Possibly naively, I would have assumed the figures might be based on things like revenue blocks where they do catch people. Then a further assumption that the split of passengers in said blocks are representative of passengers in general.
I imagine this is where the data comes from. As you say, this assumes that this is representative. For the trams, I imagine fare evasion is very time dependent - for the commuters returning to their family homes in Beckenham at 17:30 it will be almost zero, late night on a Saturday will be a different story.
There will be other sources of data like gateline assistants recording people pushing through gates (not on trams, but on tube and rail), bus drivers noting those refusing to pay, passenger complaints about others not paying (especially on buses where this causes delay) etc. The figure will be based on something more substantial than "choose a number between 1 and 10", but will have quite a few assumptions.