The Minnesota Twins and a New Way to Break Down Batted-Ball Data - Caretaker Exclusive

[Matthew Trueblood]

For the entire duration of Rocco Baldelli's tenure, the Minnesota Twins have focused on elevating the ball. Traditionally, the most available way to measure that emphasis is to break batted balls up into ground balls, line drives, and fly balls.

Let's break with tradition.

Image courtesy of © Jordan Johnson-USA TODAY Sports

Ask anyone who has worked at length with batted-ball data, detailed versions of which go back more than 20 years, and they'll roll their eyes and inveigh against the bane of every such analyst's existence: line-drive rates. What counts as a line drive? How far from home plate must it get before bouncing, in order to avoid being tabulated as a ground ball? How high can it get before it's functionally a fly ball?

For a long, long time, baseball people who wanted to get their arms around the value of various batted-ball profiles had to accept a deeply imperfect data set, stained not only with subjectivity, but with outright (and essentially unavoidable) bias. The stringers and video scouts tasked with classifying batted balls were not only forced to make a lot of tough calls (imagine if there were a half-dozen plays a game on which the official scorer could credibly call either a hit or an error), but also very likely to be swayed by the outcomes. If a ball dropped in front of an outfielder, it was always more likely to be counted as a liner than as a fly ball. If a ball that first bounced near the edge of the infield apron went through, it was frequently labeled a liner. If it was picked and the batter was retired, it was usually a ground ball.

Then, too, there was the simple problem that line drives are less frequent than either grounders or flies. The league's average breakdown fluctuates and evolves, of course, but it has tended to break down along the lines of 40-20-40. Everyone knows line drives are the best type of batted ball, but is it worth chasing them if they're only half as likely as either of the other types?

The advent of Statcast brought in the option to work around this, and both the league and the community of analysts have done that, to some extent. We now have the launch angle of every batted ball, throughout the league. By and large, though, that 40-20-40 distribution has remained. What if we did away with it?

View full article

[Rod Carews Birthday]

Interesting take on batted ball data.  The two big questions are whether the relatively positive results that it shows for the Twins ultimately correlate to better offensive production, and whether the players (and coaches) have the ability to further affect these numbers and improve upon them. 

It might be interesting to look at these numbers for some historically excellent (and historically terrible) offensive teams in order to see what those look like. 

[Karbo]

I still believe the best approach is to hit the ball hard. There will be some grounders, some liners, and some fly balls. Hitting a round ball with a rounded piece of wood is probably the hardest thing in sports.

[CCHOF5yearstoolate]

Nice article! I did a bit of a dive yesterday on what the Twins have been doing in the last couple of weeks to improve so drastically on offense, and by the stats it really seems like that answer is as simple as "make better contact." There's very little data to back up a change in hitting philosophy, which as you noted is a strategy that's been around baseball for a very long time - hit hard line drives. 

Since the Tigers series on 4/19 the biggest differences I see in batted balls are:

• +7.4 contact % (29th to 6th best)
• -3.0 whiff %  (29th to 17th)
• +2.8 hard hit % (23rd to 15th)
• -4.0 ground ball % (26th to 14th)
• +4.6 fly ball % (18th to 6th)

Approach-related changes over that timeframe include:

• +0.1 swing %
• -0.8 called strike %
• -4.4 pull % (OPS is up by 0.412 on pulled hits, +2.6 mph exit velocity & +8.3% hard hit)
• +2.9 straightaway % (OPS is up by 0.295, +1.6 mph & +6.6% hard hit)
• +1.5 oppo % (OPS is up by 0.170, but with softer hits by 2.7 mph & 12.5% hard hit -> mostly luck variance)

Additionally, they've actually been seeing more offspeed and fewer fastballs over this timeframe:

• +0.6 zone %
• -3.9 fastball % (4-seam, sinker, cutter)
• +1.5 slider %
• +2.2 changeup %

Interesting to see an attempt at equally sized bins for batted balls, and also that the upper bound of your line drive bin matches Statcast's definition on the nose. They go from 10-25 degrees, sweet spot is 8-32 degrees. For reference,

[ashbury]

I would like to see a lot less aggregation of the data.  Dividing into equal thirds is convenient for making tables, but may obscure where the real action is, or where the breakpoints between success and failure lie.  Maybe the ideal launch angle is near the center of that 2-25 degree "bucket", or maybe it's closer to one end than the other.  Conversely (somewhat), maybe the really good angles are more prevalent, because major league batters are, you know, good at their jobs, and more of the buckets would demonstrate that.

A table broken down into buckets of a single degree might be overkill, but I wouldn't assume that to be the case until I saw it.  True, there is a risk of "small sample size" if you have too many buckets, but with n=20000 or so I don't worry too much about that, and one can eyeball where the anomalies start to show up and decide that the right level of aggregation is 3 degrees or whatever, to smooth out the presentation of data.

Another thing is that when you have two metrics for success (OBP and SLG here), it's possible that the peaks for each occur at very different launch angles.  Combining into 3 buckets obscures this possibility.

Seeing the disaggregated data is a better guide for where the aggregation should take place, than simply equalizing all the buckets.  Said another way, you can always aggregate data if it's too finely divided, but you can't easily go the other way.