Free Agency II: Forecasting Chase

What’s next for the Brewers’ surprising rotation leader Chase Anderson? Can the Brewers be expected to draw lessons from Anderson into further roster moves?

Throughout the summer, I found myself repeating an argument about the Brewers front office: if the Brewers front office implements scouting, mechanical, and coaching adjustments with a given player, they have some form of prediction or forecast that forms an expectation for the impact of that adjustment. Consider the two major pitching developments for the Brewers, namely Jimmy Nelson’s delivery shift, pitch addition, and subsequent breakout, and Chase Anderson’s arsenal adjustment: given that the organization has kept biomechanical data for quite some time, it is reasonable to suspect that player development decisions about adding pitches or redesigning mechanics at the MLB level are data-driven to some extent.

Namely, if the Brewers front office understands that Nelson will change his delivery timing and throw a curveball, or Chase Anderson will shift his cutter and curve, they have some idea of the range of success expected by such a move. Even this is quite a conservative statement; given the amount of time invested in developing these players (for example, Nelson threw at least 640 innings before using a curve in 2015), a development such as adding a pitch or mechanical overhaul will not be taken lightly. It was my contention that the Brewers understood the benefits of these moves, and expected significant improvement because of these moves.

The flipside of the argument, which becomes more speculative and therefore much more interesting, is that the Brewers front office should be able to form a solid idea about the relative success or failure of a pitching mechanics or arsenal shift for a player. I gather that they should be expected to do the same for a batting mechanics adjustment, as well. This statement is not quite as radical as it sounds; it is not an inversion of “can” into “should” (where, “the Brewers can design data-driven formula to track the success of pitching mechanics and arsenal adjustments” becomes “the Brewers should be able to forecast the impact of mechanical and arsenal adjustments”). Forecasting involves the use of data collection, statistical methodology, and some form of modeling (this can come from relatively simple and straightforward measurements of change to more complicated methods such as linear regression, or ever more complicated methods still) in order to look at a series of projections and craft a statement about its most likely path; think about this in the way that Nate Silver encourages “probabilistic thinking” (providing a range of predictions instead of one), or the way PECOTA offers “percentile” projections and thousands of tests of any given season.

Granted, there are potential data issues that the ballclub could encounter (data could be incomplete and face additional questions of quality or collection errors, or the club’s analysts could make suspect decisions about the underlying concepts for their forecasts, etc.). Yet, that there are potential concerns with MLB data collection and forecasting should not be viewed as a reason to dismiss discussions of clubs’ forecasting expectations for particular mechanical adjustments; in fact, I’d argue that this is the whole point of an “analytical” movement in MLB front offices.

Stated simply, the Brewers front office probably had a very good idea of what their key pitchers’ adjustments would be worth on the field. At the very least, they should have forecast (and probably did forecast) the likely impact of those adjustments. At worst, if the Brewers forecast those mechanical adjustments and completely missed the breakout seasons, they now have additional data to expand their models. Now, they can return to their layers of data, and any previous forecasts, and investigate where their models succeeded and where they failed.

What the Brewers now have in Chase Anderson is a prototype: fans and analysts can break this prototype into any number of characteristics worth testing.

• Chase Anderson is listed as 73 inches tall and 200 pounds according to Baseball Reference; he throws with his right hand;
• he worked 418 MLB innings before his breakout;
• his breakout occurred in his age 29 season;
• he does not throw a slider according to Brooks Baseball; and so on and so forth.
• Each of these characteristics can be used to build comparisons with other MLB pitchers, in order to test lessons about Anderson’s arsenal and mechanical adjustments.

The trouble here, in terms of statistical theory, is that within the MLB there will be no randomized sample of a population of pitchers who could perform like Chase Anderson. Constructing such a sample would require meeting extremely narrow characteristics that already limit the underlying population. For example, according to Baseball Reference Play Index, 163 MLB expansion era pitchers were 73 inches tall, less than or equal to 210 pounds, and right-handed while working 100 or more innings during age-29, age-30, or age-31 seasons (limit that to age-29 seasons, and the list drops to 60; or focus instead on right-handed or left-handed pitchers, and that list expands to 231 pitchers; and so on). Luckily, there are statistical tools available for controlling for variables in a model that does not involve a randomized sample of a population.

Following the criteria listed above, here’s an example of pitchers most comparable to Anderson’s class, focusing specifically on the last two seasons. This will be helpful to begin the next installment of this series, which be an analysis of particular player acquisition targets for the Brewers:

Beyond the technical issues, an intriguing theoretical issue exists. Since one is ostensibly testing Anderson’s prototype in order to find another pitcher with similar mechanics, arsenal, or characteristics that could conform to the successful lessons passed to Anderson, we already know what we’re looking for (our pitching survey is biased). This is acceptable for one very specific reason: baseball players are extremely scarce, and pitchers especially approach the game with a relatively narrow set of strategies (the vast majority throw a fastball) and mechanical approaches (ex., the vast majority throw overhand), and rulebound constraints (delivery timing requirements, approach to the batter, etc.). Moreover, a front office is not simply “looking for baseball players,” they are “looking for baseball players that could be good,” or “looking for baseball players that could improve,” etc. What the Brewers can use their mechanical, arsenal, and coaching lessons to construct is a system, a system into which particular player prototypes can be included in order to apply lessons and logic of that system to coax future success. This is very obviously a difficult thing to do, and one should not expect an MLB front office to be successful in every case. Yet, one should expect an MLB front office to reasonably apply their successes to as many future players as possible in order to maximize those successful lessons (or even to improve upon, to build from player development failures).

In this way, assessing a pitcher’s mechanical approach is much like underwriting a real estate development deal: an MLB front office will have to make decisions based on extremely limited, or flawed, samples. Player acquisition and development is more akin to working within conditions of market failure than a perfectly competitive marketplace. Continuing the analogy, an underwriter for a multifamily project will not compare that project to single family homes, nor will an underwriter of an affordable housing multifamily building compare that project to a market rate multifamily set-up, and so on: there is a particular ideological approach and set of market constraints for each type of deal listed above, just as there are specific constraints for each class of MLB pitcher that can be evaluated. There certainly may be some scenarios in which assessing those real estate deals blindly and with an overreaching view is valuable (such as a regional land value survey), just as there may be some scenarios in which blindly assessing all baseball players would be valuable (ex., “What percentage of draft signees from the 2000s reached the MLB?”). But these conditions will not necessarily adhere to each particular transaction (and each particular player development decision is likely made with much more focused, narrow forms of knowledge, and much more problematic forms of data at that).

The Chase Anderson system can be applied to the 2017-2018 free agency, trade, and waiver classes, in order to answer one of the Brewers’ most difficult questions of the offseason: how will the club repeat rotational successes of 2017? How will the front office fill the void of the injured Jimmy Nelson? How will the club build on the success of Chase Anderson? Looking at these questions produces one of the most fun aspects of this coming offseason, for these questions can be addressed in a very particular manner by an openly analytical front office. What is especially fun is that when the Brewers eschew big name free agents (perhaps by necessity of market size, perhaps by choice) in favor of someone less well known, or perhaps more puzzling, fans and analysts should immediately look to the club’s prototypes in order to consider which lessons (from successes or failures) are being applied to the roster.

Photo Credit: Jeff Curry, USAToday Sports