Playing Time Projections

[david_hume]

I am following Todd's advice about projections and focusing on playing time, not just skills. After a quick comparison of the Mastersball set to those of two other major sites, I notice that you seem to project more ABs for many of the top players. Is this something you are aware of? Any idea as to the reason? Some examples:

Ichiro 664 Mastersball (621/612 other sites)
Pedroia 647 (604/599)
A Hill 627 (590/564)
Braun 627 (589/583)
McCutchen 624 (578/568)

The Mastersball set also has the four highest AB totals of the entire sample and projects 32 guys to have over 600 AB vs. 18 and 9 for the other sets.

This is by no means an exhaustive study, just some things I noticed while playing with the data for a few minutes.

[Todd Zola]

What would you say if I pointed out that last season, there were 30 players with >600 AB?

I'll review the guys you mentioned individually, but I do think the above observation is telling.

[david_hume]

I hope I didn't come across as critical. I'm quite unsure whether yours or the others are more accurate. I am just noticing the differences and wondering if they are due to methodology or something else.

[Todd Zola]

Ichiro 664 Mastersball (621/612 other sites)

2004 704
2005 679
2006 695
2007 678
2008 686
2009 639

I know he is getting older and missed 16 games last season after probably not missing a total of 16 since he came to the States. If there was a gradual decline, I could see dropping him. But there is no indication that last year was anything but an aberration, even at age 36. Our 664 allows for his missing a handful of games that he did not miss from 04 to 08.

[Todd Zola]

I hope I didn't come across as critical. I'm quite unsure whether yours or the others are more accurate. I am just noticing the differences and wondering if they are due to methodology or something else.

And I am hoping I didn't come off as defensive or annoyed.

It is really hard to gauge intent over the Internet, but I hope everyone understands I prefer not to do this via webcam

[Todd Zola]

Pedroia 647 (604/599)

2007 520
2008 653
2009 626

One could argue we are a mite optimistic, but what reason is there to believe he will fall below his total of the past 2 seasons?

At least you guys now know we don't rely on a 3-year average

[Todd Zola]

A Hill 627 (590/564)

2007 608
2008 205
2009 682

What you don't see here is Hill actually played in more games in 2007, but hit lower in the order.

His injury in 08 was a concussion. I think by his performance last season, it is safe to say he is over it.

We may actually be too CONSERVATIVE!!!!!

[Todd Zola]

Braun 627 (589/583)
2007 451
2008 611
2009 635

07 was not a full season. Seems to me Braun has established a level and there are no injury concerns. In fact, he has shown to be a warrior of sorts as he was a risk at most drafts last season and ended up playing 158 games.

[Todd Zola]

McCutchen 624 (578/568)

2009 443

Obviously, this one is not based at all on history but how we feel he will be used this season. The Pirate OF and 1B has a lot of moving parts, but we see McCutchen as an every day fixture in center.

Last season, he has 443 AB in 104 games. Prorate that to 155 games and that is 621.

[JP Kastner]

I know from talking with other who do projections that they commonly use a three-year model. If they had 600 two years ago and 500 last year due to some time on the DL, they would give them 550 as a safe bet. We use a number of different playing time projection apparatuses depending upon what we know about the player. If a player moves from sixth in the lineup to lead off, he is going bat more over the season.

Sometimes we compare and contrast. If we know someone is the #2 catcher and the #2 catcher before received 250 plate appearances, we would assume that the new catcher is likely to get similar playing time.

Whenever we come up with a number, there is a model behind it.

[kjduke]

I know from talking with other who do projections that they commonly use a three-year model. If they had 600 two years ago and 500 last year due to some time on the DL, they would give them 550 as a safe bet. We use a number of different playing time projection apparatuses depending upon what we know about the player. If a player moves from sixth in the lineup to lead off, he is going bat more over the season.

Sometimes we compare and contrast. If we know someone is the #2 catcher and the #2 catcher before received 250 plate appearances, we would assume that the new catcher is likely to get similar playing time.

Whenever we come up with a number, there is a model behind it.

It's good to see the crew here not shortcutting a critical aspect of projections just because there's no theoretical glory in it. That separates the professors from the professionals. Well done!

[Todd Zola]

For the bulk of my "career", I was more of a consultant with the projections. I have only been directly involved since the end of the FB.com stint through now. Something I have really come to realize lately is how undervalued the playing time aspect is. Some may say it is all a crapshoot because injuries are more prevalent than ever, and they have a point, but we are in the business of telling you what we think a player will do and not saying screw it, they're going to get hurt anyway.

A projection is composed of two elements -- the rate of performance and the amount of performance. In a recent SI.com piece, I suggested that the rate of performance between models is rather slight and if someone focused on projecting their own playing time, they were in essence doing their own projections. But please keep in mind the audience. The SI.com crowd is more mainstream than what we normally have frequent here. Back home on Mastersball, the edge derived from the skills aspect is significant. And hopefully it will be too for the SI.com readers. But you gotta learn to crawl before you walk, let alone all out sprint like some of you guys.

Gary has developed the site's methodology with respect to the skills. He has busted his butt to come up with a model that we feel best predicts the most probable future outcome based on past history. At the conclusion of each season, he does a post mortem. We don't focus on individual misses, per say, but rather if a glaring individual is representative of something more global and then he does some number crunching to incorporate that in the following year's model.

Why did a whole bunch of pitcher's within a certain age group show skills differently than others in that same group? Were they all ground ball guys? Were they all low strikeout guys? Did they all have fewer than 2 years time in the high minors? Whatever the reason, that is now factored into the model.

So if Gary is going to pour so much blood, sweat and tears into the model, we owe him the courtesy of matching his effort with the playing time aspect. ESPECIALLY since the playing time is primarily tedious busy work as opposed to anything more cerebral.

JP has a tool that incorporates each team's expected total plate appearances based on recent history with the normal breakdown of the amount of plate appearances each lineup spot gets.

Clearly, a leadoff hitter gets a higher % of the total than the #2 hitter, who gets more than the #3 hitter, etc.

For the ease of automation, probably because they feel it is "close enough for government work", some sites assume the Padres get as many plate appearances as the Yankees and the #8 hitter as many as the #2 hitter.

If you put Jacoby Ellsbury BOS 1 155 into the tool JP designed, it multiplies the expected number of plate appearances Boston is going to get by the percentage number a leadoff hitter gets then prorates it to 155/162 to render the expected plate appearances for Ellsbury. JP went so far as being able to handle players that will hit 2nd versus RHP and 6th versus LHP. Do we use this tool for everyone? No. I rely on history for those that I feel history is the best guide. Examples are Ichiro, Pedroia and Braun discussed earlier.

But McCutchen was a perfect guy for which to use the tool. So was Marco Scutaro. Scutaro went from hitting on top the Blue Jay lineup to the bottom of the Red Sox lineup. Instead of guessing how much this would impact in terms of plate appearances, I could quantify it with the tool.

We will also use what I call "logical checks" when assigning playing time.

Innings are easy. Every team gets about 9 x 162 because it is impossible to predict the number of extra inning games and number of times the home team did not need their last ups.

Plate appearances are as suggested, based on recent history.

I want to make sure that when rosters are set, each team is assigned a total number of plate appearances in the ballpark of their expectancy. If I shorted Ichiro by 60 or 70 like what was shown earlier, where do they go -- Ryan Langerhans? Why give a backup 70 more than he historically receives?

To help in this regard, I break the team into C, IF and OF. For the AL, I usually label the DH as OF. I roughly then assume 700 PA per position and see the distribution. This is more of a guide. I rely on the team's expected total for the final say, but if a team is supposed to have 6300 PA and I assign 700 to catcher, 2500 to IF and 3100 to OF, and no OF is going to play in the IF, my distribution is wrong so I will need to take away about 300 AB from the OF and give it to the IF, or leave the IF short until they sign someone we feel will pick up those PA.

NOTE -- So if Gary reads this, I will save him the time of shooting me an e-mail -- our hitting currency is PLATE APPEARANCES. I do my best to use PA (not at bat, AB) when talking about it. But occasionally I will slip and use AB.

Anyway...

In the NL, we assign 300 PA that would normally go to the pitcher in the team-wide calculation. This number varies from team to team each season and is not predictable, so we use the average to help account for the historical PA that go to the pitchers.

So again, the reason for doing this is we have learned that the playing time element is very integral to projections. And while we don't KNOW what it is going to be, if we use some logic and common sense, we can present some numbers that make sense and help give you that small edge against your competition that relies on a set of numbers that does not incorporate such robust playing time considerations.

And the best part?

You can ask us to "show our work", and if you do not agree with that, you can manually alter the projection in the CVRC and adjust the values and rankings according to how YOU feel.

[JP Kastner]

Probably the single greatest thing about the merger between CS and MB is that the strengths of each projection system (MB: Newton or NEUTS / CS: Zen Projection Engine) supplemented each other. Gary has done a heck of a job with the Skills component. I spent most of my time working out playing time models. When we merged, it was very easy putting the two together.

I'd like to follow up on the injury component. We know that there is an even 50% chance that someone picked in the first round of the MB Slow Mock draft will miss a significant amount of time due to injury. We know that if you extend it out to the third round, that there is a near 100% chance that someone will miss significant playing time. We don't know who that is. From a projection model, you just have to pretend that those injuries will not happen. We only predict what we know.

That doesn't mean we ignore injuries. If someone has a chronic condition, we can consider that and do.

[Guest]

As you can tell, we spend a decent amount of time on this process.

I would argue and I think Todd and JP would agree that going through the process of allocating playing time is probably the single best way we can think of to truly know the player pool in and out. There is an obsession with looking at draft lists position by position, but when you have to sit and look at a team like KC or Bos or whoever, and really plot out what they'll likely do, it is very helpful.

What I feel our number one responsibility is where it comes to playing time is to use the projections to outline a "most likely" outcome. Then we should be qualitatively evaluating team by team the possibilities that could exist within. Once you've done that you have at least a better feel for what the range could be for each team. Some teams have very little wiggle room in their lineups. NYY, BOS to name two. Others (KC) obviously there is significant variance. Trying to reconstruct it all is where you can find upside and value later in drafts/auctions.

I would say MINN last year was the best example of this. Kubel/Cuddyer/Young/Gomez/Span were all cheaper than otherwise would have been because of the confusion in playing time there. Well, you got huge value in Kubel, Cuddyer, and Span if you rostered them. Young probably broke even if you didn't cut him after his pathetic first half. Gomez would have hurt you. So these situations offer opportunity where leveraged properly.

[david_hume]

this has turned into a very informative thread for me. thanks guys.

our hitting currency is PLATE APPEARANCES. I do my best to use PA (not at bat, AB) when talking about it.

when calculating skill rates, do you suggest we use PA instead of AB (for example hr/pa instead of hr/ab)? AB is basically just PA-BB, correct (yes yes, sac flies and HBP and all of that)?

I would argue and I think Todd and JP would agree that going through the process of allocating playing time is probably the single best way we can think of to truly know the player pool in and out. There is an obsession with looking at draft lists position by position, but when you have to sit and look at a team like KC or Bos or whoever, and really plot out what they'll likely do, it is very helpful.

in my limited experience doing this over the past week or so, i heartily agree.

[Guest]

when calculating skill rates, do you suggest we use PA instead of AB (for example hr/pa instead of hr/ab)? AB is basically just PA-BB, correct (yes yes, sac flies and HBP and all of that)?

Absolutely. Its what we do.

[Captain Hook]

remember too that many other sites project MORE at bats for total players on a given team than that team can ever hope to see in the real world.

So they have projections for more players but the projections are not as useful.

For instance the total AB for Team X outfielders might be 1800+ or 1900+, but it is definitely NOT more than 2000 AB
(ETA - obviously only talking about AB for outfielders - not for DH AB)

[Guest]

As usual the Captain is spot on with his comments. There are a couple of sites out there that defend this practice, however, saying that while the projections make no sense in aggregate it actually makes their projections more accurate.

Since I haven't figured out yet how to determine whether a projection was accurate, and I entirely disagree with that approach, I don't take much stock in what they say.

As always this is why you should make sure you have somewhere to ask questions when you obtain information like this.

[JP Kastner]

We do use a bit of a hybrid model when you have two or more players fighting for the same job. I'll describe the scenario.

It is pre-spring training and a team has two players fighting for the utility infielder job. It is clear that neither player has an advantage. Traditionally, that team's utility infielder receives 250 PA over a season. How do you assign playing time?

We only make decisions on facts at Mastersball. The "fact" is that a utility infielder for this hypothetical team receives 250 PA.

Option 1: Guess
We could simply guess who is going to win. We have a 50% change of being completely right and a 50% chance completely wrong.

Option 2: Split Time
We should split the time and give both 125 PA. If we do that, we are completely wrong for both players.

Option 3: Give Both 250 PA
We give each player 250 but make a note in the profiles that we are doing this until we know who wins the job. While we are 100% right on one player and 100% wrong on another, we have a 100% chance of getting one player right which is better than option one.

There are cases where we have too many PA or IP because we want to get the players right first. As we gather more useful information, we transition from players' to team. We try to make a note of our decision in the profiles. In other cases, we have not assigned anyone to a position. The Marlin's first/third base job is an example. We really don't have enough information to make an educated guess. Once spring training starts, we'll have a better understanding and make a better projection.

[kjduke]

Here is a major problem with every web site's player projections, including MB:

You cannot assign a realistic value to a player based on season stats when they are projected to have less than full-time plate appearances, unless that player is likely to be rostered and started on that basis. So, while projecting realistic total PAs over the player population may seem like a worthy goal, it's useless. A few illustrations ...

#1 Let's say that Pudge is expected to start 4 games per week. You create a season projection for him and assign a dollar value on that basis. That should be a good value because he's expected to be in someone's starting lineup and likely to play 4 games per week.

#2 Juan Pierre last season. He didn't have a starting job. You project Juan to get 300 PAs between starting once a week and getting a few pinch hit PAs, plus filling in for, say 6 weeks, while Manny is hurt or otherwise being Manny.

Now to assign a dollar value to Pierre you take the stats from his 300 PAs and compare it to other players whose projections are based on various types of roles. Let's say Pierre's worth $2 on that basis (although if he was the starter you think he'd be worth $26) . Then, let's say you've projected Aaron Rowand for 550 PA's and he's worth $3 (more than Pierre). Is Rowand really worth more? Anyone see the problem?

Here is the crux, and how it should be fixed. Pierre will have no value (or negative value if starting) when he's playing once a week. But when starting, you think he's worth $1 per week ($26 in full time value divided by 26 weeks). So what is he really worth? If he gives you $1 in season value per week over 6 weeks, he's worth $6 to your roster. That's double the value of Rowand, as opposed to less. Whereas Pierre is worth $1 per week when starting, Rowand is worth just 0.12/wk. The $2 value you've assigned to Pierre based on his 300 PAs, even if you are 100% spot on with your projection, is 100% meaningless in terms of his value to a fantasy team.

For any player projected for less than full-time ABs, a useful projection/valuation system needs to include the number of full-time weeks expected for that player and a value contribution per full-time week calculation. I do this, but I don't know of any site that does it - which makes valuation and projections for about half of the player population put out by every single web site useless, at least for the fantasy player.

[viper]

I'm pretty good at figuring things out but you lost me with your example and the overall theme of your comment.

Now, I'm a believer that when you see a problem you should have an alternative you think to be better. A guy like Shawn Childs doesn't use projections but he has an amazing intuitive sense of players, their skills and what you can expect from him. I also suspect his weekly roster management is in that superstar stratosphere. He also must spend countless hours every year to augment his many years of study. People like Todd, Lawr and others have spent major portions of their waking hours studying baseball. I love this game and it is my only fantasy sport but I can devote that much time - and I think I spend more time on theory and strategy than the majority of people.

Given you seem to dislike all projections, how do you differentiate players and judge when to take pitchers or hitters in a straight draft.

[Todd Zola]

KJ is right and is something I "talk" about but do not quantify. This was a HUGE topic back in the day of usenet newsgroups, rec.sports.baseball.fantasy

Recently, I have discussed guys like Kinsler and Chipper. They are downgraded due to injury, but you are going to be filling in their roster spot for 26 weeks. My point has been the minimum you are getting is paying the discount. The maximum is their staying healthy. In between is a substitute for that time, probably adding counting stats but with a worse average.

I have a folder of CVRCs where I take each player and add in substitute stats to give everyone 700 plate appearances and rank them.

I have another folder where everyone is prorated to 700 PA to get a "normalized" rank.

I have a ton of half-finished projects like this.

It is my hope that this time next year, after we get established after the merger and are not working so blind that there will be one site that at least discusses this sort of thing.

I don't think there is an ABSOLUTE way to quantify it, as we do not know the level of substitute stats.

But since my present crusade is to soften one's reliance on absolute quantification and look at things more in a relative "intrinsic" (thanks KJ) basis, knowing Kinsler and his substitute stats are closer to Utley than Phillips is a good start.

[kjduke]

A guy like Shawn Childs doesn't use projections but he has an amazing intuitive sense of players, their skills and what you can expect from him. I also suspect his weekly roster management is in that superstar stratosphere.

I've played against Shawn many times, and partnered with him. You are correct on his feel for players, wrong on the roster mgt from what I've seen (don't think his considerable edge comes from roster mgt).

Given you seem to dislike all projections, how do you differentiate players and judge when to take pitchers or hitters in a straight draft.

Viper, I think you missed part of my post. I project everything, re-read the last paragraph of my post.

[kjduke]

I take each player and add in substitute stats to give everyone 700 plate appearances and rank them.

I have another folder where everyone is prorated to 700 PA to get a "normalized" rank.

I have a ton of half-finished projects like this.

I can relate.

I don't think there is an ABSOLUTE way to quantify it, as we do not know the level of substitute stats.

You're right on this, but knowing the value of a player's skill set assuming a full-time role, along with a projection for the number of his full-time weeks would be very valuable.

[Todd Zola]

You're right on this, but knowing the value of a player's skill set assuming a full-time role, along with a projection for the number of his full-time weeks would be very valuable.

Above I called it a good start, perhaps very valuable is a better way to put it.

[Todd Zola]

A guy like Shawn Childs doesn't use projections but he has an amazing intuitive sense of players, their skills and what you can expect from him.

Shawn absolutely uses projections. He may not fire up Excel and design a projection model, but he as a projection for each player in mind.

He talks about drafting towards targets which requires an number in mind.

And if you have ever seen him in action post draft, looking over other people's teams, he is quite good at identifying their strength and deficiencies -- which has to be based on projections.

He may come up with them differently, but in the end, he has a number in mind like most of us do.

[viper]

Given you seem to dislike all projections, how do you differentiate players and judge when to take pitchers or hitters in a straight draft.

Viper, I think you missed part of my post. I project everything, re-read the last paragraph of my post.

If you are projecting everyone to a specific AB number [which is what I think you are saying], I like that idea. I've thought of doing it myself this season for an exercise. Todd even said he has done this.

One of my first years in the NFBC, Lance Berkman was known to be out for a month. I actually projected him based on his projected AB plus what I could get from a replacement player for that month. I projected the position rather than the player. I have often wondered if that concept could be extended. The problem is that Berkman's situation was unique.

[kjduke]

Given you seem to dislike all projections, how do you differentiate players and judge when to take pitchers or hitters in a straight draft.

Viper, I think you missed part of my post. I project everything, re-read the last paragraph of my post.

If you are projecting everyone to a specific AB number [which is what I think you are saying], I like that idea. I've thought of doing it myself this season for an exercise. Todd even said he has done this.

I project PA per week, the number of useful weeks and a value per week.

So I can take an injury-prone player and say that he will play only 22 weeks, or I can take a guy who I think could be a starter for part of the season, like Wieters last season, and project his value over 18 weeks, or a guy like Pierre in my first example who might be a fill-in with 6 full-time weeks.

What's important in any of this is to look only at a players projections for the time he could be in your lineup, and what he contributes and what's he's worth over replacement value. If a guy is playing twice a week he'll be on your bench - those stats won't find their way into team totals, so they are meaningless to project for fantasy.

[kjduke]

I don't think there is an ABSOLUTE way to quantify it, as we do not know the level of substitute stats.

As I look at this again, I don't think that's true. Because embedded in the dollar value of any player is an implied replacement value. You don't need to worry about replacement stats - I used this method a few yrs ago then coded it out because it isn't necessary .....

Let's take Albert. You have a $40 value on Albert and I assume you're not projecting any DL weeks, so that is $40 over 26 weeks = $1.54/wk. Now suppose Albert was going to miss half the season but when he played had the exact same skill set. He would be worth ($1.54wk * 13wks=) $20 to your team. You don't need substitute stats to know that, right?

Continuing on, you can do this for any player by:
1) projecting his skill set based on x number of PA over a full season.
2) calculating his value
3) dividing by 26 weeks
4) multiplying by the number of useful weeks you expect him to have

Works perfectly for injured players, injury-prone players, mid-season callups and bench players who you expect to have full time roles for at least a portion of the season.

[Todd Zola]

This was the direction the discussion ended up way back when on r.s.b.f. -- that the replacement was actually the replacement for the roster spot as a whole and not just the individual player, so sometimes you may be taxing a spot multiple times if you assume multiple players taking that spot.

Perhaps one way to attack that is determine replacement on a per week basis and value on a per week basis over replacement, then multiply by the number of weeks.

Still though, if I lose a guy and have a huge cushion in batting average, a guy likely to hit .230 with high steals is worth more to me than a high average, low production guy.

But then, at the auction/draft, you want to put yourself in the best position to be able to manage these issues as they come up, so you need to just set values in the proverbial vacuum, accumulate potential value and do all you can to convert that potential value into kinetic value as the season wears on.

[Guest]

I've talked to Todd about this a decent amount, and what I would say is that KJ's contribution here literally made me a better fantasy player just for reading it (for any Billy Madison fans, his posts here are the opposite of Billy's final speech of his academic decathlon).

I have to believe that this starts from the fact that the early fantasy leagues were all the Waggoner style where only DL trips could get you a player to replace and therefore you looked at the acquisition of a player as essentially rostering him for 26 weeks. In the newer world of leagues with lots of transactions this of course should skew player value.

In essence you are changing the thought process - instead of valuing Chipper Jones' statistics you are in essence saying "I would pay $X for a roster slot in which Chipper Jones was the primary player". I like that.

The only thing I am a little fuzzy on is the population of players this would affect.

I've got the following type of players:
Fulltime Projection - no change
Injury Risk projection - they would potentially increase in value
Minor league callup - potential increase in value
Fill-in for Injured player - potential increase in value
Utility player - probably decrease in value

So how do I address the risk that these part-time fill-ins aren't useful the week I need them to be? Is there a discounting factor?

And my bigger point - I am paying for a player based on his usefulness on my team. So I pay for Juan Pierre's 6 useful weeks vs Rowand's 25 (for example). But I paid for him to put on my active roster in week 1 - when Pierre isn't useful, and Rowand is. So why is Pierre "worth" more to me? Pierre might be worth more to me as a bench player, but as an active player on my roster, I still think Rowand is, to the extent he is more likely to earn me positive value in my lineup in weeks I need him. Or maybe I am thinking through this poorly.

I am 200% on board with adjustments for injured players and minor league callup types (though the risk with both types I think to some degree would need to be factored in even more in a valuation system like is being described here). I still don't get it all the way through on the part-time players. I understand strategically preferring a more talented player to a less talented one even if they don't have the PT established, but I don't understand how it makes current valuations or projections worthless. Maybe we need more clarity on the definition of "useful week".

But I am coming around. So stay with me here and walk me through it....

[kjduke]

For a part time player's relative value to that of a full-time player, you don't need to bring replacement value or stats into the equation.

For absolute value, there are more variables than the time it's worth.

[Todd Zola]

I am literally hearkening back to the salad days of usenet.

I was thinner, no grey hair and once in awhile went on a date with a girl.

Good times.

[kjduke]

The only thing I am a little fuzzy on is the population of players this would affect.

I've got the following type of players:
Fulltime Projection - no change
Injury Risk projection - they would potentially increase in value
Minor league callup - potential increase in value
Fill-in for Injured player - potential increase in value
Utility player - probably decrease in value

correct on all fronts

So how do I address the risk that these part-time fill-ins aren't useful the week I need them to be? Is there a discounting factor?

And my bigger point - I am paying for a player based on his usefulness on my team. So I pay for Juan Pierre's 6 useful weeks vs Rowand's 25 (for example). But I paid for him to put on my active roster in week 1 - when Pierre isn't useful, and Rowand is. So why is Pierre "worth" more to me? Pierre might be worth more to me as a bench player, but as an active player on my roster, I still think Rowand is, to the extent he is more likely to earn me positive value in my lineup in weeks I need him. Or maybe I am thinking through this poorly.

All good questions, and they show the limitations of a pure quantitative approach, even a better one. The art of management still is needed. As for Pierre vs Rowand, you're correct in that it comes down to what you need. If you're in a deep lge such as an NL-only and you'll need to start Rowand, he would be worth more. If replacement players are available thru FAAB with a theoretical $0 value, you'd be better off with Pierre unless you determine the cost of holding that roster spot is too high.

I am 200% on board with adjustments for injured players and minor league callup types (though the risk with both types I think to some degree would need to be factored in even more in a valuation system like is being described here). I still don't get it all the way through on the part-time players. I understand strategically preferring a more talented player to a less talented one even if they don't have the PT established, but I don't understand how it makes current valuations or projections worthless. Maybe we need more clarity on the definition of "useful week".

It is trickier for an injury replacement type than the others , no doubt. But no matter how you slice it, you still come up with a more meaningful value than taking into consideration a part-time players stats over an entire season knowing that many of those stats are irrleevant because he'll be on your bench.

[Todd Zola]

I need to think things through.

If we value the entire roster spot, the need for replacement adjustment is eliminated so long as it is accounted for, which is the direction I was going but that whole girl thing got in the way.

Plus those files are buried on a zip-drive from my old Mac.

[Todd Zola]

It is trickier for an injury replacement type than the others , no doubt. But no matter how you slice it, you still come up with a more meaningful value than taking into consideration a part-time players stats over an entire season knowing that many of those stats are irrelevant because he'll be on your bench.

But if we simplify things to just wanting to assemble a roster at our draft or auction, only worrying about how much to discount Kinsler, Chipper or Hamilton, etc., that edge alone would be important.

Chipper's ADP is about 135. Michael Young is around 95. Jones has a discounted value due to injury, Young is expected to play all 26 weeks. It would be good to know if Chipper plus a sub would be more valuable than Young, and that is something we can quantify.

[kjduke]

I understand strategically preferring a more talented player to a less talented one even if they don't have the PT established, but I don't understand how it makes current valuations or projections worthless.

If Juan Pierre gets 27 PA per week over 6 weeks as a starter you can look at his skill set, make a projection as if those 6 week skills happened over a full 26 weeks and run his full-season value on your existing model. Then multiply by the pct of weeks he'll actually play full-time to get his value for those 6 weeks - and you'll understand what he's worth to you while playing.

Now, suppose MB is projecting 300 total PA on the season, which includes those 6 full-time weeks (27*6=162) and another 138 PA over the remaining 20 weeks. That is 7 PA per week when he isn't starting. Do those matter? No, because he'll be on your bench. Accordingly, projecting those stats are completely meaningless fantasy-wise and even worse they distort your understanding of his value.

Here's another scenario, suppose for this season you project Willie Taveras to get 300 PA, same as Pierre in our theoretical example, and Willie puts up exactly the same stats. Is he worth more or less? Well, because Willie is playing in an extremely crowded OF, the projection is that he will be a role player with playing time spread over the season rather than a viable full-time injury replacement player. Willie's value then would be the same as that of his projected season stats over 26 weeks, or $2. So he's worth $2 while Pierre's worth $6, even though they finish the season with identical stats. Which is why I believe that full season projections for part-time players, and an assigned dollar value based on those stats, are worthless.

[kjduke]

Chipper's ADP is about 135. Michael Young is around 95. Jones has a discounted value due to injury, Young is expected to play all 26 weeks. It would be good to know if Chipper plus a sub would be more valuable than Young, and that is something we can quantify.

Right, but you don't need to separately quantify replacement value, that is built into your value model at $0. All you need to do is look at Chipper's projected stats as if he played every week, then multiply by the pct of weeks he'll play to see if he should be worth more or less than Young.

[david_hume]

great discussion here about an issue that has been vaguely in the back of my mind for a while.

one thing to add: let's remember that we're not always able to predict the weeks when these part-timers will get full time action by the time our lineups need to be set. guys get hurt on mondays, minor league callups happen on tuesday mornings, and sometimes the manager says, "i'm gonna give this guy a shot for the forseeable future" and the guy gets two pinch-hit chances on the 10-game road trip. if you think juan pierre will start six weeks of the season, and all of his dollar value comes from that time, i think it's incorrect to assume you'll get him in your lineup for all six weeks. this argument would mute the adjustments kj is making, though they are definitely still important and i am fascinated to re-read this whole thread.

[Guest]

KJ - I am somewhat there. Of course my main auction league is an old school rules where you can only get rid of a guy on the DL, and no bench. So a lot of this is not as relevant to me.

But you raise a bigger point - the size of one's bench should have an impact on the value of certain players one would or wouldn't draft, and the current methodology doesnt address that. The current methods out there just say - "in a universe where no one can be replaced for whatever reason, once they are on your roster they stay on it for the season" - and that doesn't represent our reality. So there is no doubt current methods don't reflect most league rules. The question is how much of the adjustment should be made qualitatively and how much quantitatively. I'd say there is likely no better quantitative method than what you describe. I am still not at the point where I think everything else is worthless but agree that manipulation must be done by some means.

[Todd Zola]

Chipper's ADP is about 135. Michael Young is around 95. Jones has a discounted value due to injury, Young is expected to play all 26 weeks. It would be good to know if Chipper plus a sub would be more valuable than Young, and that is something we can quantify.

Right, but you don't need to separately quantify replacement value, that is built into your value model at $0. All you need to do is look at Chipper's projected stats as if he played every week, then multiply by the pct of weeks he'll play to see if he should be worth more or less than Young.

I see it more as adding the stats of the substitute guy (don't want to say replacement as that might confuse). Just prorating Chipper assumes the sub performs at the same level as the original guy.

This can obviously be done for Sheets, Harden, etc., but the sub is even more abstract as you can often choose a stronger match up from your bench.

[Todd Zola]

great discussion here about an issue that has been vaguely in the back of my mind for a while.

one thing to add: let's remember that we're not always able to predict the weeks when these part-timers will get full time action by the time our lineups need to be set. guys get hurt on mondays, minor league callups happen on tuesday mornings, and sometimes the manager says, "i'm gonna give this guy a shot for the forseeable future" and the guy gets two pinch-hit chances on the 10-game road trip. if you think juan pierre will start six weeks of the season, and all of his dollar value comes from that time, i think it's incorrect to assume you'll get him in your lineup for all six weeks. this argument would mute the adjustments kj is making, though they are definitely still important and i am fascinated to re-read this whole thread.

This is the same sort of built in vagueness as when dealing with the positions of middle infielder, corner infielder and utility. All we can do is look at things in the aggregate and get an idea.

[rotodog]

I actually like the ideas that are being presented here. I think really good Roto players actually realize most of this concept, but have never tried to calculate of quantify it. These are the same people that dont just follow a valuation list blindly . These are players that intuitively understand how to construct a roster and plan properly.

My question is this:

If one were to try and construct a quick and dirty method for finding a weekly value with only projections and dollar values, how would that go? I am not trying to spend the time to normalize everyones AB's or PA's and then project the skills on a FT basis. I am looking for quick and dirty...More for the FT half year players and FT injury risks...I am not concerned so much with utility players values...

Would it be something as simple as taking Ab's + BB and divide that by say 27 Ab a week to come out with a rough number of weeks a players yearly value is based on? I know this is very rough and a leadoff hitter and a # 7 hitter will be off a bit...

But take that yearly projection and divide by the weekly total to come out with a weekly dollar value while active?

Am I missing something here? Or could that be a quick and dirty method to identify some value? is there a better way that is quick without having a whole projection system and methodology engine at your disposal?

Would love some thoughts from KJ/ Todd/JP or anyone that has any advice..

[viper]

it seems that you should take longer looks at players with reduced ABs which could only be attributed to DL days. A question to those projecting playing time is how often is DL time actually predicted.

[aburt19]

I can understand the applicability to mixed league or shallow only leagues. But 12 team AL only is another matter. There
may be some applicability to positions that have players right at the edge of being rosterable (such as OF). But for the
middle infield and catcher, it's different. If I have Kinsler and I figure that there is going to be time on the DL, my choices
will probably be part time players like Omar Vizquel, who will probably hurt more with his BA than whatever he produces
in his two starts a week.

Also, if you use the roster argument, don't you need to go back to having a separate pool for each position because that's
the only way that you get an idea of what you will get if Kinsler is injured? In the two player pool, the replacement player
will probably be an OF, which doesn't help if a middle infielder gets hurt.

[kjduke]

Just prorating Chipper assumes the sub performs at the same level as the original guy.

Actually, If he's worth $30 playing a full season and we expect him to play 2/3rds of the season he's worth $20. So that's $10 in lost value because the replacement player is worth that much less, not the same.

The reason this works is the assumption that his replacement puts up $0 per week in value, which already is a built-in assumption to your value model for any player (zero value doesn't mean no stats, it is the expected value of stats for a replacement player). So what we end up with is an accurate valuation of Chipper at $20 in relation to other players without having to know his replacement's stats. Once you play around with assumptions in a spreadsheet it becomes much clearer.

However, you do have to plug in his replacement's stats for that 1/3rd of a season if you want to calculate out your rosters projected stats, but it isn't needed just to get a player's value.

[kjduke]

I can understand the applicability to mixed league or shallow only leagues. But 12 team AL only is another matter. There
may be some applicability to positions that have players right at the edge of being rosterable (such as OF). But for the
middle infield and catcher, it's different. If I have Kinsler and I figure that there is going to be time on the DL, my choices
will probably be part time players like Omar Vizquel, who will probably hurt more with his BA than whatever he produces
in his two starts a week.

Also, if you use the roster argument, don't you need to go back to having a separate pool for each position because that's
the only way that you get an idea of what you will get if Kinsler is injured? In the two player pool, the replacement player
will probably be an OF, which doesn't help if a middle infielder gets hurt.

In leagues where replacement stats are hard to come by such as an AL-only, injured players are worth less in relation to lges where replacement stats are better. The same would be true for positions where replacement stats are harder to come by, such as C or MI. The difference still exists and easily can be calculated, but it will be less.

[kjduke]

If one were to try and construct a quick and dirty method for finding a weekly value with only projections and dollar values, how would that go?

Quick and dirty 2 options:
1) Take a very comparable player in terms of skill that is projected to be full-time and multiply his value by the percentage of weeks for which you expect the part-time player to be full-time.

2) For a guy like Milton Bradley, who probably is full-time when he's healthy, you can take the year projection - MB has him at 360 AB - and scale that up to a full-time ABs (if you assume 540 you'd multiply all his stats by 1.5) to get what he'd do in a full season. Plug those stats into your model to get a dollar value, then multiply by his actual pct playing time (which in this case would by 67%, again using MB's numbers) and you'd have an accurate playing-time adjusted value.

[Todd Zola]

All else was already understood, to me this is the key and can be quantified, or at least estimated

However, you do have to plug in his replacement's stats for that 1/3rd of a season if you want to calculate out your rosters projected stats, but it isn't needed just to get a player's value.

Right now, we have Michael Young at $14, Jorge Cantu at $14 and Chipper at $14. But assuming he does not miss MORE than we project, Chipper's $14 is the most valuable.

If the other guys were $17 or $16, the scenario might not be as intuitively obvious to account for in your head.

So doing the value on a per week basis like KJ suggested earlier or adding in some estimated level of sub stats will quantify it. I would need to do some numbers to see which way I liked better.

[Todd Zola]

I can understand the applicability to mixed league or shallow only leagues. But 12 team AL only is another matter. There
may be some applicability to positions that have players right at the edge of being rosterable (such as OF). But for the
middle infield and catcher, it's different. If I have Kinsler and I figure that there is going to be time on the DL, my choices
will probably be part time players like Omar Vizquel, who will probably hurt more with his BA than whatever he produces
in his two starts a week.

Also, if you use the roster argument, don't you need to go back to having a separate pool for each position because that's
the only way that you get an idea of what you will get if Kinsler is injured? In the two player pool, the replacement player
will probably be an OF, which doesn't help if a middle infielder gets hurt.

In deep leagues it comes down to category management for sure, where the clusters are in the standings, etc.

But this is where having some positional flexibility comes in.

As far as needing more pools, no. The idea is the available replacements are all equally crappy in deep leagues. You use Omar Vizquel to play for Kinsler, I use Reggie Willits instead of Josh Hamilton.

six of one, half dozen of the other

[AllstonRockCity]

I can understand the applicability to mixed league or shallow only leagues. But 12 team AL only is another matter. There
may be some applicability to positions that have players right at the edge of being rosterable (such as OF). But for the
middle infield and catcher, it's different. If I have Kinsler and I figure that there is going to be time on the DL, my choices
will probably be part time players like Omar Vizquel, who will probably hurt more with his BA than whatever he produces
in his two starts a week.

Also, if you use the roster argument, don't you need to go back to having a separate pool for each position because that's
the only way that you get an idea of what you will get if Kinsler is injured? In the two player pool, the replacement player
will probably be an OF, which doesn't help if a middle infielder gets hurt.

In leagues where replacement stats are hard to come by such as an AL-only, injured players are worth less in relation to lges where replacement stats are better. The same would be true for positions where replacement stats are harder to come by, such as C or MI. The difference still exists and easily can be calculated, but it will be less.

In a very deep AL only, where there's say 12 guys on the wire in July who aren't backup catchers, and there is no bench, and guys are either active or FAs, unless they're on the DL (in which case they are reserved) aren't the underwhelming guys who don't get hurt more valuable to the construction of one's team than the guy you know won't play every day? It seems that one of the main backbones of your argument is that the underwhelming guys that don't get hurt are on your bench and therefore useless (and in the NFBC, I agree). But when there is no bench that must make a huge difference, no?

This is a very enlightening thread and I've enjoyed reading it several times over.

[Guest]

I generally agree, ARC. There are probably a couple of guys I might bump mentally in my head, but I think in no-bench, no-free transaction leagues this is not a major factor. Once you get into benches and free subs, yes, it should be included in the thought process.

[Todd Zola]

Unfortunately, I don't have the time to really explain as we are under some site content crunch, but thinking this through, I believe the general uselessness of what we and others do was a bit overstated.

Unless I am missing something or we are talking about leagues with daily transactions, we can only make adjustments for injury expectations.

If Pudge is only playing 4 games a week and we don't expect him to miss any time on the DL, he will occupy that roster spot.

Same with Ty Wigginton, Brett Gardner, Seth Smith, Erick Aybar or any useful player that will miss some non-DL games.

To say half of the projections are useless is hyperbole.

Recent data shows 1 in 3 players hits the DL, so the upper boundary of players that can logically be adjusted in 33% of the pool. And clearly, not all are "injury prone".

Something I think is very possible, and can be automated, which is key for a supplier of this information, is to "project" DL days much like we project any other stat, using a 3 year weighted average as a basis and doing some studies to see the likelihood a player returns to the DL based on his recent DL history.

We could then use this projected number of DL days to add in substitute stats.

Of course, as has been mentioned, a guy may go on the DL on a Wed and come off on a Thurs, so you are only replacing 1 weeks worth of stats, but that can be factored in, by converting the DL days to #weeks - a longer explanation is necessary, but hopefully you get the drift.

As has also been suggested, many do this intuitively with Kinsler, etc. But some like a number, so we can certainly discuss how to put a number on it.

Keep in mind, the objective is to provide some relative value to best assemble a roster to achieve maximum points.

The fact that some guys are more valuable in season as replacements is moot. What we want to quantify is their benefit of occupying an active roster spot from Day 1.