Skip to main content
Cross Validated

You are not logged in. Your edit will be placed in a queue until it is peer reviewed.

We welcome edits that make the post easier to understand and more valuable for readers. Because community members review edits, please try to make the post substantially better than how you found it, for example, by fixing grammar or adding additional resources and hyperlinks.

6
  • 1
    $\begingroup$ No offense here but something is missing. Is this treating the symptoms while ignoring the disease? Is a PC language solution a solution at all or is it problem in it's own right? Language is not the primary source of feelings, only a means of expressing them. Feelings are reactions to situations. Fix the situations, rather than paper them over with language. $\endgroup$ Commented Jan 22, 2019 at 6:06
  • $\begingroup$ @Carl: If a dataset is rotten with bias already, it’s very difficult to correct. Political correctness is an attempt at mitigating bias. In the ML case we are guiding the model away from biasing it’s predictions based on gender to those that are purely based on skill. Otherwise such a model will likely penalize females and assign very different scorings to their skills at each profession. Read Dave Harris’s answer for additional ways bias arises and how to fix them through physical changes (as opposed to data changes) $\endgroup$ Commented Jan 22, 2019 at 7:34
  • $\begingroup$ I commented on Dave's answer, so your suggestion shows you missed that. You may also be missing the point that to win a job you hate is a problem. Fix the retention problem, and the working environment. Making females more "attractive" solves nothing, it can exacerbate the problems. With respect to a job, the problem is not "getting married" to it, but "staying married". $\endgroup$ Commented Jan 22, 2019 at 12:46
  • 1
    $\begingroup$ @Carl: I'm not sure what you're arguing here, as OPs question is clearly asking about how to build a statistical model on an existing dataset. The links I provided show that language models, out-of-the-box, can already contain hidden biases. I could just as well argue that people who keep their jobs long enough are likely too mediocre to find jobs elsewhere. Regardless of what KPI you're optimizing (that's a relevant but, completely separate topic), your model may still exhibit gender biases. $\endgroup$ Commented Jan 22, 2019 at 21:06
  • 1
    $\begingroup$ Agreed. You did answer the question. However, female job retention of Tech jobs is poor and you did not identify the problems. So the answer is a disservice to females. If used it will cause misery. Statisticians have the moral responsibility to see their work in context and identify the questions that are more appropriate than those posited naively. $\endgroup$ Commented Jan 22, 2019 at 21:59