* The photo prejudice: I might have misread, but I think the authors meant that because A. females are more friended on most SNS (bottom of your post) and B. the main feature difference between Twitter and the other is the lack of photo, then photos might be the clue. That is of course a gross simplification: no body is technodeterministic enough to argue that features alone decide what a SNS pratice are like — but I don’t think that was raw sexism. On the other hand, if you associate data on friending in most SNS consider photos in male- and female-oriented magazines (and don’t challenge their editors’ sanity) you’d have an intesting way to make the point that you don’t like appear more reasonnable.

* I’ve read somewhere that when you actually look at friending back, males are followed back far more then female. 50% vs. 25%—the argument that both aren’t the same misses the point that unless you have a very odd social structure, they’d be correlated. Instead of denoncing the controversy, you need explanations (plenty, to avoid being accused on defending one) that match the data and that are different than what the authors are saying.

* Just say it: Spammers *are* using female-named accounts to lure more users (stupid, slutty ones). That alone *could* explain the measurement, but only if an outrageous number of accounts are spam. It’s possible and demands proper precautions, but it needed to be said.

* Twitter is not an exclusively tech-dominated area anymore, but I guess it still is enough geeky to have that inbalance play a role; I think middle-range celebrities are still mostly (male) tech gurus. Maybe having the gender-ratio of adopters over time would help; that and study the relation of adoption date versus number of followers. Even better: correlation between adoption date of followers, followees non-back and followee-back to describe the friending dynamic and measure the remaining role of techies.

* The central point of any argument against the paper is certainly how they tell gender. They might be able to do so with common American names, but anything foreign (including those that spell like American names but aren’t) is a problem. You are right to include Mech Turk in the loop, but I’d use both methods, and have a higher-level analysis to sort discrepencies between the twos (thank God for cheap, foreign PhD students). I’d probably use crowdsource to sort out neutral agents: e.g. if “Mike Donald’s” is the RP account for a fast food chain, co-authored by a team, it’s neither male not female, but would be interesting; same for “Venus Bot” our trustworthy probe on the third planet. All in all, you could do that properly, but you’d need so myuch efforts, only crowdsourcing could handle it.