Jan Sprenger (University of Turin)
Statistical significance tests cannot quantify evidence in favor of the null or default hypothesis. Hence, many experimental findings which do not speak "significantly" against the null do not get published and end up in the proverbial "file drawer", contributing to bias and lack of replicability. My contribution addresses this problem by explicating the concept of degree of corroboration (of the null hypothesis), combining Popperian intuitions with modern Bayesian---and frequentist---statistics. Degrees of corroboration cure most deficits of significance tests: they allow for a more nuanced and less misleading assessment of the null hypothesis than p-values or confidence intervals.
