Which Orioles Will Look Better at the Plate in 2015?

Batters rarely learn the strike zone suddenly and on the fly. Fangraphs has previously shown that O-Swing% correlates highly from one year to the next, something that I have confirmed and expanded upon. Nearly 65% of the variation in 2014 O-Swing rates across Major League Baseball can be explained solely by 2013 O-Swing rates. That jumps to nearly 70% when you include O-Swing history from 2010-2013 (using 2010-2013, if you’re keeping track at home, introduces multicollinearity, or a scenario in which the independent variables are themselves correlated¹).

As a result, we can fairly confidently assess a player’s ability to not swing at balls in the upcoming season – and to make contact when he does choose to swing.

To predict whether some Orioles – and some previous Orioles – would break out or fall back to earth in 2015, I’ve constructed a linear regression to predict leaguewide 2014 O-Swing% and compared it to each player’s observed 2014 O-Swing%. If the player outperformed their projected O-Swing%, which is actually exhibiting an O-Swing% below expected, he’s likely to swing more freely in 2015. On the other hand, if he uncharacteristically swung at balls in 2014 and his 2010-2013 history shows better restraint, we can expect him to be more patient and selective in 2015.

One thing to note moving forward is that I’ve chosen to impute rather than drop all missing data. That is, rather than omitting Jonathan Schoop from brief league-wide analysis because he didn’t record an O-Swing% in 2013, I’ve chosen to use his O-Swing rate from other seasons to estimate his 2013 O-Swing%. So, in Schoop’s case, since 2014 was his first season in the bigs, his O-Swing% from 2010-2013 is all the exact same as his 2014 O-Swing%. Without a model describing O-Swing% throughout the minors and as players age (and there might be, but I haven’t used it for this), this imputation is about as good as we can do. The 2014 rookies suffer (or benefit, if they were unusually good at taking balls) from this most; it’s easier to estimate what O-Swing rate players like Manny Machado would have recorded in 2010 using the average of their 2012-2014 seasons.

The issue of imputing comes up in the validity of my 2015 projections. In the case of players with limited Major League history – again, think Jonathan Schoop – their 5-year regression is based at least partially off of imputed, or calculated, numbers and not actual observations. Jonathan Schoop, as a rookie, only has one observed O-Swing% value, and his imputed 2010-2013 values are exactly equal to his 2014 O-Swing%. This is not particularly helpful in projecting his 2015 performance, since his previous years don’t actually exist but need to have some non-zero value to fit into the regression, and will cause the residual created by the difference between his expected and observed 2014 O-Swing%s to be pretty meaningless.

His youth and inexperience do mean something, though. With his rookie year under his belt and a little more experience, Schoop should exhibit slightly better plate discipline. As players get older, they generally also get better at recognizing the Major League strike zone, though I suspect that this has more to do with experience than actual age. As a rough proxy for experience, age works fine and shows that ball recognition should improve. Don’t expect too much from Schoop, though. He showed no reason to believe that he could restrain his swing in 2014, which doesn’t bode well for 2015, additional experience or not.

Here, again, imputation is helpful in the case of players like Steve Pearce, who recorded just 38 plate attempts in 2010. Since my cutoff for analysis was 40 PAs, Pearce registered as someone with no O-Swing% data for 2010, even though it does exist. Pearce showed impressive restraint in those 38 attempts, only swinging at 15% of balls, but that number isn’t useful for estimating his actual ability since it happened in such a small sample size. In my opinion, imputing what would have been his 2010 O-Swing% is more beneficial than using a relatively meaningless number.

Below I’ve shown the residuals of the observed 2014 O-Swing rates for each of the Orioles players meeting the plate attempt threshold, or the difference between the predicted values and observed values. If the residual is positive, the player should be expected to swing at balls less often in 2015 than they did in 2014 (which is shown on the x-axis). Conversely, if the residual is negative, expect less restraint moving forward. Also keep in mind that the residuals are in percentage points, so Wieter’s residual value of 2 projects him to lower his O-Swing% by 2 points, from 35.2% to about 33%.

The good: Matt Wieters‘ 2014 tear was in spite of a freer-swinging approach than he normally shows, and if he can continue to exhibit a power stroke and do a better job of recognizing balls, 2015 could be an even better year at the dish. Also primed to swing less often at bad pitches are De Aza and Machado, a promising sign for two players that the team will be putting a lot of stock into for the 2015 season. It’s worth noting that Machado’s residual is calculated using imputed data and a short sample size in 2014, so his might not be as meaningful. Youth, however, is on his side! Steve Pearce‘s breakout season wasn’t buoyed by an uncharacteristically discerning batting eye, which is promising moving forward. Expect to see additional patient at-bats from the Orioles’ most recent rags-to-riches story.

The bad: Chris Davis is expected to swing at bad pitches a lot more often, as are Flaherty, Lough and, somehow, Adam Jones and Delmon Young. Much has been made of Coolbaugh’s hiring as the team’s hitting coach, so maybe he can instill some patience in two of the Orioles’ most important players. This model doesn’t account for hitting coach, so you never know, right?

The departed: Nelson Cruz, a big part of the Orioles’ 2014 offense, actually swung at balls more often in 2014 than he was expected to. We’ll find out whether this is was function of age or MLB’s version of a heat check in 2015, but my suspicion is that it’s a bit of both. Cruz couldn’t miss for parts of 2014 and was willing to take some out-of-zone risks, some of which probably paid off. Markakis performed about to expectations, so no news there.

*  *  *

Like O-Swing%, Contact% tracks fairly nicely from year to year:

Again, we can determine whether we should expect Baltimore batters to make contact more or less often in 2015 than they did in 2014 by comparing what they actually did in 2014 to what we would have expected, based on their historic rate of contact from 2010-2014. In this case, a negative residual is indicative of a general expectation of an improved contact rate heading into 2015. And for what it’s worth, the coefficients of this Contact% regression were more in line with what I expected and what we see in the real world: the most influential year with respect to 2014 expectations is 2013, followed by 2012, and so on, until 2010 is the least influential year.

The good: JJ Hardy, who posted the second-lowest rate of contact in his career last season (his lowest was in 2009, when he played just 115 games for the Brewers), is the Orioles player with the most positive regression expected. Hopefully the less-than-ideal at bats Hardy gave the O’s at times last year were a product of nagging injuries than a late-career decline in ability. Also primed for a season of better contact is Chris Davis, who himself dealt with nagging injuries in 2014. De Aza and Machado are again bright spots and could have a better year of making contact in the upcoming season. Also good, in a way, is how many Orioles floated around their expected Contact%.

Again, Contact% improves slightly with age (or probably experience):

This is particularly good news for Machado, who seems likely to improve through both regression and age.

The bad: Only one: Matt Wieters. While he swung a little more than normal in 2014, he also made contact way more often than expected. Now, he did have a shortened season, so this might be a case of a small sample size causing us to talk about nothing. It’s worth keeping in mind that Wieters could have been having his own heat check before going down last year – if he makes contact with everything and that contact is good enough for a near-career high in BABIP, why not swing at everything? – and that we didn’t actually witness anything out of the ordinary in O-Swing% or Contact% for him.

The departed: Nick Markakis, having the most unassuming renaissance year known to baseball, apparently made a lot more contact than he usually does. Between a neck surgery and some regression to his normal Contact%, expect Markakis to be a little worse in a Braves uniform than he had been in black and orange for the last few years. Cruz made slightly better contact than expected, but nothing out of the ordinary based on the last four years of his career.

So in 2015, look for Machado and De Aza to improve a bit at the dish, Davis to strike out even more but make contact more often, Hardy to hit a little better, and maybe for Wieters to come back to earth a bit. The most promising part of this for me, aside from Machado and De Aza’s dual better restraint/better contact predictions along with their age-related improvements, is Steve Pearce hitting his expectations in 2014 and offering hopes for another solid year.

---

1. Multicollinearity doesn’t actually bias the results as long as the model is specified properly. Since it’s pretty clear that year-over-year O-Swing% is best expressed in a linear-relationship, I’m fine with ignoring multicollinearity in the model and not testing the individual variables for significance.