Fantasy Baseball Draft Strategy – Part II

How To Value Players All Season

Only three things go into valuing a player: (1) How many points can he score? (2) What’s his situation, and how many points would you expect him to score under those circumstances? (3) What are those points worth in your league?

1. Assessing Skills and Situation

Those individual decisions determine whether a player might grow or remain flat or regress, and if so, by how much:

A. Skills

Stats: what his lines looked like last year versus the three-year average; BABIP/strand rate (to see how reliable those stats were); home/road splits; left/right splits; steals/caught stealings; K:BB ratio; ground ball/fly ball rates, etc.

2. Scouting – pure power, speed, defense (impacts opportunity), bat-eye. For pitchers, velocity, strike-throwing capacity, repertoire, life.

3. Pedigree – Where a player was drafted, how he’s perceived in the organization, what the team already gave up to acquire him – this will buy him more leeway during a slump and more opportunities to succeed.

4. Health

B. Historical Comps/Career Arcs

A player’s own history is not always enough – consider younger players like Jorge Soler, or even younger yet with Yasiel Puig. One must consider how baseball players of a certain age, with a certain experience and skill-set do in general.

1. Age – hitters tended to reach their peak around age 27 and then ‘ail’ gradually into their mid-to-late-30s, until they generally went over a cliff. For better-growing pitchers, velocity still peaked early, but the ‘fully-seasoned’ gentleman of the game (thinly disguising the obvious, as usual) was supposed to have mastered his craft, if not his repertoire, in his early 30s.

2. His experience – A 700-1000 career at-bat hitter seemed to ‘get it’, while the light bulb for a pitcher often seemed to actually, finally, go on in a year or two. But in passing along this tip, there is still often a nod made to the idea of a ’divide’ between pitcher and hitter. Of course, there is more recent research (post-steroid era, so 2006-2013) that finds peak-and-decline curves for both pitcher and hitter that puts them at the exact same age: they peak right away, beginning their decline at, on average, about age 26. One explanation for this is that players are just generally more fucking ready physically and fundamentally than the majority of the aforementioned ‘Superstars’ – so there just isn’t any growth right when they hit the majors. Frontline colour Some ‘Age Series’ graphs such as these aren’t just ‘nice to know’ introspective historical studies.

C. Context

1. Line-ups, RBI and runs depend on them. Wins depend on how many runs you score. ERA and WHIP depend on the kind of defence you have behind you. The roster and who’s in the organization largely determine your playing time and how high in the line-up you hit and where you fit in the organization and typically the higher up that score hits, the more opportunity you’ll have.

2. Stadium dimensions – Some stadiums favor different hitters and/or pitchers significantly.

3. League-Division (NL East pitchers face much less daunting foes than AL East ones, for instance, with a ~½ Run ERA differential)

2. Creating a 2015 Stat-Line

Having extracted all the relevant facts of his talent and situation, you might make yourself a prediction of his upcoming statistical line. There are two different statistical lines you might make, and there are reasons for and against either choice.

A. Projections – a stat line is assigned to a player in the form of an average of his many possible 2013 seasons. It’s the 50th percentile season of his skill set and situation, neither better nor worse than what a given player is likely to produce given his skill set and situation. The theoretical benefit to projecting every player is that the exercise is ostensibly unbiased – that is, you give every single player his 50th percentile season. You don’t favor one player by giving him his 75th percentile season and another his 25th.

The problem with this is two-fold:

1. You think that you are giving a player 50 per cent of his seasons, but your sense of his skills and his context are more charitable than someone else’s. So a set of 50th-percentile numbers for Yasiel Puig might be someone else’s 75th-percentile numbers. Basically, your projections are no less subjective than someone else’s guesses or hunches; the subjectivity is just baked in the input phase of the process, rather than the output phase.

2. If you used a formula of three-year averages and projecting to the mean, then you’d be being less subjective, but your projections would bear almost no resemblance to the actual distribution of stats when looking at a given season. A simple example would be: I get $1 coin-flip money. Heads, you lose, tails, you win. So my expected return – my 50th percentile outcome – is 50 cents. I should project myself to win 50 cents. But this is never going to happen – I’m either going to win $1 or I’m going to suck down nothing. So by projecting 50 cents as my stat line, I guarantee that I’m going to be wrong no matter what happens.

To put an analogy in baseball terms, someone’s going to go 20-5, or bat .330 or sock 40 homers. Well, if you give everyone the 50 per cent of their performances, probably no one does any of those. But you know somebody’s gonna get their 95th percentile season, somebody’s gonna get their 25th percentile season, so it would be nice to have something that looks like that.So that’s where we landed:

B. Allocations – In this allocation, we assign players stat lines that copy as closely as possible the actual distribution of MLB player stats in a real season; the guy with his 75th-percentile season, the guy with his 25th. We throw someone a 20-win season and throw someone who is just as good a 13-win season because, well, for whatever reason, that’s the way baseball goes.

The problem with it is, it’s completely ad hoc and arbitrary to decide that, in the words of Tom Peters, greatness lies ahead for your friend, Player X, while mediocrity is your destiny as Player Y, when the two of you have similar skill sets and both have ended up in similar circumstances. Your money isn’t worth 50 cents and you can have it back. You offered to give us each 50 cents. Now you’re telling me you’re going to win a dollar while I win nothing, but in fact it’s just as likely that I win the dollar and you win nothing.

Anyway, at the end of the day your stat lines will be for everybody, and somewhere you’ll be forced – if you want to have your draft/auction lines based on something tangible – to use projections, but I like to throw in a bunch of predictions, which I like to think of as something like hunches. A hunch is some emphasis you place on a skill, stat or contextual detail that just doesn’t seem like it rises to the level of something you’d make universal, but in this one case it just jumps off the page at you. Aggregating everything disparate – not all of which is so easily stated in quantitative terms – into a projection is certainly part art as well as science, and we know that some things work synergistically with each other, say a benefits from location change, that affects players unequally or differently to varying degrees. But leave room for hunches, and be honest with yourself about how fallible they can be.

C. Volatility and reliability – You lose information when you reduce an array of diverse statistics about a player to a 50th percentile statline. By itself, a projection tells you only that the player is going to have an average season. Not all averages handle the same distribution of highs and lows. The numbers might look similar, but you’ll experience very different outcomes when you plug Gonzalez’s projection and Puig’s projection into your fantasy line-up. You’ll get fewer truly great games out of Gonzalez but fewer virtually-zero games with Puig. You might want to pay extra for reliability early on in your draft (or at auction, where you can’t wait to draft everyone because you keep overspending your budget on awesome players.)

The projected average numbers are far less important in the middle and especially at the end of your draft (paying market value at auction) – you want the player to be capable of much better in his 80th or 90th percentile season, even if the possibility that he gets cut or demoted outright, i.e., his 20th-percentile season, drags his average down a bit. That’s why the 22-year old Trea Turner, who did nothing in 2015, should be on your 12-team mixed league bench, while the 33-year old J.J. Hardy probably doesn’t.

Those straight projections can’t be used alone to create a cheat sheet – you need to have information on a player’s volatility as well as reliability.

3. Translating Performance into Value

Once you’ve come up with your projections for all of the players in 2016, you then have to find some way to turn them into a cheat sheet or dollar values list. It’s easy to look at your Miguel Cabrera projection and see it as your 20th line and say he ought to be high on your list. But how does it stack up against Andrew McCutchen and by how much? Would you rather have a .330 batting average in 500 plate appearances, or 38 home runs in that time? That doesn’t jump out at you as obvious. And the relative comparison among all the players and across the categories is not something we can eyeball a solution for.

To address that, we need two key concepts:

Authentic nfl jerseys for cheap A. Replacement (or VORP – Value Over Replacement) – How much better a player is than any player you could get on the waiver wire of your league. If you are in a 12-team mixed league, then you can say that the top 168 players on the offensive side are starting at any given time (if you start 14 offensive players). Because number 169 might be a steals specialist with zero power, or a big power bat that can’t hit .250, you can find a replacement value by averaging a few groups of the next 10-20 players. Let’s use a random number generator to find our top 168 hitters and write a little program which compresses the players to only three lines. We asked the random number generator to list the top 168 hitters in some random mixed league of 12 teams using a few standard measures of production, and then compressed them down to only three lines per player. authentic nfl jerseys for cheap

For this reason, any starter’s value is related to the degree in which his stats exceed or fall under these benchmarks (again, assuming we’re using these numbers to represent your league parameters). So Cabrera is projected for 30 HR. He’s at 20 HR over replacement. You’d subtract his RBI, runs, SB and average (pro-rated for at-bats) likewise. After subtracting replacement-stat value from all of the hitters on this list, we now have their stats in reality – as much as the stats can serve that purpose. A Ben Revere is projected for 1 HR. His true home runs value for purposes of his value is –9.

But this still doesn’t solve for the original question – is a .330 average in 500 at-bats worth more than 38 HR, in other words, is a 75 point difference in batting average in 500 at-bats worth more than 28 HR? We’re asking which stat is the bigger outlier relative to the player pool, more likely to affect the standings. For that, we need a second concept – Standard Deviation.

* There are a lot of finer points being skipped here, such as: some starting catchers will be worse than replacement, I’m ignoring position scarcity, the size of your bench and whether your bench or the waiver wire is the starting point for replacement value, that a prospect who plays half a season will sometimes be worth more than replacement value, because you get the other half of a season from someone, or that shuttling in two-start pitchers or platoon players changes the baseline, etc.

B. Standard Deviation – this represents the average amount the little suckers (data points or players’ stats in a given category) hang out away from where they’re supposed to be in that category (the mean in that category). So say that the average number of homers in the usable player pool is 15 and that the player pool is composed of four players, two of whom have 20 and two of whom are sad little fellas with 10. Then the standard deviation is five HR. But say the average were also 15, but then there are two of them who aspire mightily to get it up to 30 and two of them who are fired up about 0001. Then the standard deviation would be everything: 15.

Note that in the first case the data points are bunchier and in the second they are sparser. What does that mean for value of a stat-line? Well, if you look back at that 1921 set, the dudes ranked second through 14th in homers all had between 24 and 16 homers in baseball’s tricky-to-comp-ball era; standard deviation wasn’t that great for that player pool. But the Bambino walked on to the field, leading the league in dingers with 59! That’s what Babe Ruth does for your fantasy team in that light – he goes and crushes all of your homer opportunities essentially himself. Standard deviation and value-above-replacement tell us why. If you say that replacement value for that year was about five, then Ruth had 54 homers above replacement. And if standard deviation was say 4, then Ruth was roughly 13.5 standard deviations above replacement.

3. Translating Performance into Value

Once you’ve come up with your projections for all of the players in 2016, you then have to find some way to turn them into a cheat sheet or dollar values list. It’s easy to look at your Miguel Cabrera projection and see it as your 20th line and say he ought to be high on your list. But how does it stack up against Andrew McCutchen and by how much? Would you rather have a .330 batting average in 500 plate appearances, or 38 home runs in that time? That doesn’t jump out at you as obvious. And the relative comparison among all the players and across the categories is not something we can eyeball a solution for.

To address that, we need two key concepts:

Authentic nfl jerseys for cheap A. Replacement (or VORP – Value Over Replacement) – How much better a player is than any player you could get on the waiver wire of your league. If you are in a 12-team mixed league, then you can say that the top 168 players on the offensive side are starting at any given time (if you start 14 offensive players). Because number 169 might be a steals specialist with zero power, or a big power bat that can’t hit .250, you can find a replacement value by averaging a few groups of the next 10-20 players. Let’s use a random number generator to find our top 168 hitters and write a little program which compresses the players to only three lines. We asked the random number generator to list the top 168 hitters in some random mixed league of 12 teams using a few standard measures of production, and then compressed them down to only three lines per player. authentic nfl jerseys for cheap

For this reason, any starter’s value is related to the degree in which his stats exceed or fall under these benchmarks (again, assuming we’re using these numbers to represent your league parameters). So Cabrera is projected for 30 HR. He’s at 20 HR over replacement. You’d subtract his RBI, runs, SB and average (pro-rated for at-bats) likewise. After subtracting replacement-stat value from all of the hitters on this list, we now have their stats in reality – as much as the stats can serve that purpose. A Ben Revere is projected for 1 HR. His true home runs value for purposes of his value is –9.

But this still doesn’t solve for the original question – is a .330 average in 500 at-bats worth more than 38 HR, in other words, is a 75 point difference in batting average in 500 at-bats worth more than 28 HR? We’re asking which stat is the bigger outlier relative to the player pool, more likely to affect the standings. For that, we need a second concept – Standard Deviation.

* There are a lot of finer points being skipped here, such as: some starting catchers will be worse than replacement, I’m ignoring position scarcity, the size of your bench and whether your bench or the waiver wire is the starting point for replacement value, that a prospect who plays half a season will sometimes be worth more than replacement value, because you get the other half of a season from someone, or that shuttling in two-start pitchers or platoon players changes the baseline, etc.

B. Standard Deviation – this represents the average amount the little suckers (data points or players’ stats in a given category) hang out away from where they’re supposed to be in that category (the mean in that category). So say that the average number of homers in the usable player pool is 15 and that the player pool is composed of four players, two of whom have 20 and two of whom are sad little fellas with 10. Then the standard deviation is five HR. But say the average were also 15, but then there are two of them who aspire mightily to get it up to 30 and two of them who are fired up about 0001. Then the standard deviation would be everything: 15.

Note that in the first case the data points are bunchier and in the second they are sparser. What does that mean for value of a stat-line? Well, if you look back at that 1921 set, the dudes ranked second through 14th in homers all had between 24 and 16 homers in baseball’s tricky-to-comp-ball era; standard deviation wasn’t that great for that player pool. But the Bambino walked on to the field, leading the league in dingers with 59! That’s what Babe Ruth does for your fantasy team in that light – he goes and crushes all of your homer opportunities essentially himself. Standard deviation and value-above-replacement tell us why. If you say that replacement value for that year was about five, then Ruth had 54 homers above replacement. And if standard deviation was say 4, then Ruth was roughly 13.5 standard deviations above replacement.

D. Converted Performance into Value on the Fly, Against Actual Stats (Plus Other Considerations) — All of the above said, I am not terribly dependent on projections to evaluate my players. After all, I do the research and create a cheat-sheet slot for all those entropy factors as best I can. After enough auctions and drafts, I have a pretty solid idea of what any given statline is really worth without plugging the numbers into every individual line, and I know it’s just a made-up rendering of some arbitrary 2016 season, anyway, and just about as scientific as placing a player into the cheat-sheet slot itself.

And, besides, I’m not a prisoner of my cheat sheet format when the auction or draft gets going – perhaps I take my No 12 SS over my No 10 one, at the last minute, based on volatility concerns and the like, when the devil demands it.

And even if you go to three-year averages and regress to projected numbers to create your projections, even then you are going to make decisions about playing time, park effects, age/experience/health related growth/regression that are not hormone-free (as well as, perhaps, not perfectly logical). And even if you were to automate those factors – park effects for gap-hitting lefties that are 6-2, 190 and score a 65 on the power scale, downgrade it for a player’s schedule, his historical comps for age-related career arcs – you are still making a lot of choices on the inputs that are very, very ambiguous. And even to the extent your model ends up being more precise than it is now, and even to the extent that it has backtested those results to your satisfaction, there is no guarantee that your conscious ‘aggregation’ of the factors that make up value will be any better than the unconscious ‘snapshot assessment’ of an experienced (and, as important a factor, open-minded) evaluator – at least not unless the winds of change remain at their current velocity long enough for you to amass a much larger sample than the one we have now (basically, you would need two or three decades at your current pace to narrow the gap). (But bear in mind, I am not saying this will be impossible, because it might not.)

The point is: no amount of staterms or RSAs can give you a shot at estimating the true dollar value of a fantasy football player, but once you’ve cultivated yourself a pile of whichever material facts you like to believe in, you’ll eventually have to assigning values to those beliefs. If (1) you’re an insect-eyed stickler for cheat-sheet discipline, then you’ll need to compare those projected stats to the replacement-value baseline, then to compare those baselineless Z-scores to whatever those Z-scores are worth (i.e., their prospective categorical buy-out) and (2) you can measure that buy-out with the number of standard deviations away from the baseline they are.