[ad_1]
An article recently published in Nature Magazine calls for measures to be taken against the misleading use of "statistical significance". The document is supported by more than eight hundred academic disciplines.
Pleased to meet you, 'P values'!
Statistical significance prevails in many areas and has a profound impact on our daily lives, choices and decisions. The three scientists behind the article argue that in statistical badyzes, it happens all too often that it comes to the conclusion that there is " no difference "between two groups studied. In statistics, this phenomenon is called the "null hypothesis".
SEE ALSO: THE BEST ACADEMIC JOURNALS OF ENGINEERS
The authors claim that a study establishing such a thing based solely on the null hypothesis is dangerously misleading. Their argument is that there may be a tiny difference between two groups studied, although one of them may prove significant, while the other is insignificant. This dichotomization is due to the method that relies too strictly on a factor, from the threshold.
"Let's be clear about what needs to stop: we should never conclude that there is no difference or badociation simply because a P value is greater than a threshold such that 0.05 (…) .We also should not conclude that two studies are in conflict because one had a statistically significant result and the other no.These errors waste research efforts and skew decisions policies. "
How it works?
"For example, consider a series of badyzes of the unintended effects of anti-inflammatory drugs.2 Their results being statistically insignificant, one group of researchers concluded that drug exposure was" not badociated with a new atrial fibrillation (…) and that the results contrasted with those of a previous study with a statistically significant result. "
In reviewing the actual data not proving the above, they argue as follows: "It is ridiculous to conclude that the statistically insignificant results show" no badociation ", when the interval estimate includes an increase It is also absurd to claim that these results contrast with previous results showing the same observed effect, yet these common practices show how misleading the use of statistical significance thresholds can be. "
The consequences of the professors of Amrhein, Greenland and McShane also indicate that the problem as a whole is more humane than statistical, it is us and our cognitive processes that work so categorically. This "has led scientists and newspaper publishers to favor such results, thus distorting the literature." Statistically significant estimates are biased upwards and potentially to a large extent, while non-statistically significant estimates are skewed. on the decline".
Is there an exit? "We (…) call for the abandonment of the whole concept of statistical significance. (…) One of the reasons to avoid this" dichotomania "is that all statistics, including P-values and confidence intervals, naturally vary from one study to another and often do so to a surprising degree. "
"We must learn to accept uncertainty," they continue. "A convenient way to do this is to rename the confidence intervals to" compatibility intervals "and interpret them in such a way as to avoid overconfidence."
They are not alone
This article is important in a series of other similar warnings written by scientists in recent years, all arguing against the use of deceptive methodology. In 2016, the American Statistical Association issued a statement in The American statistician warning against the misuse of statistical significance and values of P.
The issue also included many comments on the subject. This month, a special issue of the same newspaper tries to push these reforms further. He presents more than 40 articles on "Statistical inference in the 21st century: a world beyond P <0.05". The editors introduce the collection with the following caution: "Do not say" statistically significant. "Another article with dozens of signatories also calls writers and journal editors to disavow these terms.
[ad_2]
Source link
