Skip to content

Adjusting for Type M error

Erik Drysdale discusses and gives some formulas, demonstrating on an example that will be familiar to regular readers of this blog.


  1. Ron Kenett says:

    Very nice piece by Erik Drysdale – thank you for pointing it out.

    1. A higher level view of statistical inference was discussed in this blos based on

    2. A broader perspective motivated by the often verbal representation of findings is to apply the S type error. See

  2. psyoskeptic says:

    When he got to “all statistically significant findings are biased”, in bold no less, I lost interest but I’m sure I’ll press on. Nevertheless, I don’t think everyone will be motivated to read a paper that’s clearly factually incorrect in a bolded statement that underlines the premise of the paper, in the first few paragraphs.

    • Christian Hennig says:

      I didn’t like that statement either, but it didn’t stop me from going further, and I didn’t regret it. What he means is that generally effect estimates are biased conditional on statistical significance, which is a valid point; later he doesn’t do much market shouting of that kind anymore.

Leave a Reply