Abstract:
|
The use of P-value is pervasive in medical research, to an extent that it becomes the single mostly used measure to gauge for "significance". P-value is commonly misunderstood as the probability of null hypothesis being true. It overestimates the evidence against the null hypothesis. It does not estimate the magnitude of the treatment effect. It violates the likelihood principle as it depends on unobserved data. As a result, many "positive" studies cannot be replicated or do not lead to clinically meaningful results. In this talk, I will give an overview of the "P-value overdose" problems. Some antidotes will also be discussed including switching the statistical inference paradigm from frequentist to Bayesian, considering Bayes factor and positive predictive value instead of P-value, emphasizing more on estimation of the magnitude of the treatment effect rather than hypothesis testing, etc. The Bayesian approach is complementary to the frequentist counterpart but provides a direct and coherent assessment of the evidence contained in the data. It offers a superior alternative to the frequentist approach and gives a more accurate statistical assessment of the study results.
|