Tweaking your tables: Suppressing superfluous subtotals in PROC TABULATE

Transcription

1 ABSTRACT Tweaking your tables: Suppressing superfluous subtotals in PROC TABULATE Steve Cavill, NSW Bureau of Crime Statistics and Research, Sydney, Australia PROC TABULATE is a great tool for generating cross tab style reports. It's very flexible but has a few little limitations. One is suppressing superfluous subtotals. The ALL keyword creates a total or subtotal for the categories in one dimension. However, if there is only one category in the dimension, the subtotal is still shown, which is really just repeating the detail line again. This can look a bit confusing in the final output. This talk demonstrates a method to suppress those superfluous totals by saving the output from PROC TABULATE using the OUT= option. That data set is then reprocessed to remove the undesirable totals using the _TYPE_ variable which identifies the total rows. PROC TABULATE is then run again against the reprocessed data set to create the final table. INTRODUCTION PROC TABULATE is one of the most commonly used reporting procedures in SAS and has very powerful summary and output capabilities. Sometimes PROC TABULATE does almost, but not quite, exactly what you want. One of the possibly less used options of PROC TABULATE is the OUT= option to send the output of the procedure to a data set. Using the OUT= option can enable you to take the standard output of PROC TABULATE and post process that output to create exactly what you want. This paper shows one example of modifying PROC TABULATE standard output by post processing the output data set to suppress superfluous subtotals if they are not desired. The general technique of manipulating the output data set to get the exact output you want can be extended to most situations where you want to alter the standard output. The examples in this paper are based on the SAS supplied sample data set SASHELP.CARS. (I have subset the CARS dataset to three makes only to simplify the output). The complete sample program to create all the output in this paper is included in the appendix. 1

2 PROC TABULATE SUBTOTALS The ALL= keyword of PROC TABULATE creates subtotals, for example, this code: proc tabulate data=data format=6. class origin make type var msrp tables origin *(make*(type all) all) all='grand Total',n msrp*mean*f=dollar7. produces the output in Figure 1 Because there is only one type of vehicle from Asia the same detail line appears three times. Similarly, for Jeep, there is only one vehicle type so the subtotal for Jeep is the same as the detail line. It can make the output look cleaner if those subtotals are suppressed, so subtotals only appear if there is more than one detail line in the category, as shown in Figure 2. Figure 2 also shows we have changed the default label of the subtotal to include the data label. PROC TABULATE allows you to change the default text All to any label you want, but it s static text and doesn t include the data value for the category. You can see this in the sample program where all='grand Total' has changed the label on the Grand Total line Figure 2: Modified output Figure 1: Standard output PROCEDURE The steps to create the desired output are as follows STEP 1 : OUTPUT THE TABULATE OUTPUT TO A SAS DATA SET proc tabulate data=data out=table format=6. The output data set looks like this: 2

3 Figure 3: Output data set The _TYPE_ column defines what type of row we are looking at. A 1 in position n means that class variable is included in that summary row. The positions are based on the order of the class variables in the class statement. So: 111 means all class variables are part of that row so in our table that s the rows for Origin, make and type effectively our detail rows 110 means the row contains Origin and Make, and is the summary of all Type s for that Origin and Make 100 are the totals for Origin 000 is the grand total STEP 2 REMOVE THE SUBTOTALS WE DON T WANT The subtotals we want to remove have been highlighted in Figure 3 above. These are the Make subtotals for Isuzu and Ford, and the Origin total for Asia (as highlighted in Figure 1). Obviously we don t want to hardcode these data values, as the categories to suppress will change as the data changes. What identifies those categories is that, firstly they are total lines, and secondly there is only one item in that category. i.e. there is only one Type of vehicle for Isuzu and Jeep, and there is only one Make within Asia. We can easily use SQL or PROC FREQ or SUMMARY to find these categories. Here is using SQL: proc sql create table make_suppress as select make,count (distinct type) as freq from data group by make having count (distinct type)=1 /* only one type within the make*/ create table origin_suppress as select origin,count (distinct make) as freq from data group by origin having count (distinct make)=1 /* only one make in the origin */ 3

4 quit This creates the tables below. We could also have created macro variables instead, which is really just a matter of preference. I find tables can make it easier to report on what has been done. Make freq Isuzu 1 Jeep 1 Output 1. Listing of make_suppress Origin freq Asia 1 Output 2. Listing of origin_suppress Now we can delete those rows and change the values of the labels as desired data table /* the default label widths are not wide enough to hold the desired total label */ length origin make type $20 set table /* first change the total labels (which are blank in the output data set */ if _type_='000' then origin='grand Total' /*grand total*/ /*origin total - include the value of Origin on the total line */ if _type_='100' then make=catx(' ',origin,'total') /*make total - include the value of Make on the total line */ if _type_='110' then type=catx(' ',make,'total') proc sql /*now delete the total lines where there is only one category in that dimension*/ delete from table where _type_='110' and make in (select make from make_suppress) /* will delete Jeep and Isuzu totals */ delete from table where _type_='100' and origin in (select origin from origin_suppress) /* will delete Asia subtotal */ quit The output data set now has only the rows we want to display, and has total labels modified to show the labels we want, including the data value of the relevant category. 4

5 USE PROC TABULATE TO SHOW THE FINAL TABLE /* now create the final table /* - we don't need totals because they are in the input data set, /* and the cell value is the sum of N and msrp_sum, /* where N and msrp_sum were calculated in the previous tabulate */ proc tabulate data=table format=6. class origin make type/order=data missing /* totals have missing in the subcategory*/ var N msrp_mean /* N and msrp_mean were calculated in the previous tabulate so just show it */ tables origin*make*type, N*sum msrp_mean*sum*f=dollar7. Notes on the code The all= keywords have all been removed, as don t need to calculate totals, because the output dataset already has the totals calculated by the first run of PROC TABULATE. The name of the msrp variable now contains the suffix _mean, because that is the statistic we asked for in the first PROC TABULATE. We need the missing option on the class statement, because the subtotal rows have blank values in the subcategory column We use the order= option on the class statement because the output dataset is already in the desired order. If we don t use the order=data option the subtotals will be re-sorted into alphabetical order rather than the totals appearing at the bottom. We use the sum statistic, as there is only one row per category so we can just sum the value (which is the same as the value itself) CONCLUSION The example uses a specific class statement to simplify the demonstration. This code can easily be generalized into a macro that takes the class statement as input. This macro is included in the appendix. 5

6 The out= option is supported by most SAS reporting procedures. The technique of reprocessing the out= data set can be used in lots cases where you want to modify the standard SAS output. CONTACT INFORMATION Your comments and questions are valued and encouraged. Contact the author at: Steve Cavill NSW Bureau of Crime Statistics and Research 20 Lee Street Sydney, NSW, SAS Community Page: This and other presentations can be found at my SAS Community presentations page SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. indicates USA registration. Other brand and product names are trademarks of their respective companies. 6

7 APPENDIX Source code to reproduce the examples in this paper. The sample data set is a standard data set that is part of your SAS installation data data /* subset the data to make illustration simpler */ set sashelp.cars where make in ('Isuzu','Jeep','Ford') /* create the desired tabulate table and output to a sas data set */ proc tabulate data=data out=table format=6. class origin make type var msrp tables origin *(make*(type all) all) all='grand Total',n msrp*mean*f=dollar7./ /*find the dimensions that have only one category*/ proc sql create table make_suppress as select make,count (distinct type) as freq from data group by make having count (distinct type)=1 /* only one type within the make*/ create table origin_suppress as select origin,count (distinct make) as freq from data group by origin having count (distinct make)=1 /* only one make in the origin */ quit data table length origin make type $20 /* the default label widths are not wide enough to hold the desired total label */ set table /* first change the total labels (which are blank in the output data set */ if _type_='000' then origin='grand Total' /*grand total*/ if _type_='100' then make=catx(' ',origin,'total') /*origin total - include the value of Origin on the total line */ if _type_='110' then type=catx(' ',make,'total') /*make total - include the value of Make on the total line */ proc sql /*now delete the total lines where there is only one category in that dimension*/ delete from table where _type_='110' and make in (select make from make_suppress) /* will delete Jeep and Isuzu totals */ delete from table where _type_='100' and origin in (select origin from origin_suppress) /* will delete Asia subtotal */ quit /* now create the final table 7

8 /* - we don't need totals because they are in the input data set, /* and the cell value is the sum of N and msrp_sum, where N and msrp_sum were calculated in the previous tabulate */ proc tabulate data=table format=6. class origin make type/order=data missing /* totals have missing in the subcategory*/ label origin='origin' make='make' type='type' var N msrp_mean /* N and msrp were calculated in the previous tabulate so just show it */ tables origin*make*type, N*sum msrp_mean*sum*f=dollar7. 8