TWISL: Pitcher Fatigue Test

Row color key

Same-owner pairs are highlighted in matching colors below to allow direct managerial comparison between each owner’s Control and Fatigue team. Fatigue % cells are color-coded 0–59% (red) · 60–69% (orange) · 70–79% (yellow) · 80–98% (light green) · 99–100% (green) — the same scheme used throughout the companion Marshall Fatigue Study.

Introduction

Unlike the last post on fatigue, this one is both from a league explicitly testing pitcher fatigue and also features more than one pitcher.

I won’t really be looking at in-game fatigue because overall, it wasn’t much of a factor. I will call out the few cases where in-game fatigue was a factor as a side note, similar to where I’ll call out the one case where appearance fatigue was a factor. Otherwise, the fatigue presented below is strictly pitch-based fatigue.

Aside from showing the overall fatigue from the different fatigue levels and comparing the performances from the fatigue teams to the control teams, I’ll also be answering a few questions that came up during the league about why certain teams were performing certain ways. One of the answers to these questions centers on defense, so I’ve also included a few defensive-oriented metrics to highlight those answers. Another answer centers on managerial settings, so I’ve included metrics to highlight those answers.

There were three owners with both a control and fatigue team. For comparison, each pair is highlighted in matching colors in the tables below to easily see the managerial differences and effects.

League setup

Pitching. All Group A (Control) teams and Group B (Test) teams roster identical pitching staffs with one exception: the Test teams remove 1994 Brett Saberhagen from their roster. This leaves all Control teams with 1,355 IP and all Test teams with 1,100 IP. Control teams must employ a 4-man rotation of Maddux, Saberhagen, Halladay, and Lee; Test teams must employ a 3-man rotation of Maddux, Halladay, and Lee. In order to combat early season fatigue, owners may start whomever they like the first two games, but rotations need to be set after that. For Control teams, autorest target is 90–95%; for Test teams, 70–75%. Any micromanaging or bullpen setup is up to the individual owner — if you want to give a reliever a spot start here and there to keep your guys from fatiguing too quickly, you can do so, just please consider the theme goals as you do.

Roster card for all teams:

Offense. This is where it gets really interesting. You get to select whatever offense you want. Test teams will have an additional $7,875,590 due to cutting Saberhagen. The only rule is that you may not add another pitcher to your roster. Improve your offense however you like, but don’t change the pitching staff.

The data below is only for the other three starters (Maddux, Halladay, Lee) that were in every team in the league. I make points about bullpen use and highlight where we can see that in the data below, but I did not pull each game log for every pitcher in the league.

Full team stats — offense

Defense

Next, the defense — this will help explain how some of the fatigue teams survived and/or thrived, or how some didn’t:

Pitching — basic

Finally, the pitching stats — I’ve included both basic and advanced here as the advanced data gets used for some of the other metrics later:

Pitching — advanced (full 16 teams)

Combined: all three starters, by team

Individual pitcher data: 1996 Greg Maddux

Individual pitcher data: 2011 Roy Halladay

Per-team analysis: bullpen and managerial differences

A couple of other fatigue teams dealt with bullpen management issues. One (70% of the Time…) was the other team to have a pitcher appear in more than 70% of team games and who had the second lowest share of pitches by the starters at 50.3%, but was the one that publicly acknowledged they were making sure their starters stayed above 70% to provide valuable data and letting their bullpen take the brunt of fatigue. Their starters slashed 4.76/.294/1.45 and yet their overall slash was 7.64/.332/1.85. Which goes to show how painful their late innings must’ve been to watch.

Another that appeared to also deal with bullpen issues was Tested Fatigue, whose starters alone were almost identical to the one fatigue team that was unarguably successful (TWISL - Fatigue), but their overall slash was 25-50% worse. Tested Fatigue also saw its starters share of pitches among the lowest in the league.

On the flip side, the two teams that spiraled (Swing in’ 60’s/I Need A Nap and Chicks dig the long ball) had the highest share of starter pitches and they were the only two teams that dealt with in-game fatigue, often having their starters throw beyond their allotted pitches per game, which only served to increase the rate of fatigue these pitchers encountered.

With fatigue teams, managing work load across the three fatigue types is also crucial to success, as you can see from the results of the teams that treaded into appearance or in-game fatigue on top of their pitch-based fatigue.

Observations

Two Cliff Lee with least IP combined for 81 G, 476 IP, .271 OAV, 1.2457 WHIP, and 4.56 RA/9. Two Cliff Lee with most IP combined for 104 G, 581.67 IP, .272 OAV, 1.2464 WHIP, and 4.75 RA/9. Average fatigue of two with least IP was 100%, while average fatigue of two with most was 74%. That’s 23 extra games and 105 extra IP with no real drop in performance.

The three sets of teams run by a single owner were each nearly identical on offense despite each fatigue version having a top 3 defense and two control versions having a bottom 3 defense and the third a league average defense. The TWISL Fatigue team was also almost identical to the TWISL Control team on pitching, thanks to that extra defense.

The 85-89% bucket, though the smallest sample, isn’t necessarily an aberration — the better stats there over the 90-94% bucket could be explained by defense in that sample since these buckets aren’t controlled by defense or matchups, etc.

With fatigue teams, managing work load across the three fatigue types is also crucial to success, as you can see from the results of the teams that treaded into appearance or in-game fatigue on top of their pitch-based fatigue.

Even though this is linked in the other thread, this particular point about the multiple types of pitching fatigue and how they all interplay is important to think about in terms of how fatigue impacts performance.

Notes on the data

Couple of notes on the above:

When I first started compiling the data there were two teams that clearly had spiraled out of control fatigue-wise and skewed the overall numbers significantly. So you’ll notice on the individual pitchers sections I have a sub-total for the fatigue teams that does not include those two teams labeled “F%-2”. You’ll notice that for the section that combines all three pitchers together that sub-total is now called “F%-3” because as I dug into the data it became clear that there was in fact a third team that had spiraled into out of control. So I also removed them from this data. It was harder to pick up on the individual pitcher sections because of the more limited data, though it did still stand out as an outlier for reasons I’ll get into in the next post.

The made up metrics to the right of each pitcher’s data are the number of games that pitcher pitched below the defined thresholds of 70% and 99%. That column is labeled “G<70/99%”. For those specific games below the threshold, the average fatigue % is displayed in the column labeled “AVG G<70/99%”. Then the lowest fatigue level that pitcher appeared in a game at is in the column labeled “Low %”. Next we have the more straightforward “AVG PC” which is how many pitches that pitcher threw per game — and I highlighted in red the few cases where in-game fatigue played a role.

Legend / key for advanced columns

Starter % — Percentage of pitches thrown per game by starters. Pitchers/Game — Number of pitchers to appear per team game. Defensive XO — Net outs added by defense (includes errors, +/− plays, CS, and double plays). XO/OPP — Extra outs added by defense per opportunity. OAV-+ — OAV with + plays removed. %>RL# OAV — The increase to OAV over the RL# OAV for these pitchers. %>RL# OAV-+ — The increase to OAV with + plays removed over RL# for these pitchers. %>RL# BB/9 — The increase to BB/9 over the RL# for these pitchers. %>RL# HR/9 — The increase to HR/9 over the RL# for these pitchers.

League questions

Question 1: If defense is as valuable as postulated, why with the league’s best defense did “Fatigue is Just a State of Mind” not do better?

In short, defense saved this team from a complete and total out of control death spiral. This is the third team excluded from the combined stats above because it was severely fatigued and so much so that it was skewing the average fatigue numbers so far offline.

Defining success for fatigue teams

There are multiple ways to define success from the fatigue teams here:

First, is by W/L record, of which only two of the fatigue teams managed better than .500. However, a team’s run scoring capabilities has as much to do with winning as their run prevention, and there were a few fatigue teams that performed well enough on the prevention side to have won more games, but didn’t score enough to turn that performance into a winning record.

Second, is by the overall pitching performance relative to expectations (either RL, RL#, the control teams, or some other standard). By this measure, only one fatigue team appears to have succeeded with this test.

Third, is how just the starting pitchers performed relative to expectations (either RL, RL#, the control teams, or some other standard) given that at least one team let their bullpen go to make sure their starters continued to provide useful data (and from the data, intentional or not, at least two other fatigue teams’ bullpen was their weakness that kept them from winning more games). Again, by this measure, only 1-4 fatigue teams appear to have succeeded with this test (and that depends on your standard above).

Given that, depending on how you define success for a fatigue team here, at best half the teams were successful and at worst, only one was.

That said, there is a clear importance in building a fatigue team with the first goal of winning games in mind to be successful, and that includes building a team that can score enough runs, have a defense oriented to overcoming the effects of fatigue as your pitchers tire, and drafting the right kind of pitchers to begin with. Your ballpark selection also goes a long way towards helping your team.

With 1,100 total innings and WIS Park, the pitchers were the right type — low-mid OAV, low BB, low HR pitchers. Most fatigue teams put their team in front of better than average defenses, as can be seen by the 121 net extra outs from the fatigue teams as opposed to the net 36 extra outs from the control teams.

Conclusion

I think it is still quite clear, despite a lack of overall “success” depending on how you’re defining it from above, that fatigue teams can be used successfully (using the same definition). There is definitely a balance that must be obtained between the different types of fatigue, a defense that helps control for hits (+ plays and lack of − minus plays) and helps minimize run scoring opportunities (CS, few errors, DPs, etc), a ballpark that also helps control for hits (-1B) and HR (-HR). However, if you can do the above, you can win because your pitching will be at the top of the league with average offense, or you’ll have average pitching with an offense near the top of the league.

I included the RL, RL#, and control group as ways to compare performance relative to each other, as well as at this particular cap level. At this cap ($85m) we weren’t so far removed from realistic in that we still had a team have their starters cumulatively exceed their RL and RL# stats. However, we were far enough removed that the average control team was roughly 10% worse on OAV, 27% worse on BB/9, and 94% worse on HR/9 than these same pitchers’ RL#. Comparatively, relative to the control teams, the best fatigue team was within the range of the control teams (16%, 47%, 277%), but as a group (F%-3) the fatigue teams were 12% worse than control on OAV, 27% worse on BB/9, and 148% worse on HR/9. You can see this to an extent on the fatigue groups above, and it helps gauge the linear effect of fatigue on these three stats — and, as covered above, how it can be controlled with defense, ballpark, and managerial settings.

I’d love to do this test again, but with the owners now all having a better understanding of the inter-play at work.