As a CS researcher who often assists researchers in sociology with Twitter studies I struggle with how accurate or meaningful the results are when the study dep

As a CS researcher who often assists researchers in sociology with Twitter studies I struggle with how accurate or meaningful the results are when the study depends on inferring too much about specific users. I have never conducted a study which involved actively intervening (and would like to know how IRB approval was obtained to do so), but the number of inferences which must be made are difficult enough when simply observing user activity. I would think running manipulation checks for the user's perception of the intervening profile would be nearly impossible. Ultimately what is most difficult about Twitter studies is that the data collected is rarely available for inspection by anyone other than the researcher collecting it. When so many subjective decisions must be made in analysis this is particularly troublesome.