Writers of Pro Football Prospectus 2008

09 Sep 2009

A (hypothetical) different way to make cheatsheet rankings

This is all hypothetical. Just an idea I had, and I wanted to see what people thought about it.

So cheatsheets are made by projecting a player's stats, converting those stats into points, and then making an overall list using some sort of baseline. But no one seems sure of what baseline to use. Should it be last starter, last drafted player, median player, or something else entirely? I wonder therefore if there might be a better way that avoids the sticky problem of baselines.

Let's imagine that the points scored by every player is already known going into draft day. This means that the order in which the QBs should be drafted, for example, is public knowledge. For each team, then, the only decision they need ot make is what position to draft in each round. They choose to draft QB (or RB or whatever), and then they take the highest player left at that position. If we assume that a snake draft evenly distributes talent as it is supposed to do, I believe that the best cheatsheet is the one that if everyone follows exactly, there will be the minimum amount of variation in points scored by each of the teams.
In turn, I could imagine it to be possible for one to create a computer program that finds this optimum draft order. This becomes your overall cheatsheet.

An obvious problem would be how to deal with backup players. They must be included because the relative value of each position is dependent on how many rounds the draft goes. If you imagine say an 8 round draft in a league where teams get no backups, then obviously a defense and kicker can get drafted no later than the 7th and 8th round. But if it's a 16 round draft with a bunch of backups, then those positions might move down to the 10th or the 12th or what have you.
But on the other side, obviously you have to discount the value of backups or otherwise the algorithm might say that the best strategy is to take QBs in each of the first two rounds, even though you can only start one. To me a reasonable answer is to multiply each player's projected points by the fraction of games that they are to be started. If you analyzed thousands of leagues, you could probably get pretty accurate numbers for a RB3 and a RB4 and a WR3 and so on. The average team with 2 RB starters and no offensive flex might start their 3rd RB 3 times in a season, so that 3rd RB would be worth 3/16 of their total value.

That's my idea. Does it seem to make sense? And second, is it possible? I know that it may very well be impossible to try out every possible draft order combination. I wonder if instead we could start with some sort of default draft order, and then watch players bubble up or down as the team which drafted them has projected points is higher or lower than the league average.

Posted by: Zac on 09 Sep 2009

3 replies , Last at 10 Sep 2009, 10:58am by Key19

1
Re: A (hypothetical) different way to make cheatsheet rankings
by aron7awol :: Wed, 09/09/2009 - 3:17pm

I understand the concept, but it's not realistic because there's no way you can predict the moves of the other teams. You need to take the guy who has the most value (to your team) when it's your pick. The amount of value to your team, of course, has many different variables that go into it, e.g. baseline method, number of position slots, number of teams, how many of each position you already have, etc. I believe the standard baseline systems are very flawed, but that's for a different thread :) Having accurate projections is most important, since everything is based on that. Accurate valuation based on these projections (e.g. baselines, ranking across positions) is also a good advantage, and where I think most systems can be improved. However, IMO, the way you really make hay is by accurately predicting when players will be picked and waiting on players who you have valued higher than everyone else and grabbing them right before they are picked by someone else. Of course, it's not an exact science and there's some risk involved.

2
Re: A (hypothetical) different way to make cheatsheet rankings
by tally :: Thu, 09/10/2009 - 2:11am

People are already doing this by using a dynamic baseline round by round that factors in the picks that others have made and uses some assumptions about picks they will make based on team needs.

KUBIAK's baseline adjustments try to take this into account by setting the RB baseline lower to reflect the usual early runs on RBs and the fact that backup RBs are often drafted before other starters are selected. Likewise, TE/K/D can have their values lowered to simulate draft tendencies--the latter two positions often taken in the last couple of rounds.

However, by adjusting the baseline round to round based on the results of your actual draft, you forgo the need to use these approximating adjustments in return for doing a little more work during the draft.

3
Re: A (hypothetical) different way to make cheatsheet rankings
by Key19 :: Thu, 09/10/2009 - 10:58am

There are two problems.

1: We'll never have access to the information you're referencing. No projection system will ever be perfect and time travel is not feasible at this point either.

2: Even if everyone used the same cheat sheet, even an imperfect one, (which would come close to simulating what you're describing), people would still make unpredictable picks that would throw everything off just based on team need.

I think I understand where you're going (maybe not), but I still think that no matter how advanced we become in FF, the draft will always be something that cannot be perfected. Some people and systems are good at drafting, but no person/system will ever be flawless.

Login or register to post comments