THE BOOK--Playing The Percentages In Baseball

Unfortunately (for the truth), the whole world of sports commentary revolves around what each player has done lately, from the last week or month to the second half of the season to the season itself.  It pisses me off because it literally IS the pervasive theme in sports and especially baseball and it is DEAD WRONG.

The same with teams.  “The Yankees are great, no they suck, no wait, they’re great, nope they suck, no, sorry, they’re great.”

Very well written article by Dave. One minor mistake which one of the commenters pointed out. Dave says that our “opinion” of each player should not change after the hot or cold streak.  Of course it changes since each player’s historical averages have changed slightly, the hot hitter for the better and the cold hitter for the worse.  How much that historical norm (bascially our projection for that player) changes depends on how much much data we had for that player before the hot or cold streak streak.  The less data we had, the more the streak impacts our projection. Of course, no matter what, the streak is not going to change our projection all that much.  If we have little data beforehand, then the regression, even after the streak, is going to be a lot.  If we have lots of data beforehand, then the streak is not going to make much of a dent in the projection either.

Honestly, you can have 10,000 responses and there are NOT going to be any valid rebuttals to Dave’s argument.  The math, the logic, and the research are all as sound as you can get.  It is like trying to rebut that e=mc squared or force=mass time velocity squared.

MGL is right.  It happens all the time whether they’re talking about hitting with RISP or just some hot streak.
It used to piss me off too, but I thought about what I’d say instead, “Well he has hit well w/RISP this year, but that doesn’t mean that he will continue to do so.” And that would sort of be my response to everything!  And while eventually viewers might come to understand “small samples”, I’d have been fired for being boring long before that happened.  The dummies have to say something to stay employed!
My solution:  I love the Mute button!!  The only announcers I’ll listen to are Boston’s and NYY’s.  The rest deserve the muting they get!

Honestly, you can have 10,000 responses and there are NOT going to be any valid rebuttals to Dave’s argument.  The math, the logic, and the research are all as sound as you can get.  It is like trying to rebut that e=mc squared or force=mass time velocity squared.

Well, thanks, but I’m taking exactly zero credit for the math, the logic, or the research.  That was all you guys.

FWIW, I have Vidro and Jones projected at about the same in hitting, basically zero lwts, which is about average for a 2B or CF and much below average for a corner OF or DH.  Ibanez rates slightly higher in hitting or about +2 in lwts (per 150 games).  Jones defense appears to be better so I would have to give him the edge overall over Ibanez.  Of course, Ibanez and Jones would make a great platoon, as basically Ibanez only has value versus RHP.  Jones projection is of course based mostly on his minor league stats. Minor league stats, on the average (I wish I didn’t have to always say that, but I do), is almost, but not quite, as predictive as major league stats.

One poster on Dave’s site said this:

I used Raul Ibanez’s recent streak as an example of a guy on a hot streak (a pace of nine homeruns in thirteen games is obviously unsustainable for any length of time) because it ties into the argument here. I don’t believe anyone in the clubhouse, including Adam Jones, thinks for a second that it’s a good idea to bench a guy who’s been torching the ball like Raul has lately. If the guy was legitimately tired and needed a day off you MIGHT be able to convince him of that.

Players spend years of sweat and sacrifice, hour after hour in the batting cage, watching video, lifting weights, etc. to be able to get on a roll like Raul has been on over the past two weeks. Most would rather have their teeth pulled out with rusty pliers than be pulled from the lineup when they’re hitting like that and you’ll never, ever convince them this has no value in predicting how they’ll hit today. If you sit a guy hitting like that based on nothing more than a purist adherence to statistical probability then you risk losing not only him but the ballclub.

While I don’t know that the sky would fall if you benched a player on a hot streak or kept playing a (non-star) player on a cold streak, but his point is well taken.

I don’t know how you handle this on a practical level.  There must be some common ground.  First of all, if I am advising a manager, I would tell him that if you want to give a player a rest anyway, you might as well rest him for a game or two during a cold streak rather than a hot streak even though I know that it doesn’t make any difference.

I think that even though this poster is right in that there isn’t a player, manager, or coach on the planet that does not think that a player on a torrid streak is going to continue being hot for SOME time, it IS possible to reprogram people, even baseball players.  If you explained to your players and kept reinforcing the fact that their playing time, etc., is going to based on long-term performance, I think that after a while they will begin to understand, or at least blindly follow.

The thing that also bugs me is how much time and energy managers spend on lineups.  No managers ever have a set lineup.  Some more than others of course.  I would love to be privy to the thinking (or lack of it) that goes into the making of a lineup each day.  I look at lineups every day and it is 1 out of a 100 times that a lineup from one day to the next is exactly the same even with the same players.  I means what motivates a manager to bat a certain player 7th one day and 6th or 8th the next day?  That happens ALL the time.  I mean ALL the time.  Again, if I were advising a manager or were a manager, I would have one lineup versus RHP and another versus LHP for the whole season.  (Actually you can do a little better than that by using GF platoon matchups and other nuances to optimize your lineups.) The idea that certain players like or don’t like certain spots in the order is overrated.  Most players don’t care.  Sure, if a player has a perference, I might acquiesce to it, since the difference between one slot and another is probably not worth a player being unhappy or uncomfortable (and possibly hurting his performance).  But, as I said, by and large, most players will openly say that they don’t care where they hit.

BTW, did anyone see that Griffey, the other day,
when he was taken out late in a game for defensive purposes, said, “This is the first time in my career that has ever happened, and it will be the last?” How ironic for the WORST defensive player in baseball to say that!  Either his ego is gigantic or he has no idea how bad his defense is.  Probably a little bit of both.  I lost all respect for him after hearing that. Not that I had any particular respsect for him before that.  The only “respect” I have for baseball players, as for all human beings, is when they do something productive for mankind, other than entertaining us on the ball field and making ridiculous sums of money in doing so.

Of course, Ibanez and Jones would make a great platoon, as basically Ibanez only has value versus RHP.

We’ve been suggesting since about May that Ibanez/Vidro should be the ones platooning, with Jones playing every day in left field.  Ibanez is giving Manny a run for his money as the worst defensive left fielder alive, and the outfield defense is killing the Mariners pitching staff. 

I don’t know how you handle this on a practical level.  There must be some common ground.

I think Joe Torre’s writing the bible for how to handle this right now.  Thanks to injuries and decline, Damon became a liability in center and lost his job to Melky Cabrera.  But, he’s not totally useless, so now the Yankees are using Damon to split time between left (giving Matsui rest), at first (where he’s probably their best player), and at DH.  No one’s getting benched, and Torre gets to keep his older guys rested for the stretch run. 

We’re basically suggesting the same thing in Seattle - Jones plays almost everyday in left, with Ibanez/Vidro/Guillen/Broussard/Sexson splitting the DH/1B/RF at-bats between them.  Two lefties, two righties, and a switch hitter - shouldn’t be that tough to exploit platoon advantages and keep everyone fresh. 

No managers ever have a set lineup.

John McLaren does, and it’s one of our complaints about him, actually.  He runs Ichiro-Vidro-Guillen-Ibanez-Beltre-Sexson-Johjima-Lopez-Betancourt out there on basically an everyday basis, regardless of the matchups or the pitchers G/F rate.  He starts it against lefties and righties.  He starts it whether Felix or Washburn is on the hill.  When he took over right after Hargrove’s resignation, he tinkered a little bit, but he’s clearly found a line-up he likes and he’s sticking with it.

How ironic for the WORST defensive player in baseball to say that!  Either his ego is gigantic or he has no idea how bad his defense is.

It’s definitely both, but I’d say mostly the latter.  You know better than most, I’m sure, but most of these guys still have absolutely no clue how to evaluate defenders.  The Mariners are convinced that their defense this year has been a strength of the team and their weak link is Jose Lopez.  It’s unbelievable to me that they can watch Ibanez and Sexson on a daily basis and not realize how terrible they are, but it’s been pulling teeth to try and get people to get on board with those pretty obvious conclusions.

Sure, platooning Vidro and Ibanez would be fine, although I’m not sure that platooning Jones and Ibanez would not be better since Vidro is basically equally good verus RHP and LHP I think.  If Jones is that much better defensively than Ibanez, then I supoose it makes sense to keep him in left.  A quick check of Ibanez’ UZR’s put him around -5 or so presently.  I don’t have any numbers on Jones of course, but if he is truly considered a stellar defender (and is a true CF), he would have to be at least a +5 I would think, maybe more.  That is a big difference and sounds like he is much the better option in the OF.  Ibanez does not have much overall value left, and has replacement level value (none) against a lefty pitcher.  For a team in a post-season race he should definitely NOT be playing versus any LHP.

There IS no reason to change your batting order other than versus lefty and righty opposing pitchers.  The only change I would make with the Mariners batting order would be to drop Ibanez down (or remove him as I said) versus a lefty pitcher and raise Beltre or Sexson, although I have no idea what an optimal lineup is for them in general.  For that, you need a copy of “The Book” in front of you and a pencil, paper, and calculator, or a good sim.  Anyone that thinks that they can come up with an optimal lineup off the top of their heads is fooling themselves.

Why would you have a different lineup with Felix or Washburn on the hill, unless you mean different personnel?  I was talking about batting order.  With G/F ratio, I was meant the opposing pitcher.  There is a pretty strong platoon differential between batters and pitchers with respect to their G/F ratios.  It just does not come up very often because most pitchers and batters are around average in G/F ratio, unlike L/R platoon differentials which almost always come into play for each batter.

Hey, when I was listening to the Reds broadcast on XM radio today, Brantley, in reference to platoon splits, said, “You can take that lefty/righty stuff and throw it right in the garbage!” I’ve criticized some commentators before, but Brantley is truly sickening.

A quick check of Ibanez’ UZR’s put him around -5 or so presently.

The July 17th UZR update had Ibanez at -14.  For him to be +5 presently, he’d have had to have gone +9 over the last month.  I know that there’s massive sample size issues, but I’m not sure I can wrap my head around that possibility.

Why would you have a different lineup with Felix or Washburn on the hill, unless you mean different personnel?

Right, sorry, that’s what I meant.  Three times in a weeks span, they tried sticking Vidro at second base in order to get Jones in the line-up, essentially having him replace Jose Lopez rather than one of the veterans.  Experiment one was with Miguel Batista on the mound, experiment two came with Horacio Ramirez on the mound, and experiment three came with Felix Hernandez on the mound. 

Those are, of course, the three guys in the rotation most likely to induce groundballs, and the days that you’d want to emphasize infield defense over outfield defense.  Rather than sticking Vidro in the field when Weaver (37% GB rate) or Washburn (38% GB rate) are pitching, McLaren screws the pooch even when he is tinkering with the line-up.

It’s crazy. Watching every Mariner game this year has been like watching a four year old run laps across an interstate.  You’re freaking out and waiting for the disaster at any moment, and then can’t help but laugh when probability continually gets thrown in your face and everything turns out just fine.

But did Brantley mean you can throw the concept of lefty/righty out the window or did he mean to throw the individual numbers out the window? While either would be incorrect, the latter isn’t a bad rule of thumb if you aren’t capable of regression/making a decent projection.

And I think MGL’s -5 for Ibanez is a projection, not a season total.

I read the post at USS Mariner and thought it was great.  Well done.  I was glad to see it mentioned here also. 

On a similar note, the Seattle P-I today ran a chart of the top single-month OPS numbers by individual players in Mariners history, with Ibanez current 1.415 OPS for August ‘07 at the top of the list.  After reading the entries here and at USSM (and having read The Book), I took a rudimentary look at their chart to see how these players did in the month following their “hot” months (big thanks to Baseball-Reference, of course).

Anyway, I took their top 27 list and put it into Excel.  For example, #2 on the list was:

Edgar Martinez, August ‘95, 1.346 OPS.

Then I looked at the month following each player’s hot month, plus their season OPS:

Edgar Martinez, September ‘95, .871 OPS, season OPS 1.107

In this example, Edgar had a scorching August ‘95 but a sub-par September ‘95.

With the 27 data points, I had to throw out 2 of them (Ibanez this year because September hasn’t happened yet and Edgar’s September ‘98 because there weren’t enough October ‘98 plate appearances to go on).

So, 25 hitters having the top one-month OPS numbers in Mariners’ history:

Average OPS during hot month:  1.211
Average OPS during the following month:  .937
Average OPS for the whole season:  .973

Out of the 25 players, one did a bit better (+.067) and another was nearly the same (+.002).  All others saw an OPS drop of between .102 and .503 from their hot month to the following month, for an average drop of .275.

So, Ibanez now has an OPS of .802 for the season.  In August, it’s 1.415.  Has he turned some magical corner?  Will he have an OPS of 1.000+ from here on out?

Doubtful.

Big thanks to both of you, David and MGL, for the interesting topic.

Regarding hot streaks, most athletes themselves seem to believe they are “in the zone” when they are hot.  There are many examples like this old cliche from Eric Byrnes:

When you’re really, really hot,” Byrnes says, “it can almost look like a beach ball.

While it is impossible to completely discount this experience from one’s armchair, there is another class of people that believe something very similar: inveterate gamblers, even those involved in games requiring no input from the gambler, such as craps.

It is a very real phenomenon that a shooter in craps will believe they are on a hot streak when they are shooting sevens repeatedly.  Hell, even I have been tempted to believe it while playing $5 craps, and it goes against my nature completely.  For some reason it is very hard to attribute anything to luck, even when it must be so.  After all, even Einstein said “God does not play dice.”

Bobby/11: fantastic!  Tell Baker, will you?

And, the seasonal OPS you reported (.973) includes the hot streak of 1.211.  Of course, what you are interested in is his “rest of season” OPS.  Presuming that .973 is 6 months of data and 1.211 is 1 month, .937 is another month, then the other 4 months would average to .923. 

Since you reported they’re post-streak month was .937, this shows a smidge of carryover effect (though at a sample of n=25, likely not really significant anyway).

Honestly, the best way to prove this stuff to people is to make them do the work.  Give them an assignment: prove that streaks have a carryover effect.

Do streaks exist in baseball?  That is, are there more/longer clusters of hot/cold performance than you’d expect by random variation?

I’ve heard people claim that streaks actually do exist, but you don’t know when/why they will stop.

I agree completely with the consensus.

The only difficult point is at what point “a streak” for a young player becomes indicative of development, or more accurately, what weight to attach recent performance as compared with less recent performance for younger (under 25 perhaps) players, and, (I suppose) for older (over 33 perhaps) players.

For hitters, I give a weight of 0.80 per year, or 0.9994 per day.  (Performance a year ago is weighted at .80, performance a month ago is weighted at .98, etc.)

For you techies, this is the equation in Excel:
=EXP(LN(0.8)/365.25)

Use 0.7 for pitchers.

I think it’s a fair point to discuss the weightings of players farther from the peak point, that if they are on the more steep slope that perhaps the recent performance should be weighted much more.

So, you’d want a function to also include “years from age 27”, or some such.  As well, the slope is fairly steep going up to age 26/27, and not so much until the mid-30s, at which point it drops rather quickly.

The Padres are currently hitting Geoff Blum second because of his recent hot streak (I think, anyway).

From 03-06, Blum basically put up a ~70 OPS+. This year, he’s at 83. I tried to break his season down into streaks:

Apr 5 - May 5 - .103/.182/.154 (44 PA)

May 9 - June 7 - .324/.410/.559 (40 PA)

June 8 - July 1 - .154/.184/.192 (27 PA)

July 6 - July 31 - .333/.426/.359 (47 PA)

July 31 - Aug 18 .319/.373/.420 (75 PA)

....

Apr 1 - July 1 - .192/.264/.303 (111 PA)

July 6 - Aug 18 - .333/.403/.410 (119 PA)

Does a 120 PA hot streak do anything significant to projecting future performance besides changing his overall numbers a little bit? I’m guessing that is doesn’t.

Anyway, it’s fun to see this work itself out with players. If you evaluated Blum by his streaks, you wouldn’t have a clue what kind of player he is. I’m pretty sure he’s a player I wouldn’t want batting 2nd.

I think in the book, I mentioned that a small “carryover” effect could be because hit streaks tend to occur in warmer months, and cold streaks in colder months, the health of the player is probably a factor, and even the park, as a hot streak will tend to have hitters parks in it and a wee bit of those hitters parks will “leak” into an adjacent future time period.  So we really would expect a hot or cold streak to “carry over” slightly even if it has nothing to do with a “new level of performance.”

Our estimate of a player’s talent level constantly changes, after each PA in fact (unless that PA is exactly equal to our prior projection of course).  So after a hot streak, our estimate of talent does indeed change for the better.  So this whole idea of whether we can discern if a player has taken “the next step,” or whatever you want to call it, is silly.  All we can do is recompute our estimate of that player’s talent/future perormance, using a formula (such as Tango’s) that we find from analyzing hitsory that works the best.  There is nothing more or less we can do.  Now, whether watching, knowing a player, knowing his history, style, etc., can inform us as to whether a player has increased his talent more than the projection algorithm suggests, or, whether a player had “dropped off the table” because of age or injury, or some such thing, is another story altogether.  If we are talking about just knowing the numbers, then all we can do is constantly update our projection, which is what sabermetricians do anyway.  And one has to be very careful about making judments based on observation which would trump what we glean from the numbers.  When a player, especially a pitcher, is in a hot or cold streak, even if it is competely (or mostly at least) random fluctuation, they always “look” bad or good, and you would swear that their talent has indeed changed.  Part of the randomness of baseball performance is in fact, that the player is actually performing badly (like the pitcher’s curve ball is not curving well or his velocity is down a little on his fast ball), not only that bloop hits are dropping in or line drives are being caught.  This is a concept that a lot of people don’t get.

Tango,
Thanks for the response to my comment.  After looking at the chart in the Seattle P-I, I noticed something interesting.  Within their top-27 they have 3 months from Edgar Martinez’ 1996 season:

April ‘96:  1.150 OPS
May ‘96:  1.152 OPS
July ‘96:  1.169 OPS
Season OPS:  1.059

In June of ‘96, Edgar Martinez’ OPS was 1.050.  So really, this is a guy with consistently-good numbers for 4 straight months to start the season:  1.150, 1.152, 1.050, 1.169, who then drops to .950 in August and .867 in September (also, he missed around 20 games in July and August because of injury). 

His “hot” months weren’t that much better than his full-season numbers, around 100 points of OPS and he was dealing with an injury at the start of the down months.  There are other guys on the list who were really out-performing their expected numbers (Tom Paciorek’s May ‘80 (1.198) as opposed to a season OPS of .732).

In the current case of Ibanez, maybe he shouldn’t be compared to Edgar Martinez ‘96 because Ibanez “hot” month is so far out of line with his expected numbers as compared to Edgar’s.

Beyond that, using a single month as the cut-off point, as was done in the newspaper, is arbitrary.  Obviously, a player doesn’t wake up on June 1st and hit great for 30 days, then stop hitting on July 1. 

It’s good they ran the chart yesterday, before Ibanez’ 1 for 5 dropped his OPS 70 points.  Plus, he’s 0 for 3 today.

Here’s a question bumping off of Vidro’s August. With him there’s a wealth of info to help with context. What does a FO do with a guy like Keppinger who’s value seemed to be mostly related to his projection as a a supersub (i.e. can hit well enough to carry his glove in a utility role). Given the limited major league data for him, how much should his hyperbolic 100 or so PAs with the Reds alter our evaluation of his true skill level?

We do have plenty of minor league data on Keppinger.  Based on that and his hot streak, we can be fairly confidant that he’s capable of hitting .300 and being one of the hardest guys in the league to strike out.  However, he doesn’t walk, have much power, or run very well.  Defensively, he’s a 2B without the arm for 3rd or the range for short.

Given that the Reds have Brandon Phillips, seems that Keppinger is best used in a utility role.  A possibility would be to move Phillips to short and play him at 2B, but Alex Gonzalez has a 3 year deal and is a very good defender, so that move might not make sense either.

You could see if someone wants to buy high on Keppinger in a trade, but failing that, be happy with him in a utility role.

While I would treat the MLB PA from one year ago at 0.80 weight, I’d treat the minor league PA from one year ago at 0.64.

That is, just on my guts, I’d say the reliability of a hitter’s MLB data from two years ago is the same as a hitter’s minor league data from one year ago (given the same age).  That’s another study for someone to tackle.

I’d also guess the reliability goes down as the competition level goes down.  Perhaps it follows this form in terms of weight of performance for one year ago:
0.80: MLB
0.70: AAA
0.60: AA
0.50: A
0.40: College
0.30: HS

Something along those lines.

My current projection for Keppinger is good, about that of Hatteberg and about that of an average third baseman.  I’d be real happy to have this guy as my everyday second baseman.

Perhaps the Reds should make Keppinger the 2B and trade Brandon Phillips.  I’m not sure what Keppinger’s defensive ability is.  Phillips has the 20 HR/20 steals, but Keppinger is probably just as good an overall hitter due to his superior OBP skills.  Phillips is a guy they could most likely get a lot more for in a trade.

MGL,

That projection for Keppinger refers only to his bat correct? Or does UZR already see something about his glove despite the small sample size that scouts don’t see?

No, I have no idea about his glove.  UZR is never going to “see” something in a small sample that scouts don’t see.

I only have 4 ballots for Kepp, and he comes in at fairly average across the board.  Phillips however, like last year, is considered the team’s best fielder.