Chair: Peter R. Killeen (Arizona State University)
Null Hypothesis Statistical Testing (NHST) was developed to provide an objective way to quantify inference. The result is a ritualized technique that is frequently necessary for publication despite criticisms that it is minimally informative, misleading, and produces unreproducible results. NHST tests the probability of the data given a null hypothesis that is rarely of interest and is often implausible. The result is a torturous statement of whether the data are likely to have occurred. An alternative approach, called Information Theoretic (IT) based inference, does not carry many of these problems because it returns a different probability. IT approaches ask the question of interest in model building: Of a set of models, which ones are best? And by how much? By building upon Akaike Information Criteria, IT inference returns the probability of the models considered given the data, numbers that are readily interpretable. Unlike NHST, the approach actually encourages the testing of many models in order to increase the chances of including good ones. Corrections for multiple comparisons are neither necessary nor appropriate. The tutorial will identify criticisms of NHST, offer a (relatively) nontechnical background for IT approaches, and provide examples of ITbased inference using spreadsheets.
Негізгі бет Christopher Newland, What's the Best Model for These Data? Information Theoretical Approaches, SQAB
Пікірлер