One way to create groups for analysis is a custom format. There are advantages to using a format if the "class" or "group" is based on a single variable. The first is that you do not have to create any other variables so the concerns about which data set with which recoded variable is not an issue.
It is a bit hard to tell if this suggestion is appropriate because the code you show throws errors, at least in my version of SAS:
29 proc logistic data=two_class ;
30 model class1( event='1' ) = SepalWidth SepalLength PetalWidth
30 ! PetalLength ;
31 run ;
ERROR: All observations have the same response. No statistics are
computed.
NOTE: The SAS System stopped processing this step because of errors.
NOTE: There were 150 observations read from the data set WORK.TWO_CLASS.
NOTE: PROCEDURE LOGISTIC used (Total process time):
real time 0.00 seconds
cpu time 0.00 seconds
If you are trying to model combinations of the characteristics to predict species I'm moderately sure that Proc Logistic isn't the right one.
A format approach might look like:
proc format ;
value $setosa
'Setosa' = 'Iris Setosa'
other = 'Others'
;
proc logistic data=sashelp.iris;
model species = SepalWidth SepalLength PetalWidth PetalLength ;
format species $setosa.;
run;
Which doesn't throw an error but does have separation of data issues.
A drawback to formats is creating the formats and making sure they are available in the current session (run the Proc Format code or create format catalogs in a permanent library and in the format search path).
Other advantages of formats include:
1) Time. Especially if you have very large data sets, it can take time (and storage space) to add the additional variables to data set.
2) Ease of changing ranges of definitions, especially for numeric values. (Character values and "range" using the < in the proc format are very problematic for general character values but may work some fixed length single case values). Note in the example the keyword OTHER which assigns all values not explicitly listed to a single response.
3) For a fair number of examples the code for Proc Format may be simpler than data step code.
4) Specially structured data sets can be used to create formats.
There are the limits of only a single variable can be involved. If missing values are to be excluded and not treated as the "Other" category an appropriate value clause such as . = 'Missing' or ' '='Missing' may be needed.
Relying on the stored catalogs for use and not maintaining the code to create the formats can be problematic for moving to different versions of SAS. Best is to use the Proc Format CNTLOUT= option to create data sets that can be used to recreate the formats and keep the location handy.
An example of the data set to create three formats, one for each of the species:
proc sql;
create table temp as
select distinct(species)
from sashelp.iris
;
quit;
data iriscntrl;
set temp;
by species notsorted;
fmtname= strip(species); /* check documenation for rules of format names.
making these may be the hardest part of automated process
*/
type='C';
start =species;
label =catx(' ','Iris',species);
output;
if last.species then do;
call missing (start);
label='Other';
hlo='O'; /* variable to additional informat of how format used. Capital o is for the
Other instruction*/
output;
end;
run;
proc format cntlin=iriscntrl;
run;
More complicated code is possible. In some cases you would want to sort the control set by the format name to make sure all the start values are together in the set. Otherwise the Proc format results may be odd or fail.
