THE BOOK--Playing The Percentages In Baseball

My takeaways from this piece, and please tell me if I’m wrong:

- on the right-hand image in your post, it shows that Henderson tried steals as a pretty direct function of how high his breakeven rate had to be.  It looks like Raines didn’t even bother in certain situations when it would have been a really good deal (stealing second with two outs?  stealing third with one out?) but as the situations required higher success rates, he became more picky.

- the left-hand image in your post shows that Rickey was pretty effective at piling up steals without hurting his team.  Presumably, he would have been perfect at such if his blue line was precisely overlaid on the dotted line.

- also looks like Bonds didn’t really pay much attention to game state on his steals (last image in the article)

I love the graphs. It’s like a snapshot of a player’s psyche.

Bonds attempts to steal is directly proportional to how much his team needed it, while Juan P shies away from the big steal moments. Beltran is just a wuss. Ricky just stole the time (obviously), and Raines cut his attempts back when he didn’t really need to steal. And Rose… well, most of his attempted were in the highest breakeven situations… which I’m sure didn’t help his team one bit. Very cool.

Another chart I would like to have seen is this:

x-axis: leverage index
y-axis: success rate MINUS breakeven rate

x-axis: leverage index
y-axis: attempt rate

When the breakeven rate is low, this does not mean the LI is low.  It can be very high in fact.

In order to know how much his stealing mattered, we need to compare to the leverage index (or, as Jeremy put it in his article, the “swing” between a SB and a CS, which is, basically, the same thing).

We see in his chart that of all the top basestealers, Raines had the highest swing rate, meaning that he did indeed steal more when the payoff (and cost) was greater.  Willie Wilson stole when it mattered much less.

Is the breakeven rate adjusted for the actual batters behind the runners?  I’m assuming it wasn’t.  With that assumption in place I’m not sure I fully agree with the Gardner conclusion as strongly as stated in the article.  Presumably, Gardner has better batters batting when he is on base than league average and would need a higher than league average break even rate to justify it for his particular situation.

If Garnder is on base with the meat of the Yankees lineup batting in Yankee Stadium, he needs a different break even rate than Beltran on base in Citi Field with the bottom of the Mets lineup batting.  The study is good still, I just came to a slightly different conclusion about Gardner.

I hastily assumed that the breakeven rate was directly correlating with the LI, but that’s wrong. I need to take a longer look at what exactly he did…

Rickey Henderson just stole. Period.

For better or worse, that’s Rickey.

I’m a huge Rickey fan, and the 89 ALCS was just a thing of beauty.

Doesn’t the story go something like this? “Rickey, I gave you the take sign twice.” “Yeah, I took second and I took third.”

---------------------------------

Question about Raines and stealing more in situations where it was more valuable.

I wonder how much of raines stealing in higher leverage situations is him or his coach calling for it?

Has Raines ever commented on it? Did he intentionally steal more often when his team needed it the most? Were those situations obvious?

I’m asking because it seems incredibly unlike an elite ability player to show restraint in situations where he could “pile on numbers” at a greater rate but did not because the reward wasn’t great enough for the team.

Based on the reporting at the time, Raines had the green light while he was in Montreal.  He stole when he decided to.

Raines showed restraint in stealing in low-LI situations because he cared more about his body than piling on numbers.

Mike,

“On the right-hand image in your post, it shows that Henderson tried steals as a pretty direct function of how high his breakeven rate had to be.”

I agree.

“It looks like Raines didn’t even bother in certain situations when it would have been a really good deal (stealing second with two outs?  stealing third with one out?) but as the situations required higher success rates, he became more picky.”

I didn’t get that. I think the left tail might be noisy.

Tango, I’ll try LI by Attempt rate when I get the time. I don’t follow your suggestion of LI by Success-Breakeven. Is that clutchiness?

Dmanloue, it’s true Gardner needs a higher success rate because he plays in a much higher run environment. I’d still rather see his SB% at 80% than 85%.

I find that your chart showing the success rate of Raines and Rickey to be directly linked to the breakeven point to be fascinating.

The lower the breakeven point, the lower the success rate.  Presumably, that’s because they are now running on pitchers/catchers that they normally would not have run on.

A football analogy would be the success rate on short passes being much higher than the success rate on long passes.

What we care about however is not the success rate, but the success rate above the breakeven point.

So, completing 50% of your long passes is far more impressive than completing 55% of your short passes.  Indeed, completing 55% of your short passes is a net negative.

So by Success-Breakeven you mean WPA? And charting that against LI will show how clutch a basestealer one is, right?

It wouldn’t be exactly WPA.  The scale of success and of breakeven is 0 to 1. 

If for example Raines at LI > 2 is this:
SB (breakeven = .60)
SB (breakeven = .59)
SB (breakeven = .61)
CS (breakeven = .60)

Then we have a success rate of .75, and a breakeven rate (at LI > 2) of .60, and he’s +.15 above breakeven.

We’d end up with this:
LI success minus breakeven
-- ----------------
>2 +.15 (frequency = 15%)
1.5-1.9 +.14 (frequency = 12%)
1.0-1.4 +.16 (frequency = 9%)
0.5-0.9 +.13 (frequency = 5%)
0.0-0.4 +.17 (frequency = 3%)

We see here that his stealing impact was pretty constant across leverages, but his frequency of running was far higher the higher the leverage.

(All numbers for illustration purposes only.)

The problem here, as with almost all basestealing analyses, is that caught stealings that come when the hitter swung and missed (basically, a busted hit and run) are counted, but the extra bases gained by the baserunner (and GDPs avoided) when the batter makes contact are not counted.  So all of these guys have considerably more value added than is indicated by this analysis.  Someone like Brock may well have run too often—I don’t know—but you can’t answer that question unless you know how many extra bases and avoided DPs were produced by his aggressiveness. 

It would be interesting to add count to the WPA calculations, if the data exists to do that.  A 2-out CS when the hitter is 0-2 is pretty different than when hitter is 3-0.

Here’s my attempt at attempt rate by LI.

Guy, let me know if you have ideas on how to include count or runner going data.

Jeremy, what is “swing rate” and where is LI in that chart?

The average positive gain for a SB minus the average negative gain for a CS is roughly .07 wins.  That’s the “swing rate” that Jeremy is calculating.  You see that in his linked article for each player as around 7% for each player. 

If you divide each player’s swing rate by 7%, you get his LI.  It would be much clearer if Jeremy were to do that, since we can all relate to LI.

Anyway, fantastic stuff by Jeremy!  We see that as the LI increases, Raines steals more, while Rickey steals less.  Somewhere at around LI 2.5 or so, Raines starts to lower his attempt rate.

The data has obviously been smoothed, so it’s not clear how good each fit was.

In any case, great stuff…