Tue Aug 29, 2017 4:33 am

A few words,

1-I have made some adjustments in my model. The WAR value of first basemen and outfielder have been improved compared to the other positions, because of a change in the way I determined the replacement value for these positions. Jimmie Fox now stands with the 6th best overall card. I think this change is a rejoinder with the opinion of vets. I will change the first post at the end of the week: if Foxx is not overall 6th, you have the old version.

2-I have found out and corrected a few mistakes in the diamonddope download sheet. Maddux for one is highly affected: DD listed him with only 2.05 on-base vs lh, when in fact he allows 6.05 on-base vs lh. This mistake moves him out of the first round selection. Killebrew is another one whose data was way off. I had already corrected the obvious mistakes (some players had double digits in single or homerun chances). If you are aware of other mistakes in DD, please send it to me.

3-I also made a few adjustment about WAR value for pitchers, after analyzing 22 200M seasons that were playded in 2015 and 2016 Barns tournament.

This allows me to write a few words about how WAR is estimated for pitchers. If you are quickly bored with methodology, skip the rest of this post!!!

Setting the value of pitchers in WAR is even more difficult than for offensive players.

First, my model projects an ERA and estimates how many innings were pitched for every pitcher, and then I project how many runs will be saved by using this pitcher relative to a replacement value. So basically, there are two issues: 1) am I estimating the value of pitchers correctly? 2) Am I setting the correct replacement value (for recall, WAR is about wins over replacement value, so the model needs that I establish a replacement value, but it also make sense from a GM perspective because it's akin to estimating the value of players you draft in early rounds or high on your drafing list relative to players you can get in late rounds or that you get by default in the automatic draft).

To answer the first question and check if my model was valid, I needed data so I downloaded all the 200M leagues from the 2015 and 2016 Barns tours. This amounts to 22 seasons. To simplify the discussion, I will compare my model to the data from the 22 200M seasons mostly for right-handed starting SP who start on 3-day rest--lefties and 4-day rest pitchers are more complicated to estimate---I further exclude Ford, Arrieta and Greinke among a few others from the discussion since these cards were introduced recently and I had no data for them from the 22 seasons I downloaded.

To give an example, based on a 9/9 single and 12/12 hr environment,my model estimates that Maddux ERA** is expected to be on average 5.32 (and NOT 5.09, as it was initially written in the first post, after correction of the DD error) and to throw 275 innings in a course of a full season of 200M leagues (my model did assume only 40 starts though, for ease of simplicity). 275 innings might seem low, but I thought that there are more early exits in 200M leagues, and there are usually at least one very good reliever in every team to pick up when Maddux gets fatigued later in the game.

Maddux' performance from the 22 simulated seasons on average was 40.8 starts, 294 innings, and an ERA** of 5.34. So my model nails the real ERA** but indeed underestimated a bit the number of innings (for the record, I also checked the stadium ratings, and it was indeed within units of 9/9 and 12/12 for singles and homeruns respectively for home fields of teams owning Maddux).

There are of course expected variations from one pitcher to another, so I regrouped the pool of rhp *SP pitchers into 4 categories based on the ERA estimated in my model, but I choose the delimitation also so that it broadly respects the SOM salary structure:

i) Maddux, the 11M guy and the best predicted ERA among *SP

ii) The high 10M group which includes W. Johnson, Alexander, the 1997 Pedro card, Walsh, Mathewson;

iii) The low 10M group: Joss, Tiant, 3-finger Brown, Sutton, to whom I added K. Brown and Dizzy Vance because my model expects them to have similar ERA;

iv) The rest of the pack

For the high 10M group, my model also nailed the ERA** (on average, projected 5.74 vs simulated 5.67), but also underestimated usage by roughly 20 innings, just like it did for Maddux.

For the low 10M group, my model again nailed the ERA** (on average, projected 6.17 vs simulated 6.14), but for this category, the usage in terms of innings pitched was also perfectly estimated by my model.

As for the rest of the pack. My model was also very close to reality. The three best pitchers from the rest of the pack as defined above are Clemens (the 9.53 card), Paige and Marichal. In the 22 simulated seasons, Marichal, at nearly 5.90, was the best, but he could be perceived as an outlier since his use is severely limited to low-homerun environment. Clemens had the best performance after Marichal, but Paige was a bit off, although his performance was surprisingly based on only 14 seasons.In terms of usage, my model was very close to reality. Many pitchers had 37-38 starts so cases were my model underestimated the contribution of a pitcher was very rare.

My model was not as precise for pitchers who need 4 days of rest, but this was expected since they are more likely to be spot starters. Hecker's ERA was probably the most overrated by my model, perhaps this could be because I underestimated the impact of his poor hold rating. Something to explore. As expected, lefty pitchers had, dollar for dollar, worst ERA than rhp. I estimate the difference to be 0.3

For the second question about how to determine WAR replacement for pitchers, I first had to determine a pool of pitchers that fit the definition. If the 11 other teams fulfills their starting rotation, say two teams go with a 5-men rotation for the 10 or so valuable 4-day rest pitchers, and the 9 other teams choose the 36 best cards to create 4-men rotation, then one could argue that the pool of pitched ranked between 46th and 50th or so in my model could be considered truly replacement players.If I limit again to rhp *SP, the 4 best SP after rank 46 in my model are Radborne, Dean Chance, Bradley, and Taylor. None of them went drafted in more than 16 seasons, which is a sign that they are regularly left out of starting rotation. Their combined ERA in the simulated seasons were 6.60, so this is the value I took as replacement.

**I write about ERA, but technically, both in my model and in the observed data, I include both earned and unearned runs in the calculation, so my numbers will be a bit higher than if you specifically look at your player's ERA as provided by SOM stats.