P Value : Unlocking its confusing nature
There is a worldwide problem regarding the P values. P values are not only understood poorly but misinterpreted among the scientific community.
Statistical significance is the least interesting thing about the results. You should describe the results in terms of measures of magnitude –not just, does a treatment affect people, but how much does it affect them.
-Gene V. Glass
The primary product of a research inquiry is one or more measures of effect size, not P values.
-Jacob Cohen
P value tells whether the results of an experiment is significant or not. But form that we cannot get any information about the practical application of the outcome of that experiment. On the contrary effect size conveys about the strength of the relationship between variables involved in a particular experiment. It needs to be understood that effect size is not sensitive to sample size whereas p values are influenced by sample size.
P values are confused with effect size. The measurable strength or size of a miracle isn’t the statistical significance of the discovery of the miracle from a statistical test. In numerous cases, once the presence of a particular miracle is established, the effect size caused by the detected miracle is frequently more important in assessing the implicit operation of the miracle. For illustration, in large genome-wide association studies (GWAS), numerous statistically significant complaint-applicable nucleotide variants have been detected with p values below 10-8 ; still, a maturity of these nucleotide variants have complaint odd rate below 1.5. In this environment, the odds rate is an effect size representing a fairly small increase in complaint threat, indeed though the detected complaint association is extremely statistically significant. Likewise, other statistics, similar as the E values from BLAST quests that represent anticipated number of hunt successes of the same position of similarity that would do by arbitrary chance for a given sequence database are frequently confused with p values. While p values are constantly calculated, E values are dependent on the database used in their computation, limiting their interpretation. Alternate, misapprehension of p value significance pollutes our scientific literature with significant false discovery. Numerous published scientific studies have defined “ significant ” p values in terms of weak baselines( i.e.0.05). Given the huge number of tests being performed in every scientific laboratory across the world, the selection of significance grounded on an nascence of 0.05 generates a lot of published false discovery across the combined scientific literature. This published false discovery can be buttressing when numerous others try to reproduce the same false discovery they’ve seen in the published scientific literature.
Results to this global p value problem aren’t easy. One journal has indeed taken the policy of banning the use of p values in its published papers. Also, certain scientific communities have worked together to establish guidelines that minimize false discovery. For illustration, the drugs community generally waits to accept major results until five sigmas of significance are reached. This equates to a two- tagged p value of 6 × 10-7. For mortal GWAS, p values lower than 5 × 10-8 is the standard for accepting results, which is grounded on a Bonferroni correction of an nascence of 0.05 assuming the presence of 1 million testable independent variants in the mortal genome. The field of data wisdom seeks to more understand this miracle and model it to produce a more robust dimension of significance that doesn’t limit discovery. A good launch is to report q values or other acclimated p values that help false discovery in published results, especially for published results involving high- outturn logical ways and other trial that generates high- dimensional data.
[1]. Berben, L., Sereika, S.M. and Engberg, S., 2012. Effect size estimation: methods and examples. International journal of nursing studies, 49(8), pp.1039-1047.
[2]. Nakagawa, S. and Cuthill, I.C., 2007. Effect size, confidence interval and statistical significance: a practical guide for biologists. Biological reviews, 82(4), pp.591-605.
[3]. Bioinformatics A Practical Guide to the Analysis of Genes and Proteins (Andreas D. Baxevanis, B. F. Francis Ouellette
