"But some scholars point out that Facebook is not representative of the ethnicity, educational background or income of the population at large, and its membership is self-selecting, so there are limits to research using the site."

They're all overlooking the elephant in the living room: that on-line interaction is NOT the same as genuine human interaction. How much will that simple co-factor skew the results of their research?

No it isn't exactly the same, in fact, it's better. It overcomes problems of geography and time, allowing users to have multiple forms of communication. I don't think any researcher thinks that it is an analysis on physical methods of social relationship, but I think they do realize it is a way to understand how fundamental human interactions occur.