Tag: p-value

About ASA’s Statement on p-Values

In March 2016, the American Statistical Association (ASA) published an important statement about p-value12. This statement was motivated by wrong usage and interpretation of p-value in scientific research.
Since I am passionate about this subject, I intend to write a few posts referring to some points of the ASA’s statement. You may find those posts with tag “p-value” (or “valeur-p” for the posts in french).

  1. Ronald L. Wasserstein & Nicole A. Lazar (2016) The ASA’s Statement on p-Values: Context, Process, and Purpose, The American Statistician, 70:2, 129-133, DOI: 10.1080/00031305.2016.1154108; Link : https://doi.org/10.1080/00031305.2016.1154108W↩
  2. A french unofficial translation of the ASA statement is available here on this blog↩

The curse of the p-Value

I translated the french quotations in this text as much literally as I could. I wrote them in italic just after the quotation

The french publication Pour la Science (french edition of Scientific American) published first quarter 2018 a special issue: BIG DATA : vers une révolution de l’intelligence (“BIG DATA: to a Revolution in Intelligence?”). The article titled “La malédiction de la valeur-p” (“The p-value curse”) addressed a subject I was particularly interested in: the wrong usage of p-value in scientific research. The article was interesting with a good coverage of the topic. But I noticed two statements not only wrong but also inconsistent with the article substance: the headline, titled “l’ESSENTIEL” (“The Gist”) stated in the first bullet point : “La valeur-p désigne la probabilité qu’un résultat statistique ne soit pas le fait du hasard” (“The p-value is the probability that a statistical result is not due to random”) and as a margin note later in the text :«Valeur-P : la valeur-p indique dans quelle mesure il est probable que le résultat présenté dans une étude soit vrai et ne résulte pas du hasard. Ainsi une valeur-p inférieure à 0.05 signifie que nous aurions raison 95 fois sur 100 de croire qu’un effet observé n’est pas une coincidence” (“P-Value: P-value indicates to what extent it is probable the presented result is true and doesn’t result from random. Then a p-value of 0.05 means we are right 95 times over 100 to believe that an observed effect is no coincidence”).

Puzzled, I looked on Internet for information about Professor Regina Nuzzo, mentionned as the article author. I quickly found she is teaching Statistic at Gallaudet University (Washington DC), she had been choosen by the American Statistical Association (ASA) to animate the workshop which produced the ASA Statement on p-value 1, and she had also written one of the most consulted2 article on p-value problems: Nuzzo, R. (2014), “Scientific Method: Statistical Errors,” Nature, 506, 150–152. In fact Pour la Science article seems built upon Nature article, without being a literal translation3.

I could hardly believe Professor Nuzzo had written such misinterpretations (p-value < 0.000001). Indeed, when I contacted her, she confirmed the inaccuracy of the two statements. On the other side Pour la Science did not answer my inquiry.

So I can only assume somebody, who may not be an expert, have worked on an article explaining difficulties encountered by seasoned professionals in using and interpretating p-value, and considered he may include his own p-value interpretation. An unintended ironic confirmation of the p-value curse !

  1. Ronald L. Wasserstein & Nicole A. Lazar (2016) The ASA’s Statement on p-Values: Context, Process, and Purpose, The American Statistician, 70:2, 129-133, DOI: 10.1080/00031305.2016.1154108 ; https://doi.org/10.1080/00031305.2016.1154108W↩
  2. http://www.altmetric.com/details/2115792#score↩
  3. There is also a wrong quotation attribution in Pour la Science article that does not exist in Nature one: “Charles Lambdin, d’Intel Corporation, à même proposé de rebaptiser la méthode “Statistical Hypothesis Inference Testing.” (Charles Lambdin from Intel Corporation even proposed to rename the method “Statistical Hypothesis Inference Testing.”). In fact, consulting the given reference (Lambdin 2012 Theory & Psychology 22(1) 67–90 Significance tests as sorcery: Science is empirical— significance tests are not https://doi.org/10.1177/0959354311429854) Lambdin attributes this sentence to the statistician Cohen.↩