UO issues RFP for study of gender and racial discrimination

More on this later, this post is just a placeholder. The RFP is here. The main output looks like a simple regression, for which UO will provide the data. I’m no econometrician, but it’s about an hour’s work – maybe two if you look at time trends, instead of just the proposed 2017 snapshot. It will be interesting to see how much the consultants’ bids come in at:

Bookmark the permalink.

11 Responses to UO issues RFP for study of gender and racial discrimination

  1. Old Man says:

    Three hours. Because logarithms…

    VA:F [1.9.22_1171]
    Rating: +3 (from 3 votes)
  2. uotechmatters says:

    uomatters, why not submit a bid?

    VA:F [1.9.22_1171]
    Rating: +1 (from 1 vote)
    • UO Matters says:

      Because I’d be embarrassed charging my university for doing the kind of work I’d assign as a homework exercise?

      VN:F [1.9.22_1171]
      Rating: +8 (from 8 votes)
  3. Anonymous says:

    “University reserves the right to approve the study’s final design, including variables and criteria.” Translation: We’re gonna p-hack.

    VA:F [1.9.22_1171]
    Rating: +3 (from 3 votes)
    • UO Matters says:

      The union’s MOU with the administration says they’ll consult with the faculty union on study design.

      I’m no member of the union’s executive committee, but if I were I’d certainly enjoy those discussions, particularly if the consultant is paid a flat-rate and not by the hour.

      VN:F [1.9.22_1171]
      Rating: +3 (from 3 votes)
      • Anonymous says:

        That’s good but wouldn’t really stop them. They’ve already got the data, they can take any proposed analysis, quietly run it themselves, and if it looks bad come up with a reason to change it at the proposal stage. Then a few months later boom, consultant comes back with “independent” evidence that everything is fine.

        Reasonable-sounding justifications for fiddling with models are easy to gin up and can see easily push results over or under statistical thresholds if you get to look at the data while you’re making them. Don’t take my word for it, try it here.
        https://projects.fivethirtyeight.com/p-hacking/

        Will they actually do this? Who knows, but little would shock me. The union should insist at the start of the process that the dataset is sequestered somehow and nobody can or will look at it. And any analysis proposed along the way should be run and mentioned in the report, even if all sides agree on modifying it for the “final” proposal.

        VA:F [1.9.22_1171]
        Rating: +1 (from 1 vote)
        • Hippo says:

          “Statistical significance” and “p-hacking” all assume there is a well-defined model for noise and uncertainty in the data (such as sampling error). Beyond description statistics, I am not sure what use models are here. A census says whatever it says — unless you can make a credible case for modeling noise here, no need for fancy inferences. Although that never stops econometricians…

          VA:F [1.9.22_1171]
          Rating: 0 (from 0 votes)
          • adamsmith says:

            The question in most wage discrimination cases is what you do or don’t control for.

            Control for occupation and the gender wage gap goes to zero. Is that because females select occupations with flexible hours which pay less because they want flexibility, or is that because they are shut out from leadership occupations with big wage premiums or some combination of the two?

            For this wage analysis, what you do and don’t control for is pretty always pretty key.

            VA:F [1.9.22_1171]
            Rating: +1 (from 1 vote)
            • Hippo says:

              By “control for” I suppose you mean “throw into the right-hand side of some linear model”. If, after doing so, the fitted coefficient of gender is non-zero, or “statistically significant” (I’m sure we are untroubled with the modeling assumptions producing the standard errors spit out by your favorite software package) we are supposed to believe there is discrimination? Given the model inadequacies, and in particular an inability to verify the independence of any remaining error on the right-hand side, this is pretty weak thinking. Oh right, bring on the instrumental variables. LOL.

              VA:F [1.9.22_1171]
              Rating: 0 (from 0 votes)
            • jackmccoy says:

              What are instrumenting for? Is there a causal question we would need an instrument (before we even bring up if we can find one)? By bringing up IV in a place where its of no use you’re showing your own ignorance of wage decompositions and how they’re used in court cases involving discrimination. You can laugh at econometrics, but I’m laughing at how little you understand about the thing you’re mocking.

              VA:F [1.9.22_1171]
              Rating: 0 (from 0 votes)

Leave a Reply

Your email address will not be published.