Ratings vs Results - Avoid K Rating

[dancariaz]

Just found this thread, what a great read, not just about the Avoid K rating but about hitting ratings in general. I wonder if anyone did a study like this on pitching ratings?

[DonkeyKongSr]

Originally Posted by dancariaz

Just found this thread, what a great read, not just about the Avoid K rating but about hitting ratings in general. I wonder if anyone did a study like this on pitching ratings?

Pitching ratings are really straightforward:

- Stuff = Strikeouts
- Movement = HR suppression
- Control = Walk suppression

Sprinkle in defense, park factors, and level of competition and you can highly correlate all 3 relationships.

[EZB]

Originally Posted by chazzycat

DonkeyKongSr - this was inspired by our chat on Discord yesterday.

I exported the data available from the mini-whale perfect league, each dot representing a player's entire career - but with a minimum 1,000 PA threshold. Total PA represented in the sample is around 1.6 million.

First off, as I suspected, the relationship between Contact Rating and BABIP becomes much clearer with this large of a sample:




It's not the strongest relationship, but certainly there. However the next graph is more enlightening:



As you can see, the Contact rating has a much stronger relationship with batting average than it does with BABIP.

This is important because it lends some credence to the theory that strikeouts are not separated from other outs in the game's calculation until after it's been decided whether it will be an out or not. To back up for a moment, there are two competing theories that could explain how contact and "avoid K" ratings work in the game. It's basically just a question of the order-of-operations. Here are the two theories:

1) The game calculates first whether a player will strike out, by the avoid K rating. For all the PAs where the player doesn't strike out, walk, HBP etc., then those fall into a "BIP" bucket. Only then would the game apply the BABIP calculation to determine if it's a hit or not (determined by the CON rating, luck, defense, etc.).

2) The game calculates first whether a player will get out or not. After it makes that determination, it then determines a strikeout vs. other types of outs based on the "Avoid K" rating. So having a high "Avoid K" rating in effect just moves K's to groundouts or flyouts.

If theory #1 was true, then we would expect to see the CON rating have a very strong relationship with BABIP, significantly stronger than it does with AVG. AVG includes strikeouts, which don't apply to BABIP. So removing the strikeouts from the equation should show a stronger relationship between CON & BABIP, if that theory was true.

But, instead we see the opposite. To me, this alone strongly implies that it's theory #2 which is correct.

I didn't stop there though. We can also compare the "avoid K" rating to real results for more direct observations:



Ouch. Strike two. If theory #1 was correct, we should also see a much stronger relationship between avoid K and AVG. This is close to zero relationship. However, it would be going too far to claim zero correlation. It is worth noting that there may be a very small effect here.

Recall the graph above where CON and AVG have a strong relationship, with an R-squared value of .54. This is roughly ten times as strong as the relationship AVG has with Avoid K's. So we can crudely estimate that upgrading the CON rating is about 10X more effective per point than upgrading the Avoid K rating, when it comes to batting averages.

Moving on, let's see how Avoid K rating effects overall offensive production. wRC+ is the best stat to neutralize park effects etc., so that is what I will use.



Again, there is pretty much no relationship here. It appears as though strikeouts simply do not factor into overall output whatsoever.

For anyone thinking "ratings don't matter!"...well yes, they do. That's definitely not the point I am making here. For reference, here are how the other ratings correlate with wRC+:







Compared to Avoid K, the other batter ratings show very strong correlation with overall results. Of course if we looked at more precise stats for each rating (HR rate for POW, BB rate for eye, etc.) we would see MUCH stronger correlations. But it's reassuring that we still see significant correlations once the results have been watered down to a single number. And we can make logical decisions based on the R-squared values here. For example, the EYE rating has more effect on overall output than the POW rating. That may come as a surprise.

Note: I didn't include "GAP" in the graphs above because it's very confusing. For what it's worth the R-squared is .33. However I suspect that is mostly due to the fact that high-GAP players also have high CON ratings. I certainly would not conclude that GAP is more important than EYE or POW. It's hard to say for sure.

I want to say one last thing with regards to "Avoid K": just because it doesn't impact wRC+, doesn't make it useless. That stat (nor any others) don't include the benefit of moving runners over, which could potentially be quite important. Unfortunately, that is impossible to quantify.

Any feedback is appreciated!

If I were coding this, I would do it pitch by pitch with modifiers on each for probability, the aggregate of which results in realistic numbers. So the flow would go something like this:


Pitch is selected from the pitcher and thrown to batter. Many variables would go into whether contact was made at all, but mostly that would boil down to stuff and avoid-k ratings. Strikeouts would be more common for pitchers with higher stuff because they are more likely to throw 3 missed balls, and batters would be more likely to strikeout with a lower avoid-k rating. If contact is made, numerous variables would affect the outcome. BABIP would affect the likelihood of the ball being a hit, but the zone ratings of the defender would also play into that. The rating of the specific pitch would matter. There would be a modifier based on whether the batter guessed the pitch properly, which would make a pitcher with more pitches less hittable, but this modifier would be affected by how many times the pitcher has been through the order (pitch count). Remember, the batter's contact rating is based on setting all pitcher ratings and fielder ratings to neutral and then applying the avoid-k against the BABIP. You will arrive at the same batting average regardless of the BABIP/avoid-k ratings as long as the contacts are the same. In other words, two batters may have the same contact, but one has a higher BABIP and the other a higher avoid-k rating. The one with the higher BABIP will hit fewer balls (in general and in play), but when he does they will be more likely to be hits. The one with the higher avoid K will hit more balls in play, but they will end up as outs. In the latter case, it is more desirable to be a flyball hitter, since then you can get sacrifice flys. Either way, though, the batters should bat the same average. If they face a pitcher with superior stuff, they will both strike out more, which will lower BOTH of their averages; again, they will bat the same average! Thus, I prefer the batter that strikes out more, since he will ground into a lot fewer double plays, but at the expense of sacrifice flyballs. If the player is batting in the latter part of the lineup (maybe 5th to 7th) and is a flyball hitter, it is probably better to have the higher avoid-K because the sacrifice fly will be more beneficial (think more RBIs) than the double plays will hurt (since the players after are probably weak anyway).

[jaysdailydose]

Here I am again, with a necro of another great thread... I've been away for awhile, it is what I do...

I just wanted to say that I did a lot of statistics work with previous versions, and I think you're pretty much right on the money, Chazzycat!

I also think the immediately previous poster had a very close idea of what the game is doing, but obviously doing it in two paragraphs.

You're right on the edge of things that the devs won't give away, and if they think we get too close, they'd likely switch something in a minor way.

I did want to point something out as far as your analysis goes... you really need to analyze the pitch-by-pitch logs and the spray charts for the players. I also think there may be hidden values relating to each type of pitch. "Some guys just can't hit the curveball..." "He's a dead-red fastball hitter."

There's so much helpful information in this thread as far as game theory goes, so I didn't figure anyone would mind the necro too badly (especially as I'd seen you recently active, Chazzycat!)

Great stuff. Thanks for posting it.

[jkdhoo]

I ran across this thread because I was looking into something similar. But I noticed that I had a few questions on some of the assumptions in this analysis. And I'm not an analytics guy so was hoping someone could set my thinking straight?

For Strike 1, the assumption was:

1) The game calculates first whether a player will strike out, by the avoid K rating. For all the PAs where the player doesn't strike out, walk, HBP etc., then those fall into a "BIP" bucket. Only then would the game apply the BABIP calculation to determine if it's a hit or not (determined by the CON rating, luck, defense, etc.).

However, determining whether or not a BIP is a hit, you would not use CON, since CON is an amalgamation of both BABIP and Avoid K. Since we've already consumed the Avoid K part of contact in this theory, you wouldn't be able to use the full contact rating when determining a BABIP correlation. Wouldn't we need a way to remove the AvoidK part of Contact in order to be able to correlate contact to babip?

Moving on to strike 2, the assumption is that we should see a correlation between Avoid K and AVG. But if we already took AvoidK out of the equation, wouldn't we assume there would be no correlation? We can assume that a batter with high Avoid K is going to hit the ball more, but we can't assume that more batted balls means more hits unless we know what causes a batted ball to turn into a hit. Since Avoid K wouldn't be part of that equation, I don't know why we would expect Avoid K to have a huge impact to AVG. I might expect a slight impact, since a strikeout is the first chance to lower a batting average, but two players with the same contact - one high avoid K, and one low avoid K, will have a similar average because the player with the low avoid k has a higher babip and vice versa?