## Thursday, April 8, 2010

### Equivalent ERA: A new way to gauge minor league pitchers

Two interesting tools, one from Sean Smith and another from Tom Tango, have allowed me to devise a method for gauging minor league pitchers and their relative performance between levels.

Obviously, if you were to take an A ball pitcher and vault him to MLB, he would get destroyed. And if you took the average AAA pitcher, he would do a bit better but still not so well. But how much better is the AAA pitcher than the A ball pitcher? How can you tell when a pitcher at one level may be ready to handle the next level? Or when a pitcher at a given level is a bit too far over his head?

You can look at a pitcher's batted ball rates (groundballs, line drives, flyballs) and use linear weights to determine how many runs he ought to have allowed. But the run values for those outcomes are going to be different at each level and in each league, not to mention each park. And at best, those are only going to tell you how a player is doing relative to his peers... rather than how he compares to all players in an organization across the board.

Here's my crack method for evaluating how an average pitcher from a given level would do in the bigs:

- Determine the MLE (minor league equivalent) of the composite major league average over 2005-2009 for the minor league level you want to evaluate, for the team in question. This is how an MLB hitter would do at your given level. Find these numbers for groundball, flyball and line drive averages.
- Count the batted balls, walks and strikeouts for each pitcher evaluated, and convert all of these according to the determined MLEs for the player's given level to get a new, adjusted line.
- Take that adjusted line and plug it into Tango's Run Expectancy Markov formula to get the number of runs a lineup full of said MLB player would score each game at the given level.
- Multiply that total by 0.9 to get the Expected MLB ERA of the average pitcher at that level.

You could do this for every full season level from AAA to Single A, though due to the lack of equivalent data for Rookie and Short Season A ball you can't go any farther than full season A ball unless you were willing to fully research the levels yourself and find MLEs for those.

I would surmise the reason past saberheads didn't bother is

a) Those players are so far from the bigs than their MLEs are somewhat irrelevant

b) The samples at those levels, even over a full half-season, is so small that there's too much noise for an MLE to be relevant

c) A lot of those players are still in the instructional phase and what they do at that level has so much noise regardless of sample that their MLEs become somewhat irrelevant

d) Since these players are developing physically and skill-wise, their batted ball and K/BB profiles are going to reflect this: Groundball rates will be high, a lot of balls are going to get by novice fielders and hitting/pitching rates are going to be all over the place. Often these players don't get promoted anyway unless a team up the chain has an opening on the roster that triggers a domino effect of promotions. And it's often not because of merit but because of organizational need, a live arm that gets a test run at a higher level simply because the team needed help. The best you can do if you wanted to try anyway is scale back the player's stats 20% for every level below full season A ball... 80% value for Short Season A... 64% for High Rookie ball... and 51.2% for rookie ball proper.

It's very tough to sift through the noise of a player's rookie ball or short season A campaign and figure out if his numbers are real, if he's still ironing out the kinks, if he's feasting on undeveloped inferiors, or if he just sucks. It's an exercise many feel is best left to an organization's scouts until the player gets to full season A ball.

But from there, we can begin assessing a player's value. I took a five year composite MLB average:

165849 AB, 43524 H, 8737 doubles, 949 triples, 5042 HR, 16620 BB, 33591 K.

From that, I crunched that number into MLE's for the Seattle Mariners' affiliates: Tacoma, West Tennessee, High Desert and Clinton. I ran each of those revised lines through Tango's Markov formula, multiplied by 0.9 and got the following equivalent ERAs:

Tacoma: 6.36
West Tenn: 7.30
High Desert: 12.80
Clinton: 13.13

These are the expected ERAs of an average pitcher for each respective team. How do you use this? Take the season stats of a particular prospect. Assume them as MLB stats, and convert them using the MLE formula to the player's respective level. Run the revised stats through the Markov equation to get a run total, multiply by 0.9, and that's the Equivalent MLB ERA for that player's performance.

(One challenge is that minor league pitching data is missing doubles and triples stats, each of which are needed for the formulas. You can devise equivalents by taking the total doubles and triples for a league and dividing by the number of hits, then multiplying each percentage by the number of hits the pitcher allows to get an approximate number for each.)

For example, let's take the Military Man, Nick Hill. Save for a brief stint in Tacoma, Hill pitched in West Tenn in 2009, where he had a fine year as a starter over 97.1 IP, good for a 2.93 FIP.

366 AB, 86 H, 21 doubles, 2 triples (league average: 2.5% of hits), 5 HR, 24 BB, 100 K

(Minor League Splits and many minor league sites list batters faced but not ABs. It's easy, however, to find AB's. Just take the number of hits and divide by the average. 86 H / .235 = 366)

(If you really wanted to wonk out, you could use composite batted ball averages with his batted ball data (53.5% GB, 19.4% LD, 27.1% FB) to eliminate defensive/park variance and figure out what his hit, slugging and HR totals ought to have been and then extrapolate new numbers, but for now let's keep it simple and stick with the raw data. You'd have to revise and neutralize your MLE stats for each level to do that anyway.)

MLE calculations convert these numbers to the following:

358 AB, 101 H, 27 doubles, 3 triples, 7 HR, 32 BB, 92 K

Running that revised line through the Markov equation gives us an average of 5.078 runs per game. Multiply that by 0.9 to get an Equivalent MLB ERA of 4.57. Not bad: Nick Hill's performance last year would have been good enough to be a passable MLB pitcher. And go figure people discussed the possibility of him joining the big club before spring training this year (though obviously that eventually did not happen).

Needless to say, this pitcher would have definitely held his own in Tacoma, whose average Equivalent MLB ERA was 6.36. In fact, you can run the MLE conversions between levels. Hill's Expected PCL ERA would have been a shiny 3.13.

You could do this for every regular hurler at every level of an organization, and you'd have a number that would allow you to rank every pitcher at every level from full season A ball to AAA.