Are large or small data sets more appropriate for using perecentiles to summarize data? Why?