Věcná významnost výsledků výzkumu

Abstract

This paper deals with the concept of effect size of results in scientific research, emphasizing its irreplaceable role in interpreting results that would otherwise be evaluated only on the basis of statistical significance. The paper shows the basic measures of effect size for various types of statistical tests - both parametric (e.g. Cohen's d, eta^2) and non-parametric (e.g. parameter r, point biserial correlation). These measures are subsequently applied to practical examples from the field of medicine. The interpretation of effect size and statistical significance is compared here and the degree of practical impact of the results is evaluated. The importance of choosing the size of the sample size is also pointed out. The analyses point to the fact that statistically significant results do not necessarily have to be substantively significant and vice versa. For example, the results of simulations show that with a large sample size, statistical significance can be achieved even with a negligibly small substantive significance. On the contrary, for larger effects that have practical value, the results may not be statistically significant. Including the measure of effect size in the evaluation of research results is a necessary step towards achieving higher quality of scientific works. The work emphasizes the need for more consistent adherence to publication standards (APA, AERA), which reflect this obligation.