Clive Thompson has written an article on the duration of games, and how it relates to different kinds of gamers.
I call it "the myth of the 40-hour gamer." Whenever you pick up a narrative adventure game these days, it always comes with this guarantee: This game offers about 40 hours of play.
This is precisely what I was told by Eidos -- and countless game reviewers -- when I picked up Tomb Raider: Legend earlier this year. As I gushed at the time, Legend was the first genuinely superb Lara Croft game in years, with a reinvigorated control system, elegant puzzles, and an epic storyline involving one of Lara's long-vanished colleagues. I was hooked -- and eager to finish the game and solve the mystery. So I shoved it into my PS2, dual-wielded the pistols and began playing ...
... until about four weeks later, when I finally threw in the towel. Why? Because I couldn't get anywhere near the end. I plugged away at the game whenever I could squeeze an hour away from my day job and my family. All told, I spent far more than 40 hours -- but still only got two-thirds through.
The article touches on a couple of interesting topics:
First of all, I would argue that 'game duration' is a nebulous concept related more to 'value' than to actual time spent playing. As such, it is obviously strongly linked to replayability. In the end, games have to deliver a certain kind of value, and that value is related to price. So you can make free web games that you can play for 30 seconds, but once you start asking, say, 60 dollars you better make sure the game can be played for a reasonable amount of time, whether that includes replaying, new modes etc. or just a single play-through.
Second, the general industry attitude (in so far as this can be discerned) to game duration for full-price games has changed notably in the last 5 years. Square used to have a rule of 'an hour per dollar', and for all I know they still have. But there have been some public or semi-public announcements saying 20 hours is OK, it's quality that counts, not quantity, etc. Since then, some games have tested the lower ends of the game duration spectrum - I remember Max Payne 2 got criticised for only providing 8 hours of gameplay. (Having said that, a former co-worker of mine once offered the theory that if a game is hard to finish, people tend to keep it instead of selling it back to game stores. This is a bit cynical but makes economic sense.)
Third, obviously from a development standpoint one wants to optimise development costs per entertainment hour, and also entertainment value per hour. This is why everyone is complaining about high budgets - the trend in the industry for the last decade at least has been to increase entertainment value per hour by increasing development costs per entertainment hour. In other words, bigger production values. As trends do, this will overshoot, some people will notice too late, and niches for new approaches will appear, or grow in size. This is happening right now.
Fourth, it is hard to measure or even define game duration. This is essentially what Mr. Thompson is pointing out in his article. Hard-core gamers breeze through games that he gives up on after weeks of playing. This is not a new development, and I would argue that it has by now become a convention inside the industry (surely correlated to the average age of middle management). Even the counter-movement to this has become boring already. Nevertheless, it is true. No longer being able to dedicate large chunks of time to gaming (it would interfere with my WoW), and having a low frustration threshold, I am a strong proponent of making games more accessible, playable in shorter chunks, etc. (and, hey, it sure worked for WoW, as you can read here).
The best approach to knowing how long your game takes is to take an 'average' gamer and having him or her play the entire game. I am not sure how many companies do this. Better methodology (especially regarding that 'average' gamer) should yield better results. I think what is more likely to happen is that one takes a couple of single-level play times from relatively new players and multiplies it by the number of levels. At some point, typically nobody cares that much, as long as you hit 20 rather than 5 hours for a full-price game. In the end, it's all about the perceived value of your game for your customers.
Fifth, most estimates of game duration made at the start of a game's development are, to the best of my knowledge, pure guesswork (I would love to be corrected on this by the way). Although I can think of a few developers with the analytical balls to know how long a level should take, will take, and does take (e.g. any game Mark Cerny works on), I think most multiply wishful thinking (the minimum amount of levels they think they can make, plus some more to be removed again when it becomes obvious there's no time) with an arbitrary number ('an hour' sounds like the time you'd need to finish any level, in any game... right?). At least, that's what I do, and I've never worked with anyone doing anything fancier. The trick is to guess with authority.
Sixth, one of my pet subjects in game design are the qualities of a game over time: difficulty, accessibility, fun, introduction of new elements, etc. Obviously, game duration and replayability are factors too. I think it's a fascinating subject, and it's too bad you need something resembling a complete game before you can really go to work on this stuff, because by that time, the pressure to ship is usually too high to do anything fancy. Which is a pity, as I consider these qualities to be strong factors to a game's success, and if not low-hanging, at least pickable fruit for game design improvements.
Seventh, even I managed to finish Tomb Raider: Legend in a weekend. I mean, sheesh.