VBD Question (1 Viewer)

[Hobbes]

Almost everyone uses their own baselines for the number of players to include to determine the VBD for each position based on # teams, # starting and roster spots, etc. But, in your opinion which total is better to use as the baseline?

1. The last baseline player (i.e. player 1 - player 12; player 2 - player 12; player 3 - player 12, etc.)

2. The average of all players in the baseline (i.e. player 1 - average of players 1-12; player 2 - average of players 1-12, etc.)

3. Some other baseline?

What are your thoughts on which you use and why? What have you noticed about the effect on different positions?

 

[by_the_sea_wannabe]

I like to set the baseline at how many players are likely to be drafted at each position at the point in the draft where however many players you start are drafted. We start 9 players in most of leagues so my baseline should be set at 12*9 = 108. I use 100 for simplicity sake because it is easier to look up the historical numbers. So my baseline is qb =14, rb=36, wr=40, te=10. To me that is a good point to measure by because to me that is about where the value of the draft starts to shift from taking the best player to taking more risks and filling in any holes in your roster.

 

[GregR]

Almost everyone uses their own baselines for the number of players to include to determine the VBD for each position based on # teams, # starting and roster spots, etc. But, in your opinion which total is better to use as the baseline?

1. The last baseline player (i.e. player 1 - player 12; player 2 - player 12; player 3 - player 12, etc.)

2. The average of all players in the baseline (i.e. player 1 - average of players 1-12; player 2 - average of players 1-12, etc.)

3. Some other baseline?

What are your thoughts on which you use and why? What have you noticed about the effect on different positions?

I think using average of players at the position is much better, though most don't use it. Maybe the biggest issue is you end up with negative numbers, which isn't a problem in making a cheat sheet. But it is a problem if you're doing auction values. You'd need to normalize the final VBD values back to positive numbers.

Here's why I think it's better. Imagine these two sets of projections for a 6 team league (to keep the example short):

QB1 400 400

QB2 390 395

QB3 380 395

QB4 370 395

QB5 360 395

QB6 350 350

If we use QB6 as a baseline, then in both leagues QB1 has a value of 50. However, it should be obvious that QB1 in the first league is a bigger value than in the second. If you compare team by team, QB1 in the first league is an improvement of 10, 20, 30, 40 and 50 points. That's 150 points total.

The Qb in the last league is four times a 5 point improvement and once a 50 point improvement over your opponents. That's 70 points of improvement. Obviously there's a disparity here. The way in which the scoring at a position drops off contributes to the value of the players, and a 'normal' baseline method doesn't include that. (Much later edit: It isn't the method for the baseline really, it's the method we calculate the value that is at the heart of it.)

An average baseline would take that into account. The first case QB2-QB5 average 370, so 400-370 = 30 VBD for QB1 in the first projections. The second set they average 386 so 400 - 386 = 14. That result is a lot more representative of their true value to your team.

Obviously I contrived an extreme example to make it obvious. But the same issues would happen in a normal set of projections. Not all positions fall off at the same rate, and even the same position doesn't fall off the same each year, and that is lost in a normal baseline method.

 

[Gottabesweet]

I generally get how the VBD works. I'm having difficulty in one league setting everything up. I know you guys are busy but if you can guide me in the right direction I'd appreciate it.

I have both the VBD excel and DD I plan to use for my draft. It's a 12 team league, where we start QB-RB-WR-TE-2 FLEX-K-DST and only 4 bench (12 rounds, ppr, 6 per td)

Now I pick 1.9 and assume Jimmy Graham should sky rocket up my board but he simply is not. I have a strong indication that two QB's will go in the first round.

How should I set up the DD roster spots field? Right now it's set to the starters, Should I just set it up as 1 QB, 1RB, 1WR, 2 FLEX, TE and then just fill a starting lineup since it's so flexiable? 1RB, 3WR, 2 AND 2 or 3 and 1 are all possibilities and with only 4 bench spot 3/4 or even 4/4 spots will be RB/WR

Any help would be appreciated!

 

[Hobbes]

Almost everyone uses their own baselines for the number of players to include to determine the VBD for each position based on # teams, # starting and roster spots, etc. But, in your opinion which total is better to use as the baseline?

1. The last baseline player (i.e. player 1 - player 12; player 2 - player 12; player 3 - player 12, etc.)

2. The average of all players in the baseline (i.e. player 1 - average of players 1-12; player 2 - average of players 1-12, etc.)

3. Some other baseline?

What are your thoughts on which you use and why? What have you noticed about the effect on different positions?

I think using average of players at the position is much better, though most don't use it. Maybe the biggest issue is you end up with negative numbers, which isn't a problem in making a cheat sheet. But it is a problem if you're doing auction values. You'd need to normalize the final VBD values back to positive numbers.

Here's why I think it's better. Imagine these two sets of projections for a 6 team league (to keep the example short):

QB1 400 400

QB2 390 395

QB3 380 395

QB4 370 395

QB5 360 395

QB6 350 350

If we use QB6 as a baseline, then in both leagues QB1 has a value of 50. However, it should be obvious that QB1 in the first league is a bigger value than in the second. If you compare team by team, QB1 in the first league is an improvement of 10, 20, 30, 40 and 50 points. That's 150 points total.

The Qb in the last league is four times a 5 point improvement and once a 50 point improvement over your opponents. That's 70 points of improvement. Obviously there's a disparity here. The way in which the scoring at a position drops off contributes to the value of the players, and a 'normal' baseline method doesn't include that.

An average baseline would take that into account. The first case QB2-QB5 average 370, so 400-370 = 30 VBD for QB1 in the first projections. The second set they average 386 so 400 - 386 = 14. That result is a lot more representative of their true value to your team.

Obviously I contrived an extreme example to make it obvious. But the same issues would happen in a normal set of projections. Not all positions fall off at the same rate, and even the same position doesn't fall off the same each year, and that is lost in a normal baseline method.

Greg - so how do you deal with positions that have less fall off as a whole (like say TE) vs. those that fall off more? It seems that the benefit that you are talking about above is working against you as soon as you pass the average for that position...then the value falls off just as fast in the opposite direction (i.e. more negative compared to the average than other positions). Here's an example from our league:

I set baselines at QB15, RB36, WR42, and TE6.

When you look at the overall rank using the average baseline value the overall ranking for each of these last "starters" in my league is:

QB15 - 137

RB36 - 132

WR42 - 92

TE6 - 65

When I use the "last starter" method the overall rank changes to:

QB15 - 97

RB36 - 96

WR42 - 98

TE6 - 99

Even if I eliminate all of the "backups" and just look at players within the baseline, I have a tough time valuing TE6 above RB23, WR29, and QB10 just because they are "less negative."

Thoughts?

 

[GregR]

Almost everyone uses their own baselines for the number of players to include to determine the VBD for each position based on # teams, # starting and roster spots, etc. But, in your opinion which total is better to use as the baseline?

1. The last baseline player (i.e. player 1 - player 12; player 2 - player 12; player 3 - player 12, etc.)

2. The average of all players in the baseline (i.e. player 1 - average of players 1-12; player 2 - average of players 1-12, etc.)

3. Some other baseline?

What are your thoughts on which you use and why? What have you noticed about the effect on different positions?

I think using average of players at the position is much better, though most don't use it. Maybe the biggest issue is you end up with negative numbers, which isn't a problem in making a cheat sheet. But it is a problem if you're doing auction values. You'd need to normalize the final VBD values back to positive numbers.

Here's why I think it's better. Imagine these two sets of projections for a 6 team league (to keep the example short):

QB1 400 400

QB2 390 395

QB3 380 395

QB4 370 395

QB5 360 395

QB6 350 350

If we use QB6 as a baseline, then in both leagues QB1 has a value of 50. However, it should be obvious that QB1 in the first league is a bigger value than in the second. If you compare team by team, QB1 in the first league is an improvement of 10, 20, 30, 40 and 50 points. That's 150 points total.

The Qb in the last league is four times a 5 point improvement and once a 50 point improvement over your opponents. That's 70 points of improvement. Obviously there's a disparity here. The way in which the scoring at a position drops off contributes to the value of the players, and a 'normal' baseline method doesn't include that.

An average baseline would take that into account. The first case QB2-QB5 average 370, so 400-370 = 30 VBD for QB1 in the first projections. The second set they average 386 so 400 - 386 = 14. That result is a lot more representative of their true value to your team.

Obviously I contrived an extreme example to make it obvious. But the same issues would happen in a normal set of projections. Not all positions fall off at the same rate, and even the same position doesn't fall off the same each year, and that is lost in a normal baseline method.

Greg - so how do you deal with positions that have less fall off as a whole (like say TE) vs. those that fall off more? It seems that the benefit that you are talking about above is working against you as soon as you pass the average for that position...then the value falls off just as fast in the opposite direction (i.e. more negative compared to the average than other positions). Here's an example from our league:

I set baselines at QB15, RB36, WR42, and TE6.

When you look at the overall rank using the average baseline value the overall ranking for each of these last "starters" in my league is:

QB15 - 137

RB36 - 132

WR42 - 92

TE6 - 65

When I use the "last starter" method the overall rank changes to:

QB15 - 97

RB36 - 96

WR42 - 98

TE6 - 99

Even if I eliminate all of the "backups" and just look at players within the baseline, I have a tough time valuing TE6 above RB23, WR29, and QB10 just because they are "less negative."

Thoughts?

Edit: Deleted, was off in what I said, working on it, see below.

 

[toonarmy]

Surely that just takes you back to where you were, i.e. regular VBD? What am I missing?

 

[GregR]

Surely that just takes you back to where you were, i.e. regular VBD? What am I missing?

Think you're right, I've said it wrong. The normalizing at the end is the messy part. Been a long time since I've tried it this way, when DD came out I liked all the other features it has so much I stopped doing my own spreadsheet for this.

The intermediate step looks right, as here's a sample... but then normalizing the results so they match together isn't as simple as adding the baseline back in or as you say, you lose what differentiates them:

Player FP VBD NewMethodQB1 400 50 29QB2 395 45 23QB3 380 30 5QB4 370 20 -7QB5 360 10 -19QB6 350 0 -31 Different projections: QB1 400 50 14QB2 395 45 8QB3 395 45 8QB4 395 45 8QB5 395 45 8QB6 350 0 -46So what I called NewMethod is the result of averaging the differences between each player and all others at his position. And the difference is what we want. QB1 is 50 VBD in both sets of projections. But the other method has his value at 29 and the first and 14 in the second, because he outscores all the other QBs between he and the baseline by so much less.

Edit to add: Anyone have a copy of posts on Old Yeller where we discussed it heavily back then, heh.

 

[GregR]

The more that I stare at it, the more I'm thinking the answer was that you don't normalize before comparing across positions then. 0 in one position is 0 in another. I recall having another step because you need that to get to auction values since the negatives won't work as is. But it's not necessary to make a cheatsheet.

So with the above, while I called both of them QB, let's pretend the first was QB and the second set of projections were actually my RBs in the same league and then we'll compare them.

Player FP VBD NewVBDQB1 400 50 29QB2 395 45 23QB3 380 30 5QB4 370 20 -7QB5 360 10 -19QB6 350 0 -31 Different projections: RB1 400 50 14RB2 395 45 8RB3 395 45 8RB4 395 45 8RB5 395 45 8RB6 350 0 -46
So QB3 (first data set) is worth 5, while in the second data set that we're now calling RBs, RB2 through RB5 who score identically have a value of 8. That doesn't need additional normalization. It is saying QB3 is close in value to those RBs. It is also saying that RB6 is less valuable than QB6.

In normal VBD we'd say since they are both baseline that they must be equal value, because that's just how the system is meant to work. But the reality is, they clearly are not equal value. RB6 is 50 points worse than RB1, 45 worse than RBs 2-5. QB6 is also 50 points worse than QB1, and 45 points worse than QB2. But he's not nearly as bad compared to QBs 3-5. So if you were choosing between being stuck with QB6 or RB6 as a starter... if all other things were equal, you're much worse off with RB6 than you are QB6. And so our valuation system should reflect that.

I believe the extra step I was thinking of, I just took the player with the lowest value regardless of position, and add in to every player, not just his position, to make them all positive again. That was only necessary to use it for auction values.

Let me try expanding RB to what would be our 6 team, 1 QB, 2 RB league and see if the results hold as valid.

Edit to add: Tested it with 2 starting RBs and it's behaving correctly. So I think that's it, yes.

 

[GregR]

By the way, the Excel formula for that... starting with just the regular math bit:

(QB1-QB2)+(QB1-QB3)+... (QB1-QB6) / 5

which can be written:

(5*QB1 - sum(QB2:QB6) ) / 5

Easiest way to make that a formula you can copy down the position is let it sum over all QBs and just subtract out an extra time the one you're currently looking at. So:

(6*QB1 - sum(QB1:QB6) / 5

Or in Excel formula, if you have all the projections for a position in column B, from B1 down to the baseline player (but not anyone past him)... this would be what you'd put in Cell C1 or any other cell to calculate this type of VBD value. Then you could copy and paste it all the way down the position to have it pick up the rest of the players:

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

 

[GregR]

Sorry it's taken so many posts to get coherent on this. Been years since I've done this method.

With what I describe, you aren't just choosing a baseline differently. You are using a different method of calculating the value from your baseline. Instead of just "sampling" each player against one baseline player.... you compare the player to every other meaningful player at his position, and average the results to express his value. You still have to choose a player beyond which you aren't going to do that with, and you can still call that your baseline player. But where in normal VBD that one player hugely affects the value of all players at his position, with this method he doesn't drive his position's value more than any other player at the position does.

 

[dagilfer]

I like the theory but I run into some issues when I try to get things back into the positives for auction purposes. As was mentioned, normalizing baselines to 0 brings everything back almost identical to standard VBD. But if I add the VBD value of the lowest player globally like you said, it skews the prices in weird directions. It's hard to explain for me, but it's most evident when looking at how high TE prices become. Doing it this way makes my TE3 equal to WR7. It seems like the value being added is relatively much more to a position like TE with less drop off between players in my projections, which is obviously counter-intuitive. Any ideas?

 

[GregR]

I like the theory but I run into some issues when I try to get things back into the positives for auction purposes. As was mentioned, normalizing baselines to 0 brings everything back almost identical to standard VBD. But if I add the VBD value of the lowest player globally like you said, it skews the prices in weird directions. It's hard to explain for me, but it's most evident when looking at how high TE prices become. Doing it this way makes my TE3 equal to WR7. It seems like the value being added is relatively much more to a position like TE with less drop off between players in my projections, which is obviously counter-intuitive. Any ideas?

What are the values like for TE3 and WR7 just as the output of VBD using this, before you start trying to convert them to auction prices?

 

[Chase Stuart]

Almost everyone uses their own baselines for the number of players to include to determine the VBD for each position based on # teams, # starting and roster spots, etc. But, in your opinion which total is better to use as the baseline?

1. The last baseline player (i.e. player 1 - player 12; player 2 - player 12; player 3 - player 12, etc.)

2. The average of all players in the baseline (i.e. player 1 - average of players 1-12; player 2 - average of players 1-12, etc.)

3. Some other baseline?

What are your thoughts on which you use and why? What have you noticed about the effect on different positions?

I think using average of players at the position is much better, though most don't use it. Maybe the biggest issue is you end up with negative numbers, which isn't a problem in making a cheat sheet. But it is a problem if you're doing auction values. You'd need to normalize the final VBD values back to positive numbers.

Here's why I think it's better. Imagine these two sets of projections for a 6 team league (to keep the example short):

QB1 400 400

QB2 390 395

QB3 380 395

QB4 370 395

QB5 360 395

QB6 350 350

If we use QB6 as a baseline, then in both leagues QB1 has a value of 50. However, it should be obvious that QB1 in the first league is a bigger value than in the second. If you compare team by team, QB1 in the first league is an improvement of 10, 20, 30, 40 and 50 points. That's 150 points total.

The Qb in the last league is four times a 5 point improvement and once a 50 point improvement over your opponents. That's 70 points of improvement. Obviously there's a disparity here. The way in which the scoring at a position drops off contributes to the value of the players, and a 'normal' baseline method doesn't include that. (Much later edit: It isn't the method for the baseline really, it's the method we calculate the value that is at the heart of it.)

An average baseline would take that into account. The first case QB2-QB5 average 370, so 400-370 = 30 VBD for QB1 in the first projections. The second set they average 386 so 400 - 386 = 14. That result is a lot more representative of their true value to your team.

Obviously I contrived an extreme example to make it obvious. But the same issues would happen in a normal set of projections. Not all positions fall off at the same rate, and even the same position doesn't fall off the same each year, and that is lost in a normal baseline method.

My optimal draft article deals with that issue, which is a good point: http://subscribers.footballguys.com/apps/article.php?article=stuart_optimal_draft

 

[GregR]

Thinking about this philosophically, I'm thinking maybe averaging the difference to each player might not be a good thing, particularly for auction drafts. It would be fine if we started the same number of players at each position, but I think it's probably distorting it if you have different numbers of starters. If you have a 1 QB, 2 RB league and QB1 has a net 110 points of value over the other 11 QBs and RB1 has a net 230 points of value over the other 23 RBs, then our final value should have RB1 more valuable... while if we average them, they are both a value of 10.

Or... going to give some thought to whether it would be appropriate to divide by the number of starting spots instead. Which... hmm, need to give it more thought.

I probably won't be able to play with examples for awhile, but maybe you'd want to try that and see how the auction values look then? You'd still need to normalize them to deal with the negative numbers, and I think you'd still want to shift all positions the same amount.

 

[acarey50]

My optimal draft article deals with that issue, which is a good point: http://subscribers.footballguys.com/apps/article.php?article=stuart_optimal_draft

Any chance the programming you used to create these simulations could be shared so it could be customized. For example, I would love to be able to run the simulation using FFPC scoring/lineups (or be able to plug in different league scoring/lineup rules) and also to be able to plug in my own set of projections.

 

[GregR]

Doing some quick thought experiments in my head, I'm thinking we need to divide the sum of differences by number of starting spots. Have to run, will work examples later.

Also, I love what Chase did. I did something like that years ago in trying to figure out the optimal draft strategy, compared results to VBD cheatsheet and a straight up Dynamic VBD, compared to drafting stud RB and highest points available. Optimal draft is such a tough equation to solve pretty much have to brute force it like that.

 

[Haverchuck]

I generally get how the VBD works. I'm having difficulty in one league setting everything up. I know you guys are busy but if you can guide me in the right direction I'd appreciate it.

I have both the VBD excel and DD I plan to use for my draft. It's a 12 team league, where we start QB-RB-WR-TE-2 FLEX-K-DST and only 4 bench (12 rounds, ppr, 6 per td)

Now I pick 1.9 and assume Jimmy Graham should sky rocket up my board but he simply is not. I have a strong indication that two QB's will go in the first round.

How should I set up the DD roster spots field? Right now it's set to the starters, Should I just set it up as 1 QB, 1RB, 1WR, 2 FLEX, TE and then just fill a starting lineup since it's so flexiable? 1RB, 3WR, 2 AND 2 or 3 and 1 are all possibilities and with only 4 bench spot 3/4 or even 4/4 spots will be RB/WR

Any help would be appreciated!

I have the same starting lineup in my league but with 4 points per passing TD and NON-PPR. While doing VBD which I'm new to, I've notice QB's being more valuable than with a traditional starting lineup even with only 4 points per passing TD.

Since each team only has 4 total starters of RBs and WRs, most teams can put together 4 good starters in those spots regardless of the position.

Most people also agree that there are 12 quality QBs this year so there appers to not be a huge edge there either.

Going after Jimmy Graham seems like a good idea, but I'm still trying to run a few methods to see where the most value may come from.

 

[Dinsy Ejotuz]

I've been messing around with the "best" VBD baseline quite a bit recently and have come to the conclusion that it's way more important to know how many players at each position will actually start for you during a season, make a reasonable guess at how many games they'll play, and (for redraft) get the "dynamic" part of it right than it is where you set zero.

 

[acarey50]

Doing some quick thought experiments in my head, I'm thinking we need to divide the sum of differences by number of starting spots. Have to run, will work examples later.

Would love to explore this further. however, if I am reading correctly, I believe the calc would be the same as looking at the average points scored, for the example we'd been discussing:

QB1 400 400

QB2 390 395

QB3 380 395

QB4 370 395

QB5 360 395

QB6 350 350

Scenario 1 - (10+20+30+40+50)/5 = 150/5 = 30

Scenario 2 - (5+5+5+5+50)/5 = 70/5 = 14

I'm sure I'm probably misunderstanding your thought.

 

[Short Corner]