- Do not only report p-values. State what values were compared, e.g. provides the means or proportions in each group you are comparing), and provide the df (degrees of freedom) if appropriate, and the test value (chi-square, t-test value etc).
- P is always italicized and capitalized.
- Do not use 0 before the decimal point for statistical values P, alpha, and beta because they cannot equal 1, in other words, write P<.001 instead of P<0.001
- The actual p-value* should be expressed (P=.04) rather than expressing a statement of inequality (P<.05), unless P<.001.
- P-values should not be listed as not significant (NS) since, for meta-analysis, the actual values are important and not providing exact p-values is a form of incomplete reporting.
- If P>.01 then the p-value should always be expressed to 2 digits whether or not it is significant. When rounding, 3 digits is acceptable if rounding would change the significance of a value (eg, you may write P=.049 instead of .05).
- If P<.01, it should be expressed to 3 digits.
- For p-values less than .001, report them as P<.001, instead of the actual exact p-value. Expressing P to more than 3 significant digits does not add useful information since precise p-values with extreme results are sensitive to biases or departures from the statistical model.
- P=.000 (as outputted by some statistical packages) is impossible and should be written as P<.001
-
All footnote symbols within a table are superscripted letters (a-z). The use of *, **, *** footnotes to mark significance levels (eg, P<.05, P<.01, P<.001) is discouraged. Authors are asked to provide exact p-values instead. While we prefer the exact p-values, there are exceptions to this general rule. The use of *, **, *** footnotes to indicate significance levels is allowed in the following instances:
- In tables of systematic reviews, which tend to be busy and where the original p-values can't be found in the original publications
- When Odds Ratios instead of p-values are presented
- If for any reason authors are unable to provide the exact p-values
In rare instances, p-values may be expressed to more than 3 decimal places (eg, P =1×10−5) in cases of genome-wide association studies, studies that involve the Bonferroni correction, etc.
* Why actual p-values? The traditional reporting of p-values (indicating only that P<0.05) simply indicated whether the results were "statistically significant" or not. But p-values of 0.051 and 0.049 should be interpreted similarly despite the fact that the 0.051 is greater than 0.05 and is therefore not "significant" and that the 0.049 is less than 0.05 and thus is "significant." Reporting actual p-values avoids this problem of interpretation.
For guidelines on other statistics, please see: Guidelines for reporting statistics
Authors who are not sure how to report their quantitative results should consult the following book (on which the above guidelines are based):
How to Report Statistics in Medicine: Annotated Guidelines for Authors, Editors, and Reviewers (Medical Writing and Communication) Thomas A. Lang
Comments
1 comment
If the p value refers to the area under the curve of a probability distribution model e.g. T, Z or Chi distribution each of which have a tail(s) to infinity, is it not incorrect to state a p value equal to a number between 0 and 1? There is no area under the point on a curve. The p value in such circumstances is an integral between two points one of which is an asymptote. Therefore in null hypothesis significance testing the p value should always be reported as less than some value.
If the p value refers to an instantaneous probability such as one might encounter in the graph of time to event analyses then in such a model the p value is a differential of some point on the curve. I assume then and only then would it be appropriate to state p equal to some value between 0 and 1.
Please correct me if I am wrong in my interpretation.
Please sign in to leave a comment.