Valuation of players and draft picks

Temporary forum for the discussion of the Tuesday Morning project, to enhance the Madden 08 off-the-field experience.
Drizzt_13
All-Pro
Posts: 202
Joined: Wed Apr 18, 2012 7:32 pm

Re: Valuation of players and draft picks

Postby Drizzt_13 » Thu Apr 26, 2012 11:21 am

Yeah basing stuff on stats for scouting other teams players actually makes a lot of sense and is a really good idea. That makes a lot of sense, but it seems like it's tricky to make it have much accuracy because the game tracks only a few stats, but then having it be inaccurate might be the point. Maybe have a weighted average of the player's actual OVR and the OVR based on stats, weighted more towards actual OVR based on GM's scouting ability. Developing a way to calculate OVR based on stats would also be really useful for drafting if we want to give players college stats. We should have a separate thread for working out scouting formulas, and just focus on the ratio of players as we intend to use them.

Re: Valuation of players and draft picks

Sponsor

Sponsor
 

User avatar
torontogrudlies
MVP
Posts: 837
Joined: Fri Jul 15, 2011 4:56 pm

Re: Valuation of players and draft picks

Postby torontogrudlies » Thu Apr 26, 2012 11:43 am

Cool, are we possibly talking about two separate equations then? 1) the OVR, or our own modified version of it, based on attributes (both in-game and the secondaries we create) and 2) a sort of composite score/OVR based almost entirely on stats (I say almost because there is always some info which would be generally known, such as his speed and strength.)

I'm thinking the latter would compare to the passer rating (which could probably actually be used for the QB in our formula)....the max rating for that is 158.3, which would be the equivalent of what we understand as a 99 OVR?

Drizzt_13
All-Pro
Posts: 202
Joined: Wed Apr 18, 2012 7:32 pm

Re: Valuation of players and draft picks

Postby Drizzt_13 » Sat Apr 28, 2012 9:30 pm

Here is my latest valuation formula proposal. We should produce two different OVR's one based on the player's production and one based on their ratings, and combine theseT The combined OVR would than be modified based off of team need, contract status, GM opinion of the player and other factors. These OVR formulas should take into account the tendencies of the team evaluating the player. I propose we call the one based off of stats "production OVR" and the one based off of ratings "Scheme OVR", and the weighted average of the two "perceived OVR". A formula for this would be

Percieved OVR=(Scheme ovr*(1+scout accuracy)+production ovr*(1-Scouting Accuracy))/2
Scouting Accuracy=(play time-50+team scounting-50)/100
Play time=games started+(.5(games played in-games started))*6.25 based off of games played in the previous season

Madden tracks games started so I figured playtime would be another player rating we could have in the database, I would like to be able to base it off of the previous years as well but I'm not sure if madden tracks that effectively. Ideally playtime would use the same formula as it does now but we would use not just the games played last year but a combination of the last three years weighted towards the previous year.

Team scouting could be as complicated or as simple as we want to make it. It should be a rating from 1-100 and could be a just a stat the GM has or it could be a combination of GM scouting and HC and Coordinator ratings at that position.

Now we need to break down production OVR and scheme OVR, scheme OVR is the most complicated because it is essentially the current OVR formula, but with a little tweaking based off run/pass balance and aggressiveness. The current OVR formula basically applies this formula to each of the ratings it considers relevant to the position and then sums them and subtracts a starting value.

(((99/PRDH-PRDL)*rating weight)/total weight)*player rating
PRDH and PRDL, are, as far as I can tell, constant values that are different for each position. Rating weight is a number assigned to each rating. For example for QB's awareness is rated a 3, THA at 4 and THP at 3.5. total weight is the sum of all the individual ratings weights for each position and player rating would be the actual rating the player has at that trait. So fro example for a QB with awareness 80 the formula is

(((99/(89-35))*3)/13)*80=33.846 This is performed for every rating the player has, these are summed and then a constant termed the "starting value" is subtracted.

The "scheme OVR" would be based on this formula except we would change the rating weight slightly based on run/pass and aggressiveness, and 3-4 vs 4-3. For example TE's in a pass first offense would obviously rate PBK over RBK. WR's in an aggressive scheme would probably value speed over catch. The formula would be something like rating weight + or - ((Pass-50)/100).

For the production OVR we would have to come up with some sort of rating system for each position based on per game stats and then scale it from 1-100 (or really 70-99). Ideally this would be affected by coach tendencies as well, a pass oriented team might view a tackles sacks allowed as more important part of the equation than pancake blocks, and an aggressive team might view a high yards per catch as more important then total receptions. We would also need a way to make this be higher for player who play more games so that one good game from a backup doesn't get them rated really highly.

There are more issues with this to be explored and ways it could be made more complicated. I just want to know if people think this is a good idea or if we should adopt a different model before I develop a lot of formulas for production OVR and different ways of rating players dependent on scheme. If I get some support for this I'll create some threads for making production OVR and scheme OVR and then work out how need, salary and off the field stuff should come into the equation in this thread.

User avatar
torontogrudlies
MVP
Posts: 837
Joined: Fri Jul 15, 2011 4:56 pm

Re: Valuation of players and draft picks

Postby torontogrudlies » Sun Apr 29, 2012 12:19 am

Yeah, I'd expanded out the overall ratings spreadsheet and was playing with it just for fun. I do remember that when they were discovering these fields over on FF, they could change the values but not the attributes which were used. Working outside of Madden, we escape such limitations. If we decide (just for a silly example) that the QB's jumping ability should figure in, we can throw it in.

I will also start modifying my object oriented code, so that when a particular player is loaded up, it will also pull games started and games played. Although I wonder if we need to try to differentiate between a game they had significant playing time in, and a game they appeared in for a couple plays?

In regard to the production OVR, I hear you regarding the number not being an accurate representation if the player hasn't had significant exposure. Part of our overall evaluation formula should take into account the lack of exposure, and a wiser GM would thus lend less weight to such a figure. (While an inexperienced GM might get unduly "excited" about a flash in the pan player.) I also still feel that a few of the "scheme OVR" type attributes should cross over to the production...as opined previously, general info such as how much they bench (STR) and how fast they are (SPD) would be fairly common knowledge. This would make the resulting production OVR a little more grounded, I would think.

This looks like good stuff. Might you be able to model it in a spreadsheet, perhaps with just a couple of positions? Offhand, I'd say maybe for a QB and OL, as this would give us a decent insight on how it would work when the run/pass philosophies change....

Drizzt_13
All-Pro
Posts: 202
Joined: Wed Apr 18, 2012 7:32 pm

Re: Valuation of players and draft picks

Postby Drizzt_13 » Sun Apr 29, 2012 4:12 pm

torontogrudlies wrote:Yeah, I'd expanded out the overall ratings spreadsheet and was playing with it just for fun. I do remember that when they were discovering these fields over on FF, they could change the values but not the attributes which were used. Working outside of Madden, we escape such limitations. If we decide (just for a silly example) that the QB's jumping ability should figure in, we can throw it in.


Yeah we could, It's just that then we have to mess with the ratio and starting values and stuff, and while I'm sure we can figure it out unless we think something is just a total glaring hole in the way OVR is calculated I'd rather just mess with rating weights.

torontogrudlies wrote:I will also start modifying my object oriented code, so that when a particular player is loaded up, it will also pull games started and games played. Although I wonder if we need to try to differentiate between a game they had significant playing time in, and a game they appeared in for a couple plays?

Ideally we would use snaps rather than games, which is totally doable for things like the QB and RB position where every time they rush or throw it's tracked, but I don;t know how we would do it for the other positions. Even if we just wanted to differentiate between the amount of playtime they had, I don't know how we could do this for positions like CB, where even if they are used consistently as the nickel corner they wouldn't necessarily put up a lot of stats so that we could tell based on anything but the depth chart. Maybe we could do this based on games playedin modified by position on the depth chart rather than games started. If you're the third WR even if you've played in the same number of games as the fourth wide receiver you probably saw more snaps so the games you played in should be weighted higher. Can the app keep track of depth chart position for these calculations?

torontogrudlies wrote:In regard to the production OVR, I hear you regarding the number not being an accurate representation if the player hasn't had significant exposure. Part of our overall evaluation formula should take into account the lack of exposure, and a wiser GM would thus lend less weight to such a figure. (While an inexperienced GM might get unduly "excited" about a flash in the pan player.)

I was trying to balance this by having the amount by which production OVR is weighted be decreased the better at scouting a GM (or team) is. If you have max scouting and the player has started every game in the current formula production OVR doesn't factor in at all. Also the higher the playtime the less production factors into the perceived OVR. But if you want to add another variable I'm sure we could work something out.

torontogrudlies wrote:I also still feel that a few of the "scheme OVR" type attributes should cross over to the production...as opined previously, general info such as how much they bench (STR) and how fast they are (SPD) would be fairly common knowledge. This would make the resulting production OVR a little more grounded, I would think.


I kind of want to avoid making the production OVR too grounded, as it is supposed to represent desperate or foolish teams spending big money and trading for players like Kevin Kolb or Matt Flynn (though flynn could still work out) based on little information. I'd prefer to, rather than have it become more accurate have it's weighting in the average change based on more GM factors so that teams would be willing to take a chance on splash players for a larger variety of reasons, such as being desperate for any star, having extreme need at that position, or genuinely having the time and coaching ability to groom players towards their potential. I don't know how the GM stat-line is going to end up so I won't write a definite formula now.

Incorporating physical attributes into the production OVR makes a certain amount of real world sense as it gives you an idea of how fast a player can move, but because technique is a huge part of the game, rather than just raw ability, good scouting of the player is also necessary. Good run stuffing lineman don't just bench a lot, they also know how to use leverage correctly and do a lot of other technique things. The problem is madden doesn't track any of these things, it just tracks STR. And because a number of positions in madden really only need their physical attributes because they don't often draw on awareness and TKL, positions like D-line will have their OVR accurately predicted by just their physical drills, when we know from the NFL it takes more than being good at drills to be a good lineman.

We could just accept this as a flaw of Madden's rating system, we could simply not integrate physical traits into the production OVR, or we could try to come up with a system which takes into account technique. I. If we were talking about capping abilities, then the drills could accurately represent the cap but not the players actual rating. The cap would represent the raw physical ability and the rating the player actually had would be how well they applied that ability. So a DT with a STR capped at 99 would have a really high number of bench press reps even if the actual strength they played with was 90, because they had terrible technique but a lot of raw talent. The one thing I don't like about this is that their are players who aren't physically gifted but still play really well because their technique is perfect, and while the above system allows you to overrate a player because their physical gifts are ahead of their technique, it doesn't allow you to under rate a player because their technique is ahead of their physical gifts. So another way to do this would be to have a technique modifier applied to a base ability score. So even if you only have a speed 87 you might be really good at route running and gain separation a little easier and so play as if you were a little faster. This could be tied to progression and other factors as well.

I don't know if that is doable, or worth the time it will take, I'm just throwing ideas out there


torontogrudlies wrote:This looks like good stuff. Might you be able to model it in a spreadsheet, perhaps with just a couple of positions? Offhand, I'd say maybe for a QB and OL, as this would give us a decent insight on how it would work when the run/pass philosophies change....

[/quote]

I have a spreadsheet I made which just calculates scheme OVR. It has a space where you set the pass balance and the agression for each equation and a constant for both pass and aggression. The constant is multiplied by pass or aggressiveness -50, I would suggest using .01 but you can change it, though the formula would get messed up if any of the weights ever becomes negative. Some positions don't have anything which corresponds to pass/run, QB's and WR's will almost always be used for passing and nothing else, so the pass run part of that will effect it as part of the need formula not as part of calculating OVR.

Here's what I decided was affected by what, but this is definitely up for revisions


Drizzt_13 wrote:
QB:Coaches with high Aggressiveness would weigh THP higher than it is now, and THA a little less, more conservative coaches would be the opposite.

HB: pass happy teams would favor CAT and PBK at the expense of, say CAR and STR. Big play teams would favor SPD over BTK.

FB: Pass would value CAT and PBK over STR and RBK, Big play would value SPD and ACC over BTK and CAR

TE: Pass Would value PBK over RBK. Big play would value SPD and CAT over STR and BTK.

WR: Pass wouldn't have an effect because the current OVR formula doesn't factor in RBK at all. Big Play would value speed and ACC over CAT and AGI (possession receivers vs big play).

OL: Pass would value PBK over RBK, I don't know what makes a more aggresive lineman, maybe higher SPD and ACC for pulling plays and zone blocking over a lower STR for in line blocking, I don't know line is tricky.

DL: Pass defense oriented teams would favor ACC over STR, big play teams would favor SPD over TKL as they would be less worried about consistent tackling and more into backfield penetration. (Note: for some reason TKL is Weighted higher for DT's then it is for OLB's in the madden OVR setting which seems weird)

LB's: Pass would value SPD+AGI over STR and TKL, Aggressive would Value ACC over AWR (I'm thinking aggressive teams tend to want their LB's to blitz rather than hang back in coverage or diagnose plays)

DB's Pass would want SPD+ ACC over TKL+STR, aggressive would want CAT over AWR




I made this with google docs and I'm not sure how that works with excel so I exported it as a CSV, if you have trouble viewing it I'll just share it with you on google docs.

User avatar
torontogrudlies
MVP
Posts: 837
Joined: Fri Jul 15, 2011 4:56 pm

Re: Valuation of players and draft picks

Postby torontogrudlies » Sun Apr 29, 2012 6:51 pm

Did you mean to attach a file? You mentioned a csv, or is this something you're still working on?

User avatar
torontogrudlies
MVP
Posts: 837
Joined: Fri Jul 15, 2011 4:56 pm

Re: Valuation of players and draft picks

Postby torontogrudlies » Sun Apr 29, 2012 6:56 pm

What are considered Big Play teams? Is this aggressiveness or something different?

On the OL, to learn more about those uncertain ratings like speed, a good method I've found is to globally increase or decrease the rating in question, then total control Sim an entire week or more....see what the resulting change in stats looks like...

Drizzt_13
All-Pro
Posts: 202
Joined: Wed Apr 18, 2012 7:32 pm

Re: Valuation of players and draft picks

Postby Drizzt_13 » Sun Apr 29, 2012 9:07 pm

It won't allow .xls, .tsv, .csv, .ods or .pdf files so I think you're just going to have to PM me a Gmail so I can share the document with you.

Yeah aggressiveness is big play, I'll try messing with speed and stuff, but it's going to be a while before I can really playtest this stuff. Next week is really really busy for me, and while i may have time to comment here occasionally I won't have time to sim a bunch of games until later.

User avatar
torontogrudlies
MVP
Posts: 837
Joined: Fri Jul 15, 2011 4:56 pm

Re: Valuation of players and draft picks

Postby torontogrudlies » Sun Apr 29, 2012 10:20 pm

Drizzt_13 wrote:
torontogrudlies wrote:Yeah, I'd expanded out the overall ratings spreadsheet and was playing with it just for fun. I do remember that when they were discovering these fields over on FF, they could change the values but not the attributes which were used. Working outside of Madden, we escape such limitations. If we decide (just for a silly example) that the QB's jumping ability should figure in, we can throw it in.


Yeah we could, It's just that then we have to mess with the ratio and starting values and stuff, and while I'm sure we can figure it out unless we think something is just a total glaring hole in the way OVR is calculated I'd rather just mess with rating weights.

torontogrudlies wrote:I will also start modifying my object oriented code, so that when a particular player is loaded up, it will also pull games started and games played. Although I wonder if we need to try to differentiate between a game they had significant playing time in, and a game they appeared in for a couple plays?

Ideally we would use snaps rather than games, which is totally doable for things like the QB and RB position where every time they rush or throw it's tracked, but I don;t know how we would do it for the other positions. Even if we just wanted to differentiate between the amount of playtime they had, I don't know how we could do this for positions like CB, where even if they are used consistently as the nickel corner they wouldn't necessarily put up a lot of stats so that we could tell based on anything but the depth chart. Maybe we could do this based on games playedin modified by position on the depth chart rather than games started. If you're the third WR even if you've played in the same number of games as the fourth wide receiver you probably saw more snaps so the games you played in should be weighted higher. Can the app keep track of depth chart position for these calculations?



Do you mean, keeping stats like "in Week 2, WR Barnes was WR4 on the depth chart"?

We could, but it's not likely to be very neat. I'm wondering if there's another way we could get reasonable estimates of the data.

What if, as part of the weekly tasks, the program makes some sort of educated estimate on how many snaps/minutes a particular player played. Maybe based upon his position on the roster like you were saying, plus other data such as their game stats, and if he finished with an injury. If there were 80 offensive plays, WR4 was likely involved in 20 of them (just throwing out a scenario.) So then he has some kind of "experience" attribute which gets increased by 20.

Drizzt_13
All-Pro
Posts: 202
Joined: Wed Apr 18, 2012 7:32 pm

Re: Valuation of players and draft picks

Postby Drizzt_13 » Tue May 08, 2012 10:13 pm

torontogrudlies wrote:
Do you mean, keeping stats like "in Week 2, WR Barnes was WR4 on the depth chart"?

We could, but it's not likely to be very neat. I'm wondering if there's another way we could get reasonable estimates of the data.

What if, as part of the weekly tasks, the program makes some sort of educated estimate on how many snaps/minutes a particular player played. Maybe based upon his position on the roster like you were saying, plus other data such as their game stats, and if he finished with an injury. If there were 80 offensive plays, WR4 was likely involved in 20 of them (just throwing out a scenario.) So then he has some kind of "experience" attribute which gets increased by 20.


If we're pulling data like time of possession, then it might simply be easier to just pull the actual number of total run attempts and pass attempts and use those, rather than using an average snaps/minute. This makes it really easy to track snaps for positions that are almost always on the field QB OL and positions that have all their carries logged but it's a lot more complicated for other positions.

I don't see how play testing could help us get a more accurate snap count for each position because the game doesn't count snaps. So I think our best bet is to try to make assumptions and then try to back them up with actual NFL data. Pro Football Focus has exhaustive data on that but I think you have to pay for access and it might not be easy to average things like snap percentages. We could just make educated guesses and as long as they don't cause huge balance problems it should all be fine. I'll spend some time looking for this data but if we can't find any I'll just throw up some guesses.

The next thing is to figure out a formula that takes snaps and the stats madden tracks and produces a number, than we scale that number to overall. Obviously we would just use the QB rating for QB's but we can't just have the rating correspond directly to OVR we have to make it fit into the 70-100 range. I don't know if this should scale linearally or exponentially and I think that's something we're going to have to determine through play testing but I'll get to that later. Anyway here are some proposed formulas for generating the production OVR through stats. The basic Idea was to produce an overall for each fact of the game, so for lineman I would just use Sacks/pass snap or pancakes/run snap have those two scaled to produce an ovr and then average the ovr they produce, Ideally that average would be weighted based on team tendencies. I'm going to have a spreadsheet ready for this in a little bit it's just a lot more complicated than I thought when I started writing the post so I won't do it here.

About play-testing this stuff. We should establish a standard roster, slider settings, period length, difficulty and coach slider modifications for the purpose of play testing. That way we all get roughly similar results when looking at things like average snaps and what sort of stats a 99 overall really puts up. Hopefully these would be things that majority of the community uses so that it's easiest for people to use TMM.


Return to “Tuesday Morning Manager”

Who is online

Users browsing this forum: No registered users and 10 guests