There are several cases of wrong use of statistical tests in the scientific literature. This cuts across various categories of journals (though commonest in low-quality journals). For instance, if you search the literature, there are more studies with ANOVA than Kruskall Wallis test even when assumptions for parametric tests (sample size) are violated. Another case, is LSD versus Bonferroni posthoc test. The use of LSD increases the rate of false positive errors. This can give rise to articles with a lot of false positive results. This may contribute to the increased failure of animal findings in clinical settings (That is, inability to translate preclinical research into clinical findings).
This problem arises because a lot of reviewers do not focus on statistical analyses during peer review. I am of the view that every journal should have an Editorial Board member who is knowledgeable in statistics and screens every manuscript for correctness of statistical tests. I would like to hear various views (insights) on this topic.
Any reviewer that fails to check the correctness of stat choice and interpretation is not fit to be a reviewer. Most times, the downside of a study report is largely on statistics. Statistics include estimation of sample size, and in some studies, this may be required. I once reviewed a study where intervention was based on the p-value of a T-test while the study aimed to know the cause of an outcome...with several variables, multiple regression was not done despite the robustness of the study the researchers continued to do repeated T-test. No matter how small, every reviewer must be a 'statistician' to some extent, or refer researchers to their statistician when an error is noticed in the choice of their stat analysis.
Yes, I agree with you Omotayo Erejuwa ; however, this also depends on reviewers' knowledge of the appropriate statistical tests for each study design as you mentioned. Having each manuscript checked for correctness of statistical tests by a biostatistician before sending the manuscript for peer-review can be of great help.
Dr Omotayo Erejuwa I quite agree with you. I have made similar observations in animal studies where most researchers deal with a small number of animals per group and the data is tested for normality will certainly be non parametric, yet you find such researchers using independent T test of oneway ANOVA instead of Mann-Whitney U test or Kruskal-Wallis test respectively. This is really worrisome and calls for journals to be very serious about the statistical aspect of a research. Mustapha Umar Imam Muhammad Bashir Bello Kasimu Ghandi Ibrahim Ibrahim Malami Bilyaminu Abubakar
Dr Murtala Bello Abubakar I am glad that you too have observed this same problem. Honestly, it is very worrisome. This affects the interpretation of results. The Discussion and Conclusion of every scientific paper are based on the interpretations of the Results.
Any reviewer that fails to check the correctness of stat choice and interpretation is not fit to be a reviewer. Most times, the downside of a study report is largely on statistics. Statistics include estimation of sample size, and in some studies, this may be required. I once reviewed a study where intervention was based on the p-value of a T-test while the study aimed to know the cause of an outcome...with several variables, multiple regression was not done despite the robustness of the study the researchers continued to do repeated T-test. No matter how small, every reviewer must be a 'statistician' to some extent, or refer researchers to their statistician when an error is noticed in the choice of their stat analysis.
I forgot to add that this topic is in commemoration of peer review week which is 21-25, 2020
https://peerreviewweek.wordpress.com/
Kamoru A. Adedokun this is absolutely right but I think more importantly and probably practicable is for all journals to ensure that they have a statistician who can either verify or do a quick check on the appropriateness of the statistics before sending the article out for review.
Exactly, statistical analysis is the core of original research.
Mosab Nouraldein Mohammed Hamad and Serdar Kalemci Yes, very important.
- Badges
- Science topic
- Similar topics
- Linguistics
- Philology
- Classical Philology
- Manuscripts