Very interesting article on the nature of "statistical significance". As a layman I thought it carried way more weight that apparently it should. https://www.nature.com/articles/d41586-019-00857-9?source=science20.com
Yes, there is a BIG difference between showing no association and not showing an association. Scientists certainly need to be aware of the issues in this article, but the renaming the authors suggest as a remedy seems a little ridiculous to me. And, I don't see any measurement of this as a problem that is deprecating science. I don't understand the author's concern about designing studies to more accurately dichotomize. A patient needs to decide whether to take the pill or not to take the pill. While there will always be unlikely or unknown factors, surely one would want to know the clearest possible starting point. Like in political polling, there are confidence intervals and polling companies spend serious effort in producing significant results. I think there are larger problems in science reporting, by the way.
That’s why it’s necessary to look at the overwhelming scientific consensus and not just singular studies. I can find you a study that found there is no relationship between smoking and lung cancer. However, the overwhelming scientific consensus is there is a relationship between smoking and cancer.
I debated with a guy on this forum for years that the ' 22% wage gap,' is the result of choice not discrimination, and he refused to accept it because 5% of the wage gap couldn't be quantified statistically due to the limitations of the researchers model, multivariate regression analysis. The researchers couldn't quantify benefits in their model because there was 'collinearity,' between benefits and another dependent variable in the model, experience. Basically the r-value (coefficient of correlation) was greater between benefits and experience (when those variables were run against each other in single regression) than the r-value in multivariate regression analysis between the dependent variables experience, benefits et cetera and the independent variable, gender. So benefits couldn't be quantified statistically. He took that to mean the remaining 5% was obviously due to discrimination even though women do tend to choose more compensation in the form of non-wage benefits. So even though something can't be quantified statistically it does not mean that it does not exist...
