One of the noisiest trends in the theory of how to run and play RPGs in the last few years or so is the notion of the "sandbox." Taking its terminology from the world of video game design, a sandbox is a game mode in which the players can approach their environment with considerable freedom, and do what they want to do, and approach problems the way that they want to approach them. This is often held in contrast to the "railroad", an older slang term amongst RPGers that refers to the notion that the GM has already mapped out the campaign, and the players can really only proceed when they manage to stumble across the "right" way to do so.
These common usages have gotten really sloppy, so that nearly any game is often percieved (or at least described) as one or the other, with no middle ground. And since railroads are almost consistently villified as "not a very fun way to play", sandbox has become synonomous in some people's minds with "good." I.e., any game that is good is, in fact, a sandbox, regardless of whether or not it really has any sandboxy qualities, or emphasizes them. People also frequently ask nonsensical questions, like "which campaign setting for D&D is a good sandbox?", or "which version of D&D is a good sandbox game?" or stuff like that, regardless of the rather obvious fact that "sandbox" as a playstyle refers to how the game is actually run and played at the table, and has little if any connection to how the game is written. Any game can be a sandbox if it's approached as such by the guy running the game, and any game can fail to be a sandbox by the same measure. In my experience, the casual tossing around of the term sandbox is a blight on actual discussion, as it tends to act like a trawl, pulling out all kinds of One True Wayers who can't resist chiming in with their proclamations, which often have little if any relevance to the topic that one is attempting to discuss.
Although as a GM (and as a player) I demand and insist on a great deal of freedom, I don't consider myself a sandboxer, and am actually just as disinterested in a sandbox game as I am in a railroad game. It's all well and good to say that in theory the players should drive the game by doing things that their characters are interested in, but in reality, this is a pipe dream for most gamers, at least until well into a campaign. When a game starts, the players can hardly be expected to have a strong handle on their characters or their motivations, nor can they be expected to have engaged enough with the setting yet to expect anything from it. Any campaign that doesn't start by giving players (and their characters) something immediate to latch onto and have to deal with, as well as giving a decent explanation for why the various characters in play are sticking together as a group, is one that is off to a rough start, and is likely to fail to engage the players enough to even survive. The fact of the matter is, players need some direction, especially early on in a campaign. After a time, they need much less--and they can be expected to start driving towards things that they have taken an interest in over the course of play. But when the game starts, that's not a reasonable expectation. And frankly, "wandering around aimlessly hoping to encounter a game" is not a very compelling mode of play. And for those who say that they want go-getter players who start the game already motivated to do something, well... OK... but how do you expect players to have motivations if they haven't yet had any engagement with the setting, other than really vague ones? Even moreso, how do you expect the various players to all be on the same page in that regard? Frankly, players like that seem a bit like a red flag to me. Motivated and self-starting is fine... but that can easily migrate into pushy, bossy, and "ignore everything except my preconcieved notions of what I want this game to accomplish." Which, needless to say, is not a good group dynamic. In my experience, that usually means one player like that, and the rest who are rather easy-going and willing to follow along after the leader. Any other group dynamic leads to conflict and ... well, not a fun experience in most cases.
On top of that, I, at least, demand something more from my campaigns that merely passive environments to interact with. I find the notion of a sandbox to be boring and repetitive. I came into gaming as a fan of fantasy fiction, and what appealed to me about it was its ability to resemble a collaborative, joint-venture fiction creating exercise. In order for games to be engaging to me, I require villains, situations, and setting elements that are much more active, dynamic, and well-formed and "verisimilitudinistic" than a sandbox campaign is likely to provide. And I also demand that campaigns have climaxes and denouments from time to time--a narrative structure that appeals to me greatly as a gamer as well.
Real sandbox junkies, at this point, will point out that a good sandbox GM can make all that happen and still maintain its sandbox integrity. I'll point out that a good GM can make almost any kind of game fun, so that's kind of a meaningless statement. The fact is that the sandbox ideal makes what I want out of the game less likely to happen, and significantly so. This doesn't mean, as I've been forced to explain many times, that I like railroads. I despise them just as much, and if anything, am almost over-the-top on my insistance that player choice be a major factor in how any game plays out. But a pure, ideal sandbox is not the game for me either; I find either endpoint of that ideological spectrum on how gaming should be run to be equally unappealing.