Since we don't really know how many hypotheses are behind the scene of any paper, it has been a big mistake of scientific community to cosider p=0.05 as a gold standard. It has created so many useless false postive claims. We (and journals) must set much more stringent significance level as p=0.005 or even p=0.001. (still not to the level of GWAS). This will force researchers to design large well-powered studies. It still does not solve nonrandom bias and confounding, but at least small bad studies will disappear, which will decrease the profound degree of publication bias we are seeing today.