Thought about how to answer this for quite a while. It's a bit hard to describe some of these things without sounding like David Gibson in "the Art of Mixing" but he has a point: maybe we can point out the first degree of panning and stereo image as a way to clear the picture. Strangely enough, the ghost images that appear on the sound field can still mask themselves quite a lot.
There are some really neat graphics and maths about the shadow area around a frequency, meaning how much louder do the frequencies next to another need to be to mask the first one - a not so bad wiki article about this here. Things can be neatly arranged and rearranged to make things clearer, more spread out, more interesting.
Now, some things to consider: first on a more acoustical side:
Lower frequencies take a lot more space and power in a mix. Tilting them to a side generally makes the mix seem bass heavy or just plain left/right heavy. I tend to demonstrate that to students when they submit a mix that makes you kinda feel that you have to turn and face left. Although this is more likely with lower frequencies, tilted mixed can happen with pretty much with any sources, if there a busier side of the arrangement in one side, etc.
A subtle example of mixes being tilted by low frequency could happen when you mic up a drum kit close to a wall, or closer to a corner on one of the sides. Also, it can happen if your microphone stood at a low frequency node. Although meters would probably not reflect this your overhead drum sound will be (acoustically and psychologically) tilted.
Ghosting center: the big thing when stereo hit was the capability of producing images over a space. And better, to center a source right in the center of two speakers. This has some issues though - this image is extremely fragile and subjective to any variation between both speakers, and it can get easily masked. Some really good guys like Terry Manning have been developing a theory of cardinal points where a third speaker is added to the center and other issues related to their downmix problems, etc. But we have this in surround. Those of you that had the experience in mixing in surround definitely noticed that is far more easier to achieve clarity and positioning on the source on this field (because it's bigger) but that with the center speaker, you won't have masking problems - an example would be to do your whole mix in stereo, panning kick drums, snares etc bang on center and have the vocals on your center speaker. It will cut through far easier than it would on the stereo field.
The fact that we pan in the center also gives a reinforcement - think about it: you have 2 speakers doing the same work for you! (yes, there is the pan law but we'll get there in a second).
When to use Center, halfway, or hard left/right? Well, that's really up to you! it's clear (even on this site) that a consensus is not really possible. But a few points can be made: we already talked about the center, hard left/right is usually something that you'll go for if you really have something important to put there. If you are now thinking of overhead microphones, other microphone stereo techniques or stereo samples, don't forget that these don't really have hard left and right material - they'll have something in between! and specially because recording angle vs. playback angle get's extremely distorted near the edges. Anything outside the recording angle will also be reflected into the stereo field but in a more "reverse" fashion. Examples in recent date of use of extreme left and right could be Maroon 5's "wake up call"'s ping pong guitar but tons of other bands use that space for random effects and "ear candy".
Many times, though, when you have really detailed guitars, a fairly wide panning is achieved. When done that there's a worry of loosing the signal completely on the other channel. Think Clubs, or huge venues without front-fills (more common than one imagine). The audience would loose your guitar solo completely, or your nice synth line. You can indeed go narrower, or starting to pan with delays (using Haas theorem). More on this next.
I seem to notice mixes that try to make seem louder, punchier, are usually mixed narrower. And mixes that try and be clearer, more detailed, are usually panned wider making use of the entire stereo image. There's also some mixes that have really mono centered drums, loud and punchier, and the huge arrangements spread out very very wide!
On an electronic side, we should consider several aspects:
- Pan Law. This is an level balance that occurs because of the design of the pan pots (level being split and divided, through a resistor series and - luckily - opamps; the signal will drop. A linear pot will drop you 10dB) but desks manufacturers and DAW's have been manufacturing how this drop will occur. This define how much the level drops when coming back into the center, to compensate for the acoustic of the room and the physical properties of the speakers. Now, most desks and DAW might default at -3dB but there are daw's that range from -2.5dB to 6.02dB (as far as I know, for some mastering gear or maybe some fancy software). -3dB (we are talking of voltage here) will not equate to the 6dB SPL variation (hence the 5.99dB pan laws and 6.02dB.
- Haas theorem says that, with very short delays, a signal with the same intensity than another can be perceived as coming from one speaker as being louder. Typically 20ms are used. Have you ever been in a gig where the back towers seem to be off? well, they are not. They are just delayed by the distance between the stage and the towers, then delayed further 20ms.
You can do the same in mixing. Add a really really short delay to the opposite side. If someone hears on one headphone, or is on the opposite side of the venue, or has one speaker broken (go figure!!) this person can still hear a hint (very good hint) of what's happening with your other guitarist (or triangle percussionist solo :P).
This is a technique widely used in mixing for post (surround) as sometimes extreme panning isn't an option.
This is also something I use to try and correct a balance issue, or to fatten out a sound (careful as it might generate some phasing).
Having said all this, is not hard to understand why bass, kick drum, snare, vocals are usually panned dead center. Although some nice names (mixers) that I had the pleasure to talk and work to did share that they usually pan the center stuff just a tick to the sides around the kick or vocals.
Me? I usually try and balance out a mix level and stereo image wise before starting getting deep in other things. I find that at the ended I might shift things ever so slightly and the clarity improves. And, although I like crazy stuff like Mars Volta, only one had to do a crazy trick with drums panned. It was a LOT of work trying to get a balanced stereo image that didn't feel tilted.