THE BOOK--Playing The Percentages In Baseball

$8 million?  League average starters are worth about that much, right?  We know that of course he’s not going to have an ERA under 1, or 2, or 3 even.  But I’d think he’s more likely to have an ERA below 4 than above it.  Given that we’ve seen the upside, and it’s considerable, $12M (my choice) might be too low.

Well, the yapping has stopped, and they want to pay him 8MM so far per year.  That’s less than Jake Westbrook signed in the off-season.

I wanted to put $10MM, but I figure the chances of him getting hurt are so high (by simply being a pitcher) that even if he continues to pitch well his overall value won’t be more than $32million over 4 years.

Of course, if given the chance, I’d pay him the veteran minimum. But that’s not happening.

And I can only hope that he gets shelled for the rest of the year for some vindication!

After 55 votes, people would sign him for a 4/32 deal.  That values him as a 2.4 WAR pitcher.

Presuming 180 IP in the first year, that makes him a .500 pitcher.  Jake Westbrook and Carlos Silva signed for more in the offseason, as did Ted Lilly signed for more the prior offseason.

So, for all the yapping there was about how great Cliff Lee was pitching, nobody believes it did anything other than move him from a slightly below average pitcher to a slightly above average pitcher (which was Rob Neyer’s contention all along).

How utterly pointless that BTF thread was two weeks ago, if the consensus is that after an historic start, we really believe him to be a .500 pitcher.

I would give him 12 a year (as Blue Jays GM) to replace Burnett after he opts out. I would like to see the team have a decent left handed starter or two if Purcey can join the rotation next year.

Brent, just curious, why “left-handed” starter?  Why not an equivalent (or better) right-handed starter?

MGL--Because he’s looking to pick a fight with you

Regarding Tango/5 (and I know he’s speaking about the voters in general), I believe his talent dictates $10m or $12m a year, but with the potential for injuries, I would only feel comfortable giving him $8m.

Well, the only reason I can think of to “prefer” a lefty starter over an equivalent righty (overall) is if your home park is a little more favorable to LHB or the teams you play the most are laden with lefty batters.

#1, John, we only care about his ERA from this point on.  Given his low ERA already, he is likely to have a very good ERA for the whole season.  I’ll definitely take the over on his ERA being above 4 for the remainder of the season.  Of course, with the run scoring in the AL so low this year so far, an ERA of 4.00 might be close to league average if the current rate of runs scored continues (which I doubt).

I not sure I agree with the reasoning that he would be worth less (per year) over a four year deal due to the risk of injury.

There is also IMO an offsetting factor in that the price of what he can offer (a slightly above average pitcher) will likely go up during the contracts duration.

So if he is worth 10M per season now that 10M might seem like an absolute steal in his forth year when other pitchers of similiar “production” are signing for 15M per year.

Now the figures are pure guesswork but I would expect salaries to continue to climb and pretty rapidly at that, especially for pitching.

I tried to vote “Too Small of a Sample to Conclusively Determine” but it wouldn’t let me!

Bjorn: the point remains that what fans are currently thinking he is worth (somewhere around a 34/4 deal as of right now) is still LESS than what Westbrook, Silva, and Lilly signed for.  They are valuing him as a .500 pitcher, in-line with this chart:
http://tangotiger.net/salary2008.html

This includes injury risk, attrition, and inflation.

After 72 votes, it’s a 4/33 deal for Cliff Lee.  According to the link in post 13, that puts him at just under 2.5 WAR.

Remember, a .500 pitcher in the AL is +.130 wins per 9 IP.  A 2.5 WAR pitcher would be some combination of:
win% IP
0.470 225
0.480 205
0.490 188
0.500 173
0.510 161
0.520 150

That is, getting 225 innings from a .470 pitcher is the same as getting 150 innings from a .520.  And for all of that, teams are paying 3/29, 4/35.

Tango #5

I’m almost afraid to get this topic going again, because the previous exchanges were not particularly enjoyable, but here goes:

On the topic of hot streaks, I think weather is a good analogy.  Here are the high temperatures for the past 9 days in lovely West Linn, OR:

57, 59, 62, 73, 88, 98, 91, 84, 82, 59

The average high temperature in May is 70 degrees, but on Thursday night (the day with the 88 degree high temp.) we all knew that Friday was going to be a scorcher.  Why?  Because the causes of variation from the average high temperatures are fairly transparent here in Oregon: strong easterly wind flow from the desert means hot, dry weather.  So, when predicting Friday’s high temp., no one fell back on the average, they went with the prediction based on their understanding of Oregon weather, and the well-understood models we have for short-term forecasts.  There is a cause-effect relationship driving high temperatures, and we understand it fairly well.  Today, it is raining, thus the 59 degree high temp. projected…

In baseball, there are also cause and effect relationships driving results, but IMO we don’t understand them nearly as well, or we can’t properly isolate them, which means that often times (maybe most times) we are better off falling back on the long-term averages.  It of course doesn’t mean the causes of the results don’t exist, they are just opaque to us.  But sometimes the results are so extreme that we suspect some of those unknown factors must be driving the results, rather than simple chance.  The point at which we make that switch from depending on the overall projection to projecting the hot/cold streak (maybe with some observational evidence that has helped convince us) is personal, of course (and maybe some folks never think that is appropriate). 

So, we all know that random chance CAN generate streaky results, but for some of us, Lee’s streak convinced us that some real factors must have been behind the results.  However, important point here, that doesn’t mean that we have to believe those factors are permanent.  I don’t think they are any more permanent than the weather pattern that produced our hot weekend.  But that weather pattern was REAL, and I think the factors that have led to Lee’s outstanding 6 weeks are/were real as well (though almost certainly not permanent).  So, I don’t think that you can say someone is illogical, or a crap-yapper, just because they believe that a) Lee’s start was based on something real, and b) the real thing(s) it was based on are probably not permanent.

So, here’s where I come out on all this:

Baseball results are like weather: long-term, you are advised to adhere to the long-range projections, if you wish to be as accurate as possible.  However, in extreme cases, it may be justifiable to temporarily deviate from the long-term projection to predict short-term results.

The analogy holds, I think, for a lot of what we’ve been discussing.

Marcel = a climate predictor.  Marcel can predict season results well, without knowing any personal details about the players, but it can’t tell you what will happen in tonight’s game.  Climate can be researched using only data, with no understanding of local/regional geography.  It can tell you degree-days to size your heating/cooling plant, but can’t tell you what to wear tomorrow…

To predict weather, or tonight’s results, is much more difficult, and cannot be done relying only on aggregated data.  You have to know more, and even then the predictions are only so good.  Perfect prediction ability is probably permanently out of reach, but improvements are being made all the time (with diminishing returns on the effort)… Nevertheless, I think it’s still worth pursuing, because the knowledge is very valuable…

Greg, the proof to your statement is this:
1. take 100 players in the Retrosheet years
2. 50 of them happen to have the hottest one month streak known to man
3. the other 50 happen to have the same career totals (to date) as the hot 50, but are simply performing their usual self.

Now, what is our expectation for the next week?  Is it conceivable that the weather pattern for ALL 50 hot streakers will suddenly disappear on the 31st day?  No, of course not.  If this is a real weather pattern, which will disissipate at some point, it will die for some on the 31st, some on the 32nd, some on the 37th day, etc.  That is, it is not permanent, but neither is it so specific as for it to die for every player at the exact same time.

As for the other 50 guys: do we expect all of them to start a hot streak?  Some?  A couple?

At the very least, if the weather pattern model is to hold, the AVERAGE of the 50 hot streakers will perform better than the AVERAGE of the 50 non-streakers.  That is, you’ve selected 50 dates from summer Arizona and you’ve selected 50 random dates from Kansas, Texas, NY, and Maine.

While clearly we don’t know what the weather is like the next day in all those states, by seeding them with a prior knowledge of temperature, your expectations must be that it will be hotter, on average, in those 50 Arizona dates, even if on several of them, Texas will be hotter.

But, this doesn’t happen with humans.  Baseball players anyway.

It makes for a nice story, but to the extent that it is true, it’s barely noticeable.

***

At no point, ever, will you be able to ignore everything you know about Cliff Lee.  You may weight what you know now more, and at an increasingly rapid clip.  But, your knowledge of him, and of the population, plays a huge role here.

As noted in some other thread, one can consider Cliff Lee to be a .525 pitcher at this time.  You can’t really think more than that.  And, most people are even more pessimistic about his expectations than I am.

And I agree that your weather analogy is a great analogy.  I love it actually.  It is perfect, because it shows exactly the best argument people have: that something is real not luck, but that it is transient, not persistent.  Fantastic.

But, as I said, ballplayers simply don’t behave like that.  Their performance follows what you would expect from a binomial distribution in exactly the same shape as a distribution of weighted coins whose mean follows a bell curve of mean=.340, 1 SD = .025.

Tango, what you wrote makes sense, but that wasn’t what I was saying.

I’m not arguing that the weather pattern will hold (i.e. I’m not suggesting that a warm period of weather means “global warming") What I’m suggesting is that the weather pattern was real, and not a statistical artifact, in Lee’s case - or more generally, I am suggesting that *some* stretches of outstanding play are real, since of course random chance will create some hot and cold streaks.

So, you could say I’m defending the people who wrote “Cliff Lee is pitching really well so far this year”, but I am not defending those who wrote “I’d rather have Cliff Lee on my staff than Johan Santana” or nonsense to that effect.  Actually, I don’t think anyone said that, but you get the idea.

I’m opposing the viewpoint that hot/cold streaks don’t exist.  I’m not advocating that hot/cold streaks necessarily portend a change in a player’s baseline of performance…

"I’m not arguing that the weather pattern will hold”

But, you ARE arguing that *some* weather patterns will hold, at least for the next 60 seconds.  And some weather patterns will hold, at least for the next 60 minutes.  So, there is *some* residual persistent effect.

That is weather after all.  Sometimes, the weather will simply break one minute to the next.  And sometimes, the weather will simply break one day to the next.  Or one week to the next.

That’s the entire point of your argument, that the weather pattern is real, but it is not necessarily persistent long-term.  But, neither is it so transient as for it to be a blip every single time.

And that’s the test to see if humans are like the weather. 

I’ll get a sudden fever, and two days later, it will be gone.  So, it was real, and very transient.  However, that I am in the midst of a fever makes it more likely that 60 seconds later I’ll still have the fever than some healthy person in my shape and circumstances will have a fever.

This is the argument.

To the extent that a fever is real, to the extent that the weather pattern is real, we have *some* expectation, *some* times, to seeing it one day from now.

But, that’s not what happens with baseball players.  Like I said, it makes sense that it should.  But, try to prove it.

And in fact, you don’t even have to, since MGL did that very study in The Book.

Does the comparison to the average starters really work?  The question was what *we* would pay for Lee, rather than what he would get.  I would imagine that a) Lee would get more than 4/33 if he hit the market tomorrow, and b) most people who voted wouldn’t have given Carlos Silva 4/48 last winter.

I admit to being swayed by Greg’s arguments...le me offer a different analogy, which offers a new testing method.

When I was young I used to bowl, and I was pretty good, “retiring” with a 172 avg. Being a stat-geek even back then, I kept track of my individual frames for awhile, and modeled a simple dice game (no computers yet) based on how often each of frames achieved each of the results in real life.

Bowling is very simple to model in that there’s only one person, the ball, the lanes, and the pins.

I discovered that as a virtual bowler my avg was about 10 pts higher than I did in real life. I concluded that the frame by frame results were not truly random, that I had a disporportionate amount of “bad” frames following other “bad” frames - mentally I could become untracked and my performance suffered. When I threw my 290, opening with 10 strikes, I felt “zoned in” - I was able to replicate that “perfect motion” for a short period of time.

I never had a good curveball, mostly relying on cutters and changeups (vary speeds, lots of motion). One day at schol Gene Pentz showed me how he held his curve. I threw a fantastic curve for the next couple weeks, then I forgot.

So there may be this mental part where the brain memorizes how to perform a mechanical process. Sometimes it works better, sometimes worse, in the long run it evens out.

Of course, can that dragon be detected?

If some team gave Silva 4/48, there were certainly a few of them that would have given him 4/34.

***

What “we” would have given is certainly framed by the environment in which we are discussing.  Really, the question is “what is the most you can accept your team to pay for Cliff Lee, while thinking you got fair value for him”.  That would be implied, no?  In any case, it must be understood that it’s not a nuanced or trick question I’m asking.

My salary calculator does a good job at representing what Westbrook/Lilly/Silva should have received, and that’s really the frame of reference.

The 4/34 deal that the fans are willing to pay for Cliff Lee puts him fairly similar in quality to those guys, give or take.

"Sometimes it works better, sometimes worse, in the long run it evens out.”

Yes, but then it should have a distribution wider than expected from random.

Going back to your bowling frames: if you could truly be “in the zone”, then there would be some persistence effect from one frame to another, even if you don’t expect your daily score to have a persistence effect day-to-day (or week-to-week).

Now, when it comes to bowling, I can believe it.  Clearly, persistence exists *in everything we do*.  The question is always: to what degree.  A bowler throwing 10-20 balls over a 20 minute period could show a frame-to-frame correlation that is not close to zero.  But, how close to not zero is it?  How real is it?

The closer two events are, the more likely you can find the evidence for the weather pattern.

Try to look for it in free throws too.  If a guy hits one, is he more likely to hit the second one?  Or, is his success rate in the second free throw almost exactly his “Marcel” rate?

Again, all this exists.  But, it exists barely more than what would be expected from a random distribution.  We are human after all.

Tango, let me go reread that section of the book and get back to you.  I don’t recall the details of the method MGL used to prove this, with what sampling of data, and to what level of significance.

Sounds like MGL’s analysis has convinced you to reject the “weather” model and accept the null hypothesis, i.e. the “dice” model.  Or perhaps you accept the weather model, but data has convinced you that baseball “weather” is utterly unpredictable, therefore forcing you back to the climate projections…

I’m interested to reread MGL’s work…

"Or perhaps you accept the weather model, but data has convinced you that baseball “weather” is utterly ...”

polluted with so much random variation that any signal is almost completely lost.

Just as the persistence of a particular weather pattern might last a few hours, the persistence of a ballplayer’s changing talent level probably lasts a few minutes.  From hour to hour, you can change the mean precipitation forecast from 20% to 40% to 30% to 60%.  For a ballplayer?  The most you can hope to establish a player’s true talent level is 100% of his carer norm one hour, to 105% to another hour, to 95% another hour.  That’s the extent of it.  Yes, there’s a change.  But, it’s just so minute a difference (and that’s an extreme case), compared to the weather analogy.

Otherwise, what is your expectation?  That someone can have a true talent level of 100% one month, 150% another month, and 50% another month?  I’m willing to go to a 5% change on an hour-by-hour basis, and certainly not a 50% change on a month-by-month basis.

Again, quantify your expectations here.

By the way, thank you for this discussion and the weather analogy.  I think it’s a great analogy.

***

MGL, p.56 The Book:

As you can see from these charts, hot and cold players hit around five points in wOBA better or worse than their norm, respectively, immediately following their five-game streaks. While that is a statistically significant difference (in 130,000 PA or so), we don’t know about you, but five points of wOBA doesn’t exactly send chills down our collective spines. On second thought, we’ll leave it up to you to decide how important or worthwhile five points of wOBA is.

”...polluted with so much random variation that any signal is almost completely lost.”

I agree, this is certainly the situation now, and may prove to be the case long-term.

However, I am hopeful that some closer observation of the game (via PitchFx, HitFx, better weather data and other innovations like these) may be able to transform some current noise parameters into quantifiable factors that we can then incorporate into the models, which might then make short-term predictions more worthwhile.

Either way, I agree that lots of silly commentary gets said or written that goes way beyond what the data support.  That will never change, I am willing to predict

5 points of wOBA - might be enough for a casino that plans to be around a while, but yeah, that much impact wouldn’t blow my hair back.

Greg/28: I think we’ve reached a happy agreement as I agree with the entire post.

I think so, too. 

Unfortunate that the data is so murky, but like I said, maybe some new toys will help us get a few layers deeper…

Although there were many problems and unanswered questions in my recent studies on hot and cold Aprils for batters and pitchers, here is a summary of what I believe with some degree of confidence:

Hot and cold streaks for 5 or 10 days have little or no predictive value for batters or pitchers, as we would have reasonably expected, simply given the sample sizes of those streaks.  Even if you weighted recent performance quite heavily, 20 or 30 PA is not going to change a projection very much.  Obviously the way that commentators and baseball insiders talk about hot and cold players (mostly batters, because they play every day) is incredibly silly.  But we knew that without doing any studies.

Recent performance by pitchers appears to have considerably more predictive value than by batters.  That was to be expected for various reasons.  Heck, we already have a more aggressive weighting system for pitchers in a basic Marcel.

Looking at a month or so of really great performance by a pitcher, especially if the great performance are K and BB rates, appears to have pretty good predictive value (obviously there is still plenty of regression toward both the career level and the mean).  IOW, we can weight really great recent pitching performance by quite a bit in constructing a projection.  How much to weight is not so clear.  But, as we have seen by Lee, a projection can change by .5 run in ERA, after a month of fantastic performance, even for someone with lots of history.

The same thing does not appear to be the case for really bad performance by batters and pitchers.  Batters and pitchers appear to revert to their traditional Marcel predicted performance, including the bad streak of course.  The reason is probably that players learn new things that make them a lot better from time to time, but the only things that makes players worse are age and injury.  The change in true talent, bad and good, is probably not balanced.  Even an old player who has a terrible spate of performance pretty much reverts to his basic Marcel.

BTW, I have been watching the Yankee game for like 5 minutes while writing this post. During that 5 minute post I heard a clip from Girardi wherein he said, “When you get your regular players back in the lineup, it makes all the other players better.” Then, the commentators (Kay and someone else, like maybe Paul O’neill) said, “Just like hitting is contagious, so is bad play.  When one player makes an error, the other players tend to do the same thing.”

I am paraphrasing both Girardi and the commentators.  Now, both of these things can be tested, but would we even want to waste our time?  I wouldn’t. 

Pathetic, silly, childish things coming out of managers, ex-players, and highly-paid professional analysts ALL the friggin’ time.  Do adults really think that “hitting is contagious?” Do they actually believe all the nonsense that continually comes out of their mouths?  Would they be willing to put their money where their mouths are on any of these ridiculous things they say?  If they were, and they were right, they could make a fortune from me.  Seriously.  Maybe, like (the Amazing) Randi, I will put up a bounty on dozens of things that commentators and insiders say on a regular basis.  And if any of them can be proven with statistical significance, I will dole out the prize.  What do you think?

One more thing.  Funny thing about “chemistry” is that the storylines are often so predictable.  If the Mets end up not in the post-season, of course it will be because of bad chemistry.  If they do end up in the playoffs, it will be because they overcame all the personal adversity and came together as a team.

If the Yankees have a good rest of season and make the post-season, it will be because of their experience and professionalism and because you should never count the Yankees out.  If they don’t, it will be because age finally caught up with them, and they don’t have the desire to win a championship any more.

We can write the scripts before the season starts and just have the talking heads fill in the names.

Oops, wrong thread about the chemistry. I’ll repost in the other thread.

How much do good hitters make their teammates “better” by getting on base more and therefore giving their teammates more PA with men on base?

Tango, I wouldn’t call the poll results “what the fans think”. Readers of this blog are not a random sample of baseball fans and are certainly not representative of the group as a whole, as you know, I’m sure. I would hope that we’re a bit more informed on average, if from nothing else from simply reading The Book Blog regularly.

Right, I’m not suggesting that my polls are from random fans, but more from “studious” fans.

Most people here value Lee at 8-10 million per year, but if we were running baseball teams, Lee’s market value would be 16 million.  High bid/vote is the only one that counts.

Besides the Mariners, I don’t think too many people thought Silva was worth 4/48 either.

Silva: agreed, but 4/34 sounds fine.

Anthony, sure they give the other guys more RBI opps or more run scoring opps, or whatever, but that does not make the other guys “better” from a “talent” or “context-neutral” perspective.  Plus those extra opps for the other guys are already included in the value of the good players we are adding to the lineup.  In other words, if a player has a true talent of +20 runs (offensive linear weights) per 500 PA, those +20 runs includes the extra RBI and run scoring opps for the other players in the lineup.

But, if the manager or GM means what you are saying, then that is fine by me.

From a Ron Shandler chat today:

hyattsiville, md: Who will have better stats over the rest of the season - Cliff Lee or CC Sabathia?
Ron Shandler: Wow - great question. I am going to say Lee, for several reasons. First, he is doing it right now. There will be some drop-off but he’s already shown a certain elevated skill level. Sabathia has a better track record, but he also has those three weeks in April that might pop up from time to time. When he was going through that, we were speculating to the cause—overwork, injury, etc.—and those variables still exist.

Well, count Shandler as another guy, if he is right, who could make a fortune from me.

Ya think he’d be willing to put his money where his mouth is?

I just read the chat.  Maybe he is talking about fantasy league stats, whatever those are (I don’t play).  I don’t know.  It is a good chat though.  He is a knowledgeable and insightful guy.  And writes well to boot.

If he’s referring to fantasy stats then in a standard 5x5 league the pitching stats would be K, W, SV, ERA, WHIP.  I agree that he’s a knowledgeable and insightful guy and produces some great stuff, though I think he can focus on too small of samples at times.

After 85 votes, Cliff Lee is at exactly 4/34 as the value you guys would pay him.

Since the last time this thread was commented on, Cliff Lee has thrown 65 1/3 innings, given up 67 hits, allowed 2 home runs, has walked 15 guys, and struck out 60.  His ERA over that stretch is 3.17. 

Compare with Sabathia at 67 IP, 54 hits, 5 home runs, 16 walks, 63 strikeouts, and a 2.55 ERA.  So, technically, MGL’s winning, but you could argue that Lee’s actually outpitching Sabathia thanks to the lower home run rate. 

Anyway, at this point, I think we have to admit that Lee’s early season success wasn’t some weird random variation - he’s just pitching way better than he ever has before.

I agree, Lee has become a damn good pitcher.  He’s certainly not getting lucky on hits, though probably a bit lucky on homers.  But any guy who can maintain a 4-1 K/W rate can pitch for me.

Well, we don’t KNOW that he has become a better pitcher, of course.  We also don’t know that e wasn’t already a batter pitcher than we thought before, and his historical performance was unlucky.

However, as I think I already stated in one of these forums, once we saw that his velocity was considerably higher this year than last (and his movement on certain pitches, I think), that is a very good clue that his true talent changed and that his basic Marcel was not valid, at least not without some kind of tweak.

In fact, I think the biggest clue to a pitcher significantly changing his true talent is his fastball velocity and perhaps his “mix” of pitches (for example, if he previously threw 50% FB and 10% CB and now throws 42% FB and 18% CB, there is a good chance that something has changed for the better or for the worse).

I any case, the velocity thing aside, if we go back in time on any player and compare his performance to his Marcel before that time period, of course we are going to be “wrong” some percentage of the time, that percentage depending on how we define “wrong.” Does that mean that the Marcel itself was “wrong” at the time it was computed?  No, not really, given our limitations in projecting performance and given the vagaries of random chance.  Again, he velocity thing aside, would my projection for Lee now be any different than any other pitcher, given his historical performance this year and before?  No, it would not (again, assuming no change in velocity, which is NOT the case of course).

So I really don’t see the point.  There are probably plenty of other players besides Lee who had a certain projection on April 1, played well below or above that projection from April 1 to May 1, and then STILL played above or below their Marcel (as computed on May 1) up until now.  So what?

Is that evidence that our projections are not perfect and have no uncertainty, and that random variance in player performance over finite sample sizes exists?  I think we already knew that.

I wasn’t trying to make any point about the validity of projection - i just think Lee’s transformation from guy who got banished to triple-a to Cy Young contender has been really interesting.

I agree.  I am not sure that he is good enough to maintain the kind of level that will keep him in the running for the CYA, though.

I was thinking about this today (again):

We don’t ever know the true talent level of a player of course.  We can only estimate it.  We also can never know whether a true talent level has changed.  Our estimate of it changes every time we have more data, but we have no idea whether the true underlying talent changes and if so, by how much.  We think that players’ true talents stay relatively stable other than due to normal aging and normal injuries, and we also know (or think) that sometimes players’ true talents change drastically for one reason or another, although even then we are not sure - they may have just been incredibly lucky or unlucky in the past or in the period after we think their talent changed.

There is only ONE way to really get a substantial idea as to whether a player’s true talent has changed.  And that is from observation, scouting, and measurements, such as pitch velocity and movement, that are not as subject to random fluctuation as our usual performance metrics are.

This is a very important concept.

Let’s say that we have a bunch of “special” coins such that each one of them has a slightly different heads/tails percentage on a flip.  Every once in a while a few of them (say, 10%) get all bent up such that the true odds of landing on heads or tails after a flip changes by 5%.

If we keep flipping all the coins and tracking our results, our estimate of each coin’s true heads/tails percentage is probably going to change each time we record a new flip or series of flips.  We will never, however, be able to tell if all of a sudden a coin got bent and changed its true odds by 5%.  There is going to be too much random noise.  But, if we ever bother to look at the coins, we will see which ones got bent and therefore which ones did change their “true talent.” And even if we didn’t know that th ebent coins had their true talents change by 5%, if we had a coin whose sample performance seemed to change from one time period to another (which is going to happen by chance all the time) AND we see that it is severely bent, that is a good clue that its true talent HAS changed.

Same with players, although it is harder to identify something about that player (without being influenced by his stats) that suggests that his true talent has changed, especially with batters.