Last update:

   29-Jul-2010
 

Arch Hellen Med, 27(4), July-August 2010, 691-707

APPLIED MEDICAL RESEARCH

The wrong application of P values and hypotheses test in biomedical research

P. GALANIS
Center for Health Services Management and Evaluation, Department of Nursing, University of Athens, Athens, Greece

Data analysis and interpretation of results in biomedical research continues even today to present problems. In particular, in the majority of cases, conclusions are based on hypotheses test and P values without taking into account the biological settings and the evidence from previous studies concerning a specific hypothesis. This form of deductive inference does not permit increase of knowledge about the natural world, and for this reason the abandonment of this type of inference and the adoption of Bayesian (inductive) methods is recommended, and specifically, the calculation of a measure known as the Bayes factor. The P value is a measure of discrepancy between the data derived from a study and the null hypothesis, and represents the probability, assuming that the null hypothesis is true, of obtaining a result as far as, or further than, what was actually obtained in a particular study. The application of hypotheses test aimed at the constraint of P values, has led in the opposite direction, with P value being used as a criterion for decision making. The P value does not constitute a component of typical inference and is used wrongly by most health scientists for the ascertainment of relationship between a determinant and frequency of occurrence of an outcome. The application of hypotheses test rejects, in practice, the application of inductive inference in each study separately for extracting conclusions, since the hypothesis test is a deductive method for restriction of error rate after the iteration of a particular study. It is encouraging that lately the efforts for the restriction of P values and the application of confidence intervals in the presentation of the results of a study have yielded fruit. Confidence intervals have considerable disadvantages, the most important being that they do not relate the evidence from a particular study with that from previous studies. This disadvantage of confidence intervals is overcome using Bayesian methods and especially with the calculation of Bayes factor.

Key words: Bayes factor, Confidence interval, Explanation, Hypothesis test, Inference, P value.


© Archives of Hellenic Medicine