Perceived Highly Analytical Teams and Pitch Type Linear Weights

In July 2004, Jim Hickey was hired as the interim pitching coach by the Houston Astros. After a lengthy minor league run as a pitching coach, Hickey finally got his chance in the major leagues. After the 2004 season, Hickey had his interim tag removed to solidify his position with the Astros.

In his first full season in Houston, Hickey worked with stars Roger Clemens, Andy Pettitte and Roy Oswalt, among others. The Astros rode their pitching staff into the playoffs and made the World Series, only to be swept by the White Sox.

The following season was less successful for the Astros, as they finished just two games above the break even point and one game behind the NL Central champion and eventual World Series champion St. Louis Cardinals.

In the off season prior to the 2007 season, Hickey moved on from Houston, signing on with the Tampa Bay Devil Rays as their pitching coach. Tampa Bay was coming off a 101-loss season in 2006, with a pitching staff that on paper at least performed like you might expect for a team that hit the century mark in the loss column. The staff finished with the fourth worst ERA and fifth worst FIP in the league.

The pitching staff in Tampa Bay did not realize any significant improvement in Hickey's first season at the helm, falling to a league-worst ERA and seventh-worst FIP.

Heading into the 2008 season, Tampa Bay dropped the Devil from their name, seeking to re-invent themselves as the Tampa Bay Rays: winning organization. Having been released from the Devil, the Rays had an incredible season, making it all the way to the World Series before bowing out to the Phillies. The complete turn around in the performance of the pitching staff had a great deal to do with their success. The Rays jumped to the third best ERA in the league, while placing eleventh on the FIP leaderboard.

Since then, the Rays have managed to compete on a yearly basis in the tough AL East division, largely on the back of their pitching staff under the tutelage of Hickey. Their success as a staff culminated with a couple of impressive records in 2012. The Rays struck out more hitters as a staff than any other American League team in baseball history. The staff also allowed the fewest hits of any team in either league since the last Major League expansion occurred.

There are certainly countless factors contributing to the success that the Rays pitchers have enjoyed in recent years. One of these is health, as having players that can stay on the field means that you have to test your depth less and have the guys you actually want playing performing more frequently.

As a quick point of reference, consider the number of starters that the Rays have used per season compared to the number of starters used by their AL East brethren vying to knock off the powerful Yankees and Red Sox:

Year	Rays	Blue Jays	Orioles
2007	8	11	13
2008	9	8	13
2009	7	12	12
2010	7	11	9
2011	8	12	12
2012	8	12	12

To be fair, some of the additional pitchers used by the Jays and Orioles would be due to their being out of contention in September, and using call-ups for spot starts down the stretch. Certainly there is some health or lack thereof entering the picture here as well, favoring the Rays.

Based on my recent formation of a list of players who have undergone Tommy John surgery, the Rays also come out on top over this period. In this six year period, the Rays have had only one Major League player undergo Tommy John surgery: Jason Isringhausen (2009). The Blue Jays have had six: B.J. Ryan (2007), Shaun Marcum (2008), Jesse Litsch (2009), Kyle Drabek (2012), Luis Perez (2012), Drew Hutchison (2012). The Orioles have had five: Chris Ray (2007), Danys Baez (2007), Alfredo Simon (2009), Cla Meredith (2011), Randy Wolf (2012).

In looking at the types of pitches thrown by Hickey's pitching staffs, a very interesting trend materialized. Consider the following graph showing the percentage of each pitch type thrown by Hickey's pitchers compared to the league average.

To construct this graph, I first calculated the difference between the Rays' staff pitch use percentages and the league average percentages in 2007. This became the baseline, and then moving forward the differences between the Rays' staff pitch use percentages from the league average in a given year are relative to the baseline in 2007. This allows us to see how the Rays have been altering the frequency with which they are throwing various pitch types over time, while controlling for league wide changes over time.

Rays pitchers as a group have been moving away from the fastball and especially the slider in favor of the curveball and the changeup. In fact, according to Baseball Info Solutions (BIS) pitch categorizations, Hickey's pitching staffs have been gradually deprecating the use of the slider since his arrival as a pitching coach at the big league level with Houston in 2005.

Since 2007, Pitch F/X, a second set of data that classifies pitches, has been available. While the decline and difference from the league average are not as pronounced with the Pitch F/X data, the trend is still easily visible.

I would imagine that pitchers that are available to an organization and the types of pitches that they throw will play a large role in the decision of which pitches will be thrown on a year-to-year basis. Teams would likely not expect a pitcher to abandon a plus pitch once they've arrived in the big leagues unless they have serious reasons to believe it is necessary. While this would be rare, teams can obviously draft and sign players based on any criteria that they like, and then develop them in the way that they feel is best.

The Rays are thought of as one of, if not the most highly analytical organization in the league. Given this status, I wondered if this move away from slider utilization was a calculated decision based on evidence of its use heightening the potential for arm injury. At least once study published in 2002 found that slider usage, in youth pitchers albeit, led to an 86% increased risk of elbow soreness.

While league-wide slider usage has stayed fairly static over the six years that Hickey has been in Tampa Bay, I wondered if other teams thought of as being "saber-leaning" had also been switching away from the slider.

To categorize the teams, I found Bradley Woodrum's work on Fangraphs from earlier this year. He attempted to slot all teams into one of three categories based on the perceived use of advanced analytics in each organization.

With respect to the attempt to bin teams in this fashion, I would expect that every organization today employs intelligent people at least largely if not solely for the purpose of making sense of updated metrics. Despite this, I believe that there would be a relatively significant difference between when organizations started to adopt these new numbers and to which extent they are able to convert information learned from them into practice on the field. While I cannot speak for him, based on his comments about the article, I believe this was the basic belief of the author as well.

If we look at the average use of slider over the six year period for teams in each of the three categories, the correlation is striking.

Of the 14 organizations considered to be "highly analytical", 11 of them appear in the 14 teams that have used the slider the least over the past six seasons. Only the Indians, Padres and Cubs use the slider more often than average of these statistical-minded teams, with the Cubs dealing sliders at a rate that leaves others in the dust.

Of the 11 teams considered to be "old school", 10 of them appear in the 16 teams that have used the slider the most over the past six years. Of these teams, only the Royals, at 12th least, is clearly in the lower half of the league in slider usage.

I think everyone understands that this sort of classification of teams is a moving target, and does in fact deal with the perception of how analytic a team is rather than inside knowledge of its workings. It is remarkable, however, how teams that are perceived as being comfortable and interested in looking at new information in new ways have tended to use sliders less than the rest of the teams.

One question would be: have the teams that have used the slider less experienced less injuries from their pitching staffs?

As one quick test, we can look at Tommy John surgeries undergone by Major League players of each team over this six year period.

From these results, there does not appear to have been a benefit to throwing less sliders, as it pertains to avoiding the need for Tommy John surgery. As a reminder, these surgery counts include players who might have been playing in another organization for many seasons just prior to the year of the surgery, so they need to be taken with a grain of salt.

While looking at one type of arm surgery is hardly exhaustive with respect to pitching injuries, perhaps the seemingly lower use of sliders has not been done with injury prevention in mind. Another way of looking at this data would be to see how effectively these pitches were used by these teams.

To do so, we can use pitch type linear weights from the same BIS data source. These metrics represent the total number of runs saved (or allowed) on each pitch type. The numbers are normalized so that 0 is average, negative scores are below average and positive scores are above average. In this case, we must look at the standardized version, which scales the values on a "per 100 pitch" basis.

Theoretically, if a team is saving more runs per chance on a particular pitch type, it would make sense that they utilize that pitch more frequently, and vice versa.

I should note that there is a debate over the usefulness of the pitch type linear weights. Some of the issues with the metric are that it does not control for things like base/out states, nor does it capture pitch sequencing that can lead to subsequent pitches becoming more effective. While the data is not perfect, it is the best measure for these purposes that we have for which I'm aware, and by averaging the data for each team over six seasons I hope that these in game situational differences even out.

I separated the teams into "highly analytical" and "old school", based on the above grouping. I expanded my focus to include all pitch types again in this test. I then ran linear regressions on pitch frequency and the linear weighted runs saved/allowed per 100 chances for each group and pitch type independently. The R-squared results in the following table are astounding.

Group	FB	SL	CT	CB	CH
Highly Analytical	0.02	0.47	0.31	0.13	0.32
Old School	0.02	0.02	0.19	0.07	0.00

Fangraphs, BIS Data, 2007-2012

Note that for offspeed pitch types, the "highly analytical" group had fairly strong correlations between the success of the pitch and the amount that they used the pitch. In other words, teams that we perceive as being more likely to look at this type of information appear to be using it to tailor the frequency of the pitches that they throw. The relationship for slider, cutter and chanegup pitch types all were statistically significant at the 95% confidence level.

The "old school" group, on the other hand, shows almost no correlation between rate of success and rate of usage, other than perhaps in the case of the cut fastball. None of the pitch types had relationships that were statistically significant at the 95% confidence level.

From this information, it would appear that teams that are perceived as having been more highly analytical have been using pitch type linear weights, or something else more or less equivalent, to help them determine which pitches should be used a which rates. Teams perceived as being less astute at incorporating new statistics into their organization appear to be throwing pitches at rates completely unrelated to how successful the outcomes have been for each type of pitch.

As a cross-check, I ran the same test using Pitch F/X data, which provides both pitch frequencies by type as well as linear weights. I tried just the three pitch types that were statistically significant at the 95% confidence level for "highly analytical" teams based on the BIS data.

Group	SL	CT	CH
Highly Analytical	0.25	0.44	0.14
Old School	0.19	0.12	0.00

Fangraphs, Pitch F/X Data, 2007-2012

The relationships overall are not as strong for the "highly analytical" group of organizations using the Pitch F/X data. Only the cutter remainder significant at the 95% confidence level, although the p-value for the slider was just 0.07.

Once again, none of these pitch types showed statistically significant relationships for the "old school" grouping. Overall, the results are not nearly as telling as when looking at the BIS data, although the highly analytical group still looks to be more likely to use pitches based on the success that they are achieving.

From my understanding, Pitch F/X is an automated system that uses a variety of camera angles. It classifies pitches based on measured characteristics of the pitch, such as horizontal and vertical movement as well as release and end velocities. BIS relies on video scouts that watch pitches on television screens to classify pitches. Apparently each game is watched twice in an effort to ensure quality of data, and since Pitch F/X has been available, scouts can refer to it in cases where they are unsure.

Which one has been more accurate in pitch type classifications over the past six years? I cannot say, although it would appear the classification of the cutter may not have been as accurate in BIS data in the first two years of this period. Perhaps more importantly in the context of this line of study: which one is being used or trusted more significantly by Major League teams?

Unless someone convinces me that it is not worthwhile, in a follow up study I would like to remove the three admittedly-best-educated-guess analytical categories from the equation, and simply look at how closely each team has lined up pitch frequency and pitch success based on pitch type over time.

I would love to hear your feedback on this discovery. Do you believe that this demonstrates the adoption of pitch type linear weights into pitch selection decisions by teams that have been more advanced analytically? Is there a flaw in the test? Or is this all just a coincidence?

You can follow me on Twitter at @MLBPlayerAnalys. <a href="https://twitter.com/MLBPlayerAnalys" class="twitter-follow-button" data-show-count="false">Follow @MLBPlayerAnalys</a>
<script>!function(d,s,id){var js,fjs=d.getElementsByTagName(s)[0];if(!d.getElementById(id)){js=d.createElement(s);js.id=id;js.src="//platform.twitter.com/widgets.js";fjs.parentNode.insertBefore(js,fjs);}}(document,"script","twitter-wjs");</script>

Credit and thanks to Fangraphs for data upon which this analysis was based.