So what happens? Are they compiled statistically and then addressed in terms of what is most prevelant or what? Or some other method?
Good questions. You are correct in that not every single blob of crash data is perused by a human being. This would be a daunting task. Then again, this doesn't mean that data is ignored. In fact, as you guessed, statistics are run on the data and the most prevelant type of problem is addressed first. Now, priority is based on the type of problem. That is, an app hang will garner less attention by definition than an app or, especially, a system crash.
It's always a good idea to send crash data to Microsoft. Now, if an app hangs because of something you've done (like written infinite recursion into one of your algorithms) causing you to kill the process imediately which in turn invokes the error reporting wizard, then this data isn't really a problem with the system or an app running on the system so it's not a big deal if you choose not to send the data... But, again, when an app just crashes for seemingly no reason or, of course, the system crashes, fire that data off to us please!