PING: Single-Season Historical Replay-ers

[pstrickert]

I have a feeling that more people than we might imagine prefer single-season historical replays than other types of gameplay. Though OOTP does not incorporate the painstaking detail of other baseball text sims (e.g., DMB, Strat, APBA, etc.), it does, IMO, produce highly-credible statistical results. I also think that -- with a few more options -- OOTP would be a viable choice for people who prefer single-season replays. What do you all think? What options would improve OOTP from the single-season replay perspective?

I'll offer a couple of ideas . . .

1. an option to control player usage
2. an option to use (available) historical transaction files
3. an option to use (available) as-played lineups
4. an option to use (available) most common lineups

The first option is, perhaps, the most important. In OOTP, the AI has the mystical ability to see the future. Whereas IRL, the manager might have waited patiently for a player to produce (and only later took him out of the lineup), the AI manager "knows" which of the two players produced the best results. The AI manager, therefore, does not have to wait and see, before replacing the original starter. So, in an OOTP replay, the real life starter may see limited action, while the real life bench player may see a lot more action. (Who's to say, though, that IRL the bench player would have produced the same numbers if his manager had made him the starter for the whole season? The AI manager, of course, can assume he will. But that's not how it works IRL.)

[SteveP]

I'll put in my two cents, just to say that you have an interesting idea about lineups. I can readily appreciate the difficulty of programming an AI to formulate a sensible lineup, and so I accept that sometimes it makes some eccentric decisions. And I have also accepted that any similarity between lineups in OOTP and what really happened is partly happenstance. Moreover -- and I think this is important -- the AI may sometimes have a better approach to a lineup than the real manager (partly because the AI doesn't have to deal with player egos, etc.)

That said, having data available about how often players appeared in what slots in the lineup IRL, and providing an option to have the AI use that data as part of the lineup calculation, is very attractive. May even be feasible (but what do I know ). Still, I wouldn't hold my breath.

[Garlon]

If the AI takes the best hitter at each position and puts them in the lineup, it won't matter where they are in the batting order in terms of generating runs for the team. You could take the 1927 Yankees and bat Ruth and Gehrig 1-2 in the order and they'll score just as many runs as if you hit them 3-4. Or you could put your worst hitter at leadoff and still end up with the same offensive output.

Having said that, OOTP does a very good job at formulating the lineups IMO. Basically the 2 best OPS guys will end up 3-4 in the lineup, the next 2 best available OBP guys will hit 1-2, and the remaining 5-8 hitters will be in descending order of projected OPS. That's basically a good way to create a realistic lineup and seems to be what OOTP tries to do.


pstrickert - "Who's to say, though, that IRL the bench player would have produced the same numbers if his manager had made him the starter for the whole season?" That's why in the neutralized DB we make sure that all players have at least 251 AB in each of their season's batting lines. In cases where they didn't have that many IRL we pro-rate the missing AB with the average from their career or their surrounding seasons. Now, when you use 3-yr recalc, the AI is always going to be looking at a minimum of 502+ AB (rookies) and 753+ AB for anyone else. That's a good sample size to make decisions with. Yes, hindsight is 20/20, but the AI should have the same advantage as the human manager.

[pstrickert]

Garlon: I've heard the same thing said about lineups before. It doesn't matter who bats where. I keep thinking: What would Tony LaRussa say about that?

It's not a big deal, I know, but using a "most common lineups" file would prevent the occasionally head-scratcher. I mean, should Darrell Porter, the catcher, be leading off? The AI might think so, but if I'm interested in replaying the 1974 season, I may recoil when I see Porter batting lead off for Milwaukee. It just "seems" right when an OOTP historical lineup matches its real life counterpart. I think that counts for something. It enhances the "immersion" effect. I suppose the same could be said for transaction files and as-played lineups. It would be up the user, though, to decide which, if any, of these options to choose. For some people, it matters; for others, it does not. (Plus, these options might help OOTP attract people who play SOM, APBA, Action, etc.)

[Garlon]

One big problem with using "real" lineups is the fact that in OOTP there are trades and injuries. After a couple trades and injuries there is no more "real" lineup. The AI has to be able to make decisions based on ratings and other forms of logic. Also consider that the "most used" lineup IRL sometimes just doesn't occur that often. I remember that a few years back that the "most used" Cubs lineup occured maybe 8 times the entire season....I don't think that's very meaningful in any way.

Having said that, let's look at the 1974 Brewers and pick out their starting position players based on games played by position and let's look at their PA/G:

Pos: Player - Plate Apearances per Game
C: Porter - 3.740
1B: Scott - 4.253
2B: Garcia - 3.504
3B: Money - 4.459
SS: Yount - 3.402
LF: Briggs - 4.078
CF: Coluccio - 3.261
RF: May - 3.815
DH: Mitchell - 2.205

Guesstimate of typical lineup based on descending order of PA/G:

1. Money
2. Scott
3. Briggs
4. May
5. Porter
6. Garcia
7. Yount
8. Coluccio
9. Mitchell

Maybe you think this is a typical lineup for the 1974, and maybe not. This is an imperfect approach, but I think this system would generally get two things correct - the leadoff man, and the 8th hitter (9th hitter for AL teams). On the 1974 Brewers Mitchell was the DH more than anyone else with 88 games as DH. There is a guy like that on just about every team. That means for h Brewers in the best possible scenario they couldn't hae put the same batting order out there more than 88 times. I think it'd be difficult to find the same batting order even a dozen times that season. You can search thru the game logs on retrosheet if you want. Even with the same 9 players, you have 362880 possible lineup combinations. Then consider the fact that you have bench players and guys being called up and sent down to the minors and such, and managers changing the lineups vs left handed starters or ealing with players who have minor injurues. I think when you actualy go look at the data you will see that there really isn't much of a "real" lineup for any team.

Money was the leadoff man for the 1974 Brewers.

[pstrickert]

Most common lineups: This (again, as an option), I believe, could do a couple of things: (1) help the user quickly devise a credible lineup/batting order; and (2) prevent the occasional "head scratcher" (the AI batting George Scott second would be a perfect example, BTW). The user could "go with it" or simply use it as a point of departure. Perhaps it would be most helpful if the user wanted to set up a quick exhibition game, since it would save time.

Trades and injuries: The user could keep these off, if as-played lineups and transaction files were available. I know it's not for everyone, but MANY people prefer to use these files (if available).

Player usage: I still think this is the most important option. Other single-season replay-type games rely on this feature. It helps to maintain the proper balance between starters and bench players (according to historical usage for a particular season). This might be most helpful to on-line leagues, since it would prevent some managers from over- or under-playing certain players. (Think: Bob Hazel for the '57 Braves.) Come to think of it, this option might also help with relief pitcher usage. Often, in OOTP, I'll see5 or 6 relief pitchers with roughly the same number of appearance at the end of the season. In real life, though, some relief pitchers were used far more than others. (Think: Mike Marshall for the '74 Dodgers.) This option would help the AI allocate IPs to different pitchers. Maybe? :P

[pstrickert]

baseball-reference.com tells us which players appeared most often at each position in the batting order. So, for the '74 Brewers, we could easily create a basic starting lineup based on the number of times each player appeared in positions 1-9 in the batting order:

1. Money - 111
2. Berry - 45, May - 38
3. Scott - 70, Briggs - 56
4. Scott - 82, Briggs - 62
5. Porter - 60
6. Hegan - 33, Porter - 30
7. Coluccio - 57
8. Garcia - 68, Johnson - 39
9. Yount - 66, Garcia - 50

[SteveP]

I'll post on this thread again just to say that I think both you guys are right to some extent.

Yes, the AI does a pretty good job of creating lineups. In fact, I think it does remarkably well considering that it's an AI.

Yes, the AI does some things with lineups, from time to time, that would never happen in real life.

Yes, when the AI does these eccentric things, they sometimes make some sense, and probably should have been considered by real life managers (LaRussa's reputation, notwithstanding ). Otherwise they are just anomalies.

Yes, it sure seems like the game engine could factor in, and give some weight to, data on how often a player started in each slot of a lineup (at the game player's option). I think that would be a great compromise. But maybe that data isn't in the Lahman DB, so it couldn't be done. Don't know, but that might end the debate right there.

[pstrickert]

I'm not sure, but if Markus allowed us to import the files from this site . . .

Mark Miller's Baseball Goodies

. . . we'd be OK. Maybe.

[pstrickert]

Here's an APBA-related PDF file. It describes the tendencies of dozens of different micromanagers. Wouldn't it be great if OOTP allowed us to choose from such an assortment of MMs? I'm especially impressed with the differences in relief pitcher usage. To my knowledge, the strategy settings in OOTP do not allow for this type of variety. (Come on, Markus! If APBA has it, why not OOTP? )

[swampdragon]

I don't think lineups are among the top 10 AI problems. I'm very critical of the in-game AI, but the lineups are OK.