AVT vs. Projections (1 Viewer)

[jswalker1981]

I'm debating between Average Value Theory and using Projections to base my VBD board on. And I'm wondering what everyone likes to use and why. Thanks for all your help.

 

[bostonfred]

As the inventor of AVT, I'm obviously partial to it. But really, no tool is going to be perfect. If you think you're better at making projections, go for it. If you think you're better at stack-ranking players, use AVT.

 

[LHUCKS]

Both are fundamentally flawed.

 

[GregR_2]

I'm debating between Average Value Theory and using Projections to base my VBD board on. And I'm wondering what everyone likes to use and why. Thanks for all your help.

MT does a much better job of explaining it, but I'll see if I can't muddle through the explanation until he shows up. To point out the major flaw in AVT with a normal league, let me use an analogy of predicting how many yards each QB will pass for in a single week using AVT or using projections, as the issue is made much clearer with it.If you use AVT, then you'd put Peyton at say, 350 yards, because that's how many yards the best QB averages on any given week, and you have Peyton as your QB1.If you instead project stats for Peyton based on what you think is most likely, you'd know he has a career average of 261 yards per game, and your projection would probably be much closer to this, adjusted up or down slightly for other factors like who he plays.I'm not saying the error in AVT for a season is as big as this analogy. But I'm saying the same general issue exists in a full season that exists here. AVT will overproject Peyton Manning's value because it is giving him QB1's most-likely projection when in fact it should be giving him Peyton Manning's most likely projection. Those two aren't the same. If we had a league where we drafted QB5 or RB13 instead of "Hasselbeck" or "Clinton Portis", AVT would be a great fit. But in a league where we have to draft the guy by name, you want to use the projection for that player, not the projection for where you think he might finish.

 

[Maurile Tremblay]

Both are fundamentally flawed.

I wouldn't say that doing projections is flawed. I would say that drafting strictly off a set of projections without taking other things into account is flawed. But doing projections is still worthwhile.AVT . . . not worthwhile, IMO. (Look for an FBG article on the subject in the near future.)

 

[Maurile Tremblay]

I'm debating between Average Value Theory and using Projections to base my VBD board on. And I'm wondering what everyone likes to use and why. Thanks for all your help.

MT does a much better job of explaining it, but I'll see if I can't muddle through the explanation until he shows up. To point out the major flaw in AVT with a normal league, let me use an analogy of predicting how many yards each QB will pass for in a single week using AVT or using projections, as the issue is made much clearer with it.If you use AVT, then you'd put Peyton at say, 350 yards, because that's how many yards the best QB averages on any given week, and you have Peyton as your QB1.If you instead project stats for Peyton based on what you think is most likely, you'd know he has a career average of 261 yards per game, and your projection would probably be much closer to this, adjusted up or down slightly for other factors like who he plays.I'm not saying the error in AVT for a season is as big as this analogy. But I'm saying the same general issue exists in a full season that exists here. AVT will overproject Peyton Manning's value because it is giving him QB1's most-likely projection when in fact it should be giving him Peyton Manning's most likely projection. Those two aren't the same. If we had a league where we drafted QB5 or RB13 instead of "Hasselbeck" or "Clinton Portis", AVT would be a great fit. But in a league where we have to draft the guy by name, you want to use the projection for that player, not the projection for where you think he might finish.

That is a good explanation.I'll add that the real problem isn't just that the AVT over-projects the top guys at each position (and under-projects the bottom starters), but that the effect is greater at some positions than at others. So it overvalues the top guys at some positions more than it overvalues the top guys at other positions, which screws up inter-positional comparisons.

 

[Maurile Tremblay]

AVT . . . not worthwhile, IMO.

This may be overstating it. I should say "less worthwhile" rather than "not worthwhile" since going by AVT is certainly better than picking among positions randomly.

 

[LHUCKS]

Both are fundamentally flawed.

I wouldn't say that doing projections is flawed.

I wouldn't say that either, I was referring to the statement in the first post "using Projections to base my VBD board on."

 

[bostonfred]

We've all noticed how the number of running backs projected to get "20 carries a week" during the offseason is way out of line with reality. But if you cut back that number simply because you feel you have too many guys with 320 carries, to some extent, you're using the same flawed concepts as AVT.

At the start of the season, there SHOULD be more players projected to get 320 carries than who actually get it, unless you get into the business of predicting injuries. The reality is that some running back is going to struggle because his offensive line gets decimated, another couple will have lingering injuries and underproduce, another will end up having an unexpectedly brutal schedule that means their team can never seem to get the running game going because they're always playing catch up. You can't predict all of those things.

More often than not, people under-project in an attempt to make their projections look more realistic, but that's not necessarily a better predictor than projecting what they could do if everything worked out right.

 

[Maurile Tremblay]

We've all noticed how the number of running backs projected to get "20 carries a week" during the offseason is way out of line with reality.

There are two very different concepts here that should not be confused.1. What's the over/under on the number of RBs who will get at least 20 carries per week? Nobody actually projects this on purpose, as far as I know. I don't see what use it would be. This is the question that AVT gives a good answer to, though.

2. How many RBs are there whose over/under for number carries per week is at least 20? This is what the FBG projections seek to answer (if we take each projection as an over/under, which is close enough for these purposes).

To illustrate how different these questions are, I will use a dice example.

1. If I roll ten six-sided dice, how many should we expect to land on 5 or 6? The answer is 3.33.

2. If I roll ten six-sided dice, how many of them, individually, are as (or more) likely to land on 5 or 6 as they are to land on 1, 2, 3, or 4? The answer is precisely zero.

Only by confusing which type of question we are interested in does AVT start to look relevant.

 

[radballs]

This is the way I see it, so please correct me if I'm wrong or not. I like to first start by taking an individual's historical production and then factor in changes to his situation as well as his growth as an NFL player to come up with individual NFL statistical projections. Repeat for every draftable player.

Then, use AVT as a check on these statistics. How do you do this? By making sure that your individual statistics when you add them all up aren't way out of whack with what has historically happenned as a whole at those positions.

Generally, if I see issues with people's projections, it's that there is no double checking going into their individual player projections as a whole within the league.

In other words, think of player projections as a bottom up procedure to come up with player performance. Think of AVT as a top down check to make sure that you don't have some kind of unreasonable player projection inflation going on with your numbers.

Use AVT as a way to check your work to make sure that the complilation of all of your projections wouldn't propose a historical statistical anomaly. For instance, if you do all of your player projections and then the total of those projections using AVT as a way to double check them, has you showing an 8% increase in the last 3 years of QB production, an increase of 13% in RB production, a decrease of 15% in WR production, etc., then either your projections need to be re-examined or you need a terrific explanation for why you see such large changes in historical overall player projections across the league.

 

[bostonfred]

We've all noticed how the number of running backs projected to get "20 carries a week" during the offseason is way out of line with reality.

There are two very different concepts here that should not be confused.1. What's the over/under on the number of RBs who will get at least 20 carries per week? Nobody actually projects this on purpose, as far as I know. I don't see what use it would be. This is the question that AVT gives a good answer to, though.

2. How many RBs are there whose over/under for number carries per week is at least 20? This is what the FBG projections seek to answer (if we take each projection as an over/under, which is close enough for these purposes).

To illustrate how different these questions are, I will use a dice example.

1. If I roll ten six-sided dice, how many should we expect to land on 5 or 6? The answer is 3.33.

2. If I roll ten six-sided dice, how many of them, individually, are as (or more) likely to land on 5 or 6 as they are to land on 1, 2, 3, or 4? The answer is precisely zero.

Only by confusing which type of question we are interested in does AVT start to look relevant.

I think you misunderstood my post. I wasn't arguing in favor of AVT. I was arguing against people deliberately discounting their projections to make them look more like AVT. Let me see if I can restate it:1. What's the over/under on the number of RBs who will get at least 320 carries per season? This is what AVT tells us. And most projections I see end up projecting pretty close to this number.

2. How many RBs are there whose over/under for number carries per season should be 320? More than the number that will actually get 320. In fact, it could be as many as twice as many and it would still have a positive expectation.

To illustrate how different these questions are, I will use a dice example.

1. If I roll six six-sided dice, how many should I expect to roll greater than 1.5? The answer is five.

2. If I roll six six sided dice, and I predict first five to be greater than one, should I take the over or the under on the sixth? The answer is still the over.

Yet most people look at their projections, then tweak them downwards so that they predict enough ones. That's just as wrongheaded as your AVT example.

 

[Bri]

As the inventor of AVT,

what means this?

 

[Maurile Tremblay]

I think you misunderstood my post. I wasn't arguing in favor of AVT. I was arguing against people deliberately discounting their projections to make them look more like AVT.

I did misunderstand your first post. I agree with your latest one.

 

[Maurile Tremblay]

Then, use AVT as a check on these statistics.

The problem is that good projections shouldn't look anything like AVT. So by using AVT as a check, you're doing yourself a disservice.Good projections for each of six fair six-sided dice will look like this:Red die: 3.5Blue die: 3.5Orange die: 3.5Black die: 3.5White die: 3.5Green die: 3.5AVT numbers, however will look more like thisie1: 5.5Die2: 4.6Die3: 3.8Die4: 3.2Die5: 2.4Die6: 1.5If you check the first numbers against the second and try to "correct" them by making them look more like the second, you're just throwing yourself off. The first numbers are already perfect even though they don't look anything like AVT. There's no reason that the projection for "Red die" should match the projection for "Die1," just like there's no reason that the projection for "LaDainian Tomlinson" should match the projection for "RB1." Matching them up is likely to make your projections less accurate, not more accurate.

 

[radballs]

More often than not, people under-project in an attempt to make their projections look more realistic, but that's not necessarily a better predictor than projecting what they could do if everything worked out right.

You pretty much have to under-project though because there has to be some kind of risk factor (whatever that might be) to discount numbers back toward historical normalcy. I understand your point about how do you discount it, because projections should be based off of assuming a guy is going to produce for you for 16 games. Still, you can use AVT to temper your overall player projections for a given position and you can be much more accurate than someone who assumes that everyone should be projected at 16*(their relevant, historical FPG).

 

[bostonfred]

Then, use AVT as a check on these statistics.

The problem is that good projections shouldn't look anything like AVT. So by using AVT as a check, you're doing yourself a disservice.Good projections for each of six dice will look like this:Red die: 3.5Blue die: 3.5Orange die: 3.5Black die: 3.5White die: 3.5Green die: 3.5AVT numbers, however will look more like thisie1: 5.5Die2: 4.6Die3: 3.8Die4: 3.2Die5: 2.4Die6: 1.5If you check the first numbers against the second and try to "correct" them by making them look more like the second, you're just throwing yourself off. The first numbers are already perfect even though they don't look anything like AVT. There's no reason that the projection for "Red die" should match the projection for "Die1," just like there's no reason that the projection for "LaDainian Tomlinson" should match the projection for "RB1." Matching them up is likely to make your projections less accurate, not more accurate.

Yup. If Ollie's saying he double checks his projections to make sure he doesn't project the die to roll a seven, that's one thing. If he's checking to make sure that he predicts four dice over 2.5 and two dice over 4.5, that's not as good. To get back to football terms, Greg had a good post earlier about how AVT could get you to overproject Peyton Manning. But if you adjust your projections so that you only predict one Tomlinson, one Steven Jackson, one Gore, and so on, you may actually be underprojecting some of the top guys, in the name of making your projections look more realistic.

 

[bostonfred]

More often than not, people under-project in an attempt to make their projections look more realistic, but that's not necessarily a better predictor than projecting what they could do if everything worked out right.

You pretty much have to under-project though because there has to be some kind of risk factor (whatever that might be) to discount numbers back toward historical normalcy. I understand your point about how do you discount it, because projections should be based off of assuming a guy is going to produce for you for 16 games. Still, you can use AVT to temper your overall player projections for a given position and you can be much more accurate than someone who assumes that everyone should be projected at 16*(their relevant, historical FPG).

Can you explain your methodology for this?

 

[Maurile Tremblay]

More often than not, people under-project in an attempt to make their projections look more realistic, but that's not necessarily a better predictor than projecting what they could do if everything worked out right.

You pretty much have to under-project though because there has to be some kind of risk factor (whatever that might be) to discount numbers back toward historical normalcy.

I don't think people should under-project or over-project. I think people should try to project as accurately as possible.I think a good check is to look at league totals. NFL teams rushed for a total of 57,575 yards in 2005 and 60,061 in 2006. So if the total rushing yards you are projecting for all players on all teams isn't somewhere between, say, 55,000 and 62,000, you should probably re-think things.I don't think you should purposely under-project or over-project that range.

I understand your point about how do you discount it, because projections should be based off of assuming a guy is going to produce for you for 16 games.

I don't assume this. I assume there's an X% chance that someone will play 16 games, a Y% chance he'll play 15 games, a Z% chance he'll play 14 games -- and so on -- based in part on league-wide historical averages at a certain position, and in part on an individual player's current health status and (in some cases) prior injury history.

Still, you can use AVT to temper your overall player projections for a given position and you can be much more accurate than someone who assumes that everyone should be projected at 16*(their relevant, historical FPG).

Can you give a concrete example of how this would work?

 

[radballs]

Then, use AVT as a check on these statistics.

The problem is that good projections shouldn't look anything like AVT. So by using AVT as a check, you're doing yourself a disservice.Good projections for each of six fair six-sided dice will look like this:Red die: 3.5Blue die: 3.5Orange die: 3.5Black die: 3.5White die: 3.5Green die: 3.5AVT numbers, however will look more like thisie1: 5.5Die2: 4.6Die3: 3.8Die4: 3.2Die5: 2.4Die6: 1.5If you check the first numbers against the second and try to "correct" them by making them look more like the second, you're just throwing yourself off. The first numbers are already perfect even though they don't look anything like AVT. There's no reason that the projection for "Red die" should match the projection for "Die1," just like there's no reason that the projection for "LaDainian Tomlinson" should match the projection for "RB1." Matching them up is likely to make your projections less accurate, not more accurate.

I feel like I'm the AVT devil's advocate here, but it's a small portion of what I do. I agree that the heavy lifting is really in evaluating individual expected player performance. However, using all of one's knowledge to project individual projections, do you see absolutely zero value in using a simple system that helps you see whether or not your numbers (as a whole) are a bit off factoring in past NFL seasons? In other words, what else can be used to make sure that "your projections" aren't a bit inflated over what would show to be a historical anomaly. I still haven't heard anyone counter the notion that NFL statistics (by team and all season long) can be used to set upper and lower limits as to what YOUR 2007 TOTAL PLAYER PROJECTIONS might be. By the way, I can be easily swayed in this discussion as long as I hear something other than all of the previously mentioned counter arguments. Are there any others?

 

[bostonfred]

More often than not, people under-project in an attempt to make their projections look more realistic, but that's not necessarily a better predictor than projecting what they could do if everything worked out right.

You pretty much have to under-project though because there has to be some kind of risk factor (whatever that might be) to discount numbers back toward historical normalcy.

I don't think people should under-project or over-project. I think people should try to project as accurately as possible.I think a good check is to look at league totals. NFL teams rushed for a total of 57,575 yards in 2005 and 60,061 in 2006. So if the total rushing yards you are projecting for all players on all teams isn't somewhere between, say, 55,000 and 62,000, you should probably re-think things.I don't think you should purposely under-project or over-project that range.

Again, I'd like to know the methodology you use. Let's say you're trying to project ten guys that, if healthy, would all run for 1000 yards. Let's also say that you know that, on average, one will get hurt. Do you project all of them for 900 yards? 1000? Something in between? You could make a case for an expected value calculation here, but let's say I draft the guy that gets injured. I don't really get a zero from his position; I'd just start another guy instead. Let's say you're almost done with your projections, and you have found about 57,000 yards between all the teams in the league. Let's also say that you project Alexander to get 1500 yards if healthy and Morris to get 0, but you project Morris to get 1000 if Alexander gets hurt. How many yards would you project Morris for? Alexander?

 

[radballs]

I'll get back to this thread in a bit.

 

[bostonfred]

I'll get back to this thread in a bit.

Thanks for keeping us updated.

 

[GregR_2]

...

I feel like I'm the AVT devil's advocate here, but it's a small portion of what I do. I agree that the heavy lifting is really in evaluating individual expected player performance. However, using all of one's knowledge to project individual projections, do you see absolutely zero value in using a simple system that helps you see whether or not your numbers (as a whole) are a bit off factoring in past NFL seasons? In other words, what else can be used to make sure that "your projections" aren't a bit inflated over what would show to be a historical anomaly.

I still haven't heard anyone counter the notion that NFL statistics (by team and all season long) can be used to set upper and lower limits as to what YOUR 2007 TOTAL PLAYER PROJECTIONS might be. By the way, I can be easily swayed in this discussion as long as I hear something other than all of the previously mentioned counter arguments. Are there any others?

Radballs, is maybe where this is all getting confused is that you are talking about using leaguewide totals for historical checks, but then referring to using AVT to do it? Because using leaguewide totals as your bounds, and using AVT, would be two unconnected topics.To apply the leaguewide numbers you would never go through the steps you have to do for AVT. If someone talks about using AVT numbers as a historical check, we're all going to assume you are taking the AVT average for RB1, or at least his min/max, and compare them to your RB1. Because that's what AVT produces.

If you're just talking about what MT did about adding up all your rushing yards and comparing them to the NFL's results, which I gather is what you're saying, you don't need to have ever gone through the AVT methodology, and even if you had you'd never use the results to do that kind of historical double check.

So I don't think anyone is disagreeing, I think the role (or lack thereof) of AVT in what is being advocated is maybe where it's getting confusing?

 

[Maurile Tremblay]

However, using all of one's knowledge to project individual projections, do you see absolutely zero value in using a simple system that helps you see whether or not your numbers (as a whole) are a bit off factoring in past NFL seasons?

There would be tremendous value if there were a system that could tell me whether my numbers were a bit off.AVT doesn't do this, unless you can think of a way to do it that I can't.

In other words, what else can be used to make sure that "your projections" aren't a bit inflated over what would show to be a historical anomaly.

As I said in my last post, you can look at historical league-wide totals, as a first step, to see if you're over- or under-projecting across the board.

I still haven't heard anyone counter the notion that NFL statistics (by team and all season long) can be used to set upper and lower limits as to what YOUR 2007 TOTAL PLAYER PROJECTIONS might be.

Why would anyone want to counter that notion? Seems perfectly reasonable to me.

By the way, I can be easily swayed in this discussion as long as I hear something other than all of the previously mentioned counter arguments. Are there any others?

The previously mentioned counter-arguments are iron-clad. What else are you looking for?

 

[Maurile Tremblay]

Radballs, is maybe where this is all getting confused is that you are talking about using leaguewide totals for historical checks, but then referring to using AVT to do it? Because using leaguewide totals as your bounds, and using AVT, would be two unconnected topics.

That could be it.

 

[Kleck]

As the inventor of AVT...

ORLY? Clayton advised me to think up a a catchy three word slogan, but no, I wouldn't listen. One year after I wrote that article folks started throwing around the term AVT.

Actually, I still feel that old article is valid and gives a solid purpose for averaging statistics just to check and see if your projections are realistic. It does a great job farther down your cheatsheets, but no so much near the top.

 

[Maurile Tremblay]

From Shick!'s article:

The #20 RB since 1990 has averaged about 142 points per season. Does your #20 RB in your projections for the upcoming season score about 142 points? Better yet, should your #20 RB be projected to score 142 points? That’s a tough one. Wish I knew the answer to that one…

For the #20 RB, it is kind of a tough one. For the #1 RB, however, it is an easy question. It is obvious that the projection for RB1's points should be higher than the projection for LaDainian Tomlinson's points. (Tomlinson can't possibly outscore RB1, but he can fall short of RB1. So RB1 is free-rolling against Tomlinson.)

 

[Chase Stuart]

From Shick!'s article:

The #20 RB since 1990 has averaged about 142 points per season. Does your #20 RB in your projections for the upcoming season score about 142 points? Better yet, should your #20 RB be projected to score 142 points? That’s a tough one. Wish I knew the answer to that one…

For the #20 RB, it is kind of a tough one. For the #1 RB, however, it is an easy question. It is obvious that the projection for RB1's points should be higher than the projection for LaDainian Tomlinson's points. (Tomlinson can't possibly outscore RB1, but he can fall short of RB1. So RB1 is free-rolling against Tomlinson.)

That statement is probably a bit stronger than you mean. It's certainly possible that your LT projections > #1RB projections, at least in theory.But yea, everything else you've written about AVT =

 

[Maurile Tremblay]

From Shick!'s article:

The #20 RB since 1990 has averaged about 142 points per season. Does your #20 RB in your projections for the upcoming season score about 142 points? Better yet, should your #20 RB be projected to score 142 points? That’s a tough one. Wish I knew the answer to that one…

For the #20 RB, it is kind of a tough one. For the #1 RB, however, it is an easy question. It is obvious that the projection for RB1's points should be higher than the projection for LaDainian Tomlinson's points. (Tomlinson can't possibly outscore RB1, but he can fall short of RB1. So RB1 is free-rolling against Tomlinson.)

That statement is probably a bit stronger than you mean. It's certainly possible that your LT projections > #1RB projections, at least in theory.But yea, everything else you've written about AVT =

I mean that projections for LT in 2007 must necessarily be lower than projections for RB1 in 2007.LT's projection may be higher than the historical RB1 numbers, but that has to do with LT as an individual, so it involves making an individual projection instead of just going by AVT. That's different.

 

[Bri]

As the inventor of AVT...

ORLY? Clayton advised me to think up a a catchy three word slogan, but no, I wouldn't listen. One year after I wrote that article folks started throwing around the term AVT.

You? Carlton?Someone in RSFF used to throw out stats like that 19th RB ran for this many yards, 23rd WR had this many receiving etc. amidst discussions.

I thought it was MB.

Anyhow, I don't think it's so unique that others couldn't have also thought they invented such a theory. Same with projections and lots of stuff in FF. Like that FF was invented by someone in the midwest theory. We just played rotisere baseball but for football and never heard of that guy. We invented it too then and so did thousands of other people. If it's a logical obvious step to take, alot of people will take it. "Invent" doesn't seem like the right word but I have no clue what is.

BTW nice to see ya around

 

[Sonny Lubick Blowup Doll]

I'll get back to this thread in a bit.

Thanks for keeping us updated.

 

[deadstick]

I use both - projections, averaged with AVT over the past 3 years (so if my #20 RB is projected to score way more than the #20 RB has in the past 3 years - he gets brought down in comparison to other positions, but is still the #20 RB - helps keep VBD in check). I've been doing this for 15 years.

 

[cmigs]

Coming to this discussion late, but I've been starting to think about projections methods for next year.

One thing that got me interested in AVT - in all the articles written about it, the projections for players end up being several fewer standard deviations away from the actual stats than expert projections. However, these articles are all relatively old (most recent had stats from maybe 04). Has anyone done anything more recent comparing AVT to projections?

 

[GregR_2]

Coming to this discussion late, but I've been starting to think about projections methods for next year.One thing that got me interested in AVT - in all the articles written about it, the projections for players end up being several fewer standard deviations away from the actual stats than expert projections. However, these articles are all relatively old (most recent had stats from maybe 04). Has anyone done anything more recent comparing AVT to projections?

To save time in hunting, can you link to the articles you're referring to? ETA: Yes I know you're asking about new ones, which I don't know if there are any or not, but I'm interested in seeing these old ones.

 

[bostonfred]

As the inventor of AVT, I'm obviously partial to it. But really, no tool is going to be perfect. If you think you're better at making projections, go for it. If you think you're better at stack-ranking players, use AVT.

I just wanted to clarify that I was kidding here.

 

[cmigs]

Coming to this discussion late, but I've been starting to think about projections methods for next year.

One thing that got me interested in AVT - in all the articles written about it, the projections for players end up being several fewer standard deviations away from the actual stats than expert projections. However, these articles are all relatively old (most recent had stats from maybe 04). Has anyone done anything more recent comparing AVT to projections?

To save time in hunting, can you link to the articles you're referring to? ETA: Yes I know you're asking about new ones, which I don't know if there are any or not, but I'm interested in seeing these old ones.

http://www.footballguys.com/avt.htmhttp://forums.footballdiehards.com/index.php?showtopic=12497

 

[fatkid]

I understand the problem with using AVT versus projections as the following.

Projections are a reflection of how you think players will do for a season, and generally you don't really factor in injuries very much. AVT factors in those injuries, by accounting for average injuries year by year.You know someone like Larry Johnson or Ronnie Brown will get hurt, but you don't know who. If you use AVT, you lower your projection for the 15th or 20th ranked player, but they may not be the one getting hurt.

Assuming that is correct so far, does projecting only Points Per Game and using AVT for Point Per Game comparison increase our accuracy? The PPG stats should be relatively protected from major injuries. Has anyone looked at how consisten PPG is from year to year?

 

[Maurile Tremblay]

The problem with AVT really doesn't have much of anything to do with injuries.

Go back to GregR's example of projecting Manning for 350 yards in a game instead of 266 yards. Nothing to do with injury.

 

[cmigs]

The problem with AVT really doesn't have much of anything to do with injuries.Go back to GregR's example of projecting Manning for 350 yards in a game instead of 266 yards. Nothing to do with injury.

I don't think it's that simple. First, why exactly would AVT project anyone at 350 a game? No one has averaged that the last few years - not even Brady, who would be the outlier that would pull the QB1 projection up for next year in terms of AVT.Second, and more to the heart of the question, the argument implicitly assumes the quality of projections. To rephrase the above argument - "why would I use a system that randomly assigns player values when I know EXACTLY how someone will perform?" The problem, of course, is that you don't know how players will perform. AVT simply produces a competing set of projections, which, at least in the articles I've seen, has been closer to actual performance.I'm actually willing to believe that projections have gotten better over the last few years (due to increased scrutiny and stronger statistical analysis), but I'd have to see it to believe it. Does anyone want to suggest a simple way to study this? I've got all off season to put in some work, but I'm not a math guy, so I'm not sure what I'd want to be looking for. Maybe standard deviation in the top 100 players?

 

[Maurile Tremblay]

First, why exactly would AVT project anyone at 350 a game? No one has averaged that the last few years - not even Brady, who would be the outlier that would pull the QB1 projection up for next year in terms of AVT.

The top passer in a given week will generally throw for about 350 yards. So if you wanted to do weekly projections AVT style, you'd project whoever your #1 QB is that week to throw for ~350 yards.

Second, and more to the heart of the question, the argument implicitly assumes the quality of projections.

Anti-AVT arguments don't have to assume anything about the quality of projections. I can show that APVT works better than AVT, and neither relies on projections any more than the other does. (While AVT projects the top-rated RB to get what the EOY RB1 typically gets, APVT projects the top-rated RB to get what the top-rated RB [as of the preseason] usually gets.)

AVT simply produces a competing set of projections, which, at least in the articles I've seen, has been closer to actual performance.

Do you have a link to any such article?

I'm actually willing to believe that projections have gotten better over the last few years (due to increased scrutiny and stronger statistical analysis), but I'd have to see it to believe it. Does anyone want to suggest a simple way to study this? I've got all off season to put in some work, but I'm not a math guy, so I'm not sure what I'd want to be looking for. Maybe standard deviation in the top 100 players?

How about summing the total errors or summing the squares of the errors? (Or both, since once you've done one the other is easy.) Another skeptical message board poster did that in a previous offseason thread and found that projections were more accurate than AVT for that year; but it never hurts to repeat an experiment.

 

[cmigs]

First, why exactly would AVT project anyone at 350 a game? No one has averaged that the last few years - not even Brady, who would be the outlier that would pull the QB1 projection up for next year in terms of AVT.

The top passer in a given week will generally throw for about 350 yards. So if you wanted to do weekly projections AVT style, you'd project whoever your #1 QB is that week to throw for ~350 yards.

Oh, agreed, I wouldn't do weekly projections using AVT for exactly that reason. But season projections would cover this problem.

Anti-AVT arguments don't have to assume anything about the quality of projections. I can show that APVT works better than AVT, and neither relies on projections any more than the other does. (While AVT projects the top-rated RB to get what the EOY RB1 typically gets, APVT projects the top-rated RB to get what the top-rated RB [as of the preseason] usually gets.)

AVT simply produces a competing set of projections, which, at least in the articles I've seen, has been closer to actual performance.

Do you have a link to any such article?

As per my above post:http://www.footballguys.com/avt.htm

http://forums.footballdiehards.com/index.php?showtopic=12497

Also, what is APVT?

I'm actually willing to believe that projections have gotten better over the last few years (due to increased scrutiny and stronger statistical analysis), but I'd have to see it to believe it. Does anyone want to suggest a simple way to study this? I've got all off season to put in some work, but I'm not a math guy, so I'm not sure what I'd want to be looking for. Maybe standard deviation in the top 100 players?

How about summing the total errors or summing the squares of the errors? (Or both, since once you've done one the other is easy.) Another skeptical message board poster did that in a previous offseason thread and found that projections were more accurate than AVT for that year; but it never hurts to repeat an experiment.

I'll give it a try over the next few weeks.Thanks for the dialogue, it's been enlightening.

 

[Maurile Tremblay]

First, why exactly would AVT project anyone at 350 a game? No one has averaged that the last few years - not even Brady, who would be the outlier that would pull the QB1 projection up for next year in terms of AVT.

The top passer in a given week will generally throw for about 350 yards. So if you wanted to do weekly projections AVT style, you'd project whoever your #1 QB is that week to throw for ~350 yards.

Oh, agreed, I wouldn't do weekly projections using AVT for exactly that reason. But season projections would cover this problem.

No, the exact same problem exists. It's just easier to intuitively grasp in weekly projections for some people.

Anti-AVT arguments don't have to assume anything about the quality of projections. I can show that APVT works better than AVT, and neither relies on projections any more than the other does. (While AVT projects the top-rated RB to get what the EOY RB1 typically gets, APVT projects the top-rated RB to get what the top-rated RB [as of the preseason] usually gets.)

AVT simply produces a competing set of projections, which, at least in the articles I've seen, has been closer to actual performance.

As per my above post:http://www.footballguys.com/avt.htm

http://forums.footballdiehards.com/index.php?showtopic=12497

Thanks, but neither of those article measure the accuracy of a given set of projections compared to the accuracy of AVT.

Also, what is APVT?

Let's say that historically, the top RB as measured by end-of-year results averages 350 fantasy points. Let's also say that, historically, whichever RB is ranked #1 on most people's cheat sheets before the season starts averages 300 fantasy points.Let's say that during the 2008 preseason, LaDainian Tomlinson is ranked #1 on most people's cheat sheets.

AVT would project Tomlinson for 350 fantasy points, while APVT would project Tomlinson for 300 fantasy points.

Thanks for the dialogue, it's been enlightening.

Sure thing.

 

[BusterTBronco]

This is a flawed argument. You don't use AVT for a single game, you use it for a full season. The week to week variances are baked into the AVT number for the season.

If AVT projects that QB1 will have 4200 yards passing for the season and you believe Peyton Manning to be QB1 going into next season, it is perfectly reasonable to project that Manning will have 4200 yards.

 

[Maurile Tremblay]

If AVT projects that QB1 will have 4200 yards passing for the season and you believe Peyton Manning to be QB1 going into next season, it is perfectly reasonable to project that Manning will have 4200 yards.

If you project Manning to get 4200 yards, how many yards would you project QB1 to get?If Vegas were making a line for who would get more passing yards on the season between Manning and QB1, who would be the favorite (or would they be even)?I submit that QB1 would be a huge favorite over Manning, and rightly so. If that's true, shouldn't our projected passing yards for QB1 be higher than they are for Manning?

 

[Bri]

If AVT projects that QB1 will have 4200 yards passing for the season and you believe Peyton Manning to be QB1 going into next season, it is perfectly reasonable to project that Manning will have 4200 yards.

If you project Manning to get 4200 yards, how many yards would you project QB1 to get?If Vegas were making a line for who would get more passing yards on the season between Manning and QB1, who would be the favorite (or would they be even)?I submit that QB1 would be a huge favorite over Manning, and rightly so. If that's true, shouldn't our projected passing yards for QB1 be higher than they are for Manning?

I understood that as Manning was QB1, not two separate players

 

[Maurile Tremblay]

If AVT projects that QB1 will have 4200 yards passing for the season and you believe Peyton Manning to be QB1 going into next season, it is perfectly reasonable to project that Manning will have 4200 yards.

If you project Manning to get 4200 yards, how many yards would you project QB1 to get?If Vegas were making a line for who would get more passing yards on the season between Manning and QB1, who would be the favorite (or would they be even)?I submit that QB1 would be a huge favorite over Manning, and rightly so. If that's true, shouldn't our projected passing yards for QB1 be higher than they are for Manning?

I understood that as Manning was QB1, not two separate players

"Peyton Manning" means Peyton Manning."QB1" means whoever ends up being the top QB at the end of the year.Maybe Peyton Manning will be QB1 and maybe he won't be. But if your projections for Manning are the same as your projections for QB1, you're implying that there's a 100% chance that Manning will finish as QB1 -- which is obviously not realistic. Maybe he has a 25% chance, but it's far from 100%.Therefore, your projections for QB1 ought to be higher than your projections for Manning.AVT sets them equal to each other, which is conceptually wrong.

 

[Bri]

If AVT projects that QB1 will have 4200 yards passing for the season and you believe Peyton Manning to be QB1 going into next season, it is perfectly reasonable to project that Manning will have 4200 yards.

If you project Manning to get 4200 yards, how many yards would you project QB1 to get?If Vegas were making a line for who would get more passing yards on the season between Manning and QB1, who would be the favorite (or would they be even)?I submit that QB1 would be a huge favorite over Manning, and rightly so. If that's true, shouldn't our projected passing yards for QB1 be higher than they are for Manning?

I understood that as Manning was QB1, not two separate players

"Peyton Manning" means Peyton Manning."QB1" means whoever ends up being the top QB at the end of the year.Maybe Peyton Manning will be QB1 and maybe he won't be. But if your projections for Manning are the same as your projections for QB1, you're implying that there's a 100% chance that Manning will finish as QB1 -- which is obviously not realistic. Maybe he has a 25% chance, but it's far from 100%.Therefore, your projections for QB1 ought to be higher than your projections for Manning.AVT sets them equal to each other, which is conceptually wrong.

:head spinning:Thanks for clarifyingYou projection guys are so wacky it's fascinatingI don't know why you a math wiz, do them. Maybe it's the theory of it all and figuring out the probability with tangible numbers. I just figure you'd be someone to recognize unpredictability and roll with something simpler.Preseason projections for the most part are quite inaccurate.Preseason gut rankings for the most part are quite inaccurate.Everyone does "their thing" I just am surprised you go with the one that's so time consuming when the results are similarly flawed.Anyhow, rambling, it's fascinating. The lengths to prove or disprove theories on performance(projections) that are destinned to fail anyway is...just fascinating.

 

[Maurile Tremblay]

You projection guys are so wacky it's fascinating

I don't know why you a math wiz, do them. Maybe it's the theory of it all and figuring out the probability with tangible numbers. I just figure you'd be someone to recognize unpredictability and roll with something simpler.

Preseason projections for the most part are quite inaccurate.

Preseason gut rankings for the most part are quite inaccurate.

Everyone does "their thing" I just am surprised you go with the one that's so time consuming when the results are similarly flawed.

I give some reasons here:
[T]here are some good reasons for doing your own projections.

First, a big part of being a successful fantasy owner is staying informed about NFL teams and players. Doing your own projections will force you to look closely at each team's situation this year, including any changes in coaching or personnel that you may have otherwise missed. Most of us don't have the discipline to go through all thirty-two NFL teams, one by one, and think about how the passing yards are likely to be distributed among the tight ends and wide receivers, given the impact of this year's first-round draft pick or big free-agent acquisition or whatever. I guarantee you that if you sit down and project out numbers for each player on each team, you will know a lot more after the exercise than you did before. In fact, I believe the process of doing projections is more important than the results. After doing all your projections, you could throw them in the garbage and never look at them again, and it would still have been a tremendously worthwhile exercise. In going through the process of looking closely at last year's performance by each team and then considering how things might change this year, you will have picked up much more useful information along the way than you would have by simply reading other people's opinions.

Second, doing your own projections may force you to appropriately reconsider any self-contradictory opinions you currently hold about certain teams. For example, if you see the Cardinals' WRs and RBs catching 25 touchdowns as a group, but believe that the Cardinals' QBs will throw for only 18 touchdowns . . . something has to give. It is a sign that you don't have a clear grasp of what you think the Cardinals' offense will accomplish this season, and your incoherent view of the matter may never be discovered (or remedied) if you do not write out your projections.

 

[Bri]

You projection guys are so wacky it's fascinating

I don't know why you a math wiz, do them. Maybe it's the theory of it all and figuring out the probability with tangible numbers. I just figure you'd be someone to recognize unpredictability and roll with something simpler.

Preseason projections for the most part are quite inaccurate.

Preseason gut rankings for the most part are quite inaccurate.

Everyone does "their thing" I just am surprised you go with the one that's so time consuming when the results are similarly flawed.

I give some reasons here:
[T]here are some good reasons for doing your own projections.

First, a big part of being a successful fantasy owner is staying informed about NFL teams and players. Doing your own projections will force you to look closely at each team's situation this year, including any changes in coaching or personnel that you may have otherwise missed. Most of us don't have the discipline to go through all thirty-two NFL teams, one by one, and think about how the passing yards are likely to be distributed among the tight ends and wide receivers, given the impact of this year's first-round draft pick or big free-agent acquisition or whatever. I guarantee you that if you sit down and project out numbers for each player on each team, you will know a lot more after the exercise than you did before. In fact, I believe the process of doing projections is more important than the results. After doing all your projections, you could throw them in the garbage and never look at them again, and it would still have been a tremendously worthwhile exercise. In going through the process of looking closely at last year's performance by each team and then considering how things might change this year, you will have picked up much more useful information along the way than you would have by simply reading other people's opinions.

Second, doing your own projections may force you to appropriately reconsider any self-contradictory opinions you currently hold about certain teams. For example, if you see the Cardinals' WRs and RBs catching 25 touchdowns as a group, but believe that the Cardinals' QBs will throw for only 18 touchdowns . . . something has to give. It is a sign that you don't have a clear grasp of what you think the Cardinals' offense will accomplish this season, and your incoherent view of the matter may never be discovered (or remedied) if you do not write out your projections.

good explanationI disagree on some but only in that I do things my way, you do things yours. Everything's cool, good post/article.

If you do your thing and draft Brady who wins you a title, I do my thing and draft Brady who wins me a title, and some little girl drafts Brady because she thinks he's cute and wins a title, it all gets the same result. I think I'll always find that fascinating