THE BOOK--Playing The Percentages In Baseball

We should be able to estimate the chance of scoring from 1st w/o a SBA, if we know how often SBAs are made and the success rate.  For example, say SBAs were made 30% of the time and succeed at 67%, and the chance of scoring on non-SBA is “X.” Then .283 = .1*.077 + .2*.406 + .7*X, so X = .277.  Hard to believe the real # is even as low as .270, much less .250-.260.

I read the article the other day, and while he is right that there are biases in the RE or WE data, since they already INCLUDE the SBA, thought that his “adjustments” were way off base and that his approximate BE points (55%?) were way too low.

Of course his point about not necessarily using the standard WE and RE tables are a good one, however, when it comes to figuring out the BE point for any particlular situation, you have to use “custom tables” anyway.  And rather than using AND standard or custom tables whatsoever, which may all have biases, why not just use a Markov model.

In any case, his assumption that since steals are successful at a much higher rate then the BE point, that the RE or WE with a runner on 1st and NO steal is going to be much lower is wrong I think.  For one thing that is a circular argument (steals are successful at around a 70% rate, and since he things the BE point is closer to 55%, it must create a bias in the tables) and for another, I think that the steal success rates are generally right around the BE point such that it WON’T significantly bias the tables, not to mention the fact that steals are attempted in such a small percetnage of the time, that any bias that there is is not going to show up very much in the tables.

For example, in Tango’s hypothetical above, it is probably like 10% rather than 30% that steals are attempted (I don’t know the exact number).  In any case, in order to see whether there is a bias in the runner on first base WE or RE numbers, simply compare what you get with a Markov and the runner not going to what you would get with a Markov and the runner going (and getting throw out 25-30% of the time, and as Tango says, weighting this by the number of times that a SBA occurs.  My guess is that the final number will be very close to the the number you get using a Markov and no SBA’s and very close to the empirical numbers we get that we put in the tables, which include both SBA and non-SBA.  IOW< I think that the author is WAY oversating this potential bias and hence way underestimating the BE point.  In any case, he should not be “guessing” at numbers when it is fairly easy to cite the real numbers and do the appropriate analysis.  His 55% (or 45%) is in fact a complete guess, which I think is way wrong to boot.

What I do think might affect the SB BE point, which hasn’t been looked into as far as I know, is the extra advanatge of the SBA for the runner being in motion when the ball is put into play, which is fewer GDP’s and more advances on a hit minus the occasional extra DP on a line drive or short fly ball or pop fly.

Good comments.  I actually did what Guy suggests, but drew some bad assumptions.  The right number is over .270 as he notes.  I used 1/7 for SB attempts (similar to the 10% MGL guessed), figuring about 420 situations/team (baseballmusings.com), and 60 SBA’s.

I “guessed” at numbers I didn’t have data for - such as the number of SBA, SB, CS with a runner on first base and 1 out.  In my past life working at STATS in the early 90’s, I would study run% and RE’s based on the actual pitch and event databases we had.  Even in the earliest days of tracking event and pitch data, STATS would record whether the runners was in motion, so studies such as suggested by MGL could be done.  Believe me, I don’t like guessing.

Any thoughts on why the win% shows a BE of 45%?  I didn’t guess at that, rather taking the data from walkoffbalk.com.

Rob