Using Contact Rates to Evaluate Pitchers

A little over a month ago, I published this piece detailing the methods that I had created to alternately assess hitter performance. I highly recommend glancing at that article before reading this one; it will make a whole lot more sense. For the lazy, here is a brief primer: I focused on using rates (contact, hard%, etc.) to create rough estimates of what would happen on any given pitch. What is the probability that Mike Trout hits a hard line drive on a pitch in the strike zone? The more a player does that, is he more likely to be a successful hitter overall? One of the advantages of this approach is that it helps to remove the actions of a hitter from his circumstance; a hard line drive is a hard line drive, but the placement of it will greatly affect whether or not the player reaches base. Poor defense, such as one may find in the minor leagues or college ball, is made less important in judging a player.

On of the questions remaining was whether or not I could apply some of these same methods to evaluating pitching. So far, the answer is a qualified yes. We already have a number of metrics to determine pitching value without regard for circumstance, but these methods still provide useful insights. Using the existing methods, such as xFIP, we can determine which rate stats are strong indicators of success.

There is one result that emerged above all else: there is no such thing as a weak-contact pitcher. There is a significant amount of talk about pitchers “keeping the ball in the park” or “getting weak ground balls.” However, this method indicates no such thing. By simply multiplying contact rates with “Soft%” for all 2015 qualified pitchers and therefore creating the “SoftXCont” statistic, I was able to search for any correlation between this rate and xFIP. Judge the results for yourself:

View post on imgur.com

Clearly, almost no correlation. However, remember that this only examines the aggregate; perhaps some specific pitchers can leverage this so-called skill to great effect. But, it appears that at least on average, generating weak contact is a poor indicator of overall pitching success.

The opposite is absolutely true. Pitchers who allowed less hard contact saw substantial increases in xFIP, as measured by my “HardXCont” number.

View post on imgur.com

The correlation is relatively strong, especially compared to the correlations seen in other baseball metrics. Clearly there is something going on here; pitchers who allow less hard contact per pitch get better results. Duh. For an even more clean-cut view of this, we can look at GoodXCont, which uses a combination of “Hard” and “Medium” contact.

View post on imgur.com

That correlation is excellent, and indicates that measuring GoodXCont would be a significantly powerful way of evaluating pitchers.

So, we see that pitchers who limit hard contact and good contact are more successful than their peers. We also see that allowing a large amount of soft contact is not indicative of overall success. The “weak contact” type pitchers (think Rick Porcello) are not necessarily succeeding thanks to any particular ability to generate soft contact; any corresponding ability comes more from being able to allow less hard contact.

For scouts, this means finding pitchers who both limit total contact and allow only poor contact. By using these metrics, rather than the outdated ERA or a radar gun, they can get a strong impression of future big-league success.

In a future piece, I plan to dive deeper into research on “soft contact” pitchers. While these initial results indicate that soft contact is not a good indicator of overall success, there is further work to be done. Stay tuned.





1 Comment
Oldest
Newest Most Voted
Inline Feedbacks
View all comments
evo34
8 years ago

It appears your goal is to evaluate true skill, which can then be used to project future performance. If that’s the case, you need to use past variables to predict future performance. I.e., check 2014 contact metrics vs. 2015 performance (ERA), or first-half vs. second half of same season. Raw correlations don’t mean much since it’s not very difficult to find two stats that correlate in-sample.

The other thing is to try to detect whether contact metrics are picking up anything that regular projection systems are not. E.g., take the top 25 most extreme contact metric guys from the previous season and see if they consistently out-perform or under-perform Steamer. (Obviously, you’ll have to look at several seasons of data). If so, you have something of value. If not, back to the drawing board.