A recent conversation over on Candlekeep prompted me to post some suggestions for running a sandbox game. Definitionally speaking, this is a campaign in which the players—not the DM—determine the course of the campaign, through their decisions. The DM’s role becomes mostly reactionary or anticipatory: adjusting the game to suit player’s needs and sometimes unexpected decisions.
Some of us already do this to some extent in our games—basing things on PC decisions, motivations, and goals. In a full-on sandbox game, you take things to the next level: let the PCs rule, and just run with it.
Here are some thoughts to keep in mind:
1) Keep open a clear line of communication with your players. If they're going to direct the course of the campaign, you need to know where they're interested in going so you can react and build things ahead of time. Maybe you're really good at improvisational DMing (some of us are), but it never hurts to be able to know about character decisions ahead of time and make plans.
2) Though the PCs are going to guide the game, you need to have a list of plot elements to throw in if your characters don't offer clear guidance. The last thing you want is the heroes wandering aimlessly and thinking "man, this is boring!" You want an exciting story that they mainly direct, but that you step in sometimes to keep things on track.
Keep this list adaptable. Write down connections that MIGHT be between enemies, but might not, depending on the needs of the story at the time. Improvise as you go along, because the flexible nature of this campaign only allows so much before planning.
You should plan a series of major plot events that will take place "in some form" as you go. For instance, there is a conspiracy of people who are after one or more PCs for some reason as yet unidentified--they launch a major attack on the PCs every five levels or so, and each foiled attack gets the PCs closer to puzzling out who's after them. At the same time, a particular priesthood is going through a major schism that could be fixed or worsened by the PCs, and several things happen at set points in the game. Etc.
Plan on a major plot point every 2-3 levels, is what I would suggest.
3) NPCs have their own lives and pursuits. If the PCs are going to be going about their business, you have to have NPCs who are doing the same thing. This is really the only way to give the PCs the choice of whether to support a given NPC or work against him--this is what's happening, take it or leave it, do what you want with it.
Some of the NPC action takes place off-stage, and the PCs can get involved if they want. For instance, to expand on the example in #2, a priestess might be speaking out against the patriarchical establishment in the church of Lathander; left unchecked, she might incite a movement in the church, or she might be captured/imprisoned/executed as a heretic. The PCs can support her or move against her--they shouldn't feel obligated to follow either course.
4) This sort of game lends itself very well to tying plot elements, villains, treasures, and destinations into the character's backgrounds, motivations, and goals. Your players need to do more work than usual to pull off a game like this: have them craft detailed backstories along with friends, family, enemies, and nemeses for their characters, then use these with abandon in the game. Make some of them what the players might expect, and invert their expectations sometimes. Also, tie some of the characters together in ways they didn't necessarily see coming.
5) D&D is a very swingy game, level-wise, and different places in the campaign are bound to have enemies of different levels. In a normal campaign, you feel justified in railroading the 1st level PCs toward the dungeon with the kobolds, rather than the haunted castle with the epic level lich. But in a sandbox, your tendency is going to be to let the heroes go where they want, and let them suffer the consequences. You need to avoid this, lest you freak your players out and make them too hesitant/timid to do anything, for fear of getting ruthlessly annihilated. You want your PCs to be brave and bold and take chances.
Have a definite "upper-limit" to where your game is going to go. Make a commitment to yourself and your players that you're going to guide the PCs on an adventure, not send them into the jaws of death. Do not send low-level PCs against epic level threats--don't involve those threats at all in the PCs' direct business.
This is not to say, however, that you should not feel free to send them up against overwhelming threats from time to time, to put the fear of the DM in them. This is just a staple of good DMing, albeit sometimes tricky to pull off: heroes should run away sometimes, and PCs who never have to run away may conceive the irrational belief that they can always handle what the DM throws at them. This attitude must be stamped out.
All I'm warning against is sending the PCs *consistently* against threats that are too powerful for them, and/or making it impossible to retreat. Big bads should just let them go sometimes, particularly seeing as they're low-level and weak and not terribly useful for the bad guy. YET.
6) Have an endgame--or rather, several endgames. You don't necessarily need to know how the campaign's going to end, but you should know roughly where it's going before you even start it. Knowing who the ultimate villains are will help you make the story coherent and keep the characters engaged in unraveling the mystery.
Those are just some of my ideas, which come from having run two long-running FR sandbox campaigns, and played in another one (which tragically fizzled from too much story damage thanks to a violation of #5). I am currently also in a sandbox-style game set in the paragon tier, hopping the planes, courtesy of your very own Farcaster. (So seriously, blame him.)