Examining Two Years of Catch Probabilities

Earlier today, my first post about catch probability was published. In that post, I mostly looked at the two years of information, together. In here, I’d like to keep the two years separate. I mean, this is a whole new data source, with a number of potential applications. Why would I ever write just the one article and stop there? I have a weekly quota to hit, and this is better than whatever else there is to analyze.

Catch probability is a new Statcast metric, which you can read about here. As complicated as it might seem, it’s actually quite simple to understand, and it can give us better answers to questions people have been asking for decades. This is where you can find all the data, so you can poke around on your own. This is all brand new, and it’s kind of a first draft. The data will improve as adjustments are made for batted-ball direction and for outfield dimensions. Already, though, we can learn from what’s been provided.

Let’s start here by looking at the five distinct buckets for 2015 and 2016. So-called 5 Star plays are the toughest plays to make, while 1 Star plays are borderline routine. I should note that the most routine plays — and there are so, so many of them — appear to be excluded. Looking at the automatic outs doesn’t tell anybody anything. In this table, you can see year-to-year frequencies, and also year-to-year catch rates.

Catch Probability Distributions
Play Type 2015 Frequency 2016 Frequency 2015 Catch% 2016 Catch%
5 Star 23% 24% 9% 8%
4 Star 14% 13% 43% 40%
3 Star 15% 15% 69% 67%
2 Star 16% 16% 85% 82%
1 Star 32% 32% 93% 93%
SOURCE: Baseball Savant

Probably nothing in here to dwell on. In the last two years, the plays have been distributed almost exactly the same. The denominator is total plays in the five buckets, not all possible opportunities. So, of the bucketed batted balls, about a third are 1 Star plays, and about a quarter are 5 Star plays. Meanwhile, this last season, the catch rates experienced little dips. I can’t interpret that for you; I don’t know what it means. It’s just something to be aware of.

But I doubt that’s what people are interested in. The players themselves are what people are interested in. Moving on to that, using all the bucket averages, I calculated estimated +/- figures for all the outfielders. I then converted those into +/- figures per 150 opportunities. This is expressed as plays, not runs. There are 75 outfielders who had at least 50 opportunities in both 2015 and 2016. Here’s how their rates have been related year-to-year:

This is important! And a crucial check, before moving forward. It’s easy to observe a reasonably strong and linear relationship in the data, which is basically what you’d expect if the data were telling you something about player ability. You might hold correlations to higher standards than other people, but here, I think the story checks out. Even given the assumptions being made, even given the adjustments that haven’t been included yet, the data has achieved an R^2 of 0.62. One negative interpretation would be that the data is supplying consistent noise. That it’s somehow biased in consistent ways. A more positive interpretation would be that, even in this simple form, the data does a fair job of informing you about talent.

So let’s think about talent! I’ve got those 75 outfielders who’ve played fairly often in each of the last two years. Here are the outfielders who made the biggest improvements in +/- per 150 opportunities:

Catch Probability +/- Improvements
Player 2015 +/- per 150 2016 +/- per 150 Change
Steven Souza Jr. -10 5 15
Giancarlo Stanton -7 5 12
Melvin Upton Jr. -3 9 12
Avisail Garcia -6 5 11
Nick Markakis -14 -6 8
Christian Yelich 0 7 8
Angel Pagan -5 2 7
Jason Heyward 11 18 7
Marcell Ozuna -8 -1 7
Ender Inciarte 13 20 6
SOURCE: Baseball Savant

All the data comes with uncertain error bars. I’ve avoided decimal points in the presentation because I don’t want to convey a greater degree of precision than I should. I also don’t know exactly how a defensive player would get better at making catches, given that this measure takes positioning into account. But, way to go, Steven Souza Jr. The more familiar advanced numbers reflect the same general thing: By DRS and UZR, Souza was most recently an above-average defender. There’s the potential for a heck of a player in there if Souza ever becomes better at tapping into his strength.

You’ll notice that, within this top 10, there are three Marlins. I’ll add a fourth. Two years ago, Giancarlo Stanton, Marcell Ozuna, Christian Yelich, and Ichiro Suzuki made 241 plays, against 250 expected plays. This past year, the same players made 262 plays, against 250 expected plays. That would be an improvement of 21 plays, worth something in the vicinity of 20 or 25 runs. Now, maybe, this is randomness, or something else. Maybe something changed about how data is recorded in Marlins Park. But the Marlins also just had a new outfield coach, in Lorenzo Bundy. Could have something to do with Lorenzo Bundy. I can’t pretend to know more about this than I do.

Every sorted leaderboard has two ends:

Catch Probability +/- Declines
Player 2015 +/- per 150 2016 +/- per 150 Change
Gregor Blanco 2 -18 -20
Billy Burns 16 -3 -19
Mark Trumbo -25 -43 -18
Yoenis Cespedes 4 -10 -15
Joc Pederson 11 -2 -13
Lorenzo Cain 23 12 -11
Jacoby Ellsbury 16 4 -11
Norichika Aoki 4 -6 -10
Mike Trout 5 -4 -9
Starling Marte 6 -2 -9
SOURCE: Baseball Savant

In the earlier scatterplot, I’m sure you noticed the data point in the lower left-hand corner. It was sort of out there by itself. That’s Mark Trumbo, in scatterplot form. Trumbo, two years ago, looked like a bad outfield defender. Trumbo, last year, looked like a catastrophic outfield defender. I don’t think it’s news to anyone that Mark Trumbo, in the outfield, isn’t very good, but, remember how 1 Star plays are converted 93% of the time on average? This past year, Trumbo had 19 1 Star play opportunities, and he converted 11 of them. I don’t even know how that happens. Looks like a bad job by Mark Trumbo. The Orioles re-signed Pedro Alvarez, to play outfield.

Gregor Blanco shows up here with the largest drop, and it might not be coincidental that he was bothered by some knee discomfort. Lower-body injuries might also help explain, say, Lorenzo Cain making an appearance. As far as I can tell, Billy Burns stayed healthy, but DRS and UZR think he was worse in 2016, too. That’s bad for a player whose year-to-year wRC+ dropped from 103 to 52. Offensive performance and defensive performance can move around. For Burns, both moved very far in the wrong direction.

There’s still so much to explore. So many new avenues to go down. I want to caution, again, that this is new and therefore unproven. There are adjustments that will be made, and not every player’s opportunities are created the same. But we are most definitely entering a new era of defensive analysis, and we’ve been waiting for this for years. One can’t help but dig into what we have. Among players last year with at least 50 opportunities, Mark Trumbo was the worst defensive outfielder by far. The best, by a small margin, was rather surprisingly Desmond Jennings, followed less surprisingly by Billy Hamilton and Kevin Kiermaier. I can’t say yet how much we should make of that fact. I just love that we get to think about it.

Jeff made Lookout Landing a thing, but he does not still write there about the Mariners. He does write here, sometimes about the Mariners, but usually not.

newest oldest most voted
C Dial
C Dial

I posted on the other article, and run value of plays to CF (on average, not by Star position)
CF 0.842
LF 0.831
RF 0.843

5 star plays are 2Bs, with a few triples in the gaps and in the RF corner. the 4 stars are probably between a single and a double (about half?) and really, those have to be averaged off of the league average on the other 92% of those plays, or you introduce “back up fielder” influence