Not to be a pedant but the relevant variable has not been manipulated here. We would need sites of (more or less) equal-in-expectation popularity to be assigned

Not to be a pedant but the relevant variable has not been manipulated here. We would need sites of (more or less) equal-in-expectation popularity to be assigned to use social media or not. The metrics the author provides are capturing effects of past social media distribution.