Some Misinterpretations of Inferential Statistics in Dental Public Health Literature Publisher Pubmed



Alshihayb TS ; Alnasser LA ; Alsoneidar WA ; Mansournia MA
Source: BMC Oral Health Published:2025

Abstract

Objectives: Inferential statistics such as p-values and confidence intervals (CIs) are ubiquitously used in research studies. Still, researchers can incorrectly interpret them, impacting the validity and utility of the respective results. The aim was to quantify how often studies commit one incorrect interpretation, dichotomization, in the dental public health literature. Methods: The authors carried out an electronic search using PubMed to extract original papers published in 2018/2019/2023 (either online or in print) in five dental public health journals. Four trained and calibrated reviewers extracted information from the abstract and main text on the following: reporting any p-value (Yes/No), reporting p-value as inequality (Yes/No/NA), reporting non-significant p-value (Yes/No/Not applicable (NA)), reporting any confidence interval (Yes/No), reporting non-significant confidence interval (Yes/No/NA), and concluding there is no association because the p-value or confidence intervals were not-significant (Yes/No). In addition, investigators extracted if studies explicitly reported that p < 0.05 was considered significant. Results: The results indicate that p-values were reported more frequently than CIs in both the abstract and main text. The majority of studies interpreted non-significant p-values or CIs to mean no association or no effect in their main text. Also, non-significant p-values and CIs were less frequently reported in abstracts compared to the main text. These findings varied across the five dental public health journals, but were less notably changed by the print publication year. Conclusion: The results of this paper showed that some common misconceptions about p-values and CIs still linger in the dental public health literature after seven years had passed since warnings against such practices. More advanced training may be needed to overcome the issues with p-values, confidence interval reporting, and interpretation. © 2025 Elsevier B.V., All rights reserved.