So a zip code is more properly represented as an ordered point string or point cloud, rather than an area. There exist zip codes with up to 11 non-contiguous areas, zip codes which are valid only for the north half of an east-west street, and other oddities. You can fudge this a little by just drawing bounding boxes, but then you have up to 5 zip codes overlapping in some places.
That bit of trivia aside, they do make sense from a real estate point of view because a) all domiciles will have a zip code and b) it is usually freely available on listings.
Generally though, it's better to translate any given address to a census tract or census block group, since those divisions are actually defined as "a geographic region defined for the purpose of taking a census" - they can be properly mapped without ambiguity, crossovers, discontinuities, and other boggling features of zip codes, (though not of course with perfect regularity) and they also coordinate well with demographic data like median income - I'd be very interested to see a graph of percent of median income per square foot - I suspect you'd find some very 'hot' areas that don't seem like it.
The first hard algorithmic coding task I worked on was to generate isoclines of UK Postcodes from point cloud address data. Fortunately I didn't have to do the entire country, only a couple of million addresses, unfortunately it had to work on a machine running Windows 3.1 with only 4MiB of RAM....
The use of ZIP codes in GIS is largely a matter of convenience. The geocoding is generally good enough, particularly when using a statistically significant sample. And the underlying data are generally available as they're attached to most of the data of interest. That said, DIY geocoding with pretty high accuracy is possible and available:
"ZIP codes designate only delivery points within the United States and its dependencies, as well as locations of its armed forces."
http://en.wikipedia.org/wiki/ZIP_code
So a zip code is more properly represented as an ordered point string or point cloud, rather than an area. There exist zip codes with up to 11 non-contiguous areas, zip codes which are valid only for the north half of an east-west street, and other oddities. You can fudge this a little by just drawing bounding boxes, but then you have up to 5 zip codes overlapping in some places.
That bit of trivia aside, they do make sense from a real estate point of view because a) all domiciles will have a zip code and b) it is usually freely available on listings.
Generally though, it's better to translate any given address to a census tract or census block group, since those divisions are actually defined as "a geographic region defined for the purpose of taking a census" - they can be properly mapped without ambiguity, crossovers, discontinuities, and other boggling features of zip codes, (though not of course with perfect regularity) and they also coordinate well with demographic data like median income - I'd be very interested to see a graph of percent of median income per square foot - I suspect you'd find some very 'hot' areas that don't seem like it.