Skip Navigation

Statistical Guidelines and Checklist

In the Methods section:

• Type and applicability of test used are stated
• Comparisons of interest are clearly defined
• Names of tests applied are clearly stated
• All statistical methods are identified unambiguously
• Justification for use of test is given
• Data meet all assumptions of tests applied (with particular attention to nonnormal
data sets or small sample sizes, which should be identified in the text as
such)
• Adjustments made for multiple testing are explained

Details about the test

• N is reported at the start of the study and for each analysis thereafter
• Sample size calculation (or justification) is given
• Unit of analysis is given for all comparisons
• Alpha level is given for all statistical tests
• Tests are clearly identified as one- or two-tailed. If one-tailed, a justification is
necessary
• Actual P values are given for primary analyses
P values shall be reported to two decimal places
Note the following exceptions: Borderline P values (0.045 ≤ p < 0.055) can
be reported to three decimal places. P values < 0.01 can be reported to one
significant digit irrespective of the decimal places (<0.01, < 0.001, < 0.0001, etc.)

Descriptive statistics summary

• N for each data set is clearly stated
• A clearly labeled measure of center (e.g., mean or median) is given
• A clearly labeled measure of variability (e.g., standard deviation or range) is given
• All numbers following a ± sign are identified as standard errors (s.e.m.) or
standard deviations (s.d.)

Anomalies

• Any unusual or complex statistical method is clearly defined and explained for
the journal’s wide readership (authors are encouraged to use Supplementary
Information for long explanations)
• Any data exclusions are stated and explained
• Any discrepancies in the value of N between analyses are clearly explained and
justified
• Any method of treatment assignment (randomization, etc.) is explained and
justified
• Any data transformations are clearly described and justified

Multivariable models

• The type of model should be clearly identified
• All variables included in the final model as adjusters shall be listed
• All variables that could have been included but were not due to a stepping
procedure should be listed, with a general description of why they were chosen
• If a stepping procedure is used, it should be explicitly described: e.g., forward,
backward, forward-and-backward stepwise, explicit hierarchical (forced
variables) or other
• Were assumptions of the model and goodness of fit assessed?
• Were interactions with the variable of primary study interest assessed?
• Main-effects estimates should not be reported if a substantial interaction is
detected unless the interaction is removed by a described method such as
stratification or centering
• When reporting parameter estimates (B, odds ratios, hazard ratios, etc.), 95%
confidence intervals as well as P values should be reported

Tables and figures

• All tables and figures shall be referred to in the Results narrative and should be
numbered in the order in which they appear
• Data reported in tables and figures can be highlighted or summarized in the
Results narrative but should not be merely duplicated
• Tables and figures must be interpretable without reference to the text. Thus,
abbreviations need to be defined (perhaps in footnote or legend) even if
defined in text. N’s, including the denominators of reported %, should be clear
• P values that are reported in the body of the table or figure shall follow the
guidelines for P values above. However, for large tables of data not central to the
hypothesis, superscripted footnotes such as denoting <0.05 or <0.01 are allowed

Interpretations

• Statistical significance should not be used as a synonym for truth. When
reporting findings that are significantly associated, keep in mind the possibility
of type I error. When reporting findings that are not significantly associated, take
into account the possibility of type II error. When “significant” or “not significant”
are used in the Results narative, the actual P value should be reported (even if in
parentheses). Usually, when reporting that variables are (or are not) associated
or correlated, the qualifier “significantly” should be included. Examples: “systolic
blood pressure and age were significantly associated (P = 0.03) ” or “diastolic
blood pressure and education were not significantly associated (P = 0.26)”.

Within individual graphs:

Distortions
• Any distorted effect sizes (e.g., by truncation of y axis) are clearly labeled and
justified


Clear labeling
• Error bars are present on all graphs, where applicable
• All error bars are clearly labeled