So, here's my basic thesis: When multiple people collaboratively create a story in an RPG, each person controls different aspects of the story. Sometimes, things that happen under one person's control suggests a complementary response in something someone else controls, but actually responding is voluntary. In a game where everyone is having fun, each person must be controlling enough of the story that they feel they are making a significant contribution, be controlling an aspect of the story they feel comfortable with, and there should be a certain amount of give and take between everyone's spheres of control.
But 'the bit you get to control' doesn't have to be 'part of the plot'. Considerable effort has been made in recent years to parcel out control of the plot to different people in different situations, with a lot of positive results. But there are more aspects of a story up for grabs. We might divide things a variety of different ways, all of which would be somewhat arbitrary, but for the purposes of this analysis, I'm going to choose to break down 'story' into four axes I've seen used in casual literary analysis: Plot, Character, Setting, and Writing. Except that there's no 'writing' in the collaborative stories that emerge from an RPG session, so I'll call it 'Tone' instead.
Let me posit that a roleplaying system is ultimately responsible for all four axes, and that the rules of the game both delegate control and place restrictions on what can be done with this control. If you're a RPG designer sitting in front of a blank piece of paper, you start off with control of everything, and must delegate the right amount of control over the right parts of the story to the people who will play your game.
In many systems, delegation of most of the 'Character' axis is delegated to the players, usually with a variety of restrictions in the form of rules that must be followed. 'Plot' is delegated to the game master, often with no restrictions at all (indeed, sometimes the GM is encouraged to ignore the restrictions given to the players if they choose to). The 'Setting' axis is the one most often reserved by the game system for itself, though the game master is usually allowed (outside of module play, at least) some freedom to develop the local setting, with heavy restrictions imposed by the more general setting. 'Tone' is interesting. Several systems simply ignore it, effectively delegating responsibility to the group. Others attempt to evoke a particular tone, but place no restrictions on its expression in story creation. A few deal with it directly, with rules that forbid or encourage certain types of statements during play.
Looking at roleplaying games this way, we can see why the supposed paradoxical statements 'the GM controls the plot' and 'the players control the main characters' (known in some circles as 'the impossible thing before breakfast') don't actually bother a large number of people. Simply put, the bits the GM controls and the bits the players control are different aspects of the same story. The GM may or may not choose to alter or adopt the plot (or even create a new one) in response to choices the players make about their characters, and some groups derive satisfying play from different choices on the part of the GM. But in these games, ultimate responsibility for the 'Plot axis' still resides with the GM, and ultimate responsibility for the 'Character axis' still resides with the players.
Examples of the separability of plot and character abound in other media. Retellings of fairy tales, for example, tend to take the opportunity to explore new characterization of the protagonists, while retaining the plot of the original. Writers for TV shows take pre-existing characters and create new plots for them to explore. Many authors report that the characters they create sometimes come so 'alive' that they begin to perform unexpected acts; sometimes the author responds by changing the plot, and other times simply adapts the plot around the new action. Setting is more obviously distinguishable from both the plot and characters--'West Side Story' as a recasting of 'Romeo and Juliet' in a new setting is a clear example, and other adaptations abound.
So, what exactly is it that is controlled in a story? Our axes are somewhat arbitrary, but they serve as a good framework for more detailed discussion. Since the character axis has been maligned in the past, let's start with that. My hope here is that by exposing some of what is happening on a mechanistic level, it can help designers understand how their design affects traditional methods of control and restrictions, and to inspire new designs that approach the issue in new ways.
'Player Characters' are a staple of roleplaying stories, with significant control of and responsibility for the protagonists almost always given to the players. Non-protagonist characters (both incidental and antagonistic) are traditionally given to the game master to control, often with a significantly different system of restrictions--the rules the players must follow when creating their protagonists are often explicitly different from the rules the game master must follow when creating everyone else. A few characters are sometimes provided by the game system, often rulers or other major players in the provided setting.
Characters are defined by their intrinsic qualities, their abilities, their history, and their actions. PC qualities and abilities are usually defined at character creation, and usually are highly constrained by a system of rules. They are often allowed to increase during play (with rules governing 'experience points' and the like), again according to a separate system of rules. PC history is defined at character creation, but often the only imposed stricture is that the events in it not violate basic truths about the setting, and that they must however indirectly conspire to create the 'starting' character.
PC actions deserve more than a cursory overview, for it is here that the players spend most of their energy and creativity during story creation. Some PC actions serve to further the plot, and here we experience a certain amount of category breakdown. But whatever action the PC may choose to perform, in a traditional RPG, they are not *responsible* for the plot, merely for their PC's actions. So a player may derive satisfaction not necessarily from the fact that their PC managed to get through the plot, but *how* they did so. Furthermore, the PC's characterization is developed through many actions, some plot-relevant, but many not. A player may decide or may even 'discover' that their PC is friendly, bitter, clever, simple-minded, dedicated, trustworthy, witty, manic, melancholy, or brave, and a lot of this will be expressed not in the direct interactions with the plot, but in auxiliary responses to events or other players.
A player whose PC is the prince in a role-playing version of Sleeping Beauty, for example, might not be all that interested in the plot, but be very interested in defeating the various obstacles (the thicket of thorns, the dragon, the witch) in their own way, whether through force, guile, wit, or luck. Another player playing a household cleaning robot in a futuristic version of 'Cinderella' may have great fun bringing their own ideas about what it means to be a robot to an otherwise familiar story.
Plot is generally either delegated by the system to the GM with no restrictions at all, or pre-defined in a module, with the GM given a certain amount of wiggle room to adjust to fit the situation and respond to player actions as they see fit. It is then up to the conventions of the group how much control the GM gives to the players, or, more likely, how much the pre-conceived plot changes in response to player choice.
Again, for the sake of giving us concrete things to talk about, let's divide plot into its traditional linear stages: setup, conflict, rising action, climax, and denouement. In a traditional system, the GM is in charge of pretty much all of these stages, with players having a certain amount of input, mostly in the 'rising action' and 'climax' section. In M. J. Young's description of four methods of GMing, he's mostly talking about the GM's control of the plot, and their response to player actions. In Illusionism and Participationism (identical except for how aware the players are of what's going on), the GM retains control of everything, and only modifies things slightly in response to player actions. In Trailblazing, the GM sets up both the 'setup' and the 'conflict', then hands over significant control to the players for the 'rising action' bit. So much so, in fact, that the players are allowed to 'derail' the plot, at which point the game winds down with no climax at all. This danger, however, allows the players to enjoy the climax that much more if they do manage to get there, since they were able to 'beat' the game.
His final technique is 'bass playing', in which (as described, at least) the GM controls the setup and conflict, then steps back and hands both control and responsibility of the rising action and climax to the players. In this scenario, player actions don't merely *affect* the plot, they *are* the plot.
There are a variety of other ways things can be divvied up here. The players can be part of creating the setup for the plot, perhaps by submitting detailed player histories to the GM who then uses them to create a plot of his own. The players might even set up the conflict, as they do with character 'bangs' in "Sorcerer" and during the game itself (I gather) in "Capes". The GM could then either take back over control of the plot, or send control back to the players. In the "No Myth" methodology (as I understand it), the GM retains control and responsibility for the plot, but creates it 'just in time' for the characters to discover it. This is different from 'bass playing' in that while in both, player actions are very important, in bass playing the player actions are the plot itself, while in No Myth, player actions enable the plot to be discovered.
A very interesting game when viewed from this perspective is "My Life With Master", because none of the above options are taken at all. Instead, the system itself retains control and responsibility for the plot, with rules in the game that help both the GM and the players progress and finally reach the climax. Similarly, "Metal Opera" has mechanics that are specifically designed to control and pace the 'rising action', and dictates where the climax must come. Metal Opera also specifically delegates creation of the setup and initial conflict to both the players and the GM, and has rules for who gets to narrate what as the game progresses.
Other, more free-form games like "Universalis" and "The Pool" dispense with the GM entirely, and thus bequeath complete control of the plot to the players. Players must then control their own pacing through responding to each others additions, not unlike what happens in improv theatre.
In general, the 'plot' axis has probably seen the most experimentation in recent years from a variety of (usually independently-produced) roleplaying games, though there are clearly more options to be explored.
We find two common models for control distribution when it comes to setting. In one, control is mostly handed to the GM, with a few restrictions in place about how things work in it. D&D, for example, allows a GM to create an entire world with the caveat that it contain the sorts of races the PCs are allowed to create, have a certain (low) level of technology, and allow for the presence of a certain type of magic. The presence of certain types of creatures and pseudo-medieval governmental structures are encouraged. Beyond that, the GM is free to create geological structures, particular governments, people, racial attitudes, attitudes towards magic, etc.
In the second model, a campaign setting is provided, at somewhat of a 'guidebook' level of detail. The main geological features are known, as are most of the governments, societies, and principle movers and shakers of the world. Technology and/or magic levels are set, as are the populace's access and attitudes to both. From there, the GM is given control of the details of the setting, with the fairly heavy restriction that these details must not contradict the big picture. The GM might create a trade route between two countries, for example, but the materials traded must be items that could reasonably be created in those countries.
The level of detail provided by the system is on a fairly loose sliding scale, however. Generic systems such as GURPS and FUDGE might provide absolutely nothing in terms of setting, though 'supplements' are available that provide more. "Puppetland" provides a very detailed 'closed system', so all stories created in it share many details of setting, even down to the same antagonist. "My Life With Master" is (again) an interesting case, because it doesn't provide a setting so much as it provides a set of restrictions on the setting that the players and GM create.
It should be noted that the job of providing the 'guidebook' level of detail of the world, when given to the GM, may actually be filled by anyone, since the work done here usually takes place before the 'game' begins. (The one exception here is any 'secret' information about the setting that might affect the plot. Setting books often have 'GM sections' that contain such information.) It certainly helps if the GM is familiar with the setting when running the game, but they need not be the creator. Once we get down to the details of the setting where the PCs begin the adventure, we again start to cross into 'Plot' territory (revealing again the somewhat arbitrary nature of these axes). It's also worth mentioning that some details of setting are contributed by the players via their histories: the place they came from is often a player invention, for example.
While many systems attempt to encourage use of a particular tone by their players, few have rules that directly affect it. Control is thus usually granted directly to the players with no particular restrictions. There are a few exceptions: in "The Dying Earth RPG", you are awarded experience points if you use particular tag-lines during play. On the 'social contract' end of things, I've heard of rules such as "Monty Python quotes will lose you experience points".
On a more indirect level, some systems have rules in the setting that encourage certain types of things to happen that are evocative of particular tones. Systems where it is easy to die (theoretically) result in more 'gritty' games, while systems where it is nigh-impossible to die can result in more 'heroic' games. This is a bit trickier, but some systems manage it better than others. But groups tend to go where they want to in terms of tone, as anyone who's played in a laugh-riot "Shadowrun" game can attest.
When designing a roleplaying system, I think working out the basic system of how you want control of the story distributed, and what sorts of restrictions you want in place can be helpful. You can also analyze existing systems using this framework, and it might help reveal why aspects of that system worked or didn't work for you. When creating a new rule for your game, you can think about how it is intended to affect these different axes, and whether it will do a good job doing what you want it to do.