« Statistics done wrong » by Alex Reinhart (O’Reilly)

« Statistics done wrong » by Alex Reinhart (O’Reilly 2015, 176p)statistic done wrong

Statistics will not give you a compile error if done wrong! On the contrary, the more wrong, the more supportive to your point statistics may be. Who will notice anyway? Those knowledgeable are so few.
Data shows that: out of genuity or malice, statistics are done wrong 65,72% of the time.

If invoking statistic used to have some kind of power on you, it will free you from believing statistic « evidence ». But this book falls not in statistics bashing. On the contrary, it is very constructive and optimistic about improving statistic usage. Because there are solutions, both technical and also on the way scientific publications select research papers.

He does a good job at explaining the various bias that can alter statistics, if not done rigorously. He gives tons of examples and references (the author is a scholar). But his aim is not to teach you statistics, you won’t learn how to calculate anything here.

He also encourage to openly publish the data and the research paper on various Open platforms.

Very eye opening books overall, statistics are a powerful tool, but it is important to see how easy it is to get it wrong without notice!


 

Donc un bon livre, je repete en francais … Les stats sont un outil puissant, mais on obtient facilement de faux résultats sans s’en rendre compte.  Et curieusement de faux résultats vont souvent dans le sens de ce que l’on veut demontrer …   Mais son but n’est pas de rejeter les stats, au contraire : il est très constructif, positif. Il donne des exemples de mauvais usages, explique comment faire mieux, comment corriger le tir, les précautions à prendre.

Il adresse aussi le processus de publication scientifique et encourage l’utilisation de plateformes ouvertes de publication.  Mais ce n’est pas un lire pour apprendre à calculer des stats.

Bref, à lire, ca permet de se rendre compte à quel point il est facile de donner de faux résultats statistics sans que personne ne réalise l’erreur.