Thursday, April 20, 2017

"Statistically Significant" doesn't mean "Right"

Due to the science reproducibility crisis, the American Statistical Association (ASA) is trying to understand what is the root of this burden. All traces lead to the misuse and misinterpretation of p-values.
P-values can often determine what studies get published and what projects get funded. So, it is very important that not only researchers know how to report it correctly, it also means that reviewers must be highly knowledgeable in statistics.

Back in 2005, JohnIoannidis published a paper stating that 50% of published biomedical research findings with a p value of <0.05 are likely to be false positives. This is a big problem! (. In the 1920’s, 17% of the studies in major journals used statistical hypothesis testing, in 1970’s around 90%... and I bet that in 2017 the number is even higher. However, the reproducibility crisis is crazier than ever and it’s probably due to the poor knowledge in statistics and how to interpret hypothesis testing.
After reading Hilda Bastian’s blog I can say that I totally agree with her “things to keep in mind to avoid p-value potholes”. Here I mention just a couple of them, these are probably the issues that are most misinterpreted or just plain crazy.
               1.     “Significant” in “statistically significant” doesn’t mean “important”Exactly! Statistically significant just means the likelihood that a relationship between two or more variables is caused by something other than random chance.

2.     A p-value is only a piece of a puzzle: it cannot prove whether a hypothesis is true or not.As the ASA states: P-values do not measure the probability that the studied hypothesis is true, or the probability that the data were produced by random chance alone. 

3.     Some potholes are deliberately hidden: shining the light only on p’s <0.05.This means that either the authors want you to believe the story they want to tell or that they just got some data and didn’t know what it meant so they work the story around it. The COMPare teamchecked every trial published in the top five medical journals, to see if they misreported their findings. They found that out of 67 trials checked, 9 trials were reported perfectly, 354 outcomes were not reported and 357 new outcomes were silently added. Data manipulation at its finest!

These are all things to keep in mind when doing our own experiments and getting data that will eventually, and hopefully, get published. The science world is getting a little dirty because of so much competition, however this should not cloud our judgment call when deciding how to publish our science stories. P-values are very “in” right now, hypothesis testing is the latest trend, but y’all…let’s use it correctly.

No comments:

Post a Comment