THE BOOK--Playing The Percentages In Baseball

I’m a young’un so by the time I could get my baseball game fix my game of choice was Ken Griffey Jr. Baseball for the SNES. I always thought with computer programs like OOTP and Diamond Mind out now these things were all but gone, besides the Strat-O-Matic leagues still floating around.

That said my dad still has a APBA in his closet that has always piqued my interest from time to time, maybe it’s time to bust it out.

Yuck on the Baseball Classics cards.  Comic Sans FTL.

I’m an agnostic on the issue, but the Baseball Classics game was the subject of one of the most heated discussions you’ll ever read on the Tabletop Sports forum at Delphi Forums.  I know, that sounds like a dust-up at the quilting bee, but still…

I played APBA religiously for as a pre-teen and into my teen years, and I learned a ton about baseball playing the game.  This was in the 80’s.  I started out over-valuing batting average until I began playing APBA and noticed how the unsung high walk, high power guys (Ken Phelps, for example) were good to have in the lineup.  Also, when the new cards came out each year, I’d go through them and see some amazing hitter’s cards for players I’d never heard of.  I vividly remember the 1983 Doug Frobel card (.283/.328/.533) with all of his power numbers.  Of course, it was in just 60 at-bats, so I learned all about “small sample size” when he hit .203/.271/.388 the following year.  It was great growing up in the 80’s with APBA, the Bill James Abstracts, and the variety of ways teams put together their offenses during that time.

Luke/#3, can you put a link up to that discussion?

As for my experiences with baseball board games, I’m kind of like Graham in that by the time I was old enough to understand them, Nintendo and RBI Baseball was around captivating our imaginations. Still, I played some All-Star Baseball, and could easily figure out those who had bigger spaces for the on-base events were better. Later, I picked up Strat-O-Matic with the 2003 season cards, and while I haven’t gone through and simulated baseball history, I still have fun playing it from time to time and enjoy the depth. I’ve wanted to try APBA to compare systems, but have yet to pick up a copy. As for Baseball Classics, the matrix system looks like it would be interesting to compare against those APBA and Strat-O-Matic.

Luke/#3, can you put a link up to that discussion?

The first post in that discussion was at:

http://forums.delphiforums.com/tabletopsports/messages?msg=6798.1

It was followed by 288 posts, until the thread was shut down.

Some folks thought the game maker, who posted in the thread, was evasive about a flaw in the game.  Others thought his critics were unnecessarily rude and condescending.

Apparently, the game flaw at issue has been addressed by the game maker.  But you will still find a thread opening up here and there on the Tabletop forum rehashing the old debate and discussing the revised version of the game.

Me, I play Diamond Mind, and mostly the projection disk, so I sat out of the Tabletop Wars.

As a participant in the forum debate and a sometime poster here, I will summarize:

Baseball Classics is a “50-50” game; that is, half the results come from the batter’s card and half from the pitcher’s card (just like Strat-o-Matic).  But, when it was originally released, the cards did not take this into account.  To use a simple example, if the league average BA is .250 and you have a batter who hit .300, the results off his card should produce an average of .350—that way, when it’s averaged with the pitchers, it will produce .300.  Baseball Classics didn’t do that—the results off the batter’s card yielded .300—thus, he’d be expected to hit only .275.  The same was true for all results (HR rate, etc.)

Apparently that flaw (pretty basic) has been fixed now.

There are a lot of good tabletop games out there—my personal favorite is Replay Baseball, which I cannot recommend highly enough in its balance between quick play and realism.

As a tabletop game player, I tend to think about many of the subjects discussed here in terms of how to “card” the player—e.g., use actual or standard platoon splits?  have pitchers control opponents’ BABIP? etc.

Craig: that is such a basic flaw, that once the manufacturer had it pointed out to him should have simply said “oops… big goof… I’ll fix asap”.

Since I can’t access that forum from the office, I can only surmise that this didn’t happen quickly enough?

When folks first questioned the cards, the designer denied there was a problem.  It may have been a natural, if unfortunate, reaction to being called out on an error.  I understand that the problem has now been fixed (although I haven’t verified this, and am not sure if the fix was made for all of the varieties of the game).

The episode did leave many with a bad taste in their mouths.  Given that the game’s promotional material seeks to show that it’s better than Strat-o-Matic, the fact that it was (initially) much less accurate than Strat was a bit of a problem.

It sure made for a heated discussion though!  Nothing seems to get the juices flowing like debates over which games are the best/worst!!

I’ve been playing tabletop baseball games since I was about 10 years old (I’m 40 now).  Have played the majority of them - APBA, Strat, Statis-Pro, Dynasty, Replay, Ballpark, etc.  Most of them have some kind of 50/50 model (though not all) as all were created years ago, before data like Retrosheet was available.  Only a couple years ago did I discover what I consider to be tabletop baseball nirvana - “Inside Pitch Baseball”.  The thing that makes it different is that it’s based on DIPS - so pitchers generally influence strikeouts, walks and homeruns allowed while batters influence BABIP and where the ball is hit (though pitchers can influence groundball vs. flyball).  Anyway, the game is fluid, accurate, and feels more like real baseball (again, due to modern research unavailable to other games at the time) than any other game I’ve played.  Check it out at insidesportsgames.com.  Note:  I’m not affiliated with them in any way - just a fan of the game.

This comment should probably be in a different thread, but I noticed Diamond Mind is now trying an online-based system, that’s free at least for the beta-testing stage…

http://www.imaginesports.com/imagine/baseball/discovery/home/dm

Not a tabletop game, but did anyone play the Lance Haffner stats based computer sports games? I seem to be the only one to remember them and spent countless hours replaying seasons on my C64.

Hi ALL!

We’ve just launched a revolutionary strategy baseball board game, ULTIMATE BASEBALL THE GAME.  We hope you’ll give us a try!
home.ultimatebaseballthegame.com

(Previous post indefinitely closed, pending verification.)

Dial it down, if you want me to not treat your post as spam.  Why don’t you explain what it is your board game does?

If you are a human being and not a spam-a-bot, I’ll keep your post up.

We have been researching and developing our game, ULTIMATE BASEBALL THE GAME, for over 20 years, and we believe we have come up with a strategy baseball board game that is revolutionary.  It plays pitch-by-pitch and affords gamers much more opportunity for strategy on nearly every pitch (vs resolving at bats all at once with die rolls, as most replay-type simulations do).  This thread piqued our interest, as respondents are commenting on the intricacies and realism of these games.  We are amazed at how ULTIMATE BASEBALL THE GAME affords gamers so many strategic choices, yet the results turn out to be so true to real life (Ted Williams really plays like he did in real life...both offensively and defensively).  As to the question of certain skills possibly leveling off slightly lower than real life, at least in our game, that is because our player pool consists of the top players of all time (240 come in the game box, and we have rated 4,000 more in our UBTG PLAYER REGISTER), meaning that Walter Johnson’s ERA might end up slightly higher because he’s having to pitch mainly to folks like Cobb and Hornsby (which, of course, he did not have to do in real life).  We would love for respondents on this thread and the authors of THE BOOK to try us out and comment on this thread:
http://www.home.ultimatebaseballthegame.com

Thanks!

Dude, PLEASE, dial it down.  I’m skipping over half your sentences because I can’t stand the marketing.

Talk to me like I’m your buddy, and not some salesman.

Take three…

As an example, explain what you do that is as good or better than Tippett’s Diamond Mind.  And explain what DMB does that yours doesn’t.

Be honest.

Hi Tom,

I am a fan of your book and research. I started out working with John Dewan when STATS Inc. first started in what was the infancy of sabermetrics. I also was the person that brought Tom Tippet aboard (at the Chicago SABR convention) to market his game under the Pursue the Pennant brand.

Much of the same metrics in your book has gone into the Computer Manager profile of the new DYNASTY League Baseball Online version. DYNASTY League Baseball (there is also a board version) is the successor to the orginal Pursue the Pennant board game which I am the designer of.

What makes the new DYNASTY League Baseball Online version so unique is that it is the first and only real time multi-player Baseball simulation game. Stats, standings and leaders are all updated in real time without having to export or import files.  There also is a new live draft room coming this Spring for draft leagues that updates bids and budgets - a first for baseball simulation games and a real boon for draft leagues.

There is a free trial at:

DYNASTYLeagueBaseball.com

- Mike

Looks interesting.

I’m interested in how the game designers choose to model each season.  Do you treat each season as the “true talent”, and therefore, if a career .350 wOBA hitter happens to hit .425, then you use .425 as the true rate?  The idea therefore is to try to replicate that particular season?

Or, do you use Bayes/regression, and treat that player as being say a .375 hitter, who happened to hit .425 that year, and so, you model him to hit .375?

As for game players: what would you prefer?

The implication is that if you treat .425 as the true, then this player will hit .350 to .500 if you simulate.

On the other hand, if you treat .375 as the true, then when you simulate, he’ll hit .300 to .450.

What do game player prefer?

Tango, I’ve never played board games, but started with Microleague on Apple 2e (then ML2, then ML4), and really got into the sweet spot with Earl Weaver Baseball on IBM. Frontpage Sports Baseball Pro was a big jump in simulation (especially ‘98) and then HighHeat really brought it with 6 levels of MiLB and great gameplay with good sim results ... then MVP 2005 took over and could still be considered co-King with The Show.

Anyway, as one that made/makes rosters (player ratings from stats), I ALWAYS prefer true talent.

It’s one of the reasons why I don’t like “all-Time great teams” using a player’s best season. You get stuff like Lajoie with a .420 average, Gibson’s 1.12, and well Bonds season would break the formula used to create the ratings. I think his power rating (on a scale of 1 to 100) on MVP would have been around 137. *grin* My preference would be to use their 162g average as base stats to plug into the formulas, but only because it’s easier than using a player’s 5 best seasons.

True Talent, especially when using current players, allows one to “sort of” play the next/current season. In other words if I want to play the 2012 season on MLB12, then using True Talent will give me the best chance of that ... rather than just “replaying 2011” with the 2012 lineups.

True talent, for me.

I play Dynasty. The game uses the in season stats as opposed to true talent. I prefer it that way except for one thing....I want the splits regressed! As it stands, if you have a guy who goes 2 for 5 with 2 HR vs LHP, he becomes the mother of all players. The game tries to put a usage limit on those players, but that does not work in the context of my league.

One other issue with the game is the fielding ratings for OF. There are a ton of guys who have the same rating in CF as they do on the corners. What this means is CF isn’t a premium position. In fact, there are so many LHH in my league that RF is actually the more important OF position.

@19: I’d prefer an option. A quick explanation of what I’d want from a season (I’ll frame things in DMB terms when need be, since that’s what I play and it’s still the standard for accuracy):

* A standard year, where hitting .330 means a player is a .330 hitter. There would need to be a small amount of regression even for this, unless you’re handcrafting cards (not a good thing for a versatile creation system like DMB, where you plug in the stats and play), since a .330 hitter probably hit in a few more favorable situations than a .240 hitter; I’d want to use batting/pitching hand, age, height, weight, playing time, start/relief and the other statistics (what does hitting singles tell us about how often he walks?) to do this, and maybe some additional to let platoon splits regress to a normal split a little more than they would otherwise in this system. Stuff represented by ratings (in DMB, this is for bunting, running, stealing and fielding, among other things) would start from multi-year stats but then be tweaked based on running multiple years to play test.
* A version based on multi-year data, to get the real skill of the player. This would have full regression, but all players would get the same playing time. Ratings would be straight multi-year.

Thus, if I replay 1986, I have two options: one where I use just 1986 and the Mets will win on their real stats, or one based on 1984-1988 and the Mets will likely win again based on this.

Do we, or do we not, have a problem that Barry Bonds could hit 82 HR, when doing a replay of his dramatic year?

Or, do we simply want to acknowledge that there was a much greater than 50/50 chance that everything went his way to allow him to hit 73 HR? And so, we’d want to model that year as him being a true 60 HR hitter, and replaying that year means he’d hit 45-75 HR?

That’s been a big point of debate for years, and people have different opinions on it.

As a player and a designer, I believe that we should use projected/regressed stats in order to give the player the option to use players as they want. However, this is generally more difficult to do achieve. You can say that Mauer has generally been a .400 hitter over his career, and his ups and downs are mostly variance.

However, there are other people who want faithful recreations. So, if they run the season where Mauer has a great year and hits .440, they expect him to hit .440 again that year, with no variance. Some engines will allow for this, by giving the option to adjust stats to counter variance (if he’s slumping early, they’ll boost the ratings so he’ll finish around .440). Some of these people will also attempt to recreate the managerial decisions of that year, and use players in a similar way, even if it isn’t optimal.

I believe the former makes for a better simulation engine, but some people are still in their old ways, and as a designer, you either need to drop one to support the other, or support both, which is essentially what 22/Charles said above.

In that scenario, you could download a 2011 ‘replay’ season, where Alonso was mostly a PH/4C guy and can’t play more than about 100 PA. Or, you can download a 2011 regressed season where the manager can choose how they use Alonso, but he isn’t likely to be a .400 hitter again.

Not only do I not have a problem with Bonds potentially hitting 82 HR, I prefer it. When I look at the stats during the season, I know if a guy has 70 HR he’s going to hit between 55-85 for me. I don’t need a rocket science degree to know how good my teams are going to be before seeing the cards. It also creates a bigger spread in talent, which is huge since we’re only dealing with MLB talent in my league. It also raises the fluke factor, which very much helps to keep things interesting on draft day.

There’s what we know and prefer: Bonds was lucky, and would not hit 72 HR in a replay most of the time. When he does, it makes it special.

We’re in the minority. Folks replaying 2001 will feel cheated if Bonds hits 53 HRs and thing the simulation engine is broken. Hence I want both options.

I don’t think anyone wants Bonds hitting 80+ HR in particular…

The straight replay people will want him to hit as close to 73 as possible, while people like us would rather a spread.

Given 73 seems pretty fluky already, 80 would be pretty unlikely. Just making up numbers, 50±10 would be a reasonable estimate, so 73 HR is probably 96th percentile, while 80 HR is around 99.7.

The main concern is when the rating is 73±10, where getting 80+ HR is decently likely (>20%). I imagine most would agree that isn’t wanted.

For comparison, a strict replay person might expect 73±2, they want to recreate the 73, where both 65 and 80 would be considered failures of the sim.

One nice thing about Dynasty League is there is no question about the sim engine. Once you have the card in front of you you can see player A has 30 HR on his card and player B has 50, you know which one is going to hit more HR over the long run. If player A hits more HR than player B, you know without a doubt it was just randomness.

I agree that the “replay” gamers are expecting 73 +/- 2.

It’s interesting there’s a game on the market that tries to force that in.

The problem with the “replay” scenario is that they often just look at “extremes” and only want those replayed.

No one looks at the 5th starter and expects him to reproduce his 4.65 ERA, but people want George Brett to hit .390, Williams to hit .406, Ryan Howard to have 56 HR and so on.

But, we’re not acknowledging that these achievements were often flukey in their BABIP, HR/FB, etc. What if the 2009 version of Mauer in our simulation experiences the same flukey HR/FB % that he expereinced during the season? Now, we’ve got a “28 HR” Mauer experiencing an increased HR/FB% and now he’s hitting 35-40 HR, which to me would comepltely ruin any sense of realism in the game.

IMO, what we want most is “realism”. That’s what I go for. And “realism” states that there’s a range of performance when “replaying” any season or using projections or “pre-play” the upcoming season.

Basically having Bonds true talent be at 73 HR may result in him having more than that. If a “true talent 45 HR” Bonds hit 73 HR, then what would a true talent (stats, ratings) 73 HR Bonds hit if he experienced any luck?

For The Show, I ran a lot of simulations on OOTP and created a fictional “universe”, a league full of fake name players using ratings that came from the “fake” simulations (3-year average stats fed into formulas used to compute ratings) on OOTP. Now, we can play seasons with players with realistic stats and not have to “judge” the players based on how they do to their real-life counterpart. You also don;t have to worry about keeping up with the real MLB season and making the same callups/downs, trades, in the season.

When you use “real players” the tendency is too look at everything and think “oh, this isn’t realistic” because Player X has 6 fewer HR than he had the season before, and pitcher Y is only 11-14 and he was 18-7 with “real stats” and things of that nature.

I think it’s ridiculous to expect a simulation to produce almost identical stats to what are used for the ratings. Essentially it’s removed the random aspects that lead to variation and make everything “pre-determined” in order to achieve a very specific number.

I’m interested in the various viewpoints, because obviously not everyone shares the same views, or even possibly even considered the various options.

So, we have three options:

1. Bonds has a true talent 73 HR

2. Bonds has a predetermined 73 HR

3. Bonds has a true talent 50-ish HR, and with good fortune (maybe in 10% of sims?) will hit 70-75 HR

In #1, it means he’ll hit 73 HR +/- 15 or 20

In #2, it means he’ll hit 73 HR +/- 2 or 3

In #3, it means he’ll hit 40-75 HR

So, what kind of a gamer you are will decide what you want.

It also depends on what type of “game” you’re playing.

If you’re replaying the 73 HR season, then you probably want to relive the magic of the 73 HR and would expect Bonds to do that.

If you’re using the 73 HR stats to play the following season (i.e., play the 2012 with the 2011 stats), then you’re probably more flexible on the range.

If you’re dumping all of the players into a draft pool and drafting players among a collection of owners to play the upcoming season’s schedule (or a league-created schedule), then you’re probably more likely to want a “true talent” type of thing, where Bonds’s true talent is around 50 HR).

It also could affect how the human players play/manager the game. If you “know” Bonds is going to hit right around 73 HR (and you know his current sim-season stats), then you’d basically walk him any time a HR could hurt you, and many would consider that “cheesing”.

In other words, if you know that the game ensures that players end up very close to their last season stats, and you see Lincecum is pitching against you and he’s performing better than his stats say he should, you may have a very good indication that the game is going to “normalize” his current stats to reflect his last season stats by having a bad performance in his next start.

I’m not sure that a game *could* even “guarantee” that Bonds would hit between 70-75 HR, even if you re-created the exact same season with the same pitchers faced, etc. The game would basically need to be able to reproduce everyone’s stats very accurately, and I don;t think that is possible.

EXCELLENT point!  You are right.. even if we tried to “fit” it for Bonds to get to 73HR, how close can we get it (because we do the same for the opponents).

@31: I desperately want to avoid #2. That isn’t a good game design. If random chance has Bonds with 12 HRs at the All-Star Break, someone is going to notice this and try to game the game to give Bonds even more ABs, as CC said in #32, like by batting him second. Which will give another set of impossible stats.

The options I want are:

#3

and

#1, with enough regression to account for otherwise unaccountable luck (again, a .290 hitter likely had a bit more luck than a .240 hitter, so there needs to be a bit of regression so they don’t come out to .292 and .238 over many simmed years) and to keep folks from gaming the game. Even without playing option #3, a game needs to encourage a player to use baseball players in the way they were actually used in real life (if replaying a season), within reason.

More than a choice between #1 and #3, I want to be playing something like an actual baseball game, not some way of playing a trick on the lords of MLB. Sure, there are what-ifs which are fun, like having Rogers Hornsby play for the 1987 Blue Jays to see if that puts them over the top, but replaying the 1991 Expos and letting Bret Barberie bat leadoff for the whole year so he can hit .350 feels cheap.

@RW

Not sure why the DYNASTY League Baseball Use ratings don’t work for your league because they do limit both the number of platoon AB’s and time on the roster to the real life numbers.  You can’t overuse the player.

As far as Range ratings, each OF is rated independently for his range, and error rating in the various OF positions and sometimes the player grades out based on his various defensive metrics as having the same range in CF as he does in LF or RF.