article slightly knocking vbd (3 Viewers)

[jswalker1981]

I use VBD, AVT, Tiering, and ADP together. I use the AVT to come up with projections (I just can't seem to trust projections, I know they will be wrong at some point, and the AVT seems to be wrong, but at a smaller level). I then make up tiers based on the projections, I break it up into groups of 32 points. That makes the top guy in the tier scoring 2 points more a week then the last guy. After I have my tiers, I make the VBD number, and then clear out the projections column. I don't want to look at that number during the draft. It is a huge distraction, and takes away from the importance of the VBD number. I then use the ADP to figure out which round a player is going in. I figure if I do it that way, I won't be so crazy about where a player goes at which pick. If I have Gates going in the third, and he goes 3.1 instead of 3.6, I'm not going to care. But I do know that I use the VBD number a lot. Lately I've been doing some mock drafts, and I am placing myself in to either the first or the last pick. Because those picks need to have the best value I can get. And you don't know who is going to be there the next time you pick. So each pick has to really count. And that is where VBD is so very important. When you need to make a clutch pick, the best value is what you should get, and the VBD tells you who that pick should be.

 

[inca911]

Lame article.

"Easily 99% of the time, the projections you feed into the formulas and use to rank the players will be incorrect." As if projections must be 100.0% accurate to be of value?

"...the more information you have the better your chances are...." But naturally you should just use these guessed-at tiers and you'll be better off.

That article was hardly worth the effort to read.

 

[SSOG]

I have a couple of problems with projections, some of which have been already been articulated, and some of which I haven't seen satisfactorily expressed.

My first big problem with projections is the degree of specificity that some people take it to. Personally, I don't see anything wrong with saying "Player X will get about 80 catches for 1350 or so yards and 7 or 8 TDs", but I think its silly when people project 82 catches for 1353 yards. Why not 1354 yards? What if he gets tackled early on a catch and only gets 1352? In science, there's a concept of "significant figures" that basically says that we can only know things to a certain degree of certainty, and that we shouldn't overrepresent that certainty- and that's what I think a lot of projectors do, they overrepresent their certainty in their own projections.

My second big problem with coming up with a set of projections is it suggests that there is really only a single possible outcome for the season. It's all well and good to say "Tomlinson will get 1500 yards and 20 TDs", but what if I don't feel like that? Maybe I think Tomlinson could get as few as 1200/12 if he's hungover from last season, or as much as 1800/26 if he keeps it rolling strong? What if I think that Tomlinson is a larger injury risk than the rest of the RBs due to his extreme workload over the past several seasons? With a single set of projections, there is no really good way to express that uncertainty. I could decide how likely each outcome is and then compile a weighted average of all of the projections, but that doesn't really encapsulate everything I'm looking to cover. In the early rounds, I'd much rather take an RB with a lower risk factor, whereas in the later rounds I'd rather gamble based on upside, risks be damned.

My general workaround is to compile several different sets of projections- a floor, a ceiling, and a "most likely outcome" projection, all of which ignore injury risk, and then add notes in the notes section regarding specific red flags or injury concerns that a player might have (including how many carries an RB had during the last year, last three years, and entire career). That way I can switch back and forth between the various projections as the situation demands. I also always keep in a little bit of leeway to go "off book" if I really feel that I have to get someone for reasons I can't quite articulate.

 

[PainTrain]

I modified the VBD sheet a bit to allow me to hedge the list it generates to the ADP by a percentage I specify. It's a good way to spot the players that are valued much differently than VBD would suggest.

I posted this in another thread earlier, but it seemed pertinent to this discussion, so if you want to give it a test drive, I'd love to hear what you think:

ADP meets VBD

(it takes a couple seconds for the download link to appear, but it'll come up before you know it)

If you change the percentage in the yellow box, it updates the VBD list dynamically, shifting players by the amount of weight you want their ADP to carry.

 

[Maurile Tremblay]

My first big problem with projections is the degree of specificity that some people take it to. Personally, I don't see anything wrong with saying "Player X will get about 80 catches for 1350 or so yards and 7 or 8 TDs", but I think its silly when people project 82 catches for 1353 yards. Why not 1354 yards?

Rounding everything to the nearest five or ten yards would be an extra step that I don't see as being worthwhile. Rounding in that manner may not make a set of projections less accurate to any noticeable extent, but neither would it make a set of projections more accurate, so what's the point?I disagree that people will infer greater certainty or greater accuracy from greater precision when it comes to projections. I think everyone understands that a projection of 1353 yards is not likely to end up being exactly correct.

My projections for rushing yards are based on projected yards per rush times projected number of rushes, and the answer isn't always a nice round number. Rounding to the nearest whole yard improves readability, but I don't think rounding any further than that has any big advantages.

My second big problem with coming up with a set of projections is it suggests that there is really only a single possible outcome for the season. It's all well and good to say "Tomlinson will get 1500 yards and 20 TDs", but what if I don't feel like that? Maybe I think Tomlinson could get as few as 1200/12 if he's hungover from last season, or as much as 1800/26 if he keeps it rolling strong? What if I think that Tomlinson is a larger injury risk than the rest of the RBs due to his extreme workload over the past several seasons? With a single set of projections, there is no really good way to express that uncertainty. I could decide how likely each outcome is and then compile a weighted average of all of the projections, but that doesn't really encapsulate everything I'm looking to cover. In the early rounds, I'd much rather take an RB with a lower risk factor, whereas in the later rounds I'd rather gamble based on upside, risks be damned.

Coming up with a set of projections does not suggest that only a single outcome is possible.It is true that projections don't encapsulate everything you're looking for. Information is lost on the way from "50% chance of 1800 yards plus 50% chance of 1200 yards" to "1500 yards." Moreover, that information can be important for precisely the reasons you mention.

But that's not a legitimate argument against doing projections. Just like the fact that a list of bye weeks doesn't encapsulate everything you'll need to dominate your draft isn't a good argument against compiling a list of bye weeks. The key is to use all the tools and information available to you.

So don't rely on projections or VBD exclusively. You should take the safer guys in the early rounds and then gamble more in the later rounds; and you won't know which guys are which if all you look at is a set of projections. So make sure that's not all you look at.

My general workaround is to compile several different sets of projections- a floor, a ceiling, and a "most likely outcome" projection, all of which ignore injury risk, and then add notes in the notes section regarding specific red flags or injury concerns that a player might have (including how many carries an RB had during the last year, last three years, and entire career). That way I can switch back and forth between the various projections as the situation demands. I also always keep in a little bit of leeway to go "off book" if I really feel that I have to get someone for reasons I can't quite articulate.

Your way falsely suggests that there are only three possible outcomes. Seriously, your approach is great, and IMO identifies the biggest advantage to doing projections. When you go through each player on each team and come up with a projection for him, you are forced to think about best-case scenarios, worst-case scenarios, how the offense may be different this year from last year, etc, etc. You get a feel for the player's prospects for the season -- one that is informed by all the factors you are systematically taking into account. When you do projections, you don't just end up with a final number like 1353 yards that you can stick into a VBD App. You end up with a final number plus all the information you've gathered and internalized while doing that projection.

As others have mentioned, doing projections isn't the only way to develop a well informed feel for each player's prospects for the season. But it is one way, and IMO is a pretty good one.

 

[SSOG]

Your way falsely suggests that there are only three possible outcomes.

3 possible outcomes is still closer to reality than 1 possible outcome. :XAnyway, I wasn't trying to disparage projections, as I am a huge fan of projections (as you can see from my post, I do a whole lot of them), I was just saying that the way that most people do them is, in my opinion, pretty shallow. They attach too much specificity to the individual projections (I also do the ypc * carries thing, I just like to round off to nice, even, easily comparable numbers, operating under the assumption that I'm not losing any specificity since I never had that extra degree of specificity to begin with), and at the same time, they don't get enough specificity with their projections as a whole (identifying only one possible outcome without dealing with all of the other possibilities).Mostly though, at this point, I'm just rambling, and as soon as I come up with a flimsy excuse to say specificity another time or two, I'll wrap this post up and hit submit. Well look at that, I just said specificity again.

 

[GregR_2]

I didn't think much of the article. Few articles criticizing VBD show an understanding of VBD, and this article definitely wasn't any different. Like most of the others that have preceded it, he spends most of his time criticizing a flawed concept of how VBD should be used. That's sort of like me criticizing the hammer because it isn't good at removing a screw.

This is a pretty common thing and we see it in discussions here on the board time and again. I think some of the problem actually goes back to the 2 main VBD articles themselves because I don't think they stress enough that VBD is one piece of input to the decision and not the decision itself.

I think most expert users of it have come to this conclusion while people who are newer to it don't truly understand the role it should play.

I did a little study a few years back comparing different methods used to guide draft picks. I compared "best player available" (BPA) based on total points, drafting straight off a VBD cheatsheet, drafting based on a single round's Dynamic VBD drop-off (where you take the position that drops in value the most from this round to the next), and a complete brute force approach that checked every combination of players available at the team's current and future picks for the best team.

VBD and DVBD were both pretty close, and both far better than BPA. But the brute force approach beat them all by a pretty significant margin, as we should expect it to.

When you come down to it the factors that affect our team in the draft are the supply of players and the distribution of points within them, and the demand of the drafting teams for players at a position at different times of the draft. VBD is a great tool at providing the first half of that, what the value is in the pool of players, and how it changes across positions and within positions.

But VBD knows nothing about the order in which players will be selected. That's obvious in that you don't take in round 1 a player you know will be there in round 3. But it also is important on a more subtle level, where the QB you take now and the WR you'd take next round you want to outscore the pairing you'd have gotten if you took the WR this round and the QB next round.

This portion of the draft decision due to demand is what Dynamic VBD (DVBD) deals with by looking at which position drops in value the most from now to a future pick. But unfortunately it also falls short in how most use it in that it only can give you snapshots between 2 particular rounds when what you really need to maximize your team is to simulatneously consider the dropoffs between all rounds in the draft and your pick now. In short, to get the best team possible you need to consider what will happen at every single pick based on who you pick now. And neither VBD nor DVBD do that. And doing so for an entire draft is far from a trivial mathematical exercise.

Let me bring this back around to my original point by saying I think the best draft strategy is the line of thinking that I see employed in Dodds series of "Perfect Draft" articles, because that is really what deals with both the supply and demand problem. It does the best I think we can do short of having a program to brute force check every lineup, in that it helps us understand the ramifications of our decision now on the rest of our draft.

You use VBD and compare it to average draft position data modified for how you think your league will draft, and determine where there are pockets of extra value in the draft. That "extra value" may be a player you think is going too late, or it might be the ability to pick a player at the top of a run so you get the best guy from your tier and the next time your pick rolls around, no one else at the next position you plan to take has been taken. Once you understand where the value exists you can try to plan a path to take advantage of the value to maximize your team.

That's the heart of the Perfect Draft article, and really I think the VBD articles should be redone to highlight VBD's use in such a system. Then once you have your Perfect Draft view of the value in the draft combining both the supply of players and how your league will deplete them, really what it comes down to is going through multiple mock drafts on your own to test different hypothesis and see what the results are. If I wait to take a 2nd RB, what effect does it have? Does it cause me to take my RB3 earlier than I would have otherwise, and if so how many points does it cost me at other positions? What happens if a player doesn't slide to my pick that I was counting on, what are my best options to recover from being wrong? Having these questions answered BEFORE the draft so you can make informed decisions is the key to good drafting.

And VBD plays the role of contributing to those decisions, but I think the articles don't do a strong enough job of pointing out that they shouldn't make the decisions, that the guy with the highest X-value, even early in the draft, is no more than the best guy to pick for a 1 man team. That you need to use the VBD info as input for an understanding of the different paths your team can take through the draft to truly get the best team possible.

 

[AB in DC]

This portion of the draft decision due to demand is what Dynamic VBD (DVBD) deals with by looking at which position drops in value the most from now to a future pick. But unfortunately it also falls short in how most use it in that it only can give you snapshots between 2 particular rounds when what you really need to maximize your team is to simulatneously consider the dropoffs between all rounds in the draft and your pick now. In short, to get the best team possible you need to consider what will happen at every single pick based on who you pick now. And neither VBD nor DVBD do that. And doing so for an entire draft is far from a trivial mathematical exercise.Let me bring this back around to my original point by saying I think the best draft strategy is the line of thinking that I see employed in Dodds series of "Perfect Draft" articles, because that is really what deals with both the supply and demand problem. It does the best I think we can do short of having a program to brute force check every lineup, in that it helps us understand the ramifications of our decision now on the rest of our draft.You use VBD and compare it to average draft position data modified for how you think your league will draft, and determine where there are pockets of extra value in the draft. That "extra value" may be a player you think is going too late, or it might be the ability to pick a player at the top of a run so you get the best guy from your tier and the next time your pick rolls around, no one else at the next position you plan to take has been taken. Once you understand where the value exists you can try to plan a path to take advantage of the value to maximize your team.

Well said. This is exactly what I was trying to say in a freelance article a few years ago, but you've said it much better than I did.