Saying Farewell to p Values and Welcoming Confidence Intervals in Data Analysis

The finding or not of statistically significant relationships between a determinant and the frequency of occurrence of an outcome relies on the comparison of p values, which are extracted from various statistical tests, using an arbitrarily selected value, determined by the researchers, known as the alpha level. The alpha level usually used is 0.05. Over the last 30 years, the application of p values for the derivation of conclusions in health sciences has been rightly criticized. The mathematical and conceptual approach of hypotheses tests and p values was an important step in statistical analysis, but there are inherent problems in their interpretation and practical application. Ideally, the data analysis and presentation of the results of a study should include the measures of association (or rarely the measures of frequency) and the corresponding confidence intervals (CIs) that indicate the precision of measurement. Statistical estimation can provide this information. The CI is a range of values around a measure of association, computed in a study, and it displays the degree of statistical precision of the estimation. A wide CI indicates lower precision, while a narrow CI higher precision. The p value, after the comparison with the alpha level, indicates only the existence or not of statistical significance without making clear whether the index category of the determinant under study increases or decreases the frequency of occurrence of outcome. For this reason, the measure of association should be reported that indicates the magnitude of the relation between the determinant and the frequency of occurrence of the outcome, along with the corresponding CI, that indicates the precision of measurement. The validity and credibility of the results of a study should not be determined by the finding of a statistically significant relationship alone, but by the research design and the reduction of errors. Although p values should not be eliminated from the presentation of the study results, it would be beneficial to introduce the measures of association and the corresponding CIs. The knowledge of a CI includes indirectly the knowledge of the presence or absence of statistical significance, rendering p values unnecessary when CIs are used.