THE BOOK--Playing The Percentages In Baseball

I agree with the win% numbers you have here. My main confusion was how these were actually converted into a replacement level FIP number. I get replacement level as a concept but it’s never something I’ve really thought about, under-the-hood wise. The main problem I was having while thinking out loud in this post is that my intuition about difficulties in starting vs relieving was clashing with my intuition about where reliever replacement level should be.

Btw, I came up with berselius as a handle years ago when trying to find a new handle. The name comes from a minor character (who was an evil talking sword!) in a replacement level JRPG (Tales of Destiny) that I played in high school. I was foiled when I signed up for twitter by a guy whose actual last name is Berselius!

I think this is one of the few areas of your “style” player evaluation that I don’t care for. Maybe you’d be willing to address your logic behind some of these so that I’m less antagonistic to the pitcher valuation.

Why do you use winning %? This frustrates me for a couple reasons. First, it’s confusing to try and explain to newcomers as wins are often berated as meaningless. I understand we’re talking about “true talent” winning % rather than what a pitcher did in a given year but it seems needlessly confusing.

Secondly, the connection between FIP and winning % has never seemed readily apparent to me (which may very well be my own shortcoming). FIP has a very intuitive scale but I’m forced to try and reconcile it with winning % that is completely non-intuitive.  Perhaps this has something to do with crediting pitchers more for what actually happened rather than what FIP says should have happened but it is perplexing.

Fair enough. 

Let’s say the average pitcher gives up 4.7 runs per game (earned and unearned).

The replacement-level pitcher, as starter, gives up 6 runs per game.  That same pitcher, as a reliever, gives up 5 runs per game.

Pythag tells us that the 5RPG pitcher has around a .470 win% and the 6RPG pitcher has around a .380 win%.

***

The average MLB pitcher gives up runs at about 2% higher than the league average, and the average MLB reliever gives up runs at about 4% lower than the league average.

So, that would mean the average starter gives up 4.8 runs per game and the average reliever gives up 4.5 runs per game.

***

The average starter (4.8 RPG) is +1.2 RPG better than the replacement-level pitcher as starter.

The average reliever (4.5 RPG) is + 0.5 RPG better than the replacement-level pitcher as reliever.

***

The average starter gets about 65% of all innings, and the average reliever gets 35%.  65% of 162 games is 105 games, and the relievers get 57 games.

***

+1.2 x 105 = +126 starter runs above replacement
+0.5 x 57 = +28.5 reliever runs above replacement

***

Let’s give the relievers a bit of a leverage boost because the good ones will get to exploit their good performances more. 

So, while we give 10 runs per win for a starter, let’s give relievers 9 runs per win.  Just for fun.

126/10 = 12.6 starter wins
28.5/9 = 3.2 reliever wins

***

3.2 / (12.6 + 3.2) = 20%

So, 20% of value comes from relievers.

***

MLB Payroll: 10% goes to relievers and 33% goes to starters.  10 / 43 = 23% of pitcher payroll goes to relievers.

***

All this was very quick, I didn’t try to check to see if I made any mistakes, or if my run/win conversions made alot of sense.  This is a rough guideline.

one thing i’ve never seen explained (i’m sure it was somewhere) is why the average starter has a .490 WP, not .500.  an average starting 9, bullpen with an average starter vs. the exact same should result in a .500 WP, right?  i assume this was empirically derived somehow…

I think the thing that I’m missing here is how we go from the work you posted above about value to a replacement level reliver FIP of 4.32.

I don’t agree with the 4.32.

***

colin: Well, you are right that in reality, the win% is .500.  But, I was talking about it in terms of pythag, where you have a 9-inning starter against a 6-inning starter and 3-inning relievers.  Basically, I was using win% in place of ERA.

Thinking of what you are saying then, then, yeah, it’s not very clear that when I use the win% scale as to exactly what I’m doing.

So, let’s just stick to the avg=100 scale, and we should have no issue.

environment = 4.7
repl-reliever = 5.0
repl-starter = 6.0

actual-reliever = 4.5
actual-starter = 4.8

(And if you want ERA, multiply all the numbers by 0.92).

That fair enough?

I don’t agree with the 4.32.

Whoops, I should have said the 4.45 for the NL that Cameron found.

Thanks for the clarification in your above post. Where does the 0.92 number come from?

ty tango

bers, i’m pretty sure that’s average league earned runs/average league runs.  that’s fairly constant.

92% of runs are earned.

So, this is why I like to win% scale.  I say .470, it’s always .470.  But, 4.45 for NL in 1968 is very different than in 1994.

Therefore, what you need to do is either use a win% scale, a avg=100 scale, or you need to give me two numbers: the league average, and whatever other number you are using.

*smacks forehead* of course. Can’t believe I blanked on that. Thanks for the explanation, this makes a lot more sense to me now.