topic Re: dividing summary statistic by noobs in proc summary in SAS Programming

dividing summary statistic by noobs in proc summary

csetzkorn — Sun, 18 Dec 2016 08:21:51 GMT

Using this code:

DATA WORK.TestDataSet;
   INPUT Uid Col1 Col2 Target $;
   DATALINES;
	1 5.1 3.5 13.12345
	2 4.9 3.15 2.1345
;

DATA WORK.TestDataSet;
    SET WORK.TestDataSet;
    MSE = (Target - ((Col1 + Col2) + exp(2.345)))**2;
RUN;

proc summary data = WORK.TestDataSet;
 var MSE;
 output out = WORK.ProcSumOut sum=;
run;

proc print data = WORK.ProcSumOut;
run;

I would like to divide the 'computed column' by noobs (the number of rows in WORK.TestDataSet). Is this possible? The result should be in WORK.ProcSumOut.

Re: dividing summary statistic by noobs in proc summary

Jagadishkatam — Sun, 18 Dec 2016 08:47:38 GMT

I think we cannot divide the computed column in the proc summary. However in a new data step it is possible.

First we need to get the n from proc summary into the WORK.ProcSumOut. Then in the new data step we need to divide the sum with n.

proc summary data = WORK.TestDataSet ;
var MSE;
output out = WORK.ProcSumOut n= sum=;
run;

data WORK.ProcSumOut2;
set WORK.ProcSumOut ;
sum2=sum/n;
run;

Hope this is what you are expecting.

Re: dividing summary statistic by noobs in proc summary

Reeza — Sun, 18 Dec 2016 10:41:39 GMT

Unless you have missing values the average is the sum divided by the number of observations so you can calculate your statistics by specifying the mean option

Re: dividing summary statistic by noobs in proc summary

Astounding — Sun, 18 Dec 2016 14:34:30 GMT

If for some reason you want what you are asking for, it's already there. The output data set automatically contains _FREQ_. It is the total number of observations (including those that have missing values, so using it in calculations may not give you the mean exactly).

You are specifying that the output data set should contain the SUM. You can also ask for other statistics, such as MEAN (as Reeza mentioned), N (number of observations with a non-missing value), and NMISS (number of observations with a missing value).

Re: dividing summary statistic by noobs in proc summary

csetzkorn — Sun, 18 Dec 2016 15:08:32 GMT

Sorry but your suggestion does not produce the desired output of 151.104. However, the following code, which makes use of _FREQ_. does the job:

DATA WORK.TestDataSet;
   INPUT Uid Col1 Col2 Target $;
   DATALINES;
	1 5.1 3.5 13.12345
	2 4.9 3.15 2.1345
;

DATA WORK.TestDataSet;
    SET WORK.TestDataSet;
    MSE = (Target - ((Col1 + Col2) + exp(2.345)))**2;
RUN;


proc summary data = WORK.TestDataSet ;
	var MSE;
	output out = WORK.ProcSumOut sum=;
run;

proc print data = WORK.ProcSumOut;
run;

data WORK.ProcSumOut2;
	set WORK.ProcSumOut ;
	MSE=MSE/_FREQ_;
run;

proc print data = WORK.ProcSumOut2;
run;

Re: dividing summary statistic by noobs in proc summary

Reeza — Sun, 18 Dec 2016 18:20:01 GMT

proc means data = WORK.TestDataSet ;
var MSE;
output out = WORK.ProcSumOut sum= Mean= / autoname;
run;