View Single Post
Old 10-15-2019, 01:57 AM   #16
joefromchicago
Hall Of Famer
 
joefromchicago's Avatar
 
Join Date: Jun 2011
Posts: 3,630
My investigations continue.

For this simulation, I chose the 1920 White Sox. The Sox had a very stable pitching staff that year, which is ideal for testing the OOTP pitching model. The Sox staff was also unusual in that it used five starters almost exclusively - Red Faber, Eddie Cicotte, Lefty Williams, Dickey Kerr, and Roy Wilkinson. Only two other pitchers - Shovel Hodge and Joe Kiefer - started for the Sox that year, and they started a total of three games and pitched a total of 24.1 innings. No other team that year used so few starters on a consistent basis, although two clubs - the Giants and the Pirates - came close. The Giants had only five games not started by their top five starters, while the Pirates had eight.

As with the previous sim, I set the league rotation size at six and the rotation mode at "start highest-rested." I set the number of relievers to one, but I also set the number of position players to 15 and the roster limit at 23. With seven men on the pitching staff and 15 position players, that meant that there was one "free" spot on each roster that the AI could fill any way it wanted. It turned out that the AI desperately wanted to fill that spot with another reliever.

The AI initially set up the White Sox rotation with only five starters - Faber, Cicotte, Williams, Kerr, and Wilkinson. That made sense, as the Sox really didn't have anybody else who could start, and it was historically accurate. As noted above, the Giants and Pirates could also have been set up with five-man rotations, but the AI stuck with the strategy pre-sets and gave them six-man rotations. That wasn't surprising, although it was a bit disappointing. What was surprising is that some AI teams routinely carried three or even four relievers during the season. I'm not sure why the AI rigidly adhered to the rotation pre-set but ignored the bullpen-size pre-set. Maybe those teams couldn't fill out their quota of 15 position players, but that's just a guess.

I wanted the rotation to match how the pitchers ranked in terms of their innings pitched, which would have meant a rotation of Faber-Cicotte-Williams-Kerr-Wilkinson. I forgot to make that change, so the initial rotation (set by the AI) had Cicotte first and Faber second. I switched them after the opening game, but it took until about mid-June for the rotation to settle into the order that I wanted. Here are the stats for games started, with real life (RL) listed first and sim second.

Faber RL-39 sim-34
Cicotte RL-35 sim-34
Williams RL-38 sim-33
Kerr RL-27 sim-29
Wilkinson RL-12 sim-23

What we see here is what we saw in my previous simulations: the pitchers at the top of the rotation start too few games, while those at the bottom start too many. But the effect is exaggerated here because there are only five pitchers in the rotation. In real life, the top three starters for the Sox started 73% of the team's games, which I would guess is the top mark in the majors in 1920 but nevertheless not unusual - the Cardinals' top three came in at 72% and Cleveland's top trio clocked in at 71%. Teams would typically rely on three or four pitchers for the vast majority of starts, while the rest of the staff would pick up starts here and there on an as-needed basis. OOTP, on the other hand, just can't recreate that method of doling out starting assignments. It doesn't do a very good job with a six-man rotation, but it seems to do even worse with a five-man rotation. That calls for a closer look...
joefromchicago is offline   Reply With Quote