50% of studies can't be repeated with the same conclusions (marginalrevolution.com)