Fantasy Football - Footballguys Forums
This is a sample guest message. Register a free account today to become a member! Once signed in, you'll be able to participate on this site by adding your own topics and posts, as well as connect with other members through your own private inbox!
Baseline for VBD projections: Why not use means (1 Viewer)
- Thread starter KCC
- Start date
It' could be an outlier based on it's distance from the mean, but either way it is the lowest (albeit, projected) score in the distribution. Your baseline is then based on ignoring most of the data (the other 11 scores) and focussing only on the extreme lowest score.
Last edited by a moderator:
Bottomfeeder Sports
Footballguy
Any average that includes the top players at a position are going to be devalued when the top tier is included in the average and there is a large gap between those players and the next tier. Basically an average waters down that gap by dividing it by the number of players in the average.
Bottomfeeder Sports
Footballguy
By the way last starter's problem is that it ignores you also have backups locked up on other team rosters. The ideal static baseline is somewhere among the backups. Doing an average puts the baseline among the starters.
and you're right, in a case like that, it probably wouldn't make a huge difference which method you use. I'm just trying to look for potential problems with either method, so i'm focussing more on downsides to either approach.
CalBear
Footballguy
There are a lot of different ways to look at baselines. Fundamentally, changing the baseline only changes your rankings based on the difference in the falloff curve for the different positions. For example, if you have a situation where RB1=200 VBD points, RB12=50, RB24=0, QB1=150, QB6=50, QB12=0, median-starter will value RB1 more than QB1 relative to worst-starter. But if RB12=100 (linear fall-off), the two systems will give identical values for RB1 and QB1.
Generally, QB fall-off is more linear than RB fall-off, so median starter as a baseline will make the top RBs look stronger.
A problem with median starter is that your baseline sets the player who you see as having zero value; everyone below that point has negative value. So median-starter can really only be used for the top few players, after that, you need to readjust the baseline.
Another possibility is to use worst-backup as a baseline (RB36, QB24, etc.). In my auction league I've been doing custom baselines based on "player I'd only spend $1 for."
Generally, QB fall-off is more linear than RB fall-off, so median starter as a baseline will make the top RBs look stronger.
A problem with median starter is that your baseline sets the player who you see as having zero value; everyone below that point has negative value. So median-starter can really only be used for the top few players, after that, you need to readjust the baseline.
Another possibility is to use worst-backup as a baseline (RB36, QB24, etc.). In my auction league I've been doing custom baselines based on "player I'd only spend $1 for."
GregR_2
Footballguy
I've read this a dozen times and I'm not sure I agree with it, though maybe I'm just not clear on what you're saying.I do think an average starter baseline shows you something the last starter doesn't. And I don't think it is watered down. For instance, if you had the following 2 distributions at QB:
QB1: 300....300
QB2: 250....280
QB3: 100....270
QB4: 98.....220
QB5: 96.....200
QB6: 96.....180
QB7: 95.....150
QB8: 94.....120
QB9: 94.....100
QB10: 93.....93
Last starter is going to show you the same value. But QB1 in the first distribution has more true value than QB1 in the second distribution. Why? Because if you find how much of an advantage QB1 is over every other QB, he's a bigger advantage against every other team in the league than is the guy in the 2nd distribution. We then divide that number by the number of players to get a normalized value for comparison to other positions, and that number is the same as what average starter baseline shows.
That's a not only valid, but potentially important view of your data that last starter doesn't give, and a good reason for at least looking at the results of average starter.
However, I think it also puts more weight on the accuracy of each individual projection. If only for purposes of discussion, it's a lot easier to use last starter, and that's probably a big part of why we use it.
I seem to recall another detriment of average starter when I thought about this in depth a few months ago, but can't recall now what it is.
GregR_2
Footballguy
Thank you for saying that, because it's actually part of the point.I made an example that would be so glaringly obvious you don't need a system to find the value... so that when you see the result of the system, you can see it will indeed show you the result you know to be true.
And as you say, it isn't realistic. A realistic distribution, it is often not glaringly obvious how much the distribution of the QB vs the RB vs the WR would lead to increased value. Which is exactly when you want a system that can make it more obvious, right?
So the fact that reality is murkier than my example is exactly why we want a system that can show you what average starter can show.
Last edited by a moderator:
CalBear
Footballguy
Well, sort of. If you set median starter as the baseline, then everyone below the median starter will be compared to the median starter. If I'm trying to decide between Derrick Mason and Fred Taylor, I don't want to be comparing their performance relative to Hines Ward and Brian Westbrook, as I would be if I were using median starter as a baseline.
Bottomfeeder Sports
Footballguy
This was done in 2003.
Using the AVG Draft Position of a 12 team league the draft went like this (for the first 100 picks) round by round-
Avg12 Team Draft QB RB WR TE DT PK 12 0 11 1 0 0 024 2 18 4 0 0 036 4 19 12 1 0 048 6 23 16 3 0 060 8 28 21 3 0 072 9 32 28 3 0 084 14 34 30 5 1 096 17 37 34 6 2 0100 17 37 37 7 2 0If you used the FBG projection as of 8/6/2003 and used straight Last Starter (not Joe's "top secret" formula) and everyone drafted straight from this cheatsheet the draft looks like-
And Avg (mean) Starter looks like
Median Starter looks like
We can debate as to whether your cheatsheet should mirror the actual draft or not? Those who use the "its just a guide" argument can that is. If you argue that the best baseline creates the cheatsheet that requires the fewest draft day decisions. The more you need to deviate from the cheatsheet, the less value the cheatsheet so in reality if your cheatsheet is dominated by place kickers and defenses before any of the other teams are considering them, then what good is it?You can always fudge the numbers. That's what Joe and David do. They use a formula that revalues positions based on how they have historically been drafted. (At leasts that is how I understood various claims over the year.) Joe and David's formula does a pretty good job - as a start.
Joes QB RB WR TE Def PK12 0 10 2 0 0 024 4 17 3 0 0 036 7 18 8 3 0 048 9 21 14 4 0 060 9 22 22 4 1 272 10 25 24 6 3 484 12 26 28 8 4 696 13 26 30 8 7 12100 14 26 30 10 8 12The next set of baselines are based on the entire draftable player population. I don't remember the actual numbers at each position I used in 2003. I did average out a bunch of drafts to guesstimate where this cutoff was.
Finally, there is my favorite of the "standard" static baselines.
And, since my drafts never actually mirror the average draft I have developed my own baseline which I have used since 2001. It's main feature is that it allows me to guess what a particular draft will look like and change one value to attempt to model this draft. I've posted this before, but, it has been a while. My static baseline is this-[Value from Joe's secret formula] - n * [sTD Deviation of Draftable players] where n is between .5 and 1 depending on how experienced the competition I am playing. If I think they are newbies I use .5, if I think they are sharks I use 1. I'm assuming n was .75 in this example, but I don't have that recorded.
So there is a forrest view of each baseline. Again my belief is that the best baseline is the one that most mirrors your draft. Of course if your competiton drafts perfectly, and you perfectly model this draft the best you can do is hold your own. But in the real world everyone has different ideas behind where which players should be drafted. To me a cheatsheet should leave the only decisions I need to make on draft day is which players I should risk waiting an extra round or two. Otherwise it should reflect the collective effort from the previous few days, weeks, months of preparation. I need to trust that work as far as within the position decisions, the baseline's purpose is to suggest when I need to take which positions.
Using the AVG Draft Position of a 12 team league the draft went like this (for the first 100 picks) round by round-
Avg12 Team Draft QB RB WR TE DT PK 12 0 11 1 0 0 024 2 18 4 0 0 036 4 19 12 1 0 048 6 23 16 3 0 060 8 28 21 3 0 072 9 32 28 3 0 084 14 34 30 5 1 096 17 37 34 6 2 0100 17 37 37 7 2 0If you used the FBG projection as of 8/6/2003 and used straight Last Starter (not Joe's "top secret" formula) and everyone drafted straight from this cheatsheet the draft looks like-
Code:
Last Starter QB RB WR TE Def PK12 0 10 2 0 0 024 4 16 3 1 0 036 7 18 6 4 1 048 9 20 9 6 4 060 9 21 14 8 6 272 9 22 18 8 9 684 10 22 23 11 9 996 12 23 24 13 12 12
Code:
AvgStarter QB RB WR TE Def PK12 2 6 3 1 0 024 5 9 5 3 2 036 5 10 7 4 6 448 5 10 9 5 9 1060 7 13 12 6 9 1372 7 13 16 8 14 1484 8 13 20 10 14 1996 9 14 24 12 17 20100 9 15 24 13 19 20
Code:
Median Starter QB RB WR TE Def PK12 2 6 3 1 0 024 5 9 6 3 1 036 5 10 9 4 4 448 6 11 13 6 6 660 7 13 15 7 8 1072 8 13 21 8 9 1384 9 14 24 11 13 1396 9 15 26 13 14 19100 9 15 28 14 14 20Joes QB RB WR TE Def PK12 0 10 2 0 0 024 4 17 3 0 0 036 7 18 8 3 0 048 9 21 14 4 0 060 9 22 22 4 1 272 10 25 24 6 3 484 12 26 28 8 4 696 13 26 30 8 7 12100 14 26 30 10 8 12The next set of baselines are based on the entire draftable player population. I don't remember the actual numbers at each position I used in 2003. I did average out a bunch of drafts to guesstimate where this cutoff was.
Code:
AvgDrafted QB RB WR TE Def PK12 0 11 1 0 0 024 4 17 3 0 0 036 6 21 6 3 0 048 8 23 14 3 0 060 9 26 20 4 1 072 10 27 27 4 3 184 13 27 30 6 4 496 15 28 32 8 7 6100 15 28 33 8 8 8Median Drafted QB RB WR TE Def PK12 0 12 0 0 0 024 3 18 3 0 0 036 5 24 4 3 0 048 9 27 9 3 0 060 9 28 18 4 1 072 11 29 24 6 2 084 13 30 28 8 3 296 15 30 30 10 6 5100 15 30 31 11 6 7Last Drafted QB RB WR TE Def PK12 0 12 0 0 0 024 5 18 1 0 0 036 7 26 3 0 0 048 14 29 5 0 0 060 20 31 9 0 0 072 23 33 16 0 0 084 24 34 25 1 0 096 25 36 32 3 0 0100 26 37 34 3 0 0
Code:
AverageBackup QB RB WR TE Def PK12 0 12 0 0 0 024 2 20 2 0 0 036 5 27 4 0 0 048 8 29 8 3 0 060 9 30 18 3 0 072 12 31 26 3 0 084 15 33 31 4 1 096 18 34 34 7 3 0100 18 34 34 8 4 2
Code:
My Baseline QB RB WR TE Def PK12 0 12 0 0 0 024 2 20 2 0 0 036 5 27 4 0 0 048 9 30 8 1 0 060 10 31 16 3 0 072 13 33 23 3 0 084 16 34 30 4 0 096 20 35 33 6 1 1100 21 36 34 6 1 2
Great information for those who are interested to chew on for awhile. Thanks, Bottomfeeder.
Bottomfeeder Sports
Footballguy
Depends on how you construct the dynamic nature of DVBD, but if it is modeled after using an average draft to predict the players that are likely to still be available a round or two from now then the results are going to be very similar to well modeled static baseline.Don't miss understand. I have my own DVBD implementations and I have tried the Draft Dominator. But those tools are not necessarily appropriate (or even allowed) for every draft.
I compute three baselines (now use DD to do it for me):
Median starter
Worst starter
Median backup
I also rank on all three.
I look at value relative to median starter for the first few rounds, worst for round 5 thru 7 or so, and median back up for the next few.
By listing the ranks right next to each other, it really obviates where anomolies in each distribution (clusters, dropoffs) happen.
Median starter
Worst starter
Median backup
I also rank on all three.
I look at value relative to median starter for the first few rounds, worst for round 5 thru 7 or so, and median back up for the next few.
By listing the ranks right next to each other, it really obviates where anomolies in each distribution (clusters, dropoffs) happen.
Marc Levin
Hangs out with Oscar Zeta Acosta
average of all starters, or average of all players playing the position?either way, you have to calculate the lone outlier (QB12, TE12, RB24, WR24/36) at some point.
Marc Levin
Hangs out with Oscar Zeta Acosta
Excellent use of the DD.
Uruk-Hai
Footballguy
I hear ya. My point was directed more at those who get confused by the mere presence of a negative number, not where the negative number comes from. unless you're using "last-drafted" as a baseline, you're going to have negative numbers at some point (assuming you're using a static baseline).
CalBear
Footballguy
Right, but part of the appeal of last-starter is that you won't go negative (won't start comparing to players rated above your potential selections) until your starting lineup is filled out (more or less). VBD's usefulness decays pretty rapidly once the starting lineup is filled out, anyway; VBD principles assume that the player will be in the starting lineup, so once you get to backups, you're looking more at factors like upside and handcuffing.
Brock Middlebrook
Footballguy
Just sitting here reading this thread,I had a great baseline thought....I really think it's worth doing an article on, but I'm going to flesh the idea out -- I could see putting some time into this one -- and have it ready for next year.
northwoods
Footballguy
:insert "tease" tagline here:
oddball
Smart
no, no, no...that just won't do. I want the entire finished article, rough draft, and bibliography pm'd to me by tomorrow morning...
WilyCoyote
Footballguy
I feel the same way. Been subscribing to FBG for years now (and before that when it was free) for David & Joe's info. I've used David's "perfect draft" for a couple of years now. Never ventured onto the message boards until Saturday - didn't see that some guys at home sitting at a computer just like me would have some valid insights. I am getting a crash course in FF now - only 120 hours until my draft!
this and look at again later.
Uruk-Hai
Footballguy
I'm with you, CB. One case, though, where using a deeper baseline than "last-starter" may be useful is for rosters where a certain number at each position has to be drafted. I've gone back & forth on this, but I think you almost have to treat your entire roster as starters in that case - either by using a "last-drafted" baseline, or a combo with that & last starter.
CalBear
Footballguy
I agree that this can be true; for the Anarchy Invitational, which is a total-points, everyone-starts format, I used "last player" as the baseline, since everyone counts. I don't think it's that important if it's a more traditional league that happens to have roster requirements (start 2RB, draft 5 RB or whatever).
Sennacherib
Footballguy
I recall several discussions of the "best static baseline" question back on the old yellow board, and, if memory serves, the one that I liked best and used successfully until I went DVBD was originally Maurile's idea. You look at your league's history over previous seasons, and determine, on average, how many players at each position actually contributed significantly for their fantasy team. Then, use either the (mean) average contributor (my preference) or the last contributor as a baseline. A significant contribution I would define as starting more than 3 fantasy games, so we're looking at everyone who is more than just a bye-week fill-in...in other words, the guys you ought to draft, as opposed to guys who should be available on the wire.
In my 12-team, start 3WR league, it looked something like 16 QBs, 36 RBs, 50 WRs, 12 Ks and 12 D/STs, I believe. This allowed for the fact that, for example, you are much more likely to have to get significant use out of your RB3 than you are out of your WR4 or QB2, and so value should be determined differently at those positions.
Again, I've moved from this to DVBD on the DD in my own work, but I'd suggest it to any who, for whatever reason, need to use a static baseline.
Happy drafting, everybody!
In my 12-team, start 3WR league, it looked something like 16 QBs, 36 RBs, 50 WRs, 12 Ks and 12 D/STs, I believe. This allowed for the fact that, for example, you are much more likely to have to get significant use out of your RB3 than you are out of your WR4 or QB2, and so value should be determined differently at those positions.
Again, I've moved from this to DVBD on the DD in my own work, but I'd suggest it to any who, for whatever reason, need to use a static baseline.
Happy drafting, everybody!
