The Gap Between Public and Private Information

This post was written by Adam Guttridge and David Ogren, the co-founders of NEIFI Analytics, an outfit which consults for Major League teams. Guttridge began his MLB career in 2005 as an intern with the Colorado Rockies, and most recently worked as Manager of Baseball of Research and Development for the Milwaukee Brewers until the summer of 2015, when he helped launch NEIFI. As part of their current project, they tweet from @NEIFIco, and maintain a blog at their site as well.

Analysts in the public space often assume a very deferential position. Surely, they may say, teams are doing similar work with far more information, using far more sophisticated tools, and know vastly more than those working in the public sphere.

We’d venture that the true size of that gap is far, far smaller than is often suspected. Injury information? Of course teams have far greater detail. But as regards primary questions like “who has pitched better?” or “how should one separate batted ball skill from variance?” — in terms of the salient data, there simply has not been a remarkable gap between what’s available to teams and what’s available to the public.

At least, perhaps until recently.

Third-party companies are supplying a wealth of data which previously didn’t exist. The most publicized forms of that have been Trackman and Statcast. The key phrase here is data, as opposed to supplying new analysis. Data is the manna from which new analysis may come, and new types or sources of data expand the curve under which we can operate. That’s a fundamentally good thing.

There’s a wave of companies providing something different than Statcast and Trackman. While Statcast and Trackman are generally providing data that’s a more granular form of information which we already have — i.e. more detailed accounts of hitting, fielding, or pitching — others are aiming to provide information in spaces it hasn’t yet been available. A startup named DeCervo is using brain-scan technology to map the relationship between cognition and athletic performance. Wearable-tech companies like Motus and Zepp aim to provide detailed, data-centric information in the form of bat speed, a pitcher’s arm path, and more. Biometric solutions like Kitman Labs are competing to capture and provide biometric data to teams as well.

The solutions which provide more granular data (Trackman, Statcast, and also ever-evolving developments from Baseball Info Solutions) are of perhaps unknown significance. They offer a massive volume of data, but it’s an open question as to whether it yet offers significant actionable information, whether it has value as a predictive/evaluative tool rather than merely a descriptive one.

PITCHF/x was supposed to revolutionize our understanding of pitching. It may or may not have, but it doesn’t appear to have meaningfully altered our evaluation of pitchers. The simple truth is, it’s probably much better to gauge the effectiveness of a pitcher by the performance of the batters against him than it is via the separation between his breaking ball and fastball. Not because those must be exclusive items, of course. Merely that if one knew absolutely nothing about the slider movement or fastball-changeup velocity disparity of any pitcher in MLB, it would have little or no effect on their ability to assess/project the effectiveness of those pitchers, which is the question most salient to an organization’s success, more so than using pitch profiles to optimize batter-pitcher matchups, or questions of similar scope.

That’s not to say, at all, that PITCHF/x brought no benefit. Just that probably the most tangible evaluative gains perhaps came from such basic elements as providing a reliably consistent backboard for all pitchers velocities. Otherwise, the gains from PITCHF/x have seem to have been largest with pitchers themselves (and their coaches), rather than within front offices. In the plot twist which most proves the point, by far the largest evaluative gains from vastly expanding the information we have about pitching came in regards to catchers and their framing abilities.

Which is all to say that by making information more granular, it’s entirely possible to gain 10,000% more data and only 1% more evaluative power. In much more detailed and eloquent terms, Russell Carleton described this with regard to Statcast.

Adding staff to manage this new data, just so that it may begin to be handled and analyzed, is the single largest driver of growth in front offices today. That’s with the hope, more than the expectation, that these expanded data sources will deliver substantial evaluative power.

It may well be a while before teams even really know. First, there are technological issues at times, as there were with PITCHF/x in the earlier years. Mostly, though, there are certain things that simply can’t or won’t be known until we have data covering a larger time period. At what rate do sliders tend to develop, compared to other pitches? Can hitters consistently generate power despite lower bat speed? Can outfielder routes be taught and learned, or is that ability fairly innate? Those questions won’t be effectively answered without multiple seasons, if not many seasons, worth of this information. Not to torture the comparison, but the wonderful revelations of catcher framing ability came five years after the advent of the PITCHF/x data.

There’s an important lesson hidden here. As teams and the sabermetric public are on the hunt for new insights, there’s a natural assumption to make: that the next answers lie within the information we can’t yet see. If only we knew the spin on the slider, we might understand the strikeouts. There are two issues with this approach. On one hand, the presumption that the new level of detail will contain those exact details which reveal further truth, for example, that significant elements which determine strikeouts are contained within the particular information Trackman is providing, and not within other areas not captured. On the other side of that coin is the simple and fundamental truth that the most valuable insights in sabermetrics have come not from new data sources, but by re-imagining elements of the performance record which already existed in sufficient detail.

Voros McCracken’s defense-independent pitching observations forever changed the way pitching is evaluated, on a remarkable scale. That was entirely an execution of novel theory, not more granular data. Win Probability Added and Leverage Index, which require nothing more than simple play-by-play data, have given us a new framework through which to understand the value of relievers. Just a reminder that massive scale isn’t required for progress. Work showing Johnny Cueto’s baserunner control in 2012 shone a light on a skill which is, for some pitchers, worth a handful of runs per year—seems small, but truly a huge difference. These revolutions, amongst the bulk of the sabermetric progress we’ve seen, have not come from big data and advanced math. They’ve come from challenging old assumptions, and most importantly, knowing the right questions to ask.

The point is, there are still plenty of discoveries yet to be unearthed in the information already available to us. The relationship between pitching and fielding in run prevention is nowhere near a settled science. Aging curves are nowhere near a settled science. Batted ball variance is nowhere near a settled science. Player projection is nowhere near a settled science. It would be as dangerous as it would be flatly incorrect to be guilty of the assumption that the answers to such issues and more rely upon adding a deeper level of detail to our dataset. These are questions of theory. These are questions of baseball, not of statistics or technology. These are the areas that determine an organization’s ability to evaluate players.

The increased detail of the existing information will present new opportunities. It’s an exciting time to be in sabermetrics for that reason among others. It is not, however, the only place, or even likely the primary place, from which further evaluative power will come. The size of the gains yet unearthed (or unearthed by only some parties, privately) in terms of baseball theory far outweigh the gains available from more granular data, by an enormous magnitude. Sabermetrics, in either the public or private space, would be imprudent to primarily rely upon further detail to provide further wisdom.

Dave is the Managing Editor of FanGraphs.

newest oldest most voted

DIPs can be considered a paradigm shift and the initial expression of a new research program. These are uncommon events. They reflect an attempt to solve puzzles which commonsense cannot manage. Tests and evidence follow in the wake of the new research program. Eventually, the research program generates new puzzles it cannot manage. A new research program may take its place.

I can’t even guess what would integrate the different kinds of research generating data today.

Perhaps the Pentagon would like to invest billions in this research!