In the great trilogy of story elements,1 the GM is primarily in charge of plot, and the players are primarily in charge of character. Yes, I'm well aware of the many exceptions and overlaps -- almost all RPGs constrain plot and character choice from Dungeons & Dragons ("your fantasy bad-asses kill monsters and take their stuff") to Unknown Armies ("your obsessed head-cases scrabble for leverage in a Darwinian occult struggle"), and licensed games even moreso; "narrativist" games and heavy-metaplot RPG worlds transfer plot control to the designer, at least initially and in the global case; games like Mortal Coil, Shock, and Diaspora deliberately put major setting-creation dials in the players' hands; and perhaps most importantly of all, every pre-written scenario privileges the designer's plot over the GM's, at least initially and in the local case.
With all those caveats out of the way, setting is still the designer's job,2 and a decade or more of reviewing RPGs has mostly taught me what a loose, iffy job many designers make of it. Sometimes as an artistic choice, sure: Joe Prince's Contenders, for example, would be actively worse if he defined the setting in concrete (as opposed to moral and thematic) terms; sometimes because all the heavy lifting has been done by H.P. Lovecraft or every 'Nam movie ever made. Much of my design work, likewise, has leaned on Captain Kirk or the Templars; to the extent I ever had a vision for presenting setting, it was "try to write really interestingly." And "cheat by using a real-life setting -- historical or ahistorical -- which is both more interesting and easier to lie about." But that's tactics, and what I should have been doing was working on strategy. Or at least ordnance.
It was reading Weapons of the Gods, and seeing how R. Sean Borgstrom monetized the setting of wuxia China that really made me realize I'd been doing it wrong. (Or at least not doing it right.) As I said it in an Out of the Box review no doubt immured behind the Great Thorn Forest that is Gaming Report:
[This is] the greatest innovation in getting your players to literally buy into the setting since, well, ever. Borgstrom breaks up all the necessary bits of the setting (80 pages!) into Lore Sheets; you buy a Lore Sheet for Destiny Points and you get a little game-fiction vignette (by Borgstrom, which makes it good) and a number of little plot hooks that you can buy for more Destiny Points. These may have mechanical benefits, treasure benefits ("You know the location of a surviving Thousand-Sword Statue Warrior"), or story benefits ("You will encounter the reclusive and dangerous Wen Li"). You can devote "down-time" to learning these hooks for free, or pay points and get them up front. Some Lore gives you a bonus on other Lore, and it's as painless a way to teach setting as I've ever seen.You can kind of impute Vampire splatbooks as the precursor of this design, as they provided mechanical (and story) benefits to those willing to read sixty-four page Lore Sheets. But the splatbook currency is gamer time, not character points, so it only rewards those already inclined to read a bunch of setting; it's a setting-focused version of what rsdancey calls "mastery" -- meaning the implied (and eventually mechanical) reward for those gamers who immerse themselves in the ruleset. What Traveller did first, and then Vampire perfected, was to provide a "mastery experience" tied to the setting -- as a purely social reward for Trav players, but as a real benefit in Camarilla LARPs and other metaplot-intensive game experiences for Vampire players. Robin Laws took a parallel step (possibly informed by collaborative-narrative games like Mortal Coil); rather than providing metaplot rewards for "secret knowledge," his Kaiin Players' Guide banishes secret knowledge entirely, presenting a setting book solely from the players' perspective and encouraging GMs to allow player initiative to drive setting exploration; his upcoming The Armitage Files extends that concept to campaign exploration. Tautology as choice: of course you're interested in the setting of the campaign, since you picked it!
A more purely mechanical example is the "life path" character generation model, as seen in ovo in Traveller and blossomed ever fuller in Warhammer FRP, FASA Trek (and our two Trek games, because we weren't idiots), and Burning Wheel. This is the other, supply-side end of Borgstrom's demand-side Lore Sheet model. Sure, the transaction is almost the same, but there's something psychologically different about spending points to learn the setting rather than learning the setting by spending points.
But all of this is only half the equation, the player half. The other half is the GM half: how do you get them to pick up the setting and wield it like a battleaxe? (Or a warhammer.) Gary Gygax gave us the answer. And then he immediately hid it from us. The answer is the Random Encounter Table, or Wandering Monster Table, or Random Dungeon Generator, and all those other wondrous time-killers in the back of the DMG. By stocking those tables, paying some attention to the probabilities, and adding modifiers here and there, you create an immediate, accessible method for GMs to understand your setting in the most visceral way possible: by co-creating it with you. They only have to read the setting bits they've generated, and they have a story and an adventure. This is an almost insanely powerful technology for setting design and presentation, and we've unaccountably left it back in its rudimentary Bronze Age form, like the Antikythera Mechanism.
Why? Because I (and I think virtually every other designer of my generation) fixated instead on Gary's other answer, the exact wrong answer to the problem: Greyhawk. (Or Glorantha.) Don't co-create, hyper-create! Don't leave randomness around in your wilderness hexes -- define the heck out of every hex to start with! Put in weather patterns, and historical chronologies, and elf pantheons!3 This is all well and good, if you're Greg Stafford. But even Greg can't get everyone to read hundreds of pages of setting material. (Except Norwegians, for some reason.)
That's why, for the longest time (and still), my fundamental setting design policy was: "Use Earth." It's better mapped, better documented, and just plain weirder than anywhere else. At least start with Earth. But more importantly, as I've said on half a hundred panels and plenty of times in print, saying "Kragar the Liberator was secretly in the pay of the drow" is just not compelling. Nobody really cares, even if they dutifully read the forty pages on Kragar the Liberator earlier in the book. But saying "Abraham Lincoln was secretly in the pay of the drow" is compelling. The players (and GM) bring something to the table when I say "Abraham Lincoln" or "King Arthur" or "Hitler" that they don't when I say "Kragar the Liberator" or "Kragar the Lost" or "Kragar the Mad." Now, forty pages on Abraham Lincoln runs the risk of turning those people right back off again -- I still don't have a sure-fire method of presenting even real setting history in games. But the technology exists!
I ripped off Scott Glancy's terrific, punchy setting design format (Short Form: "Stereotypes and Examples") in d20 Call of Cthulhu for Grim War, and Greg Stafford (him again!) has shown us all how to write four hundred pages of compelling -- and more importantly, game-useful -- setting history with The Great Pendragon Campaign. Those two (or three, if you count Grim War) books are examples of modular setting presentation: every chunk of setting is digestible and usable immediately. If you want to run a "guns and badges" game in the 1950s, Glancy gives you four or five immediate setting hooks, tied to media sources you recognize. If you want to fight Saxons, Greg gives you a war, a battle, a bunch of NPCs, and real consequences for you and your lands to drive the story forward.
Those of you who own (or will eventually buy) The Day After Ragnarok may notice that I've picked up a few of those tools. I think it's not impossible that in the future, I'll pick up (or discover) still more of them: what would a game look like if its setting was modular, Earth-based, and player-monetized, with random story generation elements to define and gamify every hex? (Or every hex where adventure happens, at any rate.) I don't think it's impossible. And I don't think I'm finished finding other ways in which I haven't been doing my job.
This began as a post explaining the design process of Tehran: Nest of Spies, my city sourcebook on Tehran for Day After Ragnarok, but laying the theoretical background kind of took over. Next time.
 Yes, I know Aristotle picked six instead of three, and aside from 'mythos' (plot) they don't really map to Creative Writing 101. That's because the only setting that existed for Greek drama was "Greek myth." Though that said, I think someone could profitably examine RPGs in the light of Aristotle's three "lesser elements" of 'lexis' (diction and word choice), 'opsis' (spectacle), and 'melos' (music and rhythm). I pick robin_d_laws.
 Indeed, if you include the world's physics as part of the setting, the setting includes the game engine itself. Even a "universal" RPG like GURPS says things about its setting by attempting (or not attempting) to model real-world physics.
 At some level this is all Tolkien's fault. Ron Edwards has a typically incisive discussion of setting in heroic fantasy in Sorcerer & Sword, in which he argues for "Here Be Monstres" as a setting design virtue; the fog of fantasy, rather than phiddly philological sub-creation.