In what is probably the biggest case, FlightAware, a lot of the data comes from a fleet of users running ADS-B[1] receivers feeding data into their system. ADS-B data is what Air Traffic Control uses to track planes, and it's receivable using using a cheap TV tuner and a raspberry pi[2]. It ends up being probably the best way to track planes very precisely, as long as you can get enough coverage, which flightaware manages to do pretty well [3]. They then package this data and sell it via an API/tools for Fixed Base Operators or apps like Just Landed.
[1] https://en.wikipedia.org/wiki/Automatic_dependent_surveillan... [2] http://flightaware.com/adsb/piaware/build [3] http://flightaware.com/adsb/coverage