Publication Bias Reconsidered