01-26-2015 10:22 AM
I have a 24 level ordinal variable that I am trying to create a 5 level variable from based on percentiles (50th, 70th, 85th, 90th, and 95th). I have not been able to find any good example online to work off of.
01-26-2015 12:38 PM
I'm not sure how an ordinal variable maps to percentiles makes a lot of sense.
In your proc freq output there should be a variable called cumulative percentage. You can use that to recode to 5 categories based on your cutoffs.
01-26-2015 12:42 PM
Me neither, but my boss wants them this and and stated she has seen it in literature before. I am only familiar with doing percentiles on continuous variables.
01-26-2015 03:30 PM
Well, 24 is kind of, sort of, like a lot of ordinal categories, and with that many observations, you should get a fairly interesting cumulative distribution. Maybe you could fit an empirical distribution function to the data, and use that to get the percentile cutpoints.