[thinkingemote]

Well I don't know, but I'd have a guess as to how they might easily do it. Pattern recognition and/or machine learning. Something like: here is what the pattern of a user moving through space when they are in a car. Here are a million other ones. Here is a pattern of a user walking through space, taking a bus, etc etc. Now here is an unknown user. What pattern of movement does this user match closest to?

A pattern would be, for example: is he ever seen moving more than 5mph? Does he use driving directions? Is he only seen moving faster along bus routes? If he is only seen along bus routes does his pattern of moving and stopping align with the bus stops? Is there usually traffic jams at the same time of the week on different days, etc.

Just build up a database of patterns and check a 1 week old account with that to see what it fits best.

Perhaps in an area with no predictable traffic and no bus routes, that a wagon of phone matches the pattern of a traffic jam caused by a blockage on a road. It's less likely to be a bus, perhaps?