Tips on Reporting P-Values
Institutional researchers often encounter "p-values" in their work, especially with the wide availability of modern statistical software . To help avoid the misuse of software-generated p-values, two researchers (Levine & Atkin) recently published a note that warns of three particular errors that researchers may make in reporting these statistics. The tips are as follows:
- "The first reporting error...is when a result is reported as significant at p = .000. Obviously, a value with a zero probability of occurrence is, by definition, an impossible value. So, no observed event can ever by p = exactly zero...An easy solution is to round up and report such results as p < .005 or p < .001. .." [p.325]
- " A second error involves rounding a result that is technically p > .05 down to p = .05 and reporting the result as p = .05 or p < .05... technically, any value even slightly larger than p = .0500 is not statistically significant at p<= .05...Reporting otherwise is factually inaccurate...To avoid this error, authors should report the p-values out to the third decimal, that is report as p = .051, or round up and report the value of p = .051 as p < .06..." [p.325]
- " The third error can be made when reporting one-tailed p-values from readouts when wrong direction effects are found. In this case, many statistical packages will calculate and report a p-value presuming that the effect was in the predicted direction...Again, simply reporting such results as p > .05 or p = ns is a simple solution..." [p.326]
This note should remind researchers to exercise rigor in the reporting and interpretation of their output from modern statistical programs. Many researchers may have missed this level of detail in their training/education, and even those who have heard this advice before may tend to forget it as the years go by...
Timothy R. Levine (Professor, Department of Communication, Michigan State University) and Charles Atkin (Professor and Chair of the Department of Communication, Michigan State University) document these points in the article ("The Accurate Reporting of Software-Generated p-Values: A Cautionary Research Note") in Communication Research Reports (Vol.21, No.3, Summer 2004, pp. 324-327). The article includes 22 references.
[Abstract done by Willard Hom, Director, Research & Planning Unit, System Office, California Community Colleges, 1/24/05]
|
