Initiating SEAM Methodology

This article introduces the SEAM( Synthetic Estimated Average Matchup) technique for describing batter-versus-pitcher matchups, both numerically and visually. We specify a Shiny app, available here, which you can use to follow along.

This app allows users to visualize synthetic scatter map disseminations for any batter-pitcher matchup that has or have had an opportunity to occurred in the past five years( which is when Statcast data exists ). Our app also reports act metrics that are calculated directly from the displayed synthetic scatter chart delivery. This includes the expected number of singles, redoubles, triples, and home runs, as well as the expected batting average on missiles in play( xBABIP) and the expected theories on contact( xBsCON ), which can be thought of as slugging percentage except the denominator is BIP+ HR instead of AB. These matchup-dependent metrics allow for any user to assess the expected operation of smashes and pitchers when they face each other.

The SEAM method thinks scatter graph dispensations in the form of heat planneds that are smoothed versions of conventional scatter planneds. We fabricate these by combining separate batter spraying graph distributions that are fabricated for each of the lurches that the pitcher hurls. The final combining is also weighted to the usage for each move.

One challenge to this approach is the sparsity of some batter-pitcher matchup data. We alleviate such concerns with the developed at synthetic smashes and pitchers with similar characteristics as the batter and pitcher under study. Our synthetic actor innovation approach is inspired by the notion of similarity scores like those motivating PECOTA and Bill James’s work. However, unlike the affinity orchestrates presented in the past, we create similarity ratings exerting a nearest neighbor approach that is based on the underlying batter and pitcher characteristics of the players under study instead of observed statistics.

For pitcher comparings, musicians are aggregated on a season and pitch-type basis. The variables considered are velocity, invent charge, horizontal shatter, horizontal freeing direction, horizontal freeing point, horizontal transgres, horizontal release direction, horizontal freeing spot, and extension. Medians of these variables are made across each pitcher-pitch type combination. For batter comparisons, participates are aggregated on a season, handedness, and pitch-type basis. The variables considered are depart velocity, open angle, pull %, midriff %, and oppo %. Norms of these variables are made across each batter-pitch type combination.

These player characteristics are obtained from Statcast and raked exercising functionality in the baseballr R package. These characteristics should wonder the underlying talent and bents of players.

For each batter-pitcher matchup we reckon three scatter graph densities. The first is the natural spraying graph density corresponding to the players under study, the second is the spray chart of the synthetic pitcher versus the original smash, and the third is the spray chart of the original pitcher versus the synthetic smash. We report a synthetic scatter planned density, which is a weighted average of these scatter plot concentrations. The loads are chosen with the purpose of minimizing estimation skepticism in accordance with the arrangements of mean squared wrongdoing. These values reflect the overall fit and sample size of these spray charts and they are computed internally.

Our app’s default matchup is between Mike Trout and Justin Verlander. The spray graph delivery and the stats for this matchup are exposed below.

View post on

The default Trout vs. synthetic Verlander spray distribution is displayed below. This visualization is complemented with the top 10 most similar pitchers to Justin Verlander who have faced Mike Trout over the past five years.

View post on

Here’s the default synthetic Trout vs. Verlander spray distribution. This one has 10 most same smashes to Mike Trout who have faced Justin Verlander over the past five years.

View post on

Details of the SEAM Method

The ended gory scientific items are included in our newspaper, which is available on our main page, but here’s a brief overview.

We will be allowed y designate the orientation of a batted projectile, x represent the obtained variables for the batters and pitchers under study, and tell f( y | x) be a probability density function corresponding to the batted clod deployment of possible locatings y provisional on batters and pitchers possessing characteristics encoded in x.

Our goal is to estimate f( y | x) in the presence of sparse matchup data.

We will think f( y | x) with g( y | x ), where g( y | x) makes the species 😛 TAGEND

g( y | x)= lm* gm( y | x)+ lsp* gsp( y | x)+ lsb* gsb( y | x)

In this equation, lm, lsp, and lsb are non-negative forces so that lm+ lsp+ lsb= 1, gm( y | x) is the estimated batted missile concentration corresponding to the batter and pitcher under study, gsp( y | x) is an estimated batted projectile density between the synthetic pitcher and the smash, and gsb( y | x) is an estimation between the synthetic smash and the pitcher. The weights( lm, lsp, lsb) are chosen to minimize the planned squared error for using g( y | x) as an estimator of f( y | x ). We now motivate construction sites of gsp( y | x) and gsb( y | x ).

Suppose that there are J accessible pitchers and K accessible batters that form the pool of actors that are comparable to the players under study, and be suggested that the pitcher under study sheds ntype peculiar slopes. For each t= 1,…, ntype, give pt be the proportion that the pitcher under study hurls each slope. For each j= 1,…, J, and t= 1,…, ntype, let wp, j, t be a non-negative load that shows the similarity of pitcher j’s slope t to that of the pitcher under study, where we require that wp, 1, t+ …+ wp, J, t= 1. With these specifications, the estimated density function for the synthetic pitcher is 😛 TAGEND

gsp( y | x)= [?] t [?] j wp, j, t gp, j, t( y | xp, j, t)

In this equation, xp, j, t are the pitch type t covariates for pitcher j. The density run gp, j, t( y | xp, j, t) is estimated nonparametrically.

A similar structure applies the estimated density function for the synthetic smash which makes the model 😛 TAGEND

gsb( y | x)= [?] t [?] k wb, k, t gb, k, t( y | xb, k, t)

Performance Metrics

For this system, we developed tale conduct metrics that are theoretically calculated as expectations with respect to the synthetic spraying planned density g( y | x ). As mentioned above, we estimate the expected number of singles, redoubles, triples, and home runs that the batter reaches versus the pitcher in a particular matchup. We too forecast the xBABIP and xBsCON as additional summary measurements. The metric xBsCON is best interpreted as slugging percentage provisional on projectiles put into play. Our implementation will not estimate these expectations exactly, as there is not enough historic batted ball data.

To theoretically estimate these parts, we firstly acquire five years of batted ball data. Next we guess the percentage of batted balls that were either an out( O ), single( 1B ), double( 2B ), triple( 3B ), or home run( HR) at locatings y on the baseball field. We signify this vector of approximated ratios at y as 😛 TAGEND

P( y)=( pO( y ), p1B( y ), p2B( y ), p3B( y ), pHR( y ))’

Next, we acquire E( x)= [?] P( y) g( y | x) dy, where 😛 TAGEND

E( x)=( eO( x ), e1B( x ), e2B( x ), e3B( x ), eHR( x ))’

Thus E( x) is the estimated expected vector of outcomes where the high expectations is taken with respect to the estimated spray chart distribution. Expected BABIP is then calculated as 😛 TAGEND

xBABIP= e1B( x)+ e2B( x)+ e3B( x)

And xBsCON is calculated as 😛 TAGEND

xBsCON= e1B( x)+ 2e2B( x)+ 3e3B( x)+ 4eHR( x)

Our app too displays the flooring of 100 e1B( x ), 100 e2B( x ), 100 e3B( x ), and 100 eHR( x ). The 100 multiplier is a normalization that is intended for ease of interpretation.

The previous clauses sketch how we would calculate our recital metrics if we could obtain P( y) for every y. In reality, we do not have enough data available to achieve this task. Therefore, we calculate discretized different versions of these action metrics over 10 -foot-by-1 0-foot bins.

Discussion Stages

The primary contribution of this work is the development of synthetic spraying map dispensations that are calculated under the hood of a Shiny app which provides users with visual and numeric summary measures of batter-pitcher matchups. Our application shows users batter-versus-pitcher partialities while providing summaries of their overall success( or scarcity thereof ), which we hope will be of interest to the broad baseball community. The SEAM method significantly improve upon the inferential dominance of scatter planneds as a visualization of a batter’s hitting propensities, as they may be uninformative for individual matchups due to a lack of data. Our synthetic participate creation assists alleviate this problem.

We are not the first to incorporate additional actors into an analysis via similarity composes with the understanding that doing so improves estimation performance. The PECOTA prediction methodology tries to forecast the ability of players utilizing aggregate appraisals obtained from other same players. To the best use of our insight, we are the firstly to base similarity compositions exclusively on Statcast data, which we conclude stipulates a truer thought of affinity in the context of individual batter-pitcher matchups.

On average, participates who hit the ball harder with a more optimized propel angle are able to obtain better projected stats, which obligates smell since these pellets tend to produce more home run( and thus make fielders out of the equation ). One should note that these variables will of course not allow us to measure the terminated knack chart of baseball players. Implements such as speed or eye at the plate would not be fully captured by our methodology.

Brief Digression on the History of Context-Free vs. Context-Rich

Baseball has had a rich statistical history dating back to the firstly carton score created by Henry Chadwick in 1859. Fans, writers, and crews have preoccupied over baseball statistics and achievement metrics ever since. This preoccupation is best summarized by the existence of Alan Schwarz’s best-selling book” The Numbers Game: Baseball’s Lifelong Fascination with Statistics ,” which is devoted solely to the statistical autobiography of baseball.

Most commonly used player evaluation metrics are functions of context-free box score totals and/ or context-free field tracking intelligence. These metrics include, and are far from limited to, adjusted earned run average( ERA +), adjusted on base plus slugging percentage( OPS +), weighted flows formed plus( wRC +), and triumphs above permutation( War ). This class of metrics is primarily boasted as context-free, means that contextual datum such as park and organization run environment upshots are taken into consideration by and situational intelligence is ignored.

The paper” Challenging Nostalgia and Performance Metrics in Baseball” showed that context-free metrics and the class of metrics that compares a player’s accomplishments immediately with that player’s peers are ill-equipped for musician comparings across ages of baseball, although they may perform well over the course of a single season or a few consecutive seasons. That being said, these context-free metrics do not offer any lead for how any specific batter will accomplish against a particular pitcher, the most important and relevant outcome in a baseball game.

Furthermore, baseball outcomes have been assumed to be independent and identically circulated( iid) actualizations in the statistics literature( Brown 2008 and Jensen, McShane, and Wyner 2009 ). The iid acceptance of outcomes may be reasonable in the prediction contexts of the previous hyperlinks, which involve long time frames, but iid is not appropriate for big time frames when the variability in excellence of batters and pitchers can be very large.

Commonly exercised context-free metrics are sanded in the iid hypothesi and are therefore essentially meaningless for studying batter-pitcher matchups since seasonal qualities are aggregated over widely varying matchups that separately have too small a sample to be reliable. Thus, this is the create in which the context-rich SEAM method shines.


This project is a collaborative effort between Charles Young, David Dalpiaz, and Daniel J. Eck at the University of Illinois Department of Statistics.

We would like to thank ATLAS Infrastructure and the Illini Analytics group for network hosting and computational resources. We would also like to thank Alan Nathan, Jim Albert, Shane Jensen, James Balamuta, John Marden, and Dave Zhao for helpful comments.var SERVER_DATA= Object.assign( SERVER_DATA || );

Read more: