There are few people I know who feel completely competent in statistics, and yet it is one of the most important parts of research. Large research teams may have a statistician on their team, but many do not. Even with a statistician on the team, there can be differences of opinion among statisticians concerning study design, analyses, and interpretation. After all, statistics is a mathematical science. While the process of descriptive statistics is fairly logical to me, inferential statistics and probability theories take me out of my comfort zone.
At JNEB, every manuscript does not receive a statistical review from our volunteer statistical reviewer team. Our editors would love this, but it would place a fairly high burden on these volunteers. We do request a statistical review for all clustered randomized controlled trials and usually for any research with hierarchical modeling, not because we feel our authors have done anything wrong but because we want to support your getting it right.
To help us do this, authors should be explicit in their description of the statistical procedures in their Methods section. Much more is needed than the software used. If the analyses were to be repeated, what would someone unfamiliar with the dataset need to know? Include how outliers and missing data were treated. There are arguments for removing influential variables as well as leaving them in or running the analyses both ways. Just tell us about your process. Remember for non-normally distributed samples, the median and interquartile ranges should be presented rather than means and standard deviations (SD).
Cross-sectional data should begin with a test of distribution. Include whether data were normally distributed and how this was ascertained. For large databases, sampling procedures and weighting methods should be included.
In our discipline, regression is not described as “predicting” an outcome, but rather explains the percentage of variance within a dependent or outcome variable. It is helpful if authors are explicit in listing their dependent and independent variables, as well as how and why they are entered into the equation. We don’t require variance inflation factors (VIF), but this process may be helpful for large or exploratory analyses to understand if or how much collinearity exists.
When presenting regression data for JNEB, include R and R2, and when appropriate, effect sizes. Standardized betas and confidence intervals (CI) are also important for the story you are telling because it is a story and your readers should be able to understand what you are saying about your analyses. I sometimes hear that authors feel word limits hamper their ability to tell their statistical story. Rest assured that editors will better appreciate a complete story and will suggest edits if needed to shorten the manuscript rather than asking for many explanations on revises.
Statistically speaking, those who love statistics could write a much more informative editorial on statistical guidelines, and those who close their eyes, click “run” and watch what comes out of an analysis will not have waded through this editorial. However, this is a very important part of your manuscript. We have a few statistics-related webinars on JNEB.org as well as additional guidelines for authors and reviewers. Help us move your manuscript through review to publication quickly by being complete and succinct in your statistical methods and interpretation. Believe me, it could be significant.
This editorial was originally published in the February 2020 issue (Vol. 53, Issue 2) of the Journal of Nutrition Education and Behavior.