STATISTICAL SIGNIFICANCE AND REPLICABILITY - RESPONSE TO CHOWS AND FALKS COMMENTARY

Commentators agree with me that statistical significance does not beto ken replicability, but not for the reasons I give. Chow (1998) defends the use of significance tests by arguing that they are useful for ass essing the role of chance in findings. I dispute this by pointing out that the Type II error rate may be as high as 1 - alpha. Falk (1998) d efends Bayesian inference, which was impugned by me. She also makes a case for the use of replications for the purpose of testing hypotheses . I complain that the supporters of Bayesian inference have not explai ned adequately what its purpose or role in science is, and for this re ason it is an impediment to understanding in this area. I also dispute Falk's view of the role of replications in that I would require a suc cessful replication to be one that produces an effect that is clearly discernible, meaning the effect is consistently noticeable with the ai d of nothing more than descriptive statistics.