THE BOOK--Playing The Percentages In Baseball

There is a problem with the concept of the break-even point (even after factoring in leverage).  Let’s assume that 70% is the appropriate break-even point for a particular team (after factoring in the base-out state of all of that teams steal attempts).  The concept of break-even percentage would make equal a team that attempted only 10 steals in a year (getting caught 3 times) with a team that made 1000 attempts a year (getting caught 300).

From a WE standpoint, the +runs for each team would be 0, but there must be some effect (good or bad - from the perspective of the running team - I am not sure which) just from the attempt.  Just to note two effects, the running team will subject both its runners and the opposing pitcher (and to a lesser extent the opposing catcher) to additional fatigue.  That, in and of itself, has to be worth something.  What, and which way it pulls, I am not sure.

I bring this up because your suggestion is that it is best to have a higher SB%.  That might be correct.  It also might be correct that the game-effects of an attempt have positive value to the team making the attempts.  If that is the case, your time from P to 2B might be better used to maximize attempts (accepting an overall SB% near the break-even point) rather than maximizing SB% (which, with that information and very conservative play, you could probably get near 100% - with far fewer attempts).

Your suggestion, ultimately, is that the team should use this information to keep the poor runners (by which I assume you mean in this case to strickly limit to base-stealers, not base running, generally) at the break-even point while allowing the better base-stealers to approach 85%.  Since the decisions made will always be trading probability of success with rate of attempts, how do you conclude that the optimal strategy is an overall success rate of around 80% with weaker runners at the break-even point and better runners well above it?

Since the decisions made will always be trading probability of success with rate of attempts, how do you conclude that the optimal strategy is an overall success rate of around 80% with weaker runners at the break-even point and better runners well above it?

I agree that there are all kinds of issues that effect WE and steal attempts, some of which we know about and some of which we don’t.  In the McClouth thread (I think), we discussed the injury risk, long and short-term, associated with stealing bases.  Researchers, including myself and Tango, have looked at the issue of whether and how a base stealer on first affects the batter’s performance (via an effect on either the batter or pitcher).  Most of those researchers came up with little or no net effect, although I think the issue is a little to complex to know for sure.

In addition, not too many people have looked into the added benefit of a stolen base attempt when the batter puts the ball in play (more hits, and better base running advancement, despite the occasional DP). But, I think most of the focus is and should be on the base runner’s chance of success.

To answer the question I highlighted above, I don’t KNOW that it is around 80%, but I do know that it is and should be well above the overall break even point, just focusing on the win value of a success or failure.

Do it on paper this way:  Think of all the potential base stealers on the team, from those that can and will occasionally run, only against the easiest pitcher and catcher combos to the best base stealers, those who can and do run against all but the hardest catcher/pitcher combos and maybe everyone (if they can achieve at least 70% versus the best).  No one should run of course against any particular pitcher/catcher combo unless their chance of success is above the break-even point in that particular instant in the game.  We’ll simplify things and say that in all instances, the BE point is 70%.

We have, say, 3 sets of pitcher/catcher combos, again to simplify things, easy, medium, and hard.  25% are easy, 50% are medium, and 25% are hard.

We have 3 classes of potential base stealers: poor (but still able to steal - not the Molinas and Thomes), average, and very good or great.  We’ll say those are 25%, 50%, and 25% also.

Well call the catcher/pitcher groups CPI, II, and III, and the base runner groups BRI, II, and III.

We’ll also assume that each group has a fixed steal success rate.

---- CP I--CP-II--CP III
BR I--71%--N/A--N/A
BR II--80%--71%--N/A
BR III--90%--80%--71%

This is the success rate of each group against the other group.  As you can see, the worst base stealers can only run against the easiest to steal on catcher/pitchers. 

So what is the overall success rate for this “team” of baserunners?

Here is the same matrix with how often they face one another:

---- CP I--CP-II--CP III
BR I--.0625--.125--.0625
BR II--.125--.25--.125
BR III--.0625--.125--.0625

Since we have to eliminate those matchups that don’t exist, we have to revise the numbers so that everything adds to 1.

---- CP I--CP-II--CP III
BR I--.0909--N/A--N/A
BR II--.1818--.3636--N/A
BR III--.0909--.1818--.0909

If we multiply those frequencies by the success rates in each cell (the first table), we get:

.0909 * .71 +
.1818 * .80 +
.3636 * .71 +
.0909 * .90 +
.1818 * .80 +
.0909 * .71 =

.760

I think teams actually do it about right.  The problem with just using SBs and CSs is that we are capturing a fair number of blown hit and runs, in which the runner is caught but is not attempting a pure steal.  To find out if teams are stealing correctly, we need to exclude these (or include the benefits of putting runners in motion).  Ideally, we’d exclude all SBAs on which the batter swings.  Until someone does that, I think we can approximate it by looking at SBAs with two outs—there is little advantage to attempting a H&R with two outs.  Here is SB% the last two years, by # of outs:
0 73%
1 71%
2 78%
The success rate is 78% with two outs, with few H&Rs contaminating the data.  The volume of SBAs is high with two outs (comparable to 1 out, and more than 0 outs), so I don’t think this just reflects a pool of very fast runners.

Another way to look at this is by count.  With two strikes I’d expect many fewer H&Rs, and the success rate with two strikes is 77%.

Interestingly, success rate plunges on a full count to just 53%, as many slow runners must be sent.  Just exluding 3-and-2 SBAs raises the overall SB% rate by about 1 point.

One thing I’ve always wondered about is the SB “using up” baserunning opps. If J Reyes steals 2nd, his opportunity for contributing baserunning value is (essentially) limited to scoring from 2nd on a single.  But if he stays at 1st, he can try to go to 3rd on a 1B, or score on a 2b. Plus, if he is moved to 2B (by normal advancement), he can still try to score on a 1B.

So given that, shouldn’t a Reyes SB be worth less than an average SB? And shouldn’t his CS be more costly than an average CS? If so, the break even point is not some fixed value such as 70%, it varies with each player in such a way as to seemingly reduce the value of basestealing.

Maybe there is a flaw in my logic. Maybe the runs gained by Reyes on SBA dwarfs the loss in baserunning opps. But “inquiring minds want to know”.

Guy - As the article points out many teams have abandoned the traditional called steal play for a green light model where the runner picks the situation and pitch where attempting to steal makes the most sense for him.  Obviously, trying to separate green light steals where the batter is swinging at a hittable pitch from called hit and runs is next to impossible for any analyst without inside team information.

Another way to look at this is by count.  With two strikes I’d expect many fewer H&Rs, and the success rate with two strikes is 77%.

There are fewer hit balls where the runners were running on the pitch with 2 outs and less than a full count than less than a full count and with 1 or 0 outs, but not that many fewer.  1138 with 2 outs, 1480 with 1 out, and 1377 with 0 outs (2005-2008).

Interestingly, success rate plunges on a full count to just 53%, as many slow runners must be sent.  Just exluding 3-and-2 SBAs raises the overall SB% rate by about 1 point.

Guy - Check your math on this as I get a much higher success rate (63% overall) than you do.  Except for steals of third with no runner on first all of these plays must be on pitches thrown for strikes on which we would expect a 6% lower success rate.  The success rate is only drastically lower for the men on first and second base situation and these should probably almost all be considered blown hit and runs as the batter swings and misses 85% of the time.

Peter:  if you have data indicating whether batter swung, can you calculate the SB% for all plays on which batter did not swing? 

My results by count are from B-Ref league splits page.  Not sure why they would show different full count result.

Guy - Balls, strikes, and pitchouts? As I said in my previous post not swinging is not a good definition of a straight steal.

My results by count are from B-Ref league splits page.  Not sure why they would show different full count result.

I don’t use B-ref so I don’t know how the splits page is set up.  But you can’t query for SB and CS directly.  They are all listed under K or W. Querying under SB and CS gives only a handful of plays that occur on pickoffs.

But Peter, what is your proposed solution?  For years, almost all analysis of SBAs has simply ignored the fact that many CSs are really the result of busted hit-and-runs.  If we want to figure out whether teams are making reasonable decisions about stealing, we need either to 1) find some reasonable approximation of how often runners succeed on “true” SBAs, or 2) instead analyze all “runner going” plays as a group, and include the value of extra hits, bases gained, and net DPs avoided. 

To do #1 we don’t need a perfect measure.  We want to exclude most/all of the H&Rs—i.e. plays on which a manager would not send the runner if he knew the batter wouldn’t make contact.  (Not all of these result in CS, of course, but we can assume success rate is poor.) And it doesn’t matter if we inadvertently exclude some true SBAs—your green light scenario—as long as those omissions do not bias our sample of “true” steals. 

It seems to me that 2-strike and 2-out SBAs offer a reasonable approximation.  But if you think there are ways in which they provide a biased sample, I’d be interested in hearing that (really).  And clearly something is different about these SBAs.  If you don’t think that fewer H&Rs explains the difference, why do you think we observe such high success rates with 2 outs or 2 strikes?

* *

Dave S:  my guess is that while Reyes is more valuable than average on first, he’s also more valuable than average on 2B:  his odds of scoring on a single are higher; he’s more likely to advances to third, and his odds of then scoring on OF fly or IF out are above average.  So I’d guess it’s a wash.

I analyzed running on the pitch quite extensively a year ago using your method number 2.  I had found no good method of separating steals from hit and runs.  There may be fewer hit and runs on two strike counts or two outs, but as I stated in my earlier post the difference isn’t great enough for quantitative analysis in my opinion.  If you think about it, what we already know about running on hit balls and the lowered offensive effectiveness of getting behind in the count makes ordering the batter to take a pitch for a strike while the runner steals or to swing at a bad pitch during a hit and run a losing policy.  A batter should almost always be free to swing at a pitch that he thinks that he can hit well or take any pitch that he doesn’t think he can hit no matter what the runners are doing.  The one exception that comes to mind is swinging to spoil a reachable pitchout or other high fastball in certain favorable count situations. 

Because teams know when hit and runs have been called I am sure that they have analyzed their effectiveness as well as those steal attempts that are ordered by the manager or initiated by the runner who has a green light.  Unfortunately that information will never be available to amateur analysts.

One thing.  A runner never gets a “steal on this point” sign.  The most he’ll ever get is a steal with this batter up, if you can.  A manager never wants to force a runner to steal if he can’t get a good jump on a particular pitch.  So there are generally two types of steal signs.  The “don’t steal” and the “steal if you can with this batter up.” Some runners have the green light all the time, and they can go whenever they want unless the manager gives the no steal sign. This has always been the case.  I don’t think that that is changing.

I am not sure of all of this.  You’d have to ask a professional player, coach or manager.

Peter:  did you publish this analysis? If so, can you provide a link.

I think we’re talking past each other a bit, in that you’re focusing on what the batter is or is not required to do.  The distinction I would draw is whether the runner is being sent with the hope/expectation that the ball will be put in play (whether or not the hitter is literally commanded to swing), or as an attempted steal (whether the the manager has commanded the batter to take or is indifferent to batter swing/not-swing).  These are two different pools of runners.  And the evaluation needs to be different:  in the latter case, it’s a straight tradeoff between SB and CS; in the former case, where ball is often put in play, there are many other factors at play.

“A batter should almost always be free to swing at a pitch that he thinks that he can hit well or take any pitch that he doesn’t think he can hit no matter what the runners are doing.”

I agree on the first proposition, but am not sure about the second.  If a slow runner has been put in motion, it may be better for the hitter to try for contact.  Non-contact may mean a 50% chance of an out and lost runner.  Contact probably results in runner on 2nd and out (if GB), or runner on 1B and out (FB), with some chance at a hit even on tough pitch.  Not sure how that nets out, but doesn’t seem obvious that taking is always correct.

Peter:  Also, if you don’t believe there are significantly fewer hit and runs on two strike counts or two outs, do you have a theory about why the SB% is so much higher in those situations?

I don’t know about 2 outs, but almost all hit and runs are on 2-1 counts and a few other hitters’ counts (1-1, 2-0, and 3-1).  There are virtually NO hit and runs with 2 strikes - too easy to pitch out.

Guy - With 2 outs you don’t attempt a steal of third unless you are almost certain to make it.  The success rate of stealing third with 2 outs is 89.8% (344 out of 383) and you can bet the 39 runners that got thrown out at third with 2 out got reamed out when they got to the dugout.  Success rate of stealing 2nd with 2 outs is 74.1%; with less than 2 outs it is 69.5%.  There are more hit and runs with less than 2 outs but they are a small portion of the cases where the runners are in motion when the ball is hit.

There are virtually NO hit and runs with 2 strikes - too easy to pitch out.

MGL - Except for full count and less than 2 outs, of course, where the runners are put in motion quite often.