I read your comment as saying that you could lie in text about identifying information and things that might betray you in a f2f interaction, such as regional accents, weren't as important in textual interactions.
My intent was more that it is straightforward to be aware of several of the most blatant leakages. Combating dialect analysis is a lot tougher, but the quiz likely used 25 of the highest value differences (many of the choices are laser specific: http://www4.uwm.edu/FLL/linguistics/dialect/maps.html ).
Context also matters a lot. If there is no obvious path to an IP or browser interaction, all that stuff goes away.
What would be interesting is a look into passive elicitation of information. Its easy to falsify something when you know someone is looking to use a single data-point against you. But tryly powerful social engineering would manipulate your confidence, and extract high-information-content data passively (when you are not expecting it), and to do this they need to instill confidence that you <are not being ovserved or monitored>. Yet they still may need to promt information flow, to feed the analysis. Which is sort of the art of interviewing, but now in a bit more 21st century context.
