What happens to the standard error each time you quadruple the sample size?