THE BOOK--Playing The Percentages In Baseball

Was this Andy Andres class at Tufts?  He has had a course on sabermetrics for a few years now.  I wish that they had that when I was in school.

Surely there are better ways to determine for whom the shift is or should be beneficial (to the defensive team) than regressions.  Using batted ball location data would seem to be the way to go.  I would also think that in order for one to do a comprehensive study, one would at least want to identify those players for whom a shift is actually implemented, by watching some video.  Making assumptions about a shift from the data is a pretty dangerous way to do it.  At the very least, make those assumptions and then check the video.

In the 2008 Goldmine, Dewan and James listed the top 25 or so players who were shifted.  That would be the perfect place to go.

Tango, I’m glad you liked the article.

Gary, this is not part of professor Andres’ course. There is unfortunately no course on sabermetrics being taught this semester at Tufts.

MGL, we did clearly have a problem quantifying defensive shifts. I’m not sure whether you had a problem with us using a regression to find expected groundball average or just to find why players get shifted. The first regression I feel is standard. The second one I agree isn’t too useful. The idea was to show that values other than batted ball location were being used to determine which players have been shifted i.e. how “feared” a hitter was by his homerun and intentional walk rates. As for identifying players against whom the shift had been implemented, we had evidence of some form that the players we mentioned had been shifted at some point. Of course, assuming the shift is always on or never on is imprecise. Watching video as you suggest would be a good way to go. If we could have gone play by play to determine which at bats they were and weren’t shifted, our analysis would obviously be better. Here’s hoping defensive player positions start getting tracked. Also, the study Greg Rybarczyk did in the 2008 THT Baseball Anual uses some incredible data that would make for a much more comprehensive study if it were available for all players, or if the data were even possible to be calculated by anyone other than himself.

Jeremy, I liked the article.  I’m not crazy about regression analysis as an end-all, but that is probably because I am not a statistician and regressions tend to be too opaque to me.

I’m not sure my comment at Baseball Analysts took, so I’ll post it here too:

Nice article, a few notes:

Your batting averages on different types of batted balls is wrong, unless you’re leaving out home runs.

Including home runs, a simple rule is that batters have higher fly ball batting averages on pulled fly balls and higher ground ball batting averages on ground balls hit the other way.  This is true for both righty and lefty batters, but the ground ball difference is more extreme for lefty batters.

My belief (no stats to back this up) is that pulled ground balls are hit harder, but defenses play batters to pull ground balls—so ground balls to the opposite field are more likely to find holes even if they’re not hit as strongly. As a proof that pulled ground balls are hit more strongly, there are more doubles on pulled ground balls for both lefty and righty batters, even with the lower BA on those hits.

Also, hitting line drives has nothing to do with power.  A line drive is a trajectory.  I would have used HR/F as a power measure.

Studes, thanks for the comment. I’m posting this in the Baseball Analysts comments too.

The batting averages on different types of balls in play was taken off baseball reference. That’s my mistake not indicating that it was BABIP to which I was referring, and not total average, so yes, homeruns were excluded. I’m not even sure why I chose to do that.

Why hitters fare better on opposite field grounders than pulled grounders is something we tried to look at. 12 players in our sample actually had higher pull averages on grounders than opposite field averages, and none of them were homerun hitters, while the players with the biggest split in favor of opposite field average were Thome, Howard, Bonds, Chipper, Delgado, Gross and Giambi. Defensive positioning definitely played a part in it. It seems that the most important thing one can do to get hits on pulled grounders is be fast and to get hits on opposite field grounders is to hit homeruns, and therefore draw the defense over. Of course, logically it should be that the more pulled grounders you hit, the better your opposite field average would be as the defense moves over, but that’s not what the data says.

I’m sorry, but I don’t quite understand the last part of your comment. By hitting for power, do you mean power of groundball hit? Because we tested to see if LD% and HR/F mattered in GB Avg. My guess was that line drives might be correlated with power of groundballs, since they might follow a similar swing path or something. In that regard, I was probably wrong. As for actual slugging power, I would not use LD% to account for that either.

We should be able to identify batters with a shift using Gameday hit location data. I’ve done a little work on that in the past:

http://otherfifteen.blogspot.com/2008/04/looking-for-defensive-shifts.html