What do 80 PAs tell us?

Today marks the nominal completion of my 33rd trip around the sun, and boy, it’s definitely been the weirdest one yet. To be fair, in baseball terms, ever since the 60-game season was announced, we knew it was going to be weird. (“Fun” fact: through the Braves’ 60th game last year, the Phillies were leading the division. They finished at .500.) One of the challenges with “weird,” though, is that it’s hard to say “Okay, well, that’s weird, so let’s just not care and move on.” We make choices, either consciously or subconsciously, to care about things. Sometimes, the warping of those things is enough to make us care less, or not at all. But other times, it isn’t. I have a distinct, vivid memory of being a little kid and preparing to visit my grandpa in the hospital after he almost died. What I was scared of was that he would be different in some un-parsable way, transformed by a seven-year-old’s imagination. When I saw him, though, he was definitely still my grandpa, and that fear dissipated — sure, he wasn’t “well” and it would be a while before he could walk again, but the Ship of Theseus, this was not. I’ve gotten pretty far afield, but that’s basically the 2020 baseball season for you: it’s frightfully different, but also, pretty much the same. It’s a struggle, basically.

To that end, we’re almost halfway done with the season, and yet, individual performances through this halfway point are (understandably) weird. Silly, even.

Mike Yazstrzemski leads all position players in fWAR. Dylan Bundy has provided over 13 percent of his four-plus season career’s value in his last six starts. Ian Happ and Wil Myers are back from their 2019 netherrealms, and crushing it. You get the idea. Okay, one more: the Marlins are second in the NL East, half a game out of first place, four games ahead of the Nationals, and with baseball’s 11th-best record. The Marlins! And we’ve only got around 30 more games to play. (Haha, the Orioles have a better record than the Brewers. My head hurts.)

So, baseball is different, but really, it also isn’t. That duality brought me to a basic question, one that’s probably been the matter of some debate across every fan base in this odd, curtailed season: sure, we’re halfway through the 2020 season, such as it is at this point... but what do we actually do with the information we’ve compiled? This post isn’t really going to answer that question. That question is too expansive, and the basic principles guiding the answer haven’t changed. Instead, all I present to you below is some basic data for your consideration. What you do with it, just like how much you care about the 2020 season, is entirely up to you.

---

80 PAs seems like an okay cutoff for right now. Among the Braves’ position players, seven have hit this threshold, and a few others easily would have if not for injury or pandemic-related precautions. (If you squabble about 80 PAs, I’ve done the sensitivity analyses, nothing here really changes regardless of whether you move that threshold up or down a bit.)

I went back to our last full, legitimate baseball season (2019, feels like it was a decade ago) and pulled every player with 80 or more PAs through April. I then pulled the rest-of-season (not full-season, rest-of-season) performance for those players. I dumped out anyone wth fewer than 120 PAs after April (which was barely anyone that got 80 PAs in April, really). For simplicity, I’m looking only at wRC+.

(If you need a refresher, wRC+ is just a basic measure of how productive a guy is at the plate based on his frequency of outs/walks/singles/doubles/triples/homers relative to a league-average batter. A 105 wRC+ means the hitter is five percent better than league average; a 90 wRC+ means the hitter is ten percent worse than league average. It also kinda-sorta adjusts for park effects.)

I then tried the most basic arrangement possible to correlate April hitting with rest-of-season hitting. (Remember, this isn’t April versus full-season, which includes April. So one doesn’t inherently drive the other by being included in it.) It... uh... yeah. Just look:

I’m not saying there isn’t a pattern, but we’d expect there to be a pattern, after all. A good hitter is a good hitter and should be expected to be a good hitter, whether it’s April or the other five months. And yet, this correlation kind of sucks: in words, that 0.1378 is saying, basically, “Only around 14 percent of the variation in May-September hitting is explained by variations in April hitting.”

Honestly, though, that’s basic correlation is not a great way to think about this. The problem is that wRC+ is granular (it’s presented as a whole number), but a dataset with a bunch of April = 100; May-Sep = 102 and another bunch of April = 90; May-Sep = 88 would just get very confused, because it would try to treat every movement as meaningful. Let’s take a step back and do something that might relate a little better to how we think about whether someone is a “good” hitter.

What I did was simply take the array of all 2019 hitters with 200+ PAs, and split them into five groups (quintiles) based on their wRC+. The worst quintile (i.e., the worst 20 percent of batters) had a wRC+ of 80 or below. (Hey, this makes sense, the worst 20 percent of batters hit 20 percent below league average or worse. Insert the most obvious Mr. T quote here.) Just for transparency, the full set of quintiles is:

• 80 wRC+ or lower
• 81 to 93
• 94 to 106
• 107 to 124
• 125 wRC+ or higher

I then looked at where a batter ended up in terms of quintile placement in April, versus quintile placement for the rest of the season. It looks like this:

Actually, that table is useless, because it has raw numbers. Let’s try again, with percentages, so there’s some embedded context:

Actually, this one is kind of useless, too. It tells us that relatively rare outcomes are being awesome in April and being awful in May-September (the bottom left corner), or being below-average in April but awesome in May-September (the upper right corner). It also tells us that a relatively common occurrence is being awesome and April and good-to-awesome in May-September (bottom right corner). The rest of it, though, is kind of a mess.

One more try, then, in the pursuit of clarity:

Alright, here we go. I’m liking this one. In this chart, the percentages sum to 100 across each row — in other words, this table answers for you, “Of the players that started in each quintile in April, what was their proportion breakdown across quintiles in May-September?” I won’t belabor the point, because you can look at the table yourself. If you are absolutely killing it in April, you have a much better chance of continuing to kill it, or at least remain above-average, than you do of backsliding. Again, note that this is not a full-season value across the columns; the two are independent. So this isn’t anyone who has a 200 wRC+ April and then coasts to a 125+ seasonal line; to appear in the bottom right 40 percent, the player had to exceed 125+ in April and May-September, not in April and full-season.

Aside from that bottom-right corner, though — insert your favorite shrugging ascii text or meme here. Obvious case in point: a player who hit for an 80 wRC+ or worse in April had the same chance of hitting for an 80 wRC+ or worse in May-September as he did of putting up an above-average wRC+ between 107 and 124. This also isn’t a tiny sample-size quirk, either — that motion was the fifth-most common among the 25 “buckets” in the grid.

Again, you can take this however you want. Maybe you just don’t care. But, I see this as the joke about the Laffer Curve. In brief, internet-speak inflection: lolnothingmatters — at least outside of dudes who are raking in April, in which case it kinda matters but also not overwhelmingly so. This would also be a good time, if one were so inclined, to lecture people about the upper-left hand corner of the tables, relative to basically everywhere else. Just sayin’.

At this point, you might be thinking, “Uh, okay, but you’re not really giving me enough to go off of. Okay, April is weird. What about April and May, together? Does that help?” Oh, my sweet summer child, how much baseball have you absorbed into your veins?

When we have about 60 games’ worth of data, we can feel more confident that raking dudes gonna rake, raking dudes probably not gonna implode or even be below-average, and that sucking dudes not gonna rake. Everything else is still more or less in toss-up category. That upper-left corner, though — bruh. After 60 games, a guy that looks, for all intents and purposes, either broke or just bad at hitting, has as high of a chance of being average or above-average (so, twice as high of a chance if you combine these two) as staying broken.

Sure, there’s some lack-of-survivor bias in play — guys that actually are broken drop out, or they don’t get to play through May anyway. But, I have the data, and the dropping out doesn’t happen very frequently (and where it does, it is often due to injury, so guys across all the quintiles drop out). In other words, it’s not really a big deal. (As a side note, in running alternative correlations, all of which look pretty similar to the scatterplot above, addressing survivor bias based on PA cutoffs only helped the correlation when doing April versus rest-of-season. In other cuts with bigger “initial” periods, it was not relevant to the results.)

Alright, so one-third of the season still doesn’t make us very confident that a terrible hitter is going to stay terrible. Dare we ratchet it up one more month, such that we’ve got half the season in play? Yes, we dare.

I think I’ll stop here. Want to feel more confident that your pretty awful hitter is going to stay pretty awful than he is to do anything else? Wait till June is over (this table uses a 240 PA cutoff for the initial period), at least based on this table. (Note: none of this helps you with the fact that your awful has still given you 240 awful PAs at this point. Sorry.)

---

Really, I could stop here. Maybe I should. But I can hear you: “Ivan, you idiot. This season is only 60 games long, and we’ve already passed 30 of them. All I need to know is what a guy is going to do in his next 30 games. We don’t have time to wait until June.” Well, this is what you get:

If you have a good decision heuristic based on this table, by all means, let me know. (Beyond “play your average-to-good hitters,” I mean.) The correlation for April into May has been torn to shreds. When it comes to figuring out what your average-to-below hitters are going to do for the next 30 games, once they’ve done their first 30, you’re more or less flying blind if you’re relying just on their past results.

---

Think all of this is garbage? Skeptical, but need to put some hypothetical faces to the numbers? Let’s do the 2019 Braves!

April —> Rest of Season

• Freddie Freeman: 141 —> 138
• Josh Donaldson: 133 —> 132
• Ronald Acuña Jr.: 130 —> 126
• Ozzie Albies: 117 —> 116
• Ender Inciarte: 73 —> 109
• Nick Markakis: 128 —> 94
• Dansby Swanson: 114 —> 87

April + May —> Rest of Season

• Josh Donaldson: 118 —> 139
• Freddie Freeman: 145 —> 135
• Ozzie Albies: 88 —> 133
• Ronald Acuña Jr.: 114 —> 133
• Nick Markakis: 106 —> 99
• Dansby Swanson: 94 —> 92

April through June —> Rest of Season

• Josh Donaldson: 117 —> 148
• Freddie Freeman: 152 —> 121
• Ozzie Albies: 106 —> 128
• Ronald Acuña Jr.: 130 —> 123
• Nick Markakis: 103 —> 102
• Dansby Swanson: 107 —> 64

April —> May

• Josh Donaldson: 133 —> 100
• Freddie Freeman: 141 —> 149
• Ozzie Albies: 117 —> 53
• Ronald Acuña Jr.: 130 —> 96
• Nick Markakis: 128 —> 83
• Dansby Swanson: 114 —> 75

---

What you should do

If you came to this section and were hoping for strategic baseball advice, sorry. That’s not what this section is. Instead, here’s what you should do having seen the above.

In short, don’t accept it. There are shortcomings. Quite a lot. Here’s what you should do, instead, if you want more strategic baseball guidance, or just don’t want to take these numbers at face value:

• Expand the analysis beyond just 2019. 2019 wasn’t special. Patterns seem lacking. Will they be better if we go beyond 200 or so batters per analysis?
• Look at input metrics, not output metrics. We have xwOBA. I didn’t use xwOBA because that wasn’t the point — the point was seeing whether outcomes in 30 games mean anything for outcomes in later games. (Answer: ehhhhhhhhhhhhhhhh.) Do it anyway. Hopefully xwOBA is stickier. If it isn’t, then ugh. Then go read this and build Predictive xwOBA, and do an analysis on that.
• Do pitchers. Pitchers are people too. Don’t do relievers, though, not if you want to keep whatever sanity 2020 has left you with.
• Maybe don’t do quintiles. Do thirds, with below average/average/above average. Or something else. Maybe do a correlation on quintiles/other buckets, instead of on raw wRC+ or another measure.
• Most importantly, don’t just consider hitters in a vacuum. We have projections. The question is when a hitter’s past performance overtakes what the projection expects from him. That’s the real crown jewel here — when should we start looking at what a hitter has done this season, relative to what he was projected to do before the season? (I have my own answer from an older analysis, but that’s not the point.) Go do that one, and then you can start figuring out whether we should be souring on players after 30 disappointing games.