Seasons Are Arbitrary Endpoints
We often roll our eyes when announcers cite a player’s stats over the past 15 days. We’ll groan when they tout how many home runs he’s hit since the All-Star break. We’ll throw the remote when a pitcher’s last five starts are mentioned. And yet, when we attempt to analyze a player here, there’s nary a blink if ‘last season’ is mentioned.
Well, guess what. Seasons are also arbitrary endpoints. Yes, they are arbitrary endpoints that allow for easy analysis, and ones that we have all agreed to use. And, if we didn’t use them, statistical analysis would be rendered fantastically difficult. Our record books would look very strange. We’d have to phrase things very carefully.
But the fact remains, years can be as arbitrary as months and weeks. Consider this excellent comment by user James Lewis when confronted with the year-to-year power oscillation of Good Aubrey Huff and Bad Aubrey Huff:
Data start and end points.
If we instead turn Huff’s “seasons” into July ’04 -June ’05, July ’05 – June ’06, etc. we get a very different picture of the player:
ISOs:
Season 1: 0.176
Season 2: 0.186
Season 3: 0.186
Season 4: 0.203
Season 5: 0.222
Season 6: 0.157
Season 7: 0.173Rather than bouncing up and down we see a steady development, followed by a down year, and a rebound in the final season. This is no less a description of Huff’s performance over time than any other delineation, and yet he doesn’t appear to be as “up-and-down” as we have come to expect. My guess is a lot of players show the same career path the Huff does, except they hide it better with the standard start and end points that we use – seasons.
Well that’s a very effective salvo for the commentariat and also one of the great benefits of having an open community of bright minds discussing baseball seriously. String together a couple bad half-seasons, in other words, and you’ll look worse than if you’d alternated those good and bad half-seasons. There is no Good Aubrey Huff, and there is no bad Aubrey Huff. There is only Aubrey Huff. Post-peak, perhaps, but still Aubrey Huff.
Naturally curiosity pushes us to find other Huff-like players. 32-year-old Vernon Wells has famously alternated good and bad years schizophrenically, as he has shown ISOs of .194, .239, .158, .197, .140, .242 and .169 since 2005. The yo-yo calms down dramatically if you switch to the July-June format as we did with Huff: .228, .171, .154, .176, .203, .179. The highs aren’t quite so high and the lows not quite so low. Now his current .169 ISO looks just about right for his recent history.
A similar story emerges for the Twins’ entry into the on-again-off-again derby, 32-year-strong Michael Cuddyer. Though his ISOs since 2004 (.177, .159, .221, .157, .120, .245, .146, .192) don’t perfectly fit the yeah-and-nope swings of a Wells or Huff, they do calm down some when you start a season in July and end it the next June (.152, .220, .190, .135, .219, .209, .153).
But there’s one player that fits this analysis to a “T.” There’s a certain portly 27-year-old slugger in Milwaukee that has shown more power in odd years than even, making some wonder if the team that signs him in free agency will have to endure a poor first year. Yes, it’s true that this former major leaguer’s son has ISO’ed the following, starting in 2006: .213, .330, .231, .303, .209, .264. And yet…
Prince Fielder’s ISOs from July to June:
| “Season” | ISO |
|---|---|
| 06-07 | 0.265 |
| 07-08 | 0.266 |
| 08-09 | 0.274 |
| 09-10 | 0.264 |
| 10-11 | 0.246 |
Thanks to my many twitter friends for some of the names in this piece.
With a phone full of pictures of pitchers' fingers, strange beers, and his two toddler sons, Eno Sarris can be found at the ballpark or a brewery most days. Read him here, writing about the A's or Giants at The Athletic, or about beer at October. Follow him on Twitter @enosarris if you can handle the sandwiches and inanity.
“Well, guess what. Seasons are also arbitrary endpoints.”
I’m not sure thats true. With the offseason, it could be argued that each season is a sample from a different distribution. Its much more likely that a player’s true talent level is constant over the course of a season than it is between seasons.
This is a great article. Very curious data presented. To address both the evidence here and the many factors that commenters are leaving: the title might more accurate as, “April-September Can Be as Arbitrary of a Sample as Random 30-game hitting streak.”
This is my thought, too. The fact that games are not played for 6 months, new players and coaches come and go through signings and trades, rookie promotions, etc. make the offseason a MUCH less “arbitrary” endpoint than any other sample, in my opinion.
The data presented above is interesting and we should always look beyond season-ending stats, but we already do that quite often. I have read numerous articles pointing out Batista’s late 2009-season power surge as a harbinger of his huge 2010, and we all know not to look at season stats for a guy like Dan Uggla and expect to get the whole story.
This seems like a solution looking for a problem, but the reminder to look beyond the obvious is always welcome.
You can argue anything, but if data says that some players don’t abide by your rule of separate distributions, what does that mean? What happens with the players whose true talent is constant between seasons then? What happens is exactly what Eno outlined here. Nobody says this works for every player, just some.
Speaking of arbitrary endpoints, that should be a colon in the middle. Subtle joke or just an ironic display of improper punctuation?
Wow I must confess you make some very trcenhant points.
95pwfm zykdmswwdkai
Seasons are not arbitrary endpoints. There is an offseason. Players spend time not playing. They sign new contracts. Then they go to spring training, and potentially alter their approach at the plate.
The notion that June – July is no more arbitrary than April – October doesn’t hold up to scrutiny. There is a large space in which changes can occur between the end of one season and the start of another. There isn’t a large space for this between the end of June and the start of July.
This isn’t to say that fluctuations in season-to-season performance aren’t random in a lot of cases. But there are many clear reasons why a player’s performance may change significantly from year to year – too many to claim a season is arbitrary.
Players never make adjustments in season? They never change teams midseason?
I think the point is we overemphasize seasons and ignore that there might be an alternative view to find a player’s true talent level. At least that’s what I took out of it.
of course players can and do make adjustments midseason. and if you were to analyse a player’s performance after he changed his stance at the plate, or his position in the box, or his approach to swinging at the first pitch, that would be perfectly valid. it wouldn’t be an arbitrary endpoint. just as the start/end of the season isn’t an arbitrary endpoint, because something has just happened which only happen once a year (the offseason).
but just choosing june is by definition an arbitrary endpoint, because for most players nothing notable has happened.
does this make sense to you? i’m not arguing that the end of the season is the only non-arbitrary endpoint. there are many other non-arbitrary endpoints for a given player. but the start/end of the season is actually the only endpoint which is fundamentally non-arbitrary for all players. it is the only point each year where a significant event occurs for all players simultaneously.
Absolutely. It seems to me like a cop-out to just change the start and end points in order to change the appearance of a player’s stats. Yes, it can be an interesting new way to look at things, but the truth is, with every new season brings new teammates, possibly a new ballpark, a new pitching environment, and most of all, 4 months of no baseball. Players can gain 20 pounds, lose 20 pounds, change their swing, get contact lenses, etc. Yes, they can do some of these things during the season, and in those cases (see, Jose Bautista), it is very helpful to start wherever they made a change. But in most other cases, players are going to change the most between seasons, and it is therefore most beneficial to use seasons as endpoints.
In conclusion, using other endpoints can be very helpful when looking at a change a player made or to gain a new perspective, but to say that seasons are arbitrary is to vastly underestimate the impact of the offseason, both for the player and the playing environment around him.
Dexter Fowler added a leg kick in early July and has been on fire ever since, Troy Tulowitzki went to an upright batting stance in early June 2009 since then his WRC+ by month have been remarkably consistent. Players do make in season adjustments, and when they do it does often looks like they are playing two different seasons.
It’s also good to note that opening the door to other 162-game windows such as July-June invites selective endpoints and the resulting bias. It would be all too easy to attribute non-seasonally-bounded hot streaks to some change the player made, or to pick a 162-game window that paints your favorite player as one of the best in the game over that time period.
Crap, you’re right. Players don’t go on the DL, get sent to the minors, get traded, get released, give up on an ineffective pitch, go on rehab assignments, or get moved to the bullpen in-season. They don’t play winter ball. And Curtis Granderson didn’t completely re-invent his swing last August. Dexter Fowler didn’t recently add a leg kick.
Don’t forget Bautista, he didn’t change his swing in late 2009, it was in the ’09-’10 off-season, when MLB rules allow you to change your swing.
Of course players change during the season. I don’t think anyone was denying that. But the point is, seasons are not arbitrary. They are commonly the starting and ending points because the offseason is where the most changes are made. Of course we shouldn’t ONLY use seasons, but unless there is a compelling reason otherwise, why shouldn’t we? It prevents selection bias and is a convenient and effective method of assessing a player. I don’t disagree with a lot of the author’s points, but the title is just not true.
yes, and you can choose any of those as valid endpoints. they aren’t arbitrary for those specific players, because something happened to enable us to draw a distinction between before and after. but for 99.9% of the league, the point when curtis granderson altered his swing IS arbitrary. whereas for 100% of the league, the end/start of the season is a significant break from the ordinary routine. it is an event which happens but once a year.
I don’t know how exactly to measure it, but I have a pretty hard time believing that the offseason is a completely insignificant factor.
To echo what others have said, the end of a season isn’t “arbitrary”, at least not until MLB starts playing year-round. Case in point: How often do we hear about a pitcher learning a new pitch in the offseason? How often do we hear them doing the same during a season?
And also, 85% of players come to camp “in the best shape of their life!!”…like, every year. So there’s that.
(And to think, they were probably fit and athletic at 20, so by the time they reach 35 they should be finely honed freaks of nature, almost godlike, or at least near cyborg – kinda like Ivan Drago, but less Cold War killing machine.)
Hey, it was true of Barry Bonds.
and the other 15%, by weight, are Adam Dunn.
Pitchers add new pitches all the time – it just depends on whether we notice ie. Brian Wilson’s filthy 2 seamer. That happened in a season. Pitchers probably have way more pitches than they show in a game – it just comes down to whether they want to use it or not.
There was also Lincecum adding a slider in late August 2010 that allowed him to take things up a notch in September and October.
Jair Jurrjens changed the grip on his 2-seamer two games into the current season.
It happens more often than you think.
Very interesting article Eno. I had suspected that there were a few guys like Huff in this regard. While the offseason certainly is not insignificant, I think Jack Weiland’s hit the mark in saying that “we overemphasize seasons and ignore that there might be an alternative view to find a player’s true talent level.”
Are Huff, Wells and Fielder hit or miss from year to year? It certainly appears that way, but this should not cause us to ignore the overarching consistency that these players demonstrate across the span of their career. To characterize them as flawed players due to some streaky-ness that happens to coincide with the endpoints we have established might not be fair, especially when we know that many stats don’t even stabilize over the course of a full season.
I’m not sure that the choice of the word arbitrary is best here, and in fact when I wrote that original comment I had started it with “Arbitrary start and end points” before changing my mind. It’s not that seasons are exactly arbitrary, because there is certainly logic to using these as the point of delineation (explained well by others above). Nevertheless, we must recognize that the end points which we commonly use, arbitrary or not, do in fact influence the conclusions we draw.
While it might not make sense to totally ignore “seasons”, to classify players based on their performance in these units doesn’t always give the full picture.
This I can agree with – well said.
That makes a lot more sense to me. Thank you for the linguistic clarification. Not sure if I agree 100% with the analysis but it is a very thought-provoking article.
I’d agree that I used some sensationalist language perhaps. But I do believe there’s value in reducing seasons to chunks – good, long, accepted chunks – but chunks nonetheless. There are other chunks as well.
Very, very interesting. Now, how do we deal with this in practice? Should a metric be created that compares a few moving windows (a sample of arbitrary endpoints)? It could provide a flag for when the endpoint is impacting our perception of talent level. Or maybe just some kind of volatility factor based on variance relative to avg variance? Off hand I can’t come up with anything that seems promising…
Actually – the moving average graphs people talk about in comments below sound pretty good. Although I think it might be worthwhile to distill those graphs into a number or two for short-hand purposes.
I have to agree with the sentiments expressed here. I don’t have a link, but it’s been shown that if in midseason you’re trying to project a player for the remainder of the year, you have to weight the current year’s performance substantially more heavily than usual end-of-season weightings like 5-3-2 would imply. In other words, talent is “sticky” within a season.
Just a few more things that happen in the offseason to add to the above mentioned things of learning new pitches and making swing adjustments:
– players wait to have surgery in the offseason
– players age 6 months between the last and first day of the season (which is supposedly worth -.25WAR in true talent on the downswing of career
– perhaps related, my guess is that players change muscle mass %, weight, etc in the offseason
– players change teams, teammates and ballparks more in the offseason; therefore they probably also change positions more often in the offseason
So one important thing is that you’re going to get a much greater true-talent change between October 1st and April 1st than June 30th and July 1st. To ignore that seems like a fatal mistake.
They may be arbitrary in some sense, but the beginning and end of the season are endpoints, and just because they’re arbitrary, doesn’t mean they aren’t important endpoints.
The only one of those that I’d buy is surgery. True talent aging curves rarely work that way, and “best shapers” don’t do better (tho pablo is making me re-think that). Ballpark changes are accounted for in “ballpark adjusted” stats.
Pablo went from morbidly obese to passable athlete, so I don’t think he is a good comp.
This is really interesting. I’m glad you did this. I think those who disregard this article are missing the point a little. The point, it seems to me, is that our perceptions of players is often determined by either these very short 5 or 10 game samples mentioned by baseball commentators or by how they perform over seasons. If we change our mindset a little and break their careers up into different segments, our perceptions of players change some. It’s not that the offseason is irrelevant; it’s that we are likely to perceive players differently if we divide players’ careers differently from the way we normally divide their career numbers.
I’d also like to say how great it is that you were able to take a contributor’s comment and turn it into a post. It’s great to know that our comments are read and considered.
Even mine??
ESPECIALLY yours.
It seems to me that if you wanted something “telling” in terms of career path/trajectory, you should just take a moving average and make a graph. For the sample size of the moving average you could use these recommendations: http://www.fangraphs.com/library/index.php/principles/sample-size/ and then use game logs to develop the graph. For instance, for ISO, you just chart out a line of Prince Fielders last 550 ABs over his career – I’d be interested to know what that looked like, and REALLY interested to know what it would look like for Bautista.
Yes, I wanted to do this. Hah. Perhaps I can get a tutorial on moving averages and whip out some graphs in the future.
Just did a chart for Albert Pujols over his career – max ISO over a 550 AB period is 0.393 – mid-July 2009. Chart took a long time to make though
Seasons are anything but arbitrary. The game is played in seasons. The point of a season is to find the best team each season. It’s the whole point of baseball….
Yes I do believe they are accepted beginning and end points. However, when evaluating a player — as opposed to a team — there are different chunks you can use.
You don’t think a multi-month layoff has ANY effect?
“Yes I do believe they are accepted beginning and end points.”
Haha, this is a bit like saying “yes I do believe it is accepted that the sun comes up in the morning”.
I understand your argument about evaluating players rather than teams, but I still think it misses the mark a little bit. First, a player’s objective is to help his team win games so that his team can be the best team (the whole point of baseball). The time frame to evaluate how well a player performed his objective is the 162 game season.
Second, if you are just purely evaluating a player irrespective of his goal of winning the WS, then the order of the games doesn’t matter at all. There is no reason to consider the games consecutively. So sliding the window makes little sense. If this is what you want to do, you should just perform a bootstrap analysis of the players career (or peak, etc.) where you resample games or plate appearances from the full sample of all of their observations.
“The game is played in seasons.” Nay, the season is played in games.
whoa…
Whoa… that just totally blew my mind!
How about a completely arbitrary look at some of the best 162 game stretches (using any start and end point) ever compiled? Could be interesting.
yes, i’d love to see that. What’s the best OBP over any 162 games? Best wOBA, most HRs, highest WAR… I’m too lazy to do the work, but someone who writes for this site should totally do that.
I agree with non-arbitrary crowd to a degree, but I still love this.
All of this is in pursuit of knowing a player’s true talent – start the discussion there.
I think we can all agree that players have a lot of influence on their true talent levels at any given time, some big, some small. Injuries, conditioning, swing mechanics/tweaks, outside influences like personal lives/teams, etc. And seasons are definitely markers of when things like these reset themselves – especially with respect to injuries and conditioning (both positive and negative), which are probably the two biggest factors in this equation. In that regard, seasons make a lot of sense to look at. But when we do have flukey yoyos like Huff, this is exactly when we should take a different perspective on his year to year stats. Did he have any major injuries? Did he do anything materially different at the plate? Does anything in his peripherals give us a reason to think there is something happening year to year? If so, then we need to give some credence to the year-to-year splits, and acknowledge some outside variables on his true talent. But if there aren’t clear indicators of this, which in Huff’s case I tend to think there aren’t, then I think removing the year bias can show that Huff has had a pretty normal career arc.
I think the point Eno is trying to make is that there are so many, endless almost, variables when examining a player’s performance, it is easy to get drawn into narratives about a player when one seems to stand out. Like Huff and his yoyo performance, it made for a perfect story about a guy who is either on or off. But with no real REASON to be on or off, no major injuries, no Jose Bautista swing change, we have to acknowledge that it could be a matter of year to year end points creating a narrative here that doesn’t really match reality.
Huff is a true talent _____ wOBA hitter. Fill in the blank. He followed a normal career arc, with normal influences on his true talent along the way. And those fluctuations are imagined in Eno’s breakdown with his midseason splits much more accurately than they are in his normal year-to-year lines.
Seasons are not arbitrary, but they are also not indicative of as much as we might think, or they might seem in some cases.
I think what you are interpreting as good-year/bad-year fluctuations is actually just statistical noise, or something like it. I have only tried this for Huff’s numbers, but if you adjust the number of hits he has had each season for the number of plate appearances (since 2003, the first season he had more than 600 PAs), the distribution of those scaled numbers of hits per season is statistically consistent with a Poisson distribution.
This is why a good year always follows a bad year and vice versa: it’s just another instance of statistical regression. It’s also why if you resample with a different selection, as was done here, the apparent fluctuations go away. I bet if you resampled seemingly very consistent players with just the right window, you could make them look streaky too. Essentially, the number of hits a player gets in a single season is just not large enough for fluctuations of the size of Aubrey’s to be considered anything but statistical. A statistically significant fluctuation, in this case, would be roughly under 120 hits or over 200 hits in a season, assuming he sees his average of 632 plate appearances. So in some sense, one could say that seasons are not long enough!
I can show my calculation to those who are interested – it’s not completely rigorous and like I said I only did it for one player. But I suspect that’s probably what is going on here, at least in Huff’s case. Poor guy, his regressions just have a bad timescale. It would be interesting to try to predict how often you’d expect this to happen, given the length of the season and the average numbers of hits players get.
Funny enough, this actually provides justification for teams that are trying to decide whether to draft a seemingly hot-cold player: if they have enough data to have a firly good estimate of the player’s true average, and they have no other reason (injury, age) to believe his average is going to change by a lot, then they’re perfectly safe to forecast a bounce, in either direction, and act on that forecast.
This is laughably retarded. Here is a definition to help: “Based on random choice or personal whim, rather than any reason or system.”
slowclap
Another tremendous contribution brought to us by Garrett.
How is the end of a season based on random choice?
Or personal whim?
Thanks dipshit.
dude who pissed in your cereal?
+1 for the career moving average graph! July-June is just as prone to sample bias as any other 162-game sample. Nate had some ideas above for how to structure the graphs.
I’d also add a possibility of seeing what career average numbers look like over the course of a career. They would of course tend to fluctuate less and less as the player accumulates PA, but the upward or downward trend at any given time should yield some interesting insights.
I think the point is that the statistics we look at are fundamentally streaky. And 162 games is not so many. So we confuse streaky statistics with streaky players. If a player’s numbers are far more consistent from July to June than from April to Octobers, than there is a good choice his streakiness is just a case of observer bias. Of course this is not conclusive, but certainly one could take this as a starting point to make a true streakiness statistics. Also people who believe Prince Fielder will hit poorly next year for numerological reasons are probably idiots.
Yes…seasons are completely arbitrary endpoints. That is why I believe that anybody that mentions a single season or says that a player is a “1st half” or “2nd half” player is an idiot. That is why we have 162 game averages (which this site does not display but which are given on each player page on Baseball-Reference.com).
The general assumption of the article might be true. However, there is a good reason why a season is used as an endpoint for statistical evaluation of a players talents. Every new season the only stat that ultimately matters, team wins, is reset to zero. Those are the rules. And since teams play by those rules they have to build their teams in a way that suits their goal of winning games. In that regard it can indeed be said that there is some value in a kind of production that is somewhat consistent within single seasons, and not just within calendar years with random starting and end points.
Really interesting article, particularly if it were related to the ‘hot zone’ data on Baseball Analytics. It does beg the question, if your going to translate the co-ordinate system to creat symmetry to average over, why don’t you explore shorter time domains as well.
While I agree that seasons are not “arbitrary”, the real question is whether it’s worth exploring different endpoints.
I think what Eno is getting at is that sometimes using the seasons as endpoints can obscure the underlying truth of what a player is doing. Sticking blindly to using seasons as endpoints can create blind spots when analyzing certain players.
If switching the endpoints yields new insights worth examining further, well then, let’s keep our minds open to switching the endpoints on occasion.
That said, my guess is whatever endpoints you choose, Albert Pujols is going to look like a pretty decent player. 🙂
Selective =\= arbitrary. Using seasons as endpoints is selective, and can be somewhat misleading at times, but they are not arbitrary.
Funny, I was just posting about this on Beerleaguer. I would distinguish counting stats from luck-measuring stats. For certain stats, season distributions make sense. After all, the players themselves are motivated by contracts that have seasonal end points; there’s wear-and-tear with expected effects on stats; whatever pressure/leverage effects exist do so based on time.
But luck (or its absence) is really different. If we’re trying to see the influence of something that isn’t in the control of the player, or any player, then seasonal distributions seems really, really weird. Unless, of course, we’re actually measuring the effect of something else on the percentage of balls touched that turn into hits. Like: steroid effects on pitcher’s arms, or tighter wound baseballs, or changing rules about scoring errors, or global warming.
That is, why would we ever think about BAPiP for one season against another. If what we’re trying to control for is luck, is there any good reason to think that luck is seasonally distributed? So, some have said that Cole Hamels had a “bad BABIP in 2009” and the Vanimal has had a “good BABIP in 2011”. This, we’re told, means that a seasonal regression is due. But luck obviously isn’t delineated in April to October chunks, unless there is something about the player that is different season to season. Or, to put it another way, saying that Worley is having a “lucky season” is basically buying into the exact wrong-headedness about sample size that advanced stats are supposed to refute.
“That is, why would we ever think about BAPiP for one season against another”
Because a very large part of BABIP isn’t random at all.
This article does not add any value to the every-other-year fluctuations for the given players… Thinking about it mathematically, in order for this to work, you cannot have every other half year be up and every other half year be down because at that point your “typical year” stats wouldn’t fluctuate:
Apr ’04 – June ’04 Bad
June ’04 – Oct ’04 Good
Apr ’05 – June ’05 Bad
June ’05 – Oct ’05 Good
Apr ’06 – June ’06 Bad
June ’06 – Oct ’06 Good
Apr ’07 – June ’07 Bad
June ’07 – Oct ’07 Good
If you use standard seasons, you average a bad and a good each year to get neutral (steady neutral).
If you use the “arbitrary end points” theory, you average a bad and a good and yet again have steady neutral.
In order for the above article’s hypothesis to work, the half years MUST look like this:
Apr ’04 – June ’04 Bad
June ’04 – Oct ’04 Bad
Apr ’05 – June ’05 Good
June ’05 – Oct ’05 Good
Apr ’06 – June ’06 Bad
June ’06 – Oct ’06 Bad
Apr ’07 – June ’07 Good
June ’07 – Oct ’07 Good
The above is the ONLY way the hypothesis works, which is simply common sense…
Normal End Points:
Bad and Bad in ’04 = Bad
Good and Good in ’05 = Good
Bad and Bad in ’06 = Bad
Good and Good in ’07 = Good
“Arbitrary” End Points:
Bad and Good in ’04-’05 = Neutral
Good and Bad in ’05-’06 = Neutral
Bad and Good in ’06-’07 = Neutral
This is mathematical common sense…
At best this proves the above players are inherently streaky, nothing more…
You missed the point a little bit I think.
I said seasons were essentially random end points, if ones we that have agreed are useful in certain respects. So I changed the end points to ‘season long’ but with different dates. Nothing more. These are just a few players where it works the other way. The players were just an exercise in ‘season-long’ endpoints that weren’t April to September. There are thousands of players where it doesn’t work this way.
The author seems to make the argument that measuring July-June “seasons” could have better predictive value, at least for some players, than traditional seasons. However, this result is exactly what we would expect if the differences seen in performance from season to season were real. Basically, you’re taking half each of two samples instead of one sample – of course this provides a lower standard deviation. You’d get the same result if you took half from each of two consecutive spins of the roulette wheel instead of independently viewing each distinct result.
I disagree. I don’t think that I’m arguing that any of the work I did was better than using season, just that it was just about as random. If there’s any way forward with this, I think that some sort 162-game moving average might be interesting.
Or, especially with ISO, which takes the longest to become reliable, what about two-season samples? 200 games?
My point is that we need not be encumbered with ‘seasons’ if random ‘season-long periods’ show us something completely different in certain cases.
I’m saying that if seasons truly are at least somewhat independent samples – and I believe most commenters here agree that this view has validity – then a 162-game sample that crosses seasons is effectively *less* random. It’s less random because you’re averaging across two samples. The point that seaons are arbitrary endpoints is well-taken, but based on the above, I’m more comfortable applying that logic within a season than expanding it beyond a season.
Did you just insert ‘samples’ where ‘years’ used to be?
So we should roll our eyes when an announcer picks his random endpoints, but we should value a Fangraphs article author’s random endpoints…why again? Because he writes on Fangraphs?
It’s more and more evident every day. The stathead movement has run out of things to talk about.
I *think* I just said that years/seasons were as arbitrary as any random points.
That’s not saying nothing.
But that is completely absurd. You either don’t know what arbitrary means or you’re trolling or really stupid.
Stretch your mind just a little bit. Seasons are obviously not arbitrary in the true sense of the word, but from the statistical sense of the word there’s an argument to be made. It could be wrong, but it’s not nothing.
What is the “statistical sense”? I see “random” as the definition. In which case I don’t see how “stretching my mind” would make a clearly defined time frame “random”.
I’m asking you to stretch past the title and think about the statistical question at hand. When evaluating the true talent of a player statistically, we’ve always assumed that seasons were great natural end points to use. But, in some cases, other (yes, random) 162-game samples tell us a very different story. So maybe we should think more about samples.
In other words, is it more likely that Prince Fielder and Aubrey Huff have some sort of power on/off switch that flips from year to year, or that they’ve had very normal careers with traditional statistical arcs?
I would say that in these two cases, using seasonal end points have failed us a little.
And there’s been research that has suggested that certain stats like ISO and BABIP don’t ever become reliable over a season… Perhaps there’s a strong sample size somewhere between a season and a career that we haven’t found yet.
Are you familiar with the term “data snooping”?
That is what you’re doing. You’re writing a ex post narrative fallacy. Many writers did this about the magical ability of the Angels to out perform third order wins. And recently Nate Silver did this talking about Michigan St (when he wasn’t shilling a political agenda).
Please stop. Fangraphs has enough issues with nonsense and trivial articles to have deceitful statistical “analysis” added to the mix.
PS: Learn what random means. You did use a “random” endpoint. A random endpoint would be to choose a player then have an RNG choose a number out of 162 and you would then use that to analyze the data as the end point. You used a carefully selected and arbitrary (what it actually means) to lend credence to your post.
Personally, I think players should be evaluated as to past performance on how they do over the season as designated by MLB, since a chance at the playoffs over 162 is the meaningful goal of baseball, not just to put up good numbers.
But certainly looking at this kind of data is interesting to predict how a player might perform going forward. Just as it was pointed out, for example, that Bautista’s greatness actually started mid-season before last year. Obviously the longer a player sustains his greatness / shittiness, the more likely it is that it’s his real current ability level. So, good point.
I didn’t read all the posts above so sorry is this has been pointed out already, but the theory that seasons are arbitrary endpoints isn’t that tough to actually test. Just pick some truly arbitrary season-long endpoints, take a large, random sample of players, and compare the variance in a few core statistics between the quasi-seasons and the actual season. If there’s more between-season variance in quasi-seasons, the theory is false. If there’s the same amount of between-season variance in quasi and actual seasons, theory confirmed.
is this better than rolling 162-game averages? I’m all about testing this.
Probably not, but if I’m understanding you correctly, there would then be 162 quasi-seasons per player per year. That’s an astronomical amount of data, but if you can work with it, then it would yield stronger results than the method I suggested.
[1] All endpoints, in some degree, are “arbitrary”.
[2] Actually “arbitrary” isn’t the right word, since …
[3] Seasons as endpoints are selected for obvious, logical, reasons.
Seasons aren’t arbitrary after all.
There may be a better way of looking at the data for some players, for some reasons, but unless we detect a larger scale reason for doing so the natural, defined start and end of the yearly season serves as a fine endpoint for me.
Including a 5-month “offseason” right in the middle of an endpoint seems illogical to me.
The conclusion of the article is that when you factor in a 5-month rest period, certain players’ data is more consistent. I don;t know why we should be interested in that since no athlete will ever get such a period in their season, nor should their true talent factor in a rest period for the same reason.
its a curious argument but it’s just not correct
I have to disagree with this entirely, for similar reasons to some of the other comments above. Seasons are only arbitrary if you are not interested in the actual purpose of playing the sport, which is to win the world series. When the entire sport is geared towards a specific goal of making the post-season and winning, then using the start and end of the season as a statistical assessment for players make absolute sense, and is not arbitrary (or random) by any definition that I have ever heard for those terms.
From a manager’s perspective, you choose players *in the off season* with the intent of finding out who will give you the best chance to get to the playoffs and win the world series. To do so, knowing the stats that a given player has during that period of baseball (a season) is absolutely important and useful to you. It is not the only type of stats that you care about, but it is anything but arbitrary.
Other time frames may become important in other scenarios. For instance, knowing a batter’s performance between july-sept vs april – june could be very important to a manager nearing the trade deadline, looking for a player that will give an added boost to their team at the end of the season. These stats, which could much more easily be called arbitrary than full season stats, have their value and place as well. They do not make season stats random or arbitrary though.
I’m not sure if this point has been made but I’ve been thinking about this article for a month or two now and a basic mathematical flaw came up in my mind with the concept you’ve used in this article:
If a player performs well in every second cycle and poorly in every alternate cycle(let the cycle in this case be one season), then their production will look similar to an ordinary sin curve or cos curve or something liek this y={1 for x=1,3,5,7,9….etc 0 for x=2,4,6,8…etc}
If you change the beginning/end of the cycle to the very middle of the current cycles and take an average between the new cutoffs, of course the average will be even, because half of 1 and half of 0=.5 every time.
Wether a season is an arbitrary endpoint or not that answer would be true with this test in any circumstance.
Thank you for an additional wonderful post. Where else could anyone get that type of details in this kind of a perfect way of writing? I’ve a presentation subsequent week, and I am on the look for this kind of facts.