You could try storing the data as character and make a view that converts them to numbers.

data mydata_real ; input (flag1-flag5) ($1.) ; cards; 10101 run; proc sql noprint ; create view mydata_view as select input(flag1,1.) as flag1 ,input(flag2,1.) as flag2 ,input(flag3,1.) as flag3 ,input(flag4,1.) as flag4 ,input(flag5,1.) as flag5 from mydata_real ; quit;

On Thu, 19 Apr 2012 18:02:12 -0400, Bolotin Yevgeniy <YBolotin@SCHOOLS.NYC.GOV> wrote:

>This is a semi-theoretical question, so don't spend too much time on it >- i can just keep using numerics, if no better solution is available > > > > > >I have a dataset with a large number of 0/1 variables > > > >If variables are stored as numeric, they take up 3 bytes > >If variables are stored as character, they take up 1 byte > > > >However, i do various aggregate functions over subsets of these >variables, which take about half a second each for numerics and 5 >seconds each for characters > >(using proc sql, 5 second results for both input(var, 1.) and case >when... . proc sql does not implicitly convert char<->num) > >Aggregates = sum, average, count, etc > > > >The data is not indexed, so it's not just an index lookup vs table scan > > > > > >Is there any way for me to get this to work fast but still get the space >savings of character variables? > > > > > > > > > >Yevgeniy (Eugene) Bolotin >Report Developer, Progress Reports >NYC Department of Education > >(212) 374-7949 > >ybolotin@schools.nyc.gov <blocked::mailto:mmadow@schools.nyc.gov> > >