Cold Weather and Player Attributes

How does the weather impact players, such as Jorge Soler? (via MBD Chicago)

Yesterday, I examined how cold temperatures influence baseball’s penalties and players at each position. Today, I’ll aim to identify whether different player attributes also help to explain cold-weather performance. This article uses the same set-up as the previous piece to evaluate three subsets: region of origin, age and weight.

Region of Origin

Here’s the first question: do players who grew up in cold-weather areas have an advantage over those from warm-weather areas? This is a justification that writers and broadcasters toss out now and again as they attempt to predict cold-weather performance. The Cubs of the 2015 NLCS provide a good example. A whole narrative swirled around the first two NLCS games in New York: speculation that Kyle Schwarber’s Ohio roots gave him a leg up in handling the cold, and hypotheses that warm-weather natives like Las Vegas’ Kris Bryant, Puerto Rico’s Javier Baez, and Cuba’s Jorge Soler were at a disadvantage.

In those games at Citi Field, the Mets marched to two wins and never once trailed. Several Cubs apparently pinned their inferior play, in part, on the scary-cold temperatures. What’s interesting is that this reasoning could actually be valid: In this NBC Chicago article, writer Phil Rogers relays doctor speculation that ability to deal with cold could be “governed by the climate where an individual grew up.” Is this effect real when it comes to major league performance?

I partitioned the sample in a few ways. The first step was to find the average October temperature in all U.S. states, all applicable Latin American countries, and all of Canada from Weatherbase. American players were grouped by whether they were born in the 16 coldest U.S. states (with Canadian-born players added in here), the 16* warmest states, or the 18 “average” states in between. Where applicable, I used college states rather than birth states, banking on the assumption that players are recruited for a college program near the town in which they were raised. Birthplaces are misleading when someone born in one locale is brought up elsewhere. Arizonan Curt Schilling, for instance, is listed with Alaska as his birthplace.

Latin players are separated by whether they were born in the Caribbean or the other South American countries (with the latter subset herein called “Greater South America”). Other countries were left out due to small sample sizes. An exact list of included state/country classifications can be found in the Google doc in the resources.

In the warmer weather, the curves seem to interweave in a random manner. Before that point, the curves are way different. Hitters in the Caribbean and U.S. Hot categories performed worse in the cold weather than the U.S. Average batters, which was anticipated. But the other results run contrary to our Bayesian prior—notably, U.S. Cold players come out looking the worst in the coldest temperatures, and Greater South America players emerge as the highest-performing group. Importantly, those two subsets had the smallest samples and widest confidence intervals. And the U.S. Cold players do rally and hit much better in the 50-degree range, so it’s possible that their lackluster performance in the extreme cold temperatures is a small sample size artifact.

We’ll dig deeper by aggregating the data to yield each group’s wOBA at 55 degrees and chillier. Those figures are shown in the next table, where regions are ranked by wOBA.

COLD-WEATHER PERFORMANCE (<=55 DEG.) BY REGION OF ORIGIN

Region of Origin	PA	wOBA	St. Dev. (points)	95% Confidence Interval
Caribbean	14,853	.3266	4.1	(.319, .335)
Greater South America	4,651	.3262	7.4	(.312, .341)
U.S. Average	10,594	.3256	4.9	(.316, .335)
U.S. Hot	21,420	.3240	3.4	(.317, .331)
U.S. Cold	7,025	.318	5.9	(.306, .329)

Absent of the LOESS smoothers, we find that both groups of Latin players perform the best in the cold weather, followed closely by those in the U.S. Average and U.S. Hot groups. (The Caribbean-born group has jumped, as compared with the graph, because the smoother doesn’t consider sample sizes and wOBA totals in tandem.) Ranking at a distant last in the table are the U.S. Cold players. For all non-cold regions, wOBA totals are so tightly clustered that it looks unlikely that any single population gains an advantage.

What if we deviate from the previous 55-degree definition of cold weather by bumping it up to 61 degrees? In this particular graph, ~61 degrees is the actual convergence point. Let’s look at that combined data.

COLD-WEATHER PERFORMANCE (<=61 DEG.) BY REGION OF ORIGIN

Region of Origin	PA	wOBA	St. Dev. (points)	95% Confidence Interval
Greater South America	8,531	.3363	5.5	(.326, .347)
U.S. Hot	40,154	.3330	2.5	(.328, .338)
U.S. Average	19,879	.3329	3.6	(.326, .340)
Caribbean	27,923	.3304	3.0	(.324, .336)
U.S. Cold	13,282	.3290	4.4	(.320, .338)

The sample sizes have doubled, the confidence intervals are 25 percent tighter, and the order of the regions… still has the U.S. Cold group in last place. Greater South America has gained a few points of distance at the top; U.S. Hot and U.S. Average players continue to perform at the same high level. The only group the actually aligns with our prior is the Caribbean-born players, who have now settled into second-to-last place. By and large, the regional results oppose expectations.

Is it useful for teams to employ an origin/temperature platoon? I don’t think so; it doesn’t make much sense that the performance of Caribbean-born players is impaired while batters in the Greater South America and U.S. Hot bins do just fine. No matter the interpretation, our results show that a severe, cold-weather-induced downturn to the stats of Jorge Soler and other warm-weather players would be highly unexpected. All told, this justification looks like a myth, just as light eye color isn’t an excuse for poor daytime play.

Age

In that same NBC Chicago article, Nadera Sweiss, a doctor and professor currently working in the University of Illinois hospital system, told Rogers that older people feel colder than younger folk. Of course, “old” has a very different meaning depending on whether the subjects of interest are baseball players (for whom 40 is old) or the general populace (for whom 40 is not old). Does Sweiss’ insight apply to this study, in that older hitters are worse off in the cold than younger hitters?

We’ll partition the curves to correspond to a generic aging pattern. The “early prime” label represents players aged 24 – 27, “later prime” is for players aged 28 – 32, and “old” is for players aged 33 and up. “Young” players (aged 23 and under) were left out because my Marcel reliability requirement too drastically reduced their sample sizes. In the chart, darkening shades of pink indicate an older subset of players.

More than any previous chart, this one presents curves that are tightly entwined in the thick of the plot, and consistent: old players tend to lag behind their prime-age counterparts. That’s understandable at an initial glance, but undesirable; remember, these curves are adjusted for batter quality, so that trend shouldn’t be peeking through. With confidence bands that are among the more constricted of this study, I wouldn’t think it to be a sample size issue; rather, it seems like older hitters are underperforming their projection by two to three points. That’s small enough so that it shouldn’t pollute the other analyses, but we’ll remain cognizant of the issue as we move forward with this age-focused subset.

The story told by the cold-weather results is that Early Prime players perform at the highest level, with the Later Prime curve looking the worst before intersecting with the Old curve. (Later Prime players also appear the worst in the extreme heat.) Do the results change when we turn to our pooled cold-weather data?

COLD-WEATHER (<=55 DEG.) PERFORMANCE BY AGE

Age Cohort	PA	wOBA	St. Dev. (points)	95% Confidence Interval
Early Prime	14,591	.330	4.2	(.322, .338)
Later Prime	27,641	.323	3.0	(.317, .329)
Old	15,913	.321	4.0	(.314, .329)

The table does follow the prior—each wOBA figure is ordered by age range, showing that the older the group, the worse the players fared. But to me, these results are inconclusive, and may be biased by the age trend discerned earlier from the chart. Whether players’ ages matter in explaining cold-weather play is worth a look in future research.

Weight

Sweiss also mentioned that thinner people are more adversely affected by cold than heavier people. That means heavier people (both the portly and the muscular) are better off in chilly conditions than their slighter counterparts. How does that premise hold up for major league hitters?

Weights aren’t the most reliable data in baseball’s trove (Bengie Molina thinks it’s OK, actually), but we won’t take the numbers exactly as presented. Instead, we’ll combine heights and weights to calculate each player’s body mass index (BMI), and then divvy the hitters up into quartiles. I hope the erroneous Molina weight is a particularly large outlier, and these categories in the chart below do well enough to categorize hitters. The BMI curves are color-coded so that the darker the shade of green, the heavier the hitters.

We see that the heavy players post the best warm-weather performances but are inconsistent in the cold. Meanwhile, the lighter batters come out looking like the best cold-weather group. That all runs totally contrary to our expectation. For completeness, let’s check out the usual pooled cold-weather results.

COLD-WEATHER (<=55 DEG.) PERFORMANCE BY BMI

BMI Cohort	PA	wOBA	St. Dev. (points)	95% Confidence Interval
Lightweight	13,866	.3262	4.3	(.318, .335)
Heavyweight	15,619	.3251	4.0	(.317, .333)
Slightly Light	17,075	.3239	3.8	(.316, .331)
Slightly Heavy	12,660	.3230	4.5	(.314, .332)

In the table, heavy players do sneak up to the No. 2 ranking for cold-weather performance, but the groups still aren’t arranged as expected. As it is, the aggregated wOBA figures for every BMI cohort are highly concentrated: each is separated by just one wOBA point from its nearest neighbor. BMI appears to hold no power in explaining cold-weather play; Sweiss’ commentary would have to apply more closely to the general population.

Concluding Remarks

What have we learned? Well, if you’re projecting your team’s performance in cold weather, you’d do well to implement a control for position—especially for the DH. The positional subset rose above all the other ones examined, jumping out as an influencer of the game on the field. And in the playoffs, when teams are looking for every edge that can raise their probability of victory even slightly, that’s meaningful.

References & Resources

*Floridians fit better with the Caribbean group due to the state’s hot outlying temperatures, so that’s where they’re included.

• A list of region/origin assignments
• Individualized charts with confidence bands