Portal de Periódicos da CAPES

Stephen T. Ziliak and Deirdre N. McCloskey's The cult of statistical significance: how the standard error costs us jobs, justice, and lives. Ann Arbor (MI): The University of Michigan Press, 2008, xxiii+322 pp.

2008; Volume: 1; Issue: 1 Linguagem: Inglês

10.23941/ejpe.v1i1.12

1876-9098

Aris Spanos,

The stated objective of this book is to bring out the widespread abuse of significance testing in economics with a view to motivate the proposed solution to the long-standing problem of statistical vs. substantive significance based on re-introducing 'costs and benefits' into statistical testing.The authors strongly recommend returning to the decisiontheoretic approach to inference based on a 'loss function' with Bayesian underpinnings, intending to ascertain substantive significance in terms of "oomph, a measure of possible or expected loss or gain" (Ziliak and McCloskey 2008, 43).The idea of a 'loss function' was introduced by Wald ( 1939), but rejected later by Fisher (1955) who argued that when one is interested in the truth/falsity of a scientific hypothesis, the cost of any actions associated with the inference is irrelevant; this does not deny that such costs might be relevant for other purposes, including establishing a range of substantive discrepancies of interest.This is still the prevailing view in frequentist statistics, which, to use one of the authors' examples (Ziliak and McCloskey 2008, 48), rejects the argument that to evaluate the substantive discrepancy from the Newtonian prediction concerning the deflection of light by the sun, one needs a loss function which reflects the relevant 'costs and benefits'.How do the authors justify wedging the notion of a loss function back into econometrics?They interpret it in terms of 'economic cost' and trace the idea back to Gosset (1904); described pointedly as "a lifelong Bayesian" (pp.152, 158, 300).How do they make their case?Curiously enough, not by demonstrating the effectiveness of their recommended procedure in addressing the statistical vs. substantive significance problem using particular examples where other 'solutions'