Null Effects and Replication


Categories: Comedy, Error Statistics, Statistics

Post navigation

3 thoughts on “Null Effects and Replication

  1. Isn’t this a great cartoon? I especially like the lock on the data box.

    • Unfortunately, now that I’ve had a look at Johnson’s paper, I see it commits the same old fallacies of other attempts by Bayesians or quasi-Bayesians to come in and rescue significance testers by foisting make-believe priors of .5 on nulls (meaning in this case, 50% of them are true) and telling them to keep up the good recipe work. What a disappointment, too, to discover he hasn’t bothered to understand the methods he’s rescuing, e.g., claiming tests have no way to use information from failures to reject! And he’s alluding to N-P power! (never mind that he gives it a Bayesian distorted reading), he’d surely know we consider alternatives and that power analysis, confidence intervals, and severity assessments allow identifying discrepancies ruled out by non-rejections.

Blog at