Efficiency Issues

Quantiles

For a large sample size n, the calculation of quantiles, including the median, requires computing time proportional to nlog(n). Therefore, a procedure, such as UNIVARIATE, that automatically calculates quantiles might require more time than other data summarization procedures. Furthermore, because data is held in memory, the procedure also requires more storage space to perform the computations. By default, the report procedures PROC MEANS, PROC SUMMARY, and PROC TABULATE require less memory because they do not automatically compute quantiles. These procedures also provide an option to use a new fixed-memory, quantiles estimation method that is usually less memory-intense. For more information, see Quantiles in the PROC MEANS documentation.

Computing Statistics for Groups of Observations

To compute statistics for several groups of observations, you can use any of the previous procedures with a BY statement to specify BY-group variables. However, BY-group processing requires that you previously sort or index the data set, which for very large data sets might require substantial computer resources. A more efficient way to compute statistics within groups without sorting is to use a CLASS statement with one of the following procedures: MEANS, SUMMARY, or TABULATE.