In contrast to all of the complaints about railroading, I've found that while a poorly run railroad is bad, it is at least playable, which cannot be said for a poorly run sandbox. That is, if a DM does not have a lot of experience in effectively running a sandbox, especially when it comes to creating a narrative out of the way the players interact with the world, it is very, very easy for such a campaign to quickly devolve into "just a bunch of stuff that happens", with no real narrative momentum to speak of. This problem is exacerbated if the players have made no effort to make sure that their characters share goals or indeed have any real reason to work together apart from the fact that they're all pcs.
There's also the fact that, if everyone is more or less on the same page in terms of pcs sharing a goal which also happens to be the goal the DM has in mind for the campaign, some degree of railroading can make it much easier for the DM to go all out in designing some really cool, innovative scenes and encounters.
To use an actual example from my own DMing experience, when I know for certain that the PCs will at some point make it to the wizard's lair, then its a lot easier to justify spending half of my Saturday designing the most awesome, innovative wizard's lair I can. In that case, I decided that the wizard's lair was in fact the control room inside a giant golem, and the final showdown made heavy use of that fact--players and monsters faced off in a room full of chains and levers and pulleys, each of which affected the battle in some way. Pull one lever and the golem stoops, causing the entire room to slant so that everyone who fails a save is knocked prone, and until the room is righted the extreme angle means it gets treated as difficult terrain (which could be circumvented by swining on chains--but watch out, because pulling some of those chains has actual effects, like turning off the magical lights and plunging the room into darkness). It was a really cool, memorable encounter. My players loved it, and I loved running and designing it.
But, I don't know that I ever would have taken all of that prep time if I hadn't known that at some point, one way or another the pcs were gonna face off with the evil wizard. I didn't know exactly how or when they would get to it, but I knew that the giant golem was a major threat to the city the players had sworn to defend, and that once they did some digging they would figure out a wizard was in control of it, and if they followed up on those leads eventually they'd find their way to his lair. But if the players hadn't had that shared goal of defending the town, and the wizard had therefore been just one of a theoretically infinite numbers of challenges around the world that they could very well decide to pursue, I probably would have just jotted down a few notes on who the wizard is, what he acts like and what his goals are, maybe a couple of contingencies (e.g. if the players have found x artifact, he may seek them out to steal it from them or bargain for it), and then his stats as a monster. Call me lazy, but if I'm gonna spend hours designing a scene or encounter, I'd like to know it will be used in some way.
Or to use another example, Gabe at Penny Arcade attracted a lot of attention for posting some incredibly awesome encounters he'd designed--one of which involved actual lasers and mirrors, another was a combat that took place throughout an extended freefall. He's since moved the campaign into a more sandboxy style (which he also has a really cool and innovative approach to), and in discussing the transition he himself said that those incredibly awesome, innovative encounters were to some degree made possible by railroading. That is, he was able to spend ages creating those incredibly awesome encounters because he knew without a doubt that his players would eventually run into those situations. And, to be perfectly honest, as a player if I knew my DM had some incredibly awesome stuff like that planned, I'd be okay with some degree of railroading to get there.
The railroad/sandbox divide is tricky to navigate, and in fact most campaigns that I've really enjoyed and that have stuck with me have incorporated elements of each at varying points, just as Gabe is doing, and as I strive to do to some degree in my own campaigns (I typically attempt to create a world that has the potential for some degree of productive sandboxing, and to allow time for player-directed action to exploit that. but I also usually have a few big, overarching plots that I have a much clearer idea of how they will develop, and I work with my players to make sure that the goals of their characters would be in line with, or at least no directly conflict with, the major overarching plot).
But I do think that railroading often gets an unnecessarily bad rap--having some idea of what will happen makes it much easier for a DM to make sure that "what will happen" is something completely awesome, and it certainly makes it easier to build a really fleshed out story.
