Statistically Significant Harm Caused by Statistical Significance

I have argued before that the great importance placed on α = .05 in statistical studies is an attempt to replace educated judgment with a technical decision. But that decision itself is arbitrary: there is no particular reason to choose .05 over .04, or .06, or any other number less than .50.

It turns out it is even worse than I thought: an education that focuses on such a cutoff leads "researchers to interpret evidence dichotomously rather than continuously. Consequently, researchers may either disregard evidence that fails to attain statistical significance or undervalue it relative to evidence that attains statistical significance."

Education in statistics, at least as it is too often taught today in schools, makes one worse at likelihood judgments.

No comments:

Post a Comment