Journal of Biostatistics & Biometrics The w-value: An Alternative to t-and χ 2 Tests
Résumé
Reproducibility is central in science and statistics are one of the tool used to ensure that conclusions are supported by data. However, high false positive rate is a common problem denounced regularly when hypotheses testing using null and alternative hypotheses are used. The origin of this high false positive rate is deeply anchored in the method because the reported p-value is not what generally the researcher was thought to test: the p-value is the probability to observe the statistics summarizing the data if the null hypothesis H 0 is true, but the researcher would generally like to know what the probability is that H 0 is true given the observed data. We propose an alternative to t-test and χ 2-test, two of the most common tests, that used model comparison using Schwarz's Bayesian information criterion weights. Thus, we were able to answer the question that generally the researcher wants an answer: what is the probability that these k-series are obtained from a single set of parameters and then what is the probability that they are identical.
Domaines
| Origine | Fichiers éditeurs autorisés sur une archive ouverte |
|---|---|
| Licence |