The Big Hurst - Kr1s M3dl3n
skoormit - Clayton Kershaw
La Osa Rosa - Patrick Corbin
La Osa Rosa: I’m going to pick Patrick Corbin. Is that the desert mirage guy? Also, I hate being the loser. It’s not fair!
The Big Hurst: There's a Fister/Wang matchup? Fourteen-year-old me is laughing his tail off.
La Osa Rosa: I had a roommate in college named Kristen Medlen who was a 4’11” blonde cheerleader. I’m pretty sure she wrote her name the same way you did for Kr1s M3dl3n. She’s one of those people that probably got her letters and numbers confused with each other.
skoormit: Yes, Corbin is the desert mirage. I'm taking CK at altitude. They say that Coors Field takes 20% of the movement off of your pitches. 80% of CK's curveball is still the best curveball in the bigs.
The Big Hurst: Here's a wonky discussion. There's clearly a lot of randomness in pitcher starts because we're having a devil of a time picking the good ones. I think we'd all say it's hard to pick one of the best five starts in any given day. But our "top five" methodology is somewhat arbitrary. Instead, I kept it in my head to also check whether we're doing a good job at picking the "gems" - the 80+ or 90+ games. I'd been waiting until we had more data.
Here was my theory as to how this stuff worked. Randomness being what it is, any decent starter can throw a 70+ game. But my hypothesis was that only really good pitchers generally throw the 80+ or 90+ games. Dallas Braden aside, I think most gems are thrown by the more talented pitchers.
In my head, I model it like this. A pitcher gets a certain amount of points from being good. At its best, this might be something like - I dunno - six points. And then you roll 3d6 (three six-sided dice, if you're not a dork) to represent the randomness and for the remainder of the points. So Clayton Kershaw might be worth the full 6 points. Then, when you roll 3d6, it'll be somewhere between 3 and 18, but it'll tend to be 10 or 11. So, if you add all that together, Kershaw will tend to get like a 16-17, but it might be as good as a 24. He also might be as bad as a 9.
This kind of analysis would suggest that any average starter could, every once in a while, roll an 18 and post a good Game Score. But I think to get one of those 80+ or 90+ games, you need those extra few points from being a excellent pitcher, and the excellent pitchers can have a very good game even if they roll a 16 or 17 on those 3d6. In short, average pitchers can get an 18, but they can't ever get a 22. Excellent pitchers don't even need to roll the luckiest possible 18 to get a 22, so you'd expect them to get there more often.
What does this mean for us? Well, I wondered whether we should be doing a better job of picking the gems, because filtering out the games that are 70+ might eliminate a lot of the noise. I would have guessed that the randomness was too much of a factor because we're including lots of those 70+ scores. At least, I hoped we were doing a better job with the gems.
Does this play out in our data? No. (Or at least not yet.) Since we started playing, there have been 42 total games with a Game Score of 80+. Seven of these are games of 90+. We've picked none of the 90+ games, and only four of the 80+ games (5/14 Kershaw, 5/31 Colon, 6/5 Iwakuma, & 7/1 Fernandez). So we're batting around .095 for the gems. Not so good, and it might blow my theory right up.
A quirk of the data is that the gem rate seemed to slow considerably in June. Even in our partial May, there were 28 games of 80+ with six at 90+. In all of June, however, there were 13 games of 80+ and only one at 90+. Do pitchers do better earlier in the season and hitters better in the middle? Did we just ride through a month of hot hitting? Will there be more gems in July? What's the average gem rate?
RESULTSThe Big Hurst: My remarks above could not have been more timely, could they? Two subsequent thoughts: First, the Bailey no-hitter and the Kershaw 80+ gem were significantly unpredictable, at least based on the way I'm currently thinking about selecting pitchers. I think both of these gems are more praiseworthy because they shined under notably hostile circumstances. Bailey isn't usually thought of as dominant, he pitched against one of the best offenses in baseball, and he pitched in a park that's usually thought of as a good hitter's park. Kershaw is dominant, but he pitched against a neutral offense, but in the Colorado park that's a known killer. In my system, Bailey scored negatively and Kershaw was close to neutral or zero. Neither would have been a strong pick or even a decent pick. So what am I doing wrong? If I'm putting too little value on the pitcher's quality, it would've increased my chances of picking Kershaw, but it wouldn't have affected the Bailey pick. If I lowered pitcher quality relative to the other factors, then Kershaw would've looked even worse.
Second, these results confuse the suggestion that we're bad at picking gems. We nailed the (unpredictable?) Kershaw gem, but we missed Bailey and Price. Our leader skoormit picks two hot gems in two days. What would the best possible system look like for predicting 80+ and 90+ games? What variables would you use and how would you weight them? Am I missing a variable that correlates strongly? Is it more random than we thought? What's the theoretical maximum for points per selection, if it's not right around 1.0?
This game is maddening.