We are studying the difference in behavior between genders on an online community. We are only interested on those users who participate in the site and whose gender could be easily inferred by other users. We have two pieces of information that could help us identify a user's gender: their picture and their name. We decided to use a user's name to infer its gender.

Using the Global Name Database we try to find a user's name. If we find it and it's not labeled as "unknown gender", we add this user to our sample. Of course not all names are in there, so we don't have the entire population of users who we can identify its gender.

Our problem is: as our sample is not random, we don't know how to make sure we have a representative sample from all users whose gender is easily identifiable.

Is there a way to test if a sample is representative? Or, at least, verify if the errors my "classifier" makes does not add any bias to the sample?

  • $\begingroup$ What is the percentage of users that you are able to identify gender? $\endgroup$
    – alesc
    Commented Apr 25, 2015 at 19:39
  • $\begingroup$ @alesc Between 30% and 40% of active users. $\endgroup$ Commented Apr 26, 2015 at 22:38
  • 1
    $\begingroup$ I am not quite seeing why you want to have a representative sample of the general population. Or may be I am not understanding your problem -- what you stated as a problem is an obstacle in your attempt to solve the real problem. $\endgroup$
    – StasK
    Commented May 1, 2015 at 21:06


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.