After Sabre-school Special (August 21, 2003)
Here are my replies to various reader questions since last time. If you have any comments on these issues, or you have other things to discuss, plese email me directly. Practicality of DIPS
the practicality is that
1 - you must separate HR/BB/K from his stat line, and
2 - you have to find some other way to measure his skill on BIP... it exists, and we can't find it very well
UZR: I'm not buying it
Forget UZR, ZR, and everything else. You're a fan, and you're looking at your favorite SS. Balls are hit at him, in the hole, up the middle, short, deep, lefty batters, righty batters, guy is on 1b, sometimes 1b is empty, your 3b is covering the line, runner from 2b tries to steal 3b and you want to cover 2b for a pick, you have a 1b that scoops the ball poorly, and doesn't stretch well maybe, your pitcher gives up lots of GB, the score is close and late, the batter hits a very hard ball, the trajectory of the ball is a loop over the head, or a one-hopper, or a high bounce.
Anything else? So, that's real-life. Our job is to model real-life. And we're trying to infer from the data everything that just happened.
That's the UZR model. MGL's UZR is not there yet, but it's the closest thing available, unless Tippett did something better.
Is it better to be a GB or FB pitcher?
Assuming that Line Drives don't favor one type of pitcher over the other, it's much better to be a GB pitcher. On a per PA basis, each GB is worth .10 runs more to the pitcher than a FB. This is on average, and of course, you'll always find some great FB pitchers who are better than some GB pitchers.
How many PAs does each batting order get?
The last PA of every game follows an even distribution among the 9 batting spots. That is, the last PA of every game is: 11% leadoff hitter, 11% #2 hitter... 11% #9 hitter.
Once you've got that knowledge, you now know that
PA(1)=PA(2)+18
PA(2)=PA(3)+18
...
PA(8)=PA(9)+18
So, for any given year, take the avg number of PAs per season per team. Say, the 1986 Dodgers is 650. That's the PA of your #5 hitter. Your #8 hitter would be 650 - 18x3 = 596. Your #1 hitter would be 650 + 18*4 = 722.
Experience v youth: slumps
Remember, we are looking for variations in true talent, and not in performance/results.
Sure, I can believe that a rookie will be less consistent than an experienced hitter, but how much can that talent vary?
I mean, if Tim Raines' true talent was "130" on a "100 = average" scale when he was 35, maybe that true talent varied between 125 and 135 during the year, based on his conditioning, age progression, experience.
Raines at 23 say was "140". How much could his true talent change during the year? 130 to 150? I mean when you think about it, how much variation in true talent can you have?
Even if a rookie is more subject to a slump, was that based on his talent level not keeping pace? How much can his not keeping pace really affect his talent level? How much change can you really have?
LI and ARP
LI accounts for the inning/score/base/out. ARP accounts for the base/out. So, you can't just take the two and multiply since you'll be double-counting the leverage of the base/out.
So, for the quick and dirty, you assume that a pitcher will perform the same regardless of the inning/score/base/out situation. Therefore, you simply take his runs above average and multiply that by LI.
To figure runs above average, you would use BaseRuns to figure out what his "component" RA would be. And then subtract the league average from that.
Then, simply multiply by his LI to figure out the impact of his runs.
Even better is the following. Figure his runs above average. Then, figure out his runs-per-win converter based on that run environment. I like to use RPW=(Runs Scored + Runs Allowed) x 0.7 + 3.4. Typically, this will work out to around 10, less for the good pitchers.
Now, that RPW assumes an LI of 1.0. Take that RPW and divide by his LI. So, if you have an RPW of 10, and an LI of 2, this means that the runs-per-win converter for this pitcher, based on when he was brought in, is 5.
So, if he is +20 runs above average, he would be +4 wins above average.
Linear Weights and "average"
By the way, when you think of linear weights, don't think of it in terms of overall league average. Think of it in terms of overall average for a particular context. What context? Whatever you choose.
So, if you wanted to know the impact of getting a single, for the 2003 Redsox, batting in the #1 spot, with man on 1b, and 1 out, you would look for that EXACT situation, and figure out how often it happened, and how many runs scored to the end of the inning. You'd do the same for the double, HR,walk, steal, balk, etc, etc. And the cool thing? It'll all add up exactly for that particular context. No league averages needed.
What Palmer has done and presented with LWTS is only 5% of what you can really do with it.
Scouting or Performance?
As for scouting v performance, it depends on the sample size. If I had 10 PAs, I would rely 100% on scouting and 0% on performance. If I had 10,000 PAs, I would rely 0% on scouting and 100% on performance.
Therefore, at some point, they must both be equal. If you are talking about measuring a hitter's power, or speed, or fielding, or strike zone judgement, each requires its own "sample size breakeven" level. I have not yet determined what that point is, mostly because I've never seen a report from a scout.
--posted by TangoTiger at 03:00 PM EDT