THE BOOK--Playing The Percentages In Baseball

More extra inning games?

I think Joe’s wrong.  Based on Yahoo’s aggregate team stats, I see 4284 runs in 471 games for 9.096 per game.  This is through May 4th but I doubt it would be that much different than through April’s games only.

Peter: I see.  You are thinking that Joe is showing per game, as opposed to per 27 outs?

In 2007, there were 54.1 outs per game (both teams).  53.9 outs in 2008.

Nice try.  I’m thinking Joe simply goofed here.

The couple thoughts running through my head about this are:

1) Have there been more outs on the made base paths (caught stealing, thrown out trying to reach an extra base, etc.)?
2) Is there an increase in double plays this year? 3) More runners left of base?

I think before dismissing Joe’s calculations those need to be answered, as they hold some of the additional details not already captured by the “slash” data.

I’ll have to check these out after work, but if anyone else has time to do it before then I’ll be glad to see the answer.

Joe Sheehan’s NL 2007 R/G is wrong. There were about 8.85 runs per game in april 2007, not 9.31. The slash stats for that year appear to be correct. [His “April” stats include the March 2008 games.] There is some fluctuation between the rate stats in each league and runs/game, but it is up in the NL and down in the AL, not down in both.

Probably thanks to all the AL snow-outs in 2007, the ratio of games played between the leagues was not the same in april 2007 and mar-april 2008. [340:400 in 2007; 385:435 in 2008].  That causes a little distortion in simply aggregating to an MLB average.

.. as Peter suggested, Sheehan did use R/G, not R/9 inning. Sheehan’s “conclusion” could have been further undermined if there were a discrepancy in extra innings between the years as well.

Joe Arthur is right about Sheehan’s 2007 NL R/G being wrong.  Here’s another look at the numbers on a /27 out basis…

AL R/27 OPS
2007 4.73 .731
2008 4.57 .732

NL R/27 OPS
2007 4.37 .732
2008 4.53 .735

As Joe Arthur pointed out, the pendulum has swung in different directions for the two leagues from 07-08, but they appear to be on more common ground this year.  So what happened last year.  Don’t really know without more details, but a great mind once said…

“screws fall out all the time, the world’s an imperfect place.”

-Johnny Bender, Breakfast Club

It’s possible that the LOB% has shifted a little bit. If you look at pitchers with long careers on their fangraphs pages, you’ll see significant differences (in league average) taking place over short periods of time. Looking at Greg Maddux’s page, I see the league average was about 70% in 1996, and in 1992 it was about 73%. Could a change in LOB% explain this? (unless the data he presented it just simply wrong)

Dan/8: runs scored per game changed dramatically between 1992 and 1994.  Bad choice that you selected 1992.  Since 1994, runs per game is fairly constant, more or less.  LOB% is directly related to runs scored per game.  Loosely speaking, and this is loose, you can get
1 - LOB%
= OBP minus .003
(more or less). 

The basic point is that as you get more guys on base, the percentage of those guys who remain on base goes down.

Think about it when OBP is .010 and OBP is .990 to see that it makes sense.

According to Doug’s web page, rpg and rp27, as of May 2 are:

rpg
NL 4.55
AL 4.47

rp27
NL 4.52
AL 4.54

Using STATS data, as of May 1, here are the “slash” stats for the NL and AL:

OBA=(h+tbb+hp)/(ab+tbb+hp+sf), where tbb=ibb+non-int. bb.

NL .256/.330/.403
AL .259/.333/.397

Ok, I finally dug up part of what I suggested earlier:

CS:
Apr 2007 - 155
Apr 2008 - 199

GDP:
Apr 2007 - 701
Apr 2008 - 673

I can’t seem to find any split data for LOB in 2008, but Apr 2007 left 5388. My guess is that there isn’t a significant difference to explain.

Regardless, I still agree Joe’s numbers appear off. I used the Retrosheet game logs and matched the R/27 data from wes/#7

#7, when I run 2007 April, I get:

NL

.273/.337/.443

AL

.281/.347/.445

MGL/12: My main blog post has OPS of .732 in 2007 and .733 in 2008, basically a match to Wes/7.  Are you excluding pitchers?

(hey, interesting use of TBB to denote NIBB+IBB.  That certainly removes the ambiguity of BB.)

Matt/11

What about sacrifice bunts/flies?

My main blog post has OPS of .732 in 2007 and .733 in 2008, basically a match to Wes/7.

Is that for April of 07 only?  I’ll have to check my numbers.  They seem a little high. Obviously for the AL, no pitchers are hitting in April.

You occasionally see “TBB” in a stat line. Not too much anymore, though.  It is a little confusing when you see BB, as to whether it includes IBB or not.  It is supposed to include IBB (which I generally don’t include in any of MY stats), but not always, so it is a good idea to ask or to specify.

Dan/#14,

SH:
Apr 2007 - 205
Apr 2008 - 276

SF:
Apr 2007 - 210
Apr 2008 - 237

Such a small number of plays makes it hard to say more small ball is being played, especially with no tally for “sacrifice opportunities” readily available to me.

My bad.  For 2007:

NL
.258/.332/.400

AL
.255/.327/.404

Same as Joe in the article.

Wow, it took us this long to come up with 07 and 08 April stats!

So basically, the slash stats for April 07 and 08 ar are almost identical in both the NL and AL.  And the rp27 per team are down .16 in the AL and up .16 in the AL, but exactly the same overall.

So has anyone checked Joe’s numbers for the rpg?  Everything else has been checked and Joe’s numbers are right, but I don’t see any checking on his rpg in 07.

The rp27 in 07 are right in line with the rp27 in 08, but are the rpg?  Doesn’t matter, I guess, as all we care about is rp27.

So when using rp27, both runs and slash stats are almost identical in 07 and 07.  No evidence of anything “funny” going on.  No need to check outs on bases or anything like that.

While we are at it, can anyone tell me what the “games” column means in these B-R pages.  It has been driving me crazy for weeks now.  Everytime I want to compute something “per game” I can’t find anything on the B-R pages that tells me how many “games” were played in a league or in both leagues combined. For example, here is the March/April line for the link that Tango gives above, which is “split” data for the ML 2007:

2978 370 370 198 13 4 6670.1 4.12 6526 3360 3056 681 2609 211 4888 274

The 2978 is “games” and the 3360 is “runs.” What the hell is the 2978?  Whose games are those?  On all of the splits lines, I can’t for the life of me figure out what the number in the “games” column means?  Shouldn’t I be able to divide runs by games to get “runs per game?” 3360/2978 = 1.13.  What the hell is that?

BTW, absolutely love the site, B-R, for the content and the work that Sean does, but I absolutely hate the site for its messiness, complexity, and many other problems like the one cited above.

Right now I am trying to simply retrieve this years’ batter and pitcher stats, which I have done before.  I am trying to log in and it is not accepting my log in, I don’t think.  Or I cannot tell if I am logged in.

The difference between the PI and the regular pages are not clear at all.  The interface on the PI page is a disaster.  I really have no idea how to navigate this web site, which is a shame as I am sure there is all kinds of great stuff I would like to access but cannot. 

As Simon from American Idol would say, the web site is a nightmare, in my opinion.

Games at B-R are a bit weird.  It is the number of players that appeared in each game, summed.  If you look at 2008 Cards vs. Cubs, it says the Cubs have 44 G.  They have played 3 times: on 5/2 21 Cubs appeared in the game, on 5/3 11 Cubs appeared in the game, and on 5/4 12 Cubs appeared in the game.  21+11+12=44.  Would have to look at an AL series to figure how DH/P are handled.

Right, “games” is player games summed.

If you want team games, it’s simply W+L.

plus regulation games that are called with the score tied and not resumed (Cubs/Astros in ‘05, Yanks/O’s in ‘03, Braves/Giants in ‘02, etc.)

#9 Tangotiger

Not to be a bother, but can you provide an example of that formula’s results using actual stats? It doesn’t seem to work out that way to me even after reading your last sentence.

http://tangotiger.net/markov.html

Hit CALCULATE.

4.905 runs scored, with an OBP of .341.  With 27 batting outs, that implies .341/(1-.341)*27 = 14.0 times on base.  One of those was a HR, so he doesn’t count as a baserunner.

The LOB% = 1 - 3.905/13 = 70.0%

Go back, and put in “42” for AB.  Hit CALCULATE.

3.697 runs scored, with an OBP of .304.  That gives you 11.8 times on base per 27 outs (you could also have done 10+4 divided by 42-10 times 27).  One-fourteenth of the times on base is a HR, so 0.84 HR removed.  That leaves us with:

LOB = 1 - (3.697-.84) / (11.8 - .84) = 73.9%

As you can see, the OBP went from .341 to .304 (a change of .037 down) and the LOB% went from .700 to .739 (a change of .039 up).

Seeing how wonderfully symmetrical that is, we can have a rule of thumb as:

LOB% = 1.042 - OBP

So, 1.042 - .341 = 70.1%
and 1.042 - .304 = 73.8%

Just a matter of looking at empirical data, and coming up with the appropriate best-fit.  It’ll probably be close to 1.04 or 1.03 minus OBP.

Sweet, right?

Also look at post 5 here:
http://www.baseball-fever.com/showthread.php?p=1019793&highlight=marquis#post1019793

which I will reproduce here for my searching sanity:

In any case, LOB% will be something close to:
IP*3/TBF
with maybe some corrector factor for all pitchers.

For example, here are the LOB% on Fangraphs, and my quick estimator in parens:
Peavy 78.3% (74.6%)
Beckett 75.2% (73.2%)
CC 74.1% (74.2%)
Marquis 67.7% (68.0%)

For a quick estimator, it does a decent job. Probably much better for careers.

The logic behind this is the following: the % of baserunners that score is roughly in-line with the % of batters that get on base. And IP*3/TBF is roughly 1 minus OBP.

Support for the relationship can be found here:
http://www.tangotiger.net/markov.html

Change 37 to something else, like 57. In that case, you have almost 2 runs scoring, with a .230 OBP (which itself implies 8 guys reaching base). So, the percentage of runners scoring is 25%, meaning 75% left on base.

Change 37 to 24, and you have an OBP of .500 (27 guys reaching base), with 14 runs scored. 14/27=52% runners scoring, and 48% left on base.

OK, thanks.  That games thing is weird.

It’s useful when you look at pitching splits, such as a team had 500 relief games or 170 relief games in a season.

Otherwise, what would you show in the team pitching splits?

Obviously, he could show BOTH, as the team games, and the sum of the player games.  This way you can show a team having 156 games with a relief pitcher used (i.e., 162 minus 6 complete games), and a total of 500 games in relief.

#9 “1-LOB% = OBP - .003”

-> LOB% = 1.003 - OBP

#24 “LOB% = 1.042 - OBP”

Is there something different here (perhaps exclusion of HR somewhere) or have you just determined that 1.042 is a better fit?

#24 “Sweet, right?”

Yes it is.

I said .003, but I meant .030.

1.042 is based purely on the sample illustration provided.

My expectation is that it’s going to be somewhere around 1.03 or 1.04 or so, and I haven’t done anything to determine that.

And, clearly, a guy who gives up lots of HR, but has a low OBP will not have the same LOB%.  The best way to figure it out is to run a Markov or Baseruns for each pitcher, infer the LOB%, and then come up with an appropriate function.  The more data you use, the better the fit.

wOBA might be a good way to do it too.

Note that LOB% is 1 minus the scoreRate.  And in BaseRuns, the scoreRate is B/(B+C).  So, you can just use that, which is very roughly:
B = hits + extrabaseHits
C = ip*3

Just something to play around with.

What’s going on at BP?
http://www.baseballprospectus.com/article.php?articleid=7514

I wrote to Joe:

Joe,

Can you please quote your runs per game numbers as per 27 outs.  In the AL this year, there have been 17.8 innings per game, while in the NL it’s 18.0.  That by itself accounts for 1% of the difference, and so would raise your 8.8 runs per game up to 8.9 runs per game.

And according to b-r.com
http://www.baseball-reference.com/pi/psplit.cgi?team=TOT&lg=AL&year=2008#total-total

There are 2408 runs scored in the AL on 4816.1 innings, which (if you double that 2408) you can see is a perfect 9.0 runs per 18 innings.

I don’t know where that 8.76 came from, but perhaps there were alot of runs scored between May 10 and May 13.

Tom

Also note that if you take his OBP*SLG numbers for 2008 and 2007, that is a 4% difference.  His RPG numbers show a 7% difference.

And no mention of his previous article which we put through the griller here, where he had the AL numbers at 9.04 runs per game for 2008 (but had the 2007 numbers all wrong).

Joe wrote back, and I wrote back back:

Joe,

Your numbers are correct mostly (260 games, not 261 according to b-r.com):
http://www.baseball-reference.com/games/standings.cgi?date=2008-05-10

But, May 11 and 12 had 11 games with 127 runs scored.  2281+127= 2408, and 260+11=271 games, for an average of 8.89.  And, like I said, add in the 1% for the shorter games, and you are at 9.00 runs per game.  And, unfortunately, no story!

Such is the power of random variation.  It’s taken down guys bigger than us.

Tom

I think that literally the changes in defense on the Rays team my account for another .1 runs per game!

Anyway, I really think that the difference between the leagues is random fluctuation (BTW, there is NO SUCH THING as something being “within the normal range of random fluctuation,” as Joe wrote, and many other people have written, because there IS no “normal range of random fluctuation,” unless you want to arbitrarily define “normal” as 2 sigma, 1 sigma, 2.5 sigma, 1.5 sigma - take your pick).

And I fully expect (not with nearly 100% certainty of course) the gap in the leagues to continue to shrink and for rpg to go up, at least in the AL.

Also, any theory about a change in personnel from last year does not hold water, as I showed in a previous post.  You can easily compare the pre-season projections for all of the players who have played thus far this year and compare it to the payers who played last year, using their actual production or those same pre-season projections, and the playing time of each player last year.

If there is a change in personnel from one year to the next such that production is significantly changed for the better or for the worse, it should be reflected in those numbers.

For example, let’s say that the personnel from the NL last year is exactly the same as this year, except Bonds is gone, and Brian Bocock takes up his PA.  Well, obviously if we apply any numbers (projections or what have you) to all the players and their playing time this year to those of last year, we will find out that this year should have lesser production than last year because Bocock and his -40 projection is replacing Bond and his +60 projection for X amount of PA. We can do the same thing for every player who played last year as compared to this year and figure out how much change in production there should be in each league from one year to the next, based solely on the players and their playing time, both this year and last.

Finally, I would be careful about discounting PEDS as the cause of anything!  For one thing, it is certainly possible that more PED use (or discontinued use, in this case) might have occurred in the AL than in the NL.  I don’t know way Joe says that that is ridiculous.  While it is unlikely that there is/was a great disparity in PED use between the two leagues, it is certainly possible.

And just to reiterate what Tango is saying:

In the BP article, they give the rpg as 8.76 for the AL and 9.19 for the NL, or a difference of .43 rpg between the two leagues.

As of today, on B-R.com, it is 9.00 in the AL, and 9.26 in the NL, for a difference of .26 rpg, quite a change in just a couple of days.

Of course, the numbers I quote are “per 9 innings (27 outs), and NOT per game.” Obviously the only thing we care about is per out and NOT per game.

We should even be eliminating the last innings just to avoid those situations where the home team wins in the 9th or later and does not complete the inning.

But, as Tango says, a .26 rpg difference rather than a .43, as well as an increase of .24 runs in the AL and .07 in the NL, sure takes a lot of steam away from both arguments (offense being down overall, and the difference between the two leagues being alarming).

There was a mistake in Joe’s original article.  The data I have shows the April ‘07 NL R/G as 8.85, not 9.31 as Joe first stated.  His follow-up article (linked in Tango’s #30) now runs through May 10th of each year and has more accurate figures:

Runs per Game through May 10
AL NL
2008:  8.74 2008:  9.19
2007:  9.34 2007:  8.73
2006: 10.00 2006:  9.46
2005:  9.16 2005:  9.16

#34, it ain’t “accurate” until he gives us the runs per out and not the rpg.  That is the principle beef.  As of May 13, that is 9.00 and 9.26 per 27 outs.  Obviously the rpg figures by Joe and you are technically accurate, but “per game” is irrelevant in terms of an analysis of offense.  We want per out.

Yeah, especially since everything else he did was per PA or per BIP, etc.  Why didn’t he do times on base per game, and total bases per game?  For the same obvious reasons he didn’t do that, then you can’t just do runs per game.  Just like a pitcher’s ERA is earned runs per 9 innings, not earned runs per game.

That said, Joe said he would pay attention to this next time.

Yeah, I didn’t mean to gang up on Joe.  He is a great guy and the follow-up article was well written and reasoned.  Normally rpg is a good proxy for rp 9 innings.  It is just that after a month or so, we are dealing with such small samples anyway that the difference between rpg and rp 9 innings can mean a lot when comparing the leagues.  It could happen that one league happened to have had an inordinate number of EI or rain shortened games, or whatever.  I think the most illuminating thing about the whole “offense is down” or “offense is down in one league” thing is that the standard deviation of runs scored per game is large enough to make these differences after one month not very unusual even if nothing special is going on.

Roughly what was said 05/14/08 3:42p.m. ET on ESPNEWS’ The Hot List--

Sheehan:...the DH has usually been worth 1/2 run to AL, which has been difference last 30 yrs. This yr AL hitting ball on ground more. When hit in the air, not getting as much productivity. Look at #s like Hr/Fb or XBH/FB and the combo of 3 things has driven AL run scoring down under 9 rpg, which would be lowest since 1992. This trend got to do w/ how teams are selecting their lineups. Seeing teams select a lot more defensive players, a lot more speed players than power hitters. This could be statistical aberration but compared to not complete seasons but each yr through May 10 the AL run scoring and other metrics stand out over last 5 yrs.

Host: Wow! You really do your homework.

Sheehan: You know, I got to mention William Burkes on our staff. He’s able to put all these #s together. He helped me out w. this article. Did a fantastic job.

Graphic (titled Runs per game last 5 seasons)
year nl al
08 4.62 *4.38
07 4.71 4.90
06 4.76 4.97
05 4.45 4.76
04 4.64 5.01
* fewest since 1992

So he is not correcting his mistake even though it has been pointed out to him?

Plus, as I already said, we can easily test out the “different players this year than last” theory.  It is not that.  If it were, then the combined lwts projections for all players thus far this year would be less than last year.  It is not.  IOW, the pool of player talent do far this year is around the same as the pool of player talent last year.  At least with respect to hitters.  As many people have mentioned, if the pitchers happen to be better this year in the AL than last, then the hitting in the AL will be down of course.  I don’t think I checked the pitching this year as opposed to last.  I doubt, though, that the overall pool of pitching talent can be so different from year to year that it is going to make that much of a difference in terms of rpg.

Anyway, I am going to take back what I said about it probably being just a random blip.  Well, maybe not take it back, but attach less confidence to the statement. As we have found in 1987 and 1968 and 1969 and 2002 and 2003, trying to figure out why offense fluctuates is not an easy task.  There are probably many, many factors, and we just don’t have a handle on many of them.  The so-called X factor.  I am willing to admit that the “true” run scoring in the AL may be down this year (IOW, it is likely to continue) and that we will never know the reason why.  So far, I have not seen or heard any plausible theories whatsoever.  The “younger players in the NL” theory was laughable. Saying that the trend in baseball is towards defensive oriented players, while possible, is speculative and somewhat tautological.

OK, let’s put some of these numbers in perspective, and lets get the correct numbersas well:

This year, as of May 13:

NL: 9.26 per 9 innings
AL: 9.00 per 9 innings

We cannot compare these numbers to last year at the same time because last year was a fluke April in both leagues.  I am guessing that it was especially cold because run scoring in April last year was .7 rp9 less in the NL and .34 rp9 less in the AL than for the whole year. 

That is typically not the case.  Contrary to popular belief, rp9 in April is typically (the last 8 years at least) a little higher than the rp9 for the whole year.

So it is best to compare April this year to the entire year last year.  And I am not even sure why we are comparing anything to last year.  One year worth of runs (last year) is subject to a decent amount of fluctuation (around .1 SD per game for each league) from randomness alone, not to mention the variance of weather, and change in batter and pitcher personnel.

Anyway, rp9 in the NL last year was 9.47.  This year, as I said, it is 9.26 as of May 13, a little down from last year.  One SD of difference between last year and this year through May 13 is around .25 rp9, so the difference is not even 1 SD.

In the AL, last year it was 9.90 rpg (a lot of that was the IL where the AL feasted on NL pitching, apparently).  Anyway, this year, it is 9.00, which is .90 less than last year, or almost 4 SD, which is a lot.  That does strongly suggest that something is fundamentally different this year than last in the AL.

Anyway, that is the correct data.

You should definitely do it without the IL data.  That’s like the AL facing AAA in one year, and not in the next.  Sure, something fundamental happened: they didn’t play the AAA team yet!

JC gives us a cool graph of HR rates and temperature.

Oops:
http://www.sabernomics.com/sabernomics/index.php/2008/05/april-homers/

David said to take a look:
http://www.fangraphs.com/blogs/index.php/its-back

And what do we have?  As of today, 7973 runs in 15447 IP, or a rate of 4.65 runs per 27 outs in the AL.  9218 / 17822 for 4.66 runs in the NL.

Yet another non-story made into a story back in May.

Depends on what the story is.  Run scoring is still down quite a bit in the AL, which is typiclly around .3 runs higher than the NL, at least in recent years, and was around 9.6 and change over the last 3 years I think.

Why that is we might be able to deduce once we pick at the numbers when the season is over.

And home field wp is still high I think but certainly not as high as it was in June or so.  Whether that has anything to do with “no more greenies” as had been suggested by some “insiders” probably remains to be seen, or at least whether the higher HFA continues remains to be seen.  If it does, we will probably never know why, just like to this day, we don’t know why run scoring changes significantly from one era to another.  Sometimes we know why, but other times we dont.