

[Uncle Grandpa]

The solution - Advanced VBD: Instead of simply calculating VBD as the difference between player X and his baseline, you add the differences between Player X and all of the players below him until reaching the baseline player. Then divide that total by the number of players in the entire position minus 1. RB1 would be calculated as (400-375 + 400-365 + 400-350 + 400-345 + 400-300) / (6 - 1)

I just noticed the bolded.

What is the reasoning for dividing by the entire position rather than by how many players contributed to the baseline?

There is the number of players in the position , but the number of differences between Player X and the others is n-1. The result is the average of the differences, or average advantage.

ETA: I reread your question and should note that the original should say, "then divide that total by the number of the players in the entire position (included in the baseline calcs) minus 1.

Ok, I think we're on the same page, though the example had a result that followed the wording.

So for RB2... there are 4 RBs below him and the sum of differences is 140. You showed his final value as 28, or 140/5. So it should be 35, or 140/ (4 RBs below him that contributed differences)?

Actually, the 28 is correct. You're averaging the VBD of Player X against everyone else at that position. That includes the players above Player X which add nothing to the numerator, but normalize their total against the rest at their position.

 

[GregR]

Actually, the 28 is correct. You're averaging the VBD of Player X against everyone else at that position. That includes the players above Player X which add nothing to the numerator, but normalize their total against the rest at their position.

By the way, apologies if I'm obtuse about something, trying to multi-task and this stuff is tough enough on a brain fully concentrating on it.

I'm not sure if dividing by the total number of players is any better, but that might be a secondary question to whether the player above him SHOULD be contributing something (negative) to the numerator. If I understood you correctly that he's being excluded. Which really turns it into a discussion about whether it's better than a pure Average Starter.

What I mean is...

RB1 - 400

RB2 - 375

RB3 - 365

RB4 - 350

RB5 - 345

RB6 - 300

Your statement that just comparing RB1 to RB6 - as done by the Last Starter baseline - doesn't really express his value versus the whole RB field is absolutely correct.

But that same issue is a problem when calculating RB2 if you don't include any impact on his value of the guys above him.

At a purest form, RB1 gives us an edge of 25, 35, 50, 65 and 100 = 265 points for the year over our league.

RB2 gives us an edge of -25, 10, 25, 30, 75 = 115 points .

If RB1 scores less than 400, then RB2 does better against that team and his value to us should go up. If RB1 scores more than 400, then RB2 loses ground and his value to us should go down. If I understand you right and you're not having him contribute negatively to the numerator, it is the same issue (to a lesser degree) as Last Starter that it isn't fully sampling his position. (Edit to add: For RB5 whose value only depends on RB6, you are exactly back to the same amount of comparison vs is position that Last Starter did.)

This is the reason people tried the Average Starter baseline... that you sample each player against all starters at his position.

But Average Starter leads to the issue of the bottom half or so of the position end up with negative X-values which some people don't want to deal with, and that can be a pain for doing auction values.

 

[GregR]

And I'll add, I'm not saying it's worthless, I'm pointing out the pluses and minuses.

It is probably a truer reflection of value than Last Starter. Probably worse than Average Starter, but easier to work with as it avoids negative X-values, which might make it a good compromise.

It would do a better job reflecting values at the top of positions as it samples most of the position for those players. But the further you go down the position, more of the players at the position get left out. So it becomes more affected by the same flaw as Last Starter. But it never is worse about it than Last Starter, so still an improvement.

 

[Uncle Grandpa]

Great suggestion Uncle Grandpa.

Could you explain how to set up a formula for this purpose in excel?

Put the numbers for each position in a column (say, B1-B24). Then in the next column, C23=B23-B24, C22=B22-$B$24+C23, and copy up from C22 to C1.

Actually, this formula doesn't work. It double counts the VBD advantage of the players below.

Using the B1-24 suggestion, type this in cell C1 and then copy down through cell c23.

=((COUNT(B1:$B$24)-1)*B1-SUM(B2:$B$24))/(COUNT($B$1:$B$24)-1)

 

[Evilgrin 72]

 

[Uncle Grandpa]

Actually, the 28 is correct. You're averaging the VBD of Player X against everyone else at that position. That includes the players above Player X which add nothing to the numerator, but normalize their total against the rest at their position.

By the way, apologies if I'm obtuse about something, trying to multi-task and this stuff is tough enough on a brain fully concentrating on it.

I'm not sure if dividing by the total number of players is any better, but that might be a secondary question to whether the player above him SHOULD be contributing something (negative) to the numerator. If I understood you correctly that he's being excluded. Which really turns it into a discussion about whether it's better than a pure Average Starter.

What I mean is...

RB1 - 400

RB2 - 375

RB3 - 365

RB4 - 350

RB5 - 345

RB6 - 300

Your statement that just comparing RB1 to RB6 - as done by the Last Starter baseline - doesn't really express his value versus the whole RB field is absolutely correct.

But that same issue is a problem when calculating RB2 if you don't include any impact on his value of the guys above him.

At a purest form, RB1 gives us an edge of 25, 35, 50, 65 and 100 = 265 points for the year over our league.

RB2 gives us an edge of -25, 10, 25, 30, 75 = 115 points .

If RB1 scores less than 400, then RB2 does better against that team and his value to us should go up. If RB1 scores more than 400, then RB2 loses ground and his value to us should go down. If I understand you right and you're not having him contribute negatively to the numerator, it is the same issue (to a lesser degree) as Last Starter that it isn't fully sampling his position. (Edit to add: For RB5 whose value only depends on RB6, you are exactly back to the same amount of comparison vs is position that Last Starter did.)

This is the reason people tried the Average Starter baseline... that you sample each player against all starters at his position.

But Average Starter leads to the issue of the bottom half or so of the position end up with negative X-values which some people don't want to deal with, and that can be a pain for doing auction values.

Good points. I wasn't familiar with the average starter method which sounds like it's the equivalent of the AVBD presented here with the exception of maintaining the negative values for players above X.

I didn't think that this would have that much of an effect, but including the negative values does slightly out-optimize the VBD for the middle and lower players relative to my initial method.

I do suggest normalizing the values by dividing each player's result by the number of players at a position - 1. Otherwise the values are unevenly impacted by the numbers of players at a position (The top WR accumulates points against 35 other WRs while the RB only accumulates points vs. 23 RBs).

Here's a new formula which incorporates the negative values for players above Player X.

=(((COUNT(B$1:B$24)-1)*B1-SUM(B$1:$B24))+B1)/(COUNT($B$1:$B$24)-1)

 

[bostonfred]

You should usually project more than 10 rbs to put up "top ten numbers". Before the season starts, there's no way to know who will get injured. So rb projections should naturally skew higher until you adjust for injury risk. how you choose to adjust for injury risk had a huge impact on rankings, but logically it makes sense to consider your own draft strategy. For example, the value of McFadden is a lot different to a team that has him as their rb4 than rb1.

 

[CalBear]

You should usually project more than 10 rbs to put up "top ten numbers". Before the season starts, there's no way to know who will get injured. So rb projections should naturally skew higher until you adjust for injury risk.

I don't think this is correct from a mathematical perspective if you're using VBD. The point of VBD is to evaluate the relative value of positions against each other. If RBs get injured more often than WRs (they do), and if you project RBs as if they don't get injured, VBD won't accurately reflect the relative value of the positions.

 

[GregR]

I think to handle the injury thing well you might want to do something like Maurile's method, which incorporates how many fantasy starts on average you expect a player ranked in that spot at that position to have. If RBs get injured more they would have lower fantasy starts modifying their value. Though I imagine it's still dominated by factors like you probably start RB1 every week you can while RB20 might only be used 9 or 10 times as you play match ups.

Though coming up with the numbers for that takes a lot of work, which MT did (IIRC) by looking at a bunch of league's lineups over multiple seasons to come up with some average numbers for each ranking spot at each position.

 

[Uruk-Hai]

How would you handle this in leagues with flex positions? Especially multiple flex spots? Don't a good portion of H2H redrafts use flex spots there days?

Because then, in theory, you're comparing your player to the entire pool of flex-able players - not just those at his position. And yet, most leagues have a mandatory minimum number of starters at each spot. Now, assume that all drafters are picking "best value" (instead of "I'll fill all my RB/WR spots first" or "I'm drafting all Packers"). How do you determine that value?

My long-time (almost 30 years) local league is far from typical, I think, in either set-up, scoring, or drafting habits. But I'll use it as an example of what I mean.

We're mandated to start 1QB, 1RB, 2WR, 1TE, 1K, 1D. We then have 2 flex spots that can be any position with a cap of 2 total starters at every position except WR (where you can start up to 4). Total roster is double the mandatory starters, plus two at any position.

How would you compare players in this scenario? It's a problem I've always struggled with from a VBD standpoint.

 

[Sabertooth]

It think you are treading dangerously close to Joes secret formula.

 

[Uncle Grandpa]

How would you handle this in leagues with flex positions? Especially multiple flex spots? Don't a good portion of H2H redrafts use flex spots there days?

Because then, in theory, you're comparing your player to the entire pool of flex-able players - not just those at his position. And yet, most leagues have a mandatory minimum number of starters at each spot. Now, assume that all drafters are picking "best value" (instead of "I'll fill all my RB/WR spots first" or "I'm drafting all Packers"). How do you determine that value?

My long-time (almost 30 years) local league is far from typical, I think, in either set-up, scoring, or drafting habits. But I'll use it as an example of what I mean.

We're mandated to start 1QB, 1RB, 2WR, 1TE, 1K, 1D. We then have 2 flex spots that can be any position with a cap of 2 total starters at every position except WR (where you can start up to 4). Total roster is double the mandatory starters, plus two at any position.

How would you compare players in this scenario? It's a problem I've always struggled with from a VBD standpoint.

I use a historical method of determining how many flex spots to allocate to each position. If the starting lineup requirements are QRRWWWFF in a 12 team league, I know there will be at least 24 starting RBs and 36 starting WRs. Then I'll see how many RB and WRs would have made up the 24 flex spots in the previous year. If it was 14 and 10, then I'll just set my RB baseline at RB38 and WR baseline at 46. Sometimes I'll move the baseline down even lower to include the top reserves for things like injuries, bye weeks, etc.

 

[SaintsInDome2006]

Nice to see this kind of post.

 

[LawFitz]

Now just need to find an adjuster for positional attrition as well.

Thanks. What do you mean by "positional attrition"?

Euphemism for running backs get hurt and/or drop off more regularly than any other position.

And conversely, RBs have more annual risers.

Strict VBD analysis by itself doesn't account for this well IMO.

 

[bostonfred]

You should usually project more than 10 rbs to put up "top ten numbers". Before the season starts, there's no way to know who will get injured. So rb projections should naturally skew higher until you adjust for injury risk.

I don't think this is correct from a mathematical perspective if you're using VBD. The point of VBD is to evaluate the relative value of positions against each other. If RBs get injured more often than WRs (they do), and if you project RBs as if they don't get injured, VBD won't accurately reflect the relative value of the positions.

let's say rb20 got 160 points this year, rb10 got 240, and rb3 got 320. If I project a guy to get 160 points, that doesn't make him a top 20 rb. It just means that if he stays healthy I expect him to get 160 points. But I will generally have more than 20 people projected to get 160 points, and in fact many more than 20 may score 10 ppg, more than 10 will score 15ppg, and more than 3 may score 20ppg. So a projection of 10 ppg isn't really a top 20 projection, is it?

 

[bostonfred]

Let's say is the sixth round, and based on all our fancy baselines I have two players that stand out to me on my board, a running back with an avbd of 25 and a receiver with an avbd of 20. Who should I take? Would it matter if I told you we can only start three backs and I already have three on my roster protected for 40+ vbd? What's the value of a really good backup? How much of a dropoff in receiver production should I accept for security at a volatile position?

What if I were loaded at receiver, instead, and I have two backs with equal avbd numbers. Would it matter if I said that one of the running backs was a high upside backup like Knile Davis, and the other was a kind of boring guy like Chris ivory? Do you tweak your projections to account for davis' higher EV if Charles gets hurt? Or to get the more guaranteed production of a kinda sorta starting rb?

Single number projections always seem to fall apart when you use them in a vacuum. eventually, you fudge your list to get the results you want.

 

[CalBear]

It think you are treading dangerously close to Joes secret formula.

Joe's secret formula is the opposite of this approach. The purpose of Joe's secret formula is to make the Draft Dominator spit out players in roughly the order that people typically value them. It basically subverts the concept of VBD analysis by using arbitrary baselines to curve-fit to an expected result.

 

[Tau837]

I think to handle the injury thing well you might want to do something like Maurile's method, which incorporates how many fantasy starts on average you expect a player ranked in that spot at that position to have. If RBs get injured more they would have lower fantasy starts modifying their value. Though I imagine it's still dominated by factors like you probably start RB1 every week you can while RB20 might only be used 9 or 10 times as you play match ups.

Though coming up with the numbers for that takes a lot of work, which MT did (IIRC) by looking at a bunch of league's lineups over multiple seasons to come up with some average numbers for each ranking spot at each position.

I have never agreed with attempting to account for injuries in a general manner. If the average RB missed 3 games over some sample size, and thus it leads one to assume all RBs will miss 3 games, what is the value? Many RBs will stay healthy and miss 0 games. Those who are injured may miss a lot more than 3 games.

I think it is more reasonable to use a method that ignores injuries to come up with initial projections and then consider adjusting specific players who are viewed to have greater injury risk down by some amount.

 

[Tau837]

Nice to see this kind of post.

 

[GregR]

I think to handle the injury thing well you might want to do something like Maurile's method, which incorporates how many fantasy starts on average you expect a player ranked in that spot at that position to have. If RBs get injured more they would have lower fantasy starts modifying their value. Though I imagine it's still dominated by factors like you probably start RB1 every week you can while RB20 might only be used 9 or 10 times as you play match ups.

Though coming up with the numbers for that takes a lot of work, which MT did (IIRC) by looking at a bunch of league's lineups over multiple seasons to come up with some average numbers for each ranking spot at each position.

I have never agreed with attempting to account for injuries in a general manner. If the average RB missed 3 games over some sample size, and thus it leads one to assume all RBs will miss 3 games, what is the value? Many RBs will stay healthy and miss 0 games. Those who are injured may miss a lot more than 3 games.

I think it is more reasonable to use a method that ignores injuries to come up with initial projections and then consider adjusting specific players who are viewed to have greater injury risk down by some amount.

If you're trying to account for an individual's injury likelihood... say a Jadeveon Clowney preexisting injury with potential longer term issues, yes what you said would be a way of doing it, building it into his projections.

But that wasn't the topic being discussed above. This was about RBs as a group tend to get injured more than QBs or WRs do, and it's not limited to those with preexisting conditions.

For what was being discussed, you could drop everyone's projection at the position in an additional way unrelated to your concerns about prior existing injuries. Or you could handle it in some other fashion like I was saying... and I suggested Maurile's method would incorporate both that injury factor, and an additional element that you don't tend to start players as often the further down the positional list you go.

Edit to add: The comment RBs get injured more than QBs or WRs is based on studies people have done that found this to be the case, including trying to take prior injury history into account. A RB who missed X games last year is likely to miss more games this year than a WR who also missed X games last year... whether X is 0 or some larger number.

 

[Saint]

 

[GregR]

Nice to see this kind of post.

Yes, I miss the old days. A dozen years ago it seemed like this kind of stuff was as much a focus of discussion as was outlook for players.

But times change. More people now just want something that tells them what to do, so understanding the how and why isn't in as much demand.

 

[Uncle Grandpa]

Now just need to find an adjuster for positional attrition as well.

Thanks. What do you mean by "positional attrition"?

Euphemism for running backs get hurt and/or drop off more regularly than any other position.

And conversely, RBs have more annual risers.

Strict VBD analysis by itself doesn't account for this well IMO.

I personally use a valuation tool like VBD to provide a relative order of things based on projections, but it's just one component to my draft plan. I'm ok with leaving items like injury risk outside of the model. Those things are more difficult to quantify and I'm a believer that some guys are more injury prone than others. Whether it's the way they're built or the way they play the game, some players just seem to get injured more frequently.

To me, placing a position-wide adjustment for injury risk unfairly hurts a player's value while unfairly helping others. Do I think the same injury risk modifier should be applied to Jonathan Stewart as it is Frank Gore? No. I can certainly make an adjustment to my projections for each player's injury risk, but I'd prefer to do that before the final inputs go into the VBD calculator.

 

[BroncoFreak_2K3]

Nice to see this kind of post.

Yes, I miss the old days. A dozen years ago it seemed like this kind of stuff was as much a focus of discussion as were outlook for players.

But times change. More people now just want something that tells them what to do, so understanding the how and why isn't in as much demand.

 

[CalBear]

To me, placing a position-wide adjustment for injury risk unfairly hurts a player's value while unfairly helping others. Do I think the same injury risk modifier should be applied to Jonathan Stewart as it is Frank Gore? No. I can certainly make an adjustment to my projections for each player's injury risk, but I'd prefer to do that before the final inputs go into the VBD calculator.

The question is, should you be projecting RB24 to score 160 points or 130 points? Because if you're using your projection for RB24 to set your VBD baseline, it's going to affect how you value all the other players.

And if you set the baseline at 130 and still project RB24 to score 160 points, you're giving him 30 VBD points. Which might be the right thing to do, but you should be clear on what's going on.

Last year FFToday had McCoy projected as the #1 RB with 255 points (non-PPR). If you set the baseline at 160, he'd have had 95 VBD points. But--and this is a crucial point--if your projections for McCoy were correct, during the season he actually would deliver about 125 points above the baseline. His actual value would have been higher than what you valued him at in your draft/auction. According to VBD, you would have taken Demaryius Thomas (111 VBD points based on WR36) over McCoy, and also Calvin Johnson and Julio Jones. And if your projections on all of those players were correct, McCoy would have added the most value to your team.

That's a structural error; it seems clear that your baseline needs to be set based on what the real replacement player will provide, not based on a number that we know is a lot more than the #24 RB will score.

So then the question is, should you project the #24 RB (Ryan Mathews last year) for 160 or 130 points? There are arguments to be made both ways. I think in practice, FF players attribute more value to RBs near the baseline than other positions near the baseline; that is, we are likely to draft RB24 before WR36 or QB12 even though the VBD is equal for all of them. Is that the right thing to do? Here are the six players projected around the baseline at each position last year:

Joique Bell

Bishop Sankey

Ryan Mathews

Reggie Bush

Ben Tate

CJ Spiller

Only one of those (Bell) came close to his projections, and even he didn't beat them. At WR, it was:

Brian Hartline

Sammy Watkins

Reggie Wayne

Terrance Williams

Ruben Randle

Golden Tate

Five of those guys were at least in the ballpark, and Tate was a big winner. So at least in 2014, you would have been better off taking baseline WRs than RBs. It would be interesting to see how that plays out over time.

 

[GregR]

To me, placing a position-wide adjustment for injury risk unfairly hurts a player's value while unfairly helping others. Do I think the same injury risk modifier should be applied to Jonathan Stewart as it is Frank Gore? No. I can certainly make an adjustment to my projections for each player's injury risk, but I'd prefer to do that before the final inputs go into the VBD calculator.

Whether a specific RB is more sturdy than another specific RB is a different adjustment than we're talking about. You would do that separately and in addition to what we're talking about.

We're saying RBs as a group take more and different types of hits than WRs. Because of this they tend to miss more time than WRs.

This isn't unfairly downgrading some healthy RB because some other RBs are broken down and bring down the RB average as a whole. It's that even previously healthy RBs miss more time than previously healthy WRs. And broke down RBs miss more time than broke down WRs.

This is an old study now but the one that still stands out in my mind. Note conclusions like:

The difference between a sturdy WR and a fragile WR seems to be roughly equivalent to the difference between a sturdy RB and a fragile RB. But across all levels of injury history, WRs are more likely to stay healthy than RBs are. Again, no real surprise, but it's nice to be able to attach a numerical estimate to the difference.

For example he found that a RB with no injury history only has about a 50% chance of playing all 16 games, while a WR similarly without injury history has a 65-70% chance of playing 16 games. That is the part we're talking about factoring in, that there is an inherent difference based on the nature of the positions, regardless of the individual's injury history.

Edit to add another quote from the article, so the difficulty in getting a good numerical grip on this stuff isn't lost:

So this shouldn't be taken as anything other than a rough starting point for your assessments of injury risk.

Side note, this would be a great one to revisit. It might not be surprising if the defenseless receiver changes have widened the gap further. Though the increase in the passing game might increase number of hits a bit. Competing factors.

 

[Uncle Grandpa]

The question is, should you be projecting RB24 to score 160 points or 130 points? Because if you're using your projection for RB24 to set your VBD baseline, it's going to affect how you value all the other players.

And if you set the baseline at 130 and still project RB24 to score 160 points, you're giving him 30 VBD points. Which might be the right thing to do, but you should be clear on what's going on.

Last year FFToday had McCoy projected as the #1 RB with 255 points (non-PPR). If you set the baseline at 160, he'd have had 95 VBD points. But--and this is a crucial point--if your projections for McCoy were correct, during the season he actually would deliver about 125 points above the baseline. His actual value would have been higher than what you valued him at in your draft/auction. According to VBD, you would have taken Demaryius Thomas (111 VBD points based on WR36) over McCoy, and also Calvin Johnson and Julio Jones. And if your projections on all of those players were correct, McCoy would have added the most value to your team.

That's a structural error; it seems clear that your baseline needs to be set based on what the real replacement player will provide, not based on a number that we know is a lot more than the #24 RB will score.

So then the question is, should you project the #24 RB (Ryan Mathews last year) for 160 or 130 points? There are arguments to be made both ways. I think in practice, FF players attribute more value to RBs near the baseline than other positions near the baseline; that is, we are likely to draft RB24 before WR36 or QB12 even though the VBD is equal for all of them. Is that the right thing to do? Here are the six players projected around the baseline at each position last year:

The AVBD method, or average starter method that Gregg laid out, makes this much less crucial. Instead of the value calc being anchored to one number, your positional cut-off is now just a fraction of the equation. That's not to say the positional cut-off becomes irrelevant, but it's much less material than in the old VBD method.

I do like the idea of incorporating the injury risk component into your positional cut-offs. Let's say there's 2 starting RB positions in a 12 team league (no flex) and, on average, 6 of the top 24 RBs (per ADP) suffer a material injury each year, I'd increase my cut-off to RB30.

The issue of positional variance that you're highlighting could be managed in the same way. For example, QBs tend to be more predictable than RBs and WRs. Without quantifying anything, let's say you keep the QB cut-off at 12, but bump RBs another 10 and WRs another 5 to account for their increased year-to-year volatility.

 

[Uncle Grandpa]

To me, placing a position-wide adjustment for injury risk unfairly hurts a player's value while unfairly helping others. Do I think the same injury risk modifier should be applied to Jonathan Stewart as it is Frank Gore? No. I can certainly make an adjustment to my projections for each player's injury risk, but I'd prefer to do that before the final inputs go into the VBD calculator.

Whether a specific RB is more sturdy than another specific RB is a different adjustment than we're talking about. You would do that separately and in addition to what we're talking about.

We're saying RBs as a group take more and different types of hits than WRs. Because of this they tend to miss more time than WRs.

This isn't unfairly downgrading some healthy RB because some other RBs are broken down and bring down the RB average as a whole. It's that even previously healthy RBs miss more time than previously healthy WRs. And broke down RBs miss more time than broke down WRs.

I gotcha. See my post above with how I'd address this issue.

 

[CalBear]

The question is, should you be projecting RB24 to score 160 points or 130 points? Because if you're using your projection for RB24 to set your VBD baseline, it's going to affect how you value all the other players.

And if you set the baseline at 130 and still project RB24 to score 160 points, you're giving him 30 VBD points. Which might be the right thing to do, but you should be clear on what's going on.

Last year FFToday had McCoy projected as the #1 RB with 255 points (non-PPR). If you set the baseline at 160, he'd have had 95 VBD points. But--and this is a crucial point--if your projections for McCoy were correct, during the season he actually would deliver about 125 points above the baseline. His actual value would have been higher than what you valued him at in your draft/auction. According to VBD, you would have taken Demaryius Thomas (111 VBD points based on WR36) over McCoy, and also Calvin Johnson and Julio Jones. And if your projections on all of those players were correct, McCoy would have added the most value to your team.

That's a structural error; it seems clear that your baseline needs to be set based on what the real replacement player will provide, not based on a number that we know is a lot more than the #24 RB will score.

So then the question is, should you project the #24 RB (Ryan Mathews last year) for 160 or 130 points? There are arguments to be made both ways. I think in practice, FF players attribute more value to RBs near the baseline than other positions near the baseline; that is, we are likely to draft RB24 before WR36 or QB12 even though the VBD is equal for all of them. Is that the right thing to do? Here are the six players projected around the baseline at each position last year:

The AVBD method, or average starter method that Gregg laid out, makes this much less crucial. Instead of the value calc being anchored to one number, your positional cut-off is now just a fraction of the equation. That's not to say the positional cut-off becomes irrelevant, but it's much less material than in the old VBD method.

I do like the idea of incorporating the injury risk component into your positional cut-offs. Let's say there's 2 starting RB positions in a 12 team league (no flex) and, on average, 6 of the top 24 RBs (per ADP) suffer a material injury each year, I'd increase my cut-off to RB30.

The issue of positional variance that you're highlighting could be managed in the same way. For example, QBs tend to be more predictable than RBs and WRs. Without quantifying anything, let's say you keep the QB cut-off at 12, but bump RBs another 10 and WRs another 5 to account for their increased year-to-year volatility.

I'd say the problem is just as bad or even worse with AVBD, because not only is your projection for RB24 off, your projections for RB16-23 are off, too, and they're all off in the same direction. You're pretty significantly overstating the likely production of that group; and, you're probably understating the likely production of the top RBs. So you're systematically under-valuing top RBs and over-valuing crappy RBs. Moving the baseline down doesn't solve that problem; it might improve the under-valuing of top RBs but it will make the problem of over-valuing crappy RBs worse.

Again, the problem is that if you are projecting RB24 for 160 points, when the #24 RB typically scores about 130 points, moving the baseline down to RB30-35 where RBs score 100 points isn't going to improve the situation. Now you're saying that RB24 scores 60 more points than the baseline. If your projections are correct, he's not going to be RB24, he's going to be RB15 or something like that. Do you really want to be that wrong when you get the projections right?

Right now Dodds has the #1 RB scoring 238 points and the #24 RB scoring 140 points. I guarantee you that the spread between those two slots will be over 100 points. It might be 200 points. In the last 5 years it hasn't been less than 150 points, and in 4 out of 5 years it was above 180. So if those are the projections you're feeding in, you're pretty severely undervaluing top RBs if you're counting points above the projection baseline, whether that's VBD, AVBD, or average starter.

 

[Uncle Grandpa]

The question is, should you be projecting RB24 to score 160 points or 130 points? Because if you're using your projection for RB24 to set your VBD baseline, it's going to affect how you value all the other players.

And if you set the baseline at 130 and still project RB24 to score 160 points, you're giving him 30 VBD points. Which might be the right thing to do, but you should be clear on what's going on.

Last year FFToday had McCoy projected as the #1 RB with 255 points (non-PPR). If you set the baseline at 160, he'd have had 95 VBD points. But--and this is a crucial point--if your projections for McCoy were correct, during the season he actually would deliver about 125 points above the baseline. His actual value would have been higher than what you valued him at in your draft/auction. According to VBD, you would have taken Demaryius Thomas (111 VBD points based on WR36) over McCoy, and also Calvin Johnson and Julio Jones. And if your projections on all of those players were correct, McCoy would have added the most value to your team.

That's a structural error; it seems clear that your baseline needs to be set based on what the real replacement player will provide, not based on a number that we know is a lot more than the #24 RB will score.

So then the question is, should you project the #24 RB (Ryan Mathews last year) for 160 or 130 points? There are arguments to be made both ways. I think in practice, FF players attribute more value to RBs near the baseline than other positions near the baseline; that is, we are likely to draft RB24 before WR36 or QB12 even though the VBD is equal for all of them. Is that the right thing to do? Here are the six players projected around the baseline at each position last year:

The AVBD method, or average starter method that Gregg laid out, makes this much less crucial. Instead of the value calc being anchored to one number, your positional cut-off is now just a fraction of the equation. That's not to say the positional cut-off becomes irrelevant, but it's much less material than in the old VBD method.

I do like the idea of incorporating the injury risk component into your positional cut-offs. Let's say there's 2 starting RB positions in a 12 team league (no flex) and, on average, 6 of the top 24 RBs (per ADP) suffer a material injury each year, I'd increase my cut-off to RB30.

The issue of positional variance that you're highlighting could be managed in the same way. For example, QBs tend to be more predictable than RBs and WRs. Without quantifying anything, let's say you keep the QB cut-off at 12, but bump RBs another 10 and WRs another 5 to account for their increased year-to-year volatility.

Again, the problem is that if you are projecting RB24 for 160 points, when the #24 RB typically scores about 130 points, moving the baseline down to RB30-35 where RBs score 100 points isn't going to improve the situation. Now you're saying that RB24 scores 60 more points than the baseline. If your projections are correct, he's not going to be RB24, he's going to be RB15 or something like that. Do you really want to be that wrong when you get the projections right?

Perhaps our projection methods diverge and the reason for the confusion. My projections are based on the typical points for a given position. For example, if the RB1 has averaged 374 fantasy points over the last 5 years, regardless of who I rank as RB1, my projection for him will be 374. Same goes for RB24. Regardless of who I think the 24th ranked RB will be, his projection will be 150 points, the average RB24 each year.

Guessing whose name will ultimately fill each slot is a total crapshoot, but to take it a step further and project the RB1 to score 400 points or RB24 to score 160 points makes things even harder, imo.

ETA: Moving the positional cut-off wouldn't serve to better estimate the baseline player's value. I would use it to create a player pool more representative of those fighting for starting lineup positions. We all know the top 24 RBs aren't the only players used in lineups. Therefore, the baseline would be lowered to account for injuries, flex spots, bye weeks, etc.

 

[CalBear]

The question is, should you be projecting RB24 to score 160 points or 130 points? Because if you're using your projection for RB24 to set your VBD baseline, it's going to affect how you value all the other players.

And if you set the baseline at 130 and still project RB24 to score 160 points, you're giving him 30 VBD points. Which might be the right thing to do, but you should be clear on what's going on.

Last year FFToday had McCoy projected as the #1 RB with 255 points (non-PPR). If you set the baseline at 160, he'd have had 95 VBD points. But--and this is a crucial point--if your projections for McCoy were correct, during the season he actually would deliver about 125 points above the baseline. His actual value would have been higher than what you valued him at in your draft/auction. According to VBD, you would have taken Demaryius Thomas (111 VBD points based on WR36) over McCoy, and also Calvin Johnson and Julio Jones. And if your projections on all of those players were correct, McCoy would have added the most value to your team.

That's a structural error; it seems clear that your baseline needs to be set based on what the real replacement player will provide, not based on a number that we know is a lot more than the #24 RB will score.

So then the question is, should you project the #24 RB (Ryan Mathews last year) for 160 or 130 points? There are arguments to be made both ways. I think in practice, FF players attribute more value to RBs near the baseline than other positions near the baseline; that is, we are likely to draft RB24 before WR36 or QB12 even though the VBD is equal for all of them. Is that the right thing to do? Here are the six players projected around the baseline at each position last year:

The AVBD method, or average starter method that Gregg laid out, makes this much less crucial. Instead of the value calc being anchored to one number, your positional cut-off is now just a fraction of the equation. That's not to say the positional cut-off becomes irrelevant, but it's much less material than in the old VBD method.

I do like the idea of incorporating the injury risk component into your positional cut-offs. Let's say there's 2 starting RB positions in a 12 team league (no flex) and, on average, 6 of the top 24 RBs (per ADP) suffer a material injury each year, I'd increase my cut-off to RB30.

The issue of positional variance that you're highlighting could be managed in the same way. For example, QBs tend to be more predictable than RBs and WRs. Without quantifying anything, let's say you keep the QB cut-off at 12, but bump RBs another 10 and WRs another 5 to account for their increased year-to-year volatility.

Again, the problem is that if you are projecting RB24 for 160 points, when the #24 RB typically scores about 130 points, moving the baseline down to RB30-35 where RBs score 100 points isn't going to improve the situation. Now you're saying that RB24 scores 60 more points than the baseline. If your projections are correct, he's not going to be RB24, he's going to be RB15 or something like that. Do you really want to be that wrong when you get the projections right?

Perhaps our projection methods diverge and the reason for the confusion. My projections are based on the typical points for a given position. For example, if the RB1 has averaged 374 fantasy points over the last 5 years, regardless of who I rank as RB1, my projection for him will be 374. Same goes for RB24. Regardless of who I think the 24th ranked RB will be, his projection will be 150 points, the average RB24 each year.

Guessing whose name will ultimately fill each slot is a total crapshoot, but to take it a step further and project the RB1 to score 400 points or RB24 to score 160 points makes things even harder, imo.

ETA: Moving the positional cut-off wouldn't serve to better estimate the baseline player's value. I would use it to create a player pool more representative of those fighting for starting lineup positions. We all know the top 24 RBs aren't the only players used in lineups. Therefore, the baseline would be lowered to account for injuries, flex spots, bye weeks, etc.

I don't typically have time to do my own projections these days, so I use FBG's. And there are structural issues with them. I don't think it's correct to project someone to duplicate Murray's 2014, but I also think it's wrong to have the #1 RB projected at 1600/12 total, which is what Dodds has right now. Woods gives the #1 RB a few more points but has RBs down to #36 still scoring over 120. Last year the #36 RB scored 92 points.

I think your approach of estimating for the slot and then slotting in RBs is likely to result in a better-looking curve, if you're mapping to historical data.

 

[GregR]

Unless you feel 100% sure that your top ranked RB actually will finish the season #1, doing that you are going to end up over projecting the top guys compared to your actual beliefs.

It may look aesthetically pleasing in that it looks like an end of season positional list. But if you think there's a range of spots your top guy could actually finish, then the points of the #1 guy isn't the best expression of your beliefs on him.

 

[Uncle Grandpa]

I think your approach of estimating for the slot and then slotting in RBs is likely to result in a better-looking curve, if you're mapping to historical data.

Yep. I use the previous five year average. I think it's getting too cute to not only try to pick the order of players, but also how many points they'll score; especially if, like you've suggested, the total points of your pool diverge from history.