Clear reporting allows readers to confirm that an appropriate test was used and makes it easier to identify and fix potential errors prior to publication. Without the original data, it is difficult to determine how the test results would have been different had an appropriate test been used. Analyses that use inappropriate tests give potentially misleading results because the tests make incorrect assumptions about the study design or data and often test the wrong hypothesis. For example, if authors use an unpaired t-test when a paired t-test is needed, the failure to account for repeated measurements on the same subject will lead to an incorrect p-value. Vague statistical methods deprive reviewers, editors and readers of the opportunity to confirm that an appropriate type of t-test or ANOVA was used and that the results support the conclusions in the paper. First, there are many different types of t-tests and ANOVAs. However, statistical methods in these papers are often limited to vague statements such as: 'Data were analyzed by t-tests or ANOVA, as appropriate, and statistical significance was defined as p<0.05.' There are several problems with such descriptions. T-tests and analysis of variance (ANOVA) are the statistical bread-and-butter of basic biomedical science research ( Strasak et al., 2007).

These policies may include specifying which test was used for each analysis, reporting test statistics and exact p-values, and using dot plots, box plots or other figures that show the distribution of continuous data. Other journals, including Nature and affiliated journals ( Nature, 2013 Nature, 2017), eLife ( Teare, 2016) and The EMBO Journal ( EMBO Press, 2017) have recently implemented policies to encourage transparent statistical reporting. The journal Science has attempted to improve statistical reporting by adding a Statistical Board of Reviewing Editors ( McNutt, 2014). However, these guidelines have not been widely adopted.Ĭlear statistical reporting also allows errors to be identified and corrected prior to publication. Errors in reported p-values are also common and can sometimes alter the conclusions of a study ( Nuijten et al., 2016).Ī main principle of the SAMPL guidelines for reporting statistical analyses and methods in the published literature is that authors should 'describe statistical methods with enough detail to enable a knowledgeable reader with access to the original data to verify the reported results' ( Lang and Altman, 2013). Many studies omit essential details when describing experimental design or statistical methods ( Real et al., 2016 Lazic et al., 2018). A more recent study of papers reporting the results of experiments that examined the effects of prenatal interventions on offspring found that the statistical analyses in 46% of the papers were invalid because authors failed to account for non-independent observations (i.e., animals from the same litter Lazic et al., 2018). A study published in 2011 found that half of the neuroscience articles published in five top journals used inappropriate statistical techniques to compare the magnitude of two experimental effects ( Nieuwenhuis et al., 2011). Older studies suggest that statistical errors, such as failing to specify what test was used or using incorrect or suboptimal statistical tests, are common ( Müllner et al., 2002 Ruxton, 2006 Strasak et al., 2007), and more recent studies suggest that these problems persist. Problems with the statistical analyses used in published studies, along with inadequate reporting of the experimental and statistical techniques employed in the studies, are likely to have contributed to these concerns. The inability to reproduce key scientific results in certain areas of research is a growing concern among scientists, funding agencies, journals and the public ( Nature, 2013 Fosang and Colbran, 2015 National Institutes of Health, 2015a National Institutes of Health, 2015b Nature, 2017).

Paired t test jmp