Breaking Down the Kansas Kids’ Gold Glove Snub

On October 18, Rawlings and MLB announced this year’s Gold Glove finalists. Conspicuously absent from the list were two electric young Royals: Bobby Witt Jr. and Maikel Garcia. The pair took to social media to voice their thoughts on the selections, with Garcia labelling the process “a joke,” team captain Salvador Perez backing him up, and Witt perhaps summarizing our collective thoughts most concisely with a simple thinking face emoji:

? https://t.co/WnoqeHe9qv

— Bobby Witt Jr (@BwittJr) October 18, 2023

What led to Witt and Garcia’s exclusions? Let’s review the Gold Glove criteria. The SABR Defensive Index, or SDI, is a proprietary blend of fielding metrics that comprises about 25% of the selection process, with the rest depending on the votes of the manager and six other coaches per team. These seven votes per team can only be allocated to qualified players within the same league as the team, but not players on the team. So, for example, Royals coaches can only vote for non-Royal American League qualified players.

The qualifications are as follows: Each infielder and outfielder must have played at least 698 innings of defense through their team’s 138th game, only qualifying at the position they played at the most in that time. Catchers must have played in at least 69 games by this point and pitchers must have thrown at least 138 innings to be considered for their respective categories.

Without access to the confidential vote tallies, let’s start by estimating how Witt and Garcia fared per the SDI. The SDI includes data from Statcast, Sports Info Solutions, and Stats Perform. In lieu of the latter, which is proprietary, I used Ultimate Zone Rating (UZR) alongside Statcast’s Runs Above Average (RAA) and Sports Info’s Defensive Runs Saved (DRS). I opted for RAA instead of the more prevalent Outs Above Average since RAA is on the same scale as UZR and DRS.

Additionally, while our advanced fielding leaderboards have customizable innings tallies, we don’t have custom date ranges, so I opted to compare Witt and Garcia to players at their respective positions who played at least 698 innings of defense on the season (as opposed to 698 innings through their team’s 138th game). Witt ended up with 14 competitors while Garcia ended up with 13. Here’s how they fared among them:

Why the huge discrepancy between RAA and the other metrics? Both UZR and DRS depend on Sports Info Solutions’ positioning data while RAA relies on Statcast’s data. At the same time, Kansas City isn’t known as the most analytically-inclined team, and they ranked 20th in shift-to-no-shift ratio this year (yes, there are still shifts). Thus, Witt and Garcia could be getting dinged more for the Royals’ poor positioning by DRS and UZR than by RAA.

I also found that, among 653 qualified fielding seasons since 2021, DRS has been more predictive of finalist log odds (pseudo R-squared — which is used for logistic regression — of .29; not counting utility-category finalists) than either UZR (.19) or RAA (.15). It’s surprising that UZR, which is not incorporated into the SDI, seemingly has more of an impact on finalist log odds than RAA, but in this case it may be more about the common data source shared by DRS and UZR. Either way, the three metrics together have a pseudo R-squared of .33, which shows that UZR and RAA do little to improve upon a prediction that only includes DRS. In terms of effect size, DRS has more than twice the impact on finalist log odds compared to UZR and RAA.

Further, the ability of the three metrics to explain just 33% of the variance in finalist log odds speaks to the importance of coach voting in the process. Perhaps the coaches weigh the metrics in the SDI a bit, given that the pseudo R-squared is greater than the purported 25% of the process that depends on the SDI. But for our purposes, 33% is still not ideal for a good prediction. What else can we add to the mix in lieu of vote totals?

Last year, I found that hitting performance surprisingly factored into Gold Glove decisions. This is likely due to the halo effect, a psychological bias in which observers extend an overall positive view of a person to each of their specific qualities. Juan Soto’s 2022 nomination struck me as an especially egregious example: Consistently one of the best hitters in the majors since his debut five years ago, Soto’s defense rapidly deteriorated last year, but another solid season with the stick and the generational trade haul the Nationals received for him helped maintain the overall positive impression voters had.

Adding in wRC+ ups the pseudo R-squared to .36. Below are the average marks for finalist and non-finalist qualifiers in each of the metrics I’ve included so far:

Here is how Witt compared to the AL shortstop finalists:

And how Garcia compared to the third base finalists:

The halo effect might not have applied to Garcia given his 84 wRC+, failing to salvage his poor performance in the eyes of DRS, while DRS darling Anthony Volpe parlayed the same offensive showing into a finalist nod. But what about Witt? A former second-overall pick, the shortstop had a breakout season offensively, bashing 30 homers at just 23 years old (he also swiped 49 bags, but factoring in stolen bases didn’t improve the prediction). If the halo effect is real, Witt should be an angel. Sure, DRS didn’t look fondly on his defensive work either, but Carlos Correa was hardly a deserving finalist by that measure. Is there another reason the Royals’ franchise cornerstone wasn’t a Gold Glove finalist?

I wondered if, aside from their poor infield positioning, playing for a lowly Kansas City squad tanked his fortunes. In fact, lowly is an understatement — the 2023 Royals became one of just 59 teams to lose more than 105 games through all of baseball history. Now, adding team winning percentage to the prediction didn’t change its accuracy. But over the past three seasons, finalists have been on teams with an average winning percentage of 52.7% while non-finalist qualifiers have been on teams who’ve won 50.5% of their contests.

Further, I found that among finalists, all of DRS, UZR, and RAA were negatively correlated with team winning percentage, but especially RAA:

In other words, the worse your defense was, the more your team’s performance helped your finalist candidacy.

I also toyed with adding some standard fielding metrics to the mix, such as fielding percentage and errors. Fielding percentage improved the prediction a bit, but its impact was likely muddied by the inclusion of DRS, RAA, and UZR. Still, it indicates that coaches aren’t relying on traditional metrics to make their decision, either. They may instead be depending on the eye test, and given that they’re only seeing a small sample of defense from the players they’re eligible to vote for, biases like the halo effect and its team-wide counterpart can creep in to fill the gaps. Defensive reputation also seems to be a component, which could explain the exclusion of the young Witt and Garcia and the inclusion of 2021 Gold Glove winner Correa despite his own down year with the glove.

However, even with the potential for voting bias, defensive metrics themselves are far from perfect — they can’t even agree on Witt and Garcia’s performance, for starters. And that’s what the pair’s exclusion from the finalist slate ultimately hinges on — the verdict is still out on their defense so far in their young careers. The pair of 23-year-olds have plenty of time left to bolster their defensive credentials, with the Royals poised to give them ample run on the left side of the infield for years to come. They’ll just have to hope that their long leashes don’t continue to coincide with dark times in Kansas City.