How to incorporate a level of confidence measure into VBD? (1 Viewer)

[KCC]

As a pure hypothetical, let's assume one does their projections and, based on their projected fantasy points per player and their respective baselines, comes up with the following overall VBD rankings (FYI: these aren't my real rankings - so save the criticisms):

1. Ladainian Tomlinson

2. Steven Jackson

3. Adrian Peterson

4. Brian Westbrook

5. Joseph Addai

6. Randy Moss

7. Frank Gore

8. Tom Brady

9. Marion Barber

10. Peyton Manning

According to VBD theory, I should draft Tomlinson, Jackson, Peterson, Westbrook, and Addai ahead of Moss because I project them to have a greater positive point differential relative to other players at the RB position than Moss would have relative to other players at the WR position. Position depth at RB and WR is also a factor, but you get the general idea. My projections will, of course, be my best guess given the information that I have to work with, but given numerous factors (e.g. LT's age, Jackson's down year last season, ADP's lack of history in the league, etc.) I might have a low level of confidence about how those five will rank at the end of the year relative to one another and relative to other players at the RB position. On the other hand, what if I have a relatively high level of confidence that Moss will finish as the #1 WR and I am convinced that barring injury, Moss will have the best WR stats at the end of the season. Shouldn't there be a way to weight those different levels of confidence and factor that into my my VBD rankings? Also, should "more confidence" necessarily translate into just adding more yards and TDs to a player's projected stats even though using that approach may lead to some unrealistic projections? I'm curious as to whether people see any value in doing this and how to best approach it.

note: I am using "confidence" in the general sense. I am not talking about "confidence intervals" commonly used in statistics.

 

[Jayrod]

As a pure hypothetical, let's assume one does their projections and, based on their projected fantasy points per player and their respective baselines, comes up with the following overall VBD rankings (FYI: these aren't my real rankings - so save the criticisms):1. Ladainian Tomlinson2. Steven Jackson 3. Adrian Peterson4. Brian Westbrook 5. Joseph Addai6. Randy Moss 7. Frank Gore 8. Tom Brady9. Marion Barber10. Peyton ManningAccording to VBD theory, I should draft Tomlinson, Jackson, Peterson, Westbrook, and Addai ahead of Moss because I project them to have a greater positive point differential relative to other players at the RB position than Moss would have relative to other players at the WR position. Position depth at RB and WR is also a factor, but you get the general idea. My projections will, of course, be my best guess given the information that I have to work with, but given numerous factors (e.g. LT's age, Jackson's down year last season, ADP's lack of history in the league, etc.) I might have a low level of confidence about how those five will rank at the end of the year relative to one another and relative to other players at the RB position. On the other hand, what if I have a relatively high level of confidence that Moss will finish as the #1 WR and I am convinced that barring injury, Moss will have the best WR stats at the end of the season. Shouldn't there be a way to weight those different levels of confidence and factor that into my my VBD rankings? Also, should "more confidence" necessarily translate into just adding more yards and TDs to a player's projected stats even though using that approach may lead to some unrealistic projections? I'm curious as to whether people see any value in doing this and how to best approach it.note: I am using "confidence" in the general sense. I am not talking about "confidence intervals" commonly used in statistics.

Something I do to my projections is tweak them based on several factors (especially for dynasty where outside factors can effect future years).Essentially, I have my projected point total, and then I come up with an adjustment % that can be positive or negative based on all of the factors I've included. Each factor can either move the % up or down based on a numerical value I've given it. Some of the ones I use are age, experience, upside, injury risk, etc. I do different things for each position so I can really tailor the factors to what is important to that position.So as an example, I take a running backs age from 30 (what I figure to be the turning point in an RB's career) and multiply that by .5% (or 0.005). Therefore a 27 year old running back would have a 1.5% increase in value. A 32 year old RB would have a 1% decrease in value.I also apply my subjective opinions on upside and injury risk by giving them a value on a 1-5 scale and doing a similar calculation to the age.Sum up all of these % changes and I get my adjustment factor for a given player. I then multiply my projected scores by 1 + the % factor to get the players new value. It's worked great for the last two years.

 

[KCC]

As a pure hypothetical, let's assume one does their projections and, based on their projected fantasy points per player and their respective baselines, comes up with the following overall VBD rankings (FYI: these aren't my real rankings - so save the criticisms):1. Ladainian Tomlinson2. Steven Jackson 3. Adrian Peterson4. Brian Westbrook 5. Joseph Addai6. Randy Moss 7. Frank Gore 8. Tom Brady9. Marion Barber10. Peyton ManningAccording to VBD theory, I should draft Tomlinson, Jackson, Peterson, Westbrook, and Addai ahead of Moss because I project them to have a greater positive point differential relative to other players at the RB position than Moss would have relative to other players at the WR position. Position depth at RB and WR is also a factor, but you get the general idea. My projections will, of course, be my best guess given the information that I have to work with, but given numerous factors (e.g. LT's age, Jackson's down year last season, ADP's lack of history in the league, etc.) I might have a low level of confidence about how those five will rank at the end of the year relative to one another and relative to other players at the RB position. On the other hand, what if I have a relatively high level of confidence that Moss will finish as the #1 WR and I am convinced that barring injury, Moss will have the best WR stats at the end of the season. Shouldn't there be a way to weight those different levels of confidence and factor that into my my VBD rankings? Also, should "more confidence" necessarily translate into just adding more yards and TDs to a player's projected stats even though using that approach may lead to some unrealistic projections? I'm curious as to whether people see any value in doing this and how to best approach it.note: I am using "confidence" in the general sense. I am not talking about "confidence intervals" commonly used in statistics.

Something I do to my projections is tweak them based on several factors (especially for dynasty where outside factors can effect future years).Essentially, I have my projected point total, and then I come up with an adjustment % that can be positive or negative based on all of the factors I've included. Each factor can either move the % up or down based on a numerical value I've given it. Some of the ones I use are age, experience, upside, injury risk, etc. I do different things for each position so I can really tailor the factors to what is important to that position.So as an example, I take a running backs age from 30 (what I figure to be the turning point in an RB's career) and multiply that by .5% (or 0.005). Therefore a 27 year old running back would have a 1.5% increase in value. A 32 year old RB would have a 1% decrease in value.I also apply my subjective opinions on upside and injury risk by giving them a value on a 1-5 scale and doing a similar calculation to the age.Sum up all of these % changes and I get my adjustment factor for a given player. I then multiply my projected scores by 1 + the % factor to get the players new value. It's worked great for the last two years.

Very cool approach, Jayrod. This is just the type of input I was hoping for.

 

[dgreen]

You could make three projections per player: 1) Most Likely, 2) Optimistic, and 3) Pessimistic. Then, assign a weight to each projection. Your final projection could be something like:

(Opp + (4 x ML) + Pess)/6

That makes your ML projection (probably the projection most people use) 67% of you final projection (which happens to approximately equal 1 standard deviation).

So, let's say my ML yardage projection for Portis is 1500 total yards and my pessimistic projection is 1300 yards and my optimistic projection is 2100 yards. Plug all that into the above formula and I end up with 1567 yards for Portis. So, the fact that I see his upside (2100 yards) as being much greater than his downside (1300 yards) provides a little boost (67 yards) to what I see as his likely production (1500 yards).

 

[KCC]

You could make three projections per player: 1) Most Likely, 2) Optimistic, and 3) Pessimistic. Then, assign a weight to each projection. Your final projection could be something like:(Opp + (4 x ML) + Pess)/6That makes your ML projection (probably the projection most people use) 67% of you final projection (which happens to approximately equal 1 standard deviation).So, let's say my ML yardage projection for Portis is 1500 total yards and my pessimistic projection is 1300 yards and my optimistic projection is 2100 yards. Plug all that into the above formula and I end up with 1567 yards for Portis. So, the fact that I see his upside (2100 yards) as being much greater than his downside (1300 yards) provides a little boost (67 yards) to what I see as his likely production (1500 yards).

This is an intriguing approach. When you say the 67% ML projection is roughly one standard deviation, what data set would it be a standard deviation from?

 

[kethnaab]

i just use tiers

if it's someone I have absolute confidence in, I bump them up a tier. IF its someone that I have little faith in, I bump them down a tier

math is nice, but it's all based on estimates anyway.

 

[tomarken]

i just use tiersif it's someone I have absolute confidence in, I bump them up a tier. IF its someone that I have little faith in, I bump them down a tiermath is nice, but it's all based on estimates anyway.

I am actually the biggest math-lover you'll find, but really it all comes down to instinct. I love messing with stats and VBD and whatnot, but you have to remember to stay realistic. When we get to the point where we're subjectively adjusting projections that were already largely subjective, we end up with a meaningless mess of numbers. Instead of multiplying by X% or adding (1-Y)/N or anything, just go with your gut. If you're confident that Moss is a better play than Gore, take Moss. Don't overthink it. You don't need a number to tell you that.I use DraftDominator at all of my drafts, and yet most of the time I don't take the player that DD recommends. It's a good guideline, it helps you keep an eye on relative values and track the draft and whatnot, but in the end I'm not going to pick a guy because it says "VBD 78" next to his name, I'm going to take him because I feel he's the best player available.

 

[inca911]

The projections for a player should take into account the various "factors" and "adjustments" from age, potential injury, mysterious upside, etc. Once you have these "tweaked" projections, you can then calculate a single specific value based on relative player projections vs. a baseline. Adjust your estimates once, not multiple times, and you have a much simpler system. Players for whom you expect a partial season (suspension, PUP, etc) need to be evaluated based on projections that are normalized relative to the number of games they play.

dgreen's approach in establishing an average based on weighted factors is used in project scheduling (PERT), and is a reasonable approach to capture players with larger variabilities in potential outcomes. However, the weighting of the likely value to 66% of the final value used does mean the effort of likely, pessimistic, and optimistic projection have limited benefit compared to the effort and the overall error of the estimates.

VBD in a draft setting definitely has a limited benefit compared to VBD in an auction setting, where each player has a specific value. Tiers work well in drafts where a dozen players pass between picks. In an auction, "saving" $5 on every player you draft is a significant competitive advantage that wins leagues. Is your league a draft or an auction?