THE BOOK--Playing The Percentages In Baseball

The OBP for innings 2 through 8: .343 bottom half, .331 top half.  So, the home advantage is 12 points in OBP.

The OBP for the 9th inning only, tied game, 0 outs, bases empty: .350 for the bottom half and .322 for the top half.  That difference is 28 points, of which 12 is the home advantage, leaving us 16 points unexplained otherwise.

This is based on about 4000 PA for each half.  One SD is 7.5 points, which for the 16 point difference above means 2.1 SD.  There is some sort of 9th inning effect, and this has nothing to do with runners even being on base.  It COULD be for pitcher personnel (I didn’t look).

In extra innings: bottom half OBP is .356 and .340 in top half.  That 16 point difference, which after you account for the 12 point home advantage is meaningless.

I then looked by score, for top and bottom half.  (Doing this as we speak, so I don’t even know the results.)

First the top half, and again limiting it to no runners on, no outs.  I set the run difference as down by 4 (or worse), -3,-2,-1,0,+1,+2,+3, up by 4 (or more), based on the home team.

Each grouping has around 3000-4000 PA.

-4 0.336
-3 0.339
-2 0.319
-1 0.310

0 0.322

1 0.307
2 0.307
3 0.294
4 0.319

So we see a nice smooth transition except for one hiccup.  We expect that slope because of the talent of the reliever.  When the home team is ahead by 1 to 3 runs, then we expect the better reliever.

But look at how the tied scenario sticks out from that progression.

One SD is about 8 points in wOBA.  If we extrapolate the line, we’d see that tied should be around .309 or so, which means that there’s a 13 point difference, or a bit over 1.5 SD.  Nothing really to get excited about, but it’s something to look more into.

This is for the bottom half, where the sample size is roughly the same:

-4 0.322
-3 0.318
-2 0.319
-1 0.316
0 0.350

In this case, the visiting team is pitching, and they are either ahead by a little (1 to 3 runs, and so we have the ace), ahead by alot (4 runs or more and so we can have anyone there), or tied (in which case, well, something in-between).

Whereas the down by 4 or worse has close to the same wOBA for down by 1, 2, 3, when tied, we see a huge difference.

That’s a 30+ point difference, which compared to the 8 points for one SD gives us 4 standard deviations.  Something is going on here.

In my next post, I’ll see what is going on!

Continuing with the same scenarios (meaning tied game, no runners on, no outs), I’m going to look at the bottom of the 9th.

I will compare down by 4 runs or more (where we expect to see any kind of reliever, including some ace reliever in tune up situations) with a tied game (where we expect to see some decent reliever, but not necessarily the ace).

When down by 4 or more, the K rate is 18.9%.  When tied, the K rate is 19.4%.

Here’s where it happens: the walk+hitbatter rate is 7.5% when down by 4 or more, but jumps to a huge 10.5% when it’s a tied game.  That’s a significant jump.  That’s +.030 more walks.  That is, 30 points in OBP!  This explains what we are seeing.

The HR rate is the same, whether down by 4 or more, or tied.

The non-HR hits (1B+2B+3B) is the same.  HOWEVER:
The 1B v 2B+3B rate also changes substantially.  Whereas 25% of the nonHR hits are extrabase hits, it drops down to 19% when the game is tied.

So, defense is protecting the line in a tied game.

But the biggest impact is that walks jump like crazy.

This walk situation in a tied game does NOT happen in the top of the 9th.  The changes there are a bit more subtle.

Next, I will look at the bottom of the 9th, tied games, for each base-out state.  There’s 24 of them!  I don’t think I want to do 24 posts, so before I go ahead and do that, I’ll wait for some comments on the above research first.

How can protecting the line result in an equivalent hit rate but reduced 2B+3B?  If the defense could do that, shouldn’t they do it all the time? 

The only theory I can think of is that there are actually better fielders in the game at this point.  But why would you have your defensive replacements in for a tie game?

Kinda lost in Tango/3, because all those walks removes balls in play, is the BABIP.

So, here is:
OBP
BABIP
wBABIP

wBABIP is weighted BABIP where I give 0.9 units for non-extrabase hits and 1.3 units for an extrabase hit.

This is for bottom of 9th, bases empty, 0 outs.

So, the last line shows the following:
1. OBP is huge at .350, as we discussed, with all almost all of the gain coming from OBP.

2. BABIP is also higher when the score is tied.

3. weighted BABIP is lower when the score is tied or home team is down by 1.  This means that there is a smaller proportion of extra base hits.

These last two points are saying this: the defense is willing to concede more singles, as long as they allow fewer extra base hits (i.e., guarding the line).

The net payoff seems to work well enough when down by 1, but there’s still something over and above that when the score is tied.

So, something is happening with the score tied, in the bottom of the 9th.

Would love to see the PITCHf/x-ers give us that breakdown.

I repeated the above, but this time with 1 out (bases empty, bottom of 9th):

Once again, a huge jump in OBP.  While a big part of that is the walk, there’s also a huge jump in BABIP as well.

***

And with 2 outs:

Same type of pattern.

This is runner on 1B, 2 outs, bottom 9th:

Once more, OBP is high when score is tied.  Walks jump, singles jump.  Comparing the down by 4 or more, to the tied scenario, walks and singles jump account for 26 of the 30 point difference.

Wow, fascinating stuff!  Who said that sabermetrics is dead!  I’ll have to look at this stuff later…

I’d like to group the base-out states, so that I can increase the sample size, and not have to run 24 reports.

I’m thinking these groupings:
A. Bases Empty, any outs
B. Any runner on base, 2 outs
C. Runner on 1B; 2B; 1B+2B, 0 outs (i.e, bunt sit)
D. Runners on 3B, less than 2 outs
E. Runner on 1B; 2B; 1B+2B, 1 out

Looks like this:

Does that look like a reasonable grouping?

Ok, I used the above groupings.  Starting off with bottom of the 9th only, here we go:

A_Empty (meaning all bases empty, for 0,1,2 outs):

There’s around 8000 PA, so one SD is 5 points in OBP or BABIP.

Understanding that when the home team is down by 1 to 3 they are facing a tougher pitcher, the comparison points are tied and down by at least 4.

28 points gap in OBP is over 5 standard deviations. 23 points in BABIP is almost 5 SD.

Just a huge huge effect here.

***

For all the other scenarios, I’m removing IBB.

B_2outs:

Much tighter.  Around 3000 PA, so one SD is 9 points.

I’m thinking now that the tied scenario with bases empty is filled with “unintentional intentional” walks.

D_SFsit

In this scenario, the outfield is playing in, and they are giving up the long hit.  It simply is a completely incomparable scenario, and should be treated completely separately from all scenarios.

E_rest:

Similar to the bases empty scenario.

***

Now, I didn’t do the C_BuntSit scenario.  50% of the PA when tied is a bunt.  So, even if I split out the bunts from non-bunts, the non-bunts are going to have the advantage that the defense is playing in.

This is what it looks like if I remove all bunts for this category (i.e., runners on 1b, 2b, 1b/2b, 0 outs):

The interesting comparison point is down by 1 and down by 2.  Down by 1 has tons of bunts (36% of PA).  Down by 2 does not.  So we see a big gap there in OBP, notably BABIP.  Basically, defense is expecting bunt, they play in, and it’s easier to get a hit if you swing away.

There’s just so much going on here.

***

Anyway, if you are going to look at the 9th inning, tied, bottom, then my suggestion is to remove all SF situations (runner on 3b, less than 2 outs), and all bunt situations (runners on 1b and/or 2B, 0 outs).

You’ll get a much clearer picture.

***

Thanks to Guy for bringing up the idea about playing differently in tied scenarios in the 9th inning.

Another in the series of fascinating things that any baseball sim programmer should heed. I’m quite curious to see the differences in events in score/inning situations, and then use that to come up with a better such table for base/out situations, since there has to be some correlations between the two.

I’m just waiting for the PITCHf/x-ers on this one.  It could be that the major difference is the unintentional-intentional walk.

And the rest of the difference explained by trading extrabase hits for singles (focus on wBABIP for that; it’s up and down, which tells me that there may be alot of different variations of moving fielders around here).

It’s possible that if you break it up by whether a hitter is a power hitter or not, then that might explain the whole thing.

That is, if you have yourself a singles hitter, then the pitcher is going to pitch to him normally.  But with a power hitter, he’s going to walk him alot more, since the defense has already established that they’re going to guard the line to begin with.

Just some fascinating stuff frankly.

Yes definitely fascinating. I’m not sure however of the practical significance other than how we started - which was to see if depressed wOBA by starters in the 9th was an illusion which it turns out it was.

Right, I don’t know if there’s any practical significance.  Well, there might be, but we’ll only learn from it via PITCHf/x: how do pitchers and batters behave tied in the bottom of the 9th.

It’s possible we get a ton of IBB/NIBB and a fairly large shift in fielder positioning.

The only real significance I can see is to better dial in the optimal strategy for hitters, pitchers, and fielders, although it is so complex and so batter specific I’m not sure we can do that…

Followup here:

http://www.insidethebook.com/ee/index.php/site/article/performance_in_tied_games/