Abstract:
|
Introduction. Improving statistical analyses is important for rigor and reproducibility (R&R) in research. The journal Nature is trying to improve R&R by expanding methods sections and requiring research data. This paper examines how well basic biomedical scientists describe statistical analyses (i.e., type of error bars, program used) and if they provide data. Methods. A search on the Nature web site for cancer found 605 papers in 2019. When restricted to research, 187 remained (21% of research papers in 2019). A random sample of 50 was used. Data collected include types of statistical tests. Results. The most common program is GraphPad Prism (40% of papers) then R (32%); 34% did not list one. Most papers (87%) did tests for 2 independent groups; 40% did Kaplan-Meier survival analyses; 38%, a one-way ANOVA; and 38%, a two-way ANOVA. Multiple comparisons without adjustment occurred in 57% of papers. The SEM is the most common error bar used (48%), followed by SD (40%). Only 15% used CI. Most papers (87%) provided data. Discussion. These results can identify deficiencies in statistical education to basic biomedical scientists and help judge part of Nature’s efforts to improve R&R
|