A 'power law' based method to reduce size-related bias in indicators of knowledge performance: An application to university research assessment