I don't know if this is the right thing to do but if you add a FREQ statement specifying NUMBER you get results. I don't know if they are correct. Dale?

proc npar1way data = KW_test wilcoxon; by category notsorted; class market; var percent; freq number; run;

On 7/30/10, Andy Satchwell <asatchwell@lbl.gov> wrote: > Hi SASheads, > > I am trying to run a Kruskal-Wallis test to look for statistical differences > in saturation of energy efficiency measures across different markets. My > data--in SAS--is structured as: > > data KW_test; > input category $ market $ percent number; > cards; > controls federal 0.41 219 > controls mush 0.63 1387 > controls private 0.25 148 > chillers federal 0.28 127 > chillers mush 0.21 460 > chillers private 0.14 83 > cogen federal 0.04 20 > cogen mush 0.03 74 > cogen private 0.03 16 > run; > > The percent value represents the saturation of each category (i.e., energy > efficiency measure) for all projects within a particular market (e.g., > federal, etc.). For example, the data shows that controls occur in 41% of > all federal market projects (219/534). I am interested in seeing whether > there is a statistical difference for the percent saturation across the > three markets . For example, I want to know if, for chillers, the percent > saturations of 28%, 21%, and 14% are statistically different across the > federal, mush, and private markets, respectively. I used the following > command (it's a Mann-Whitney, but outputs Kruskal-Wallis, because we need to > compare more than two categories): > > proc npar1way data = KW_test wilcoxon; > by category; > class market; > var percent; > run; > > The code produces no statistical results and instead the log stated, for > each category, that it was skipping the analysis because of "not enough > data." I think that the code is failing, because the project-level data is > already aggregated to percentages. > > Am I using the wrong test? If not, what is the correct code to test if > there are statistically different percentages across markets. Please note > that I have also included the sample size of the number of projects (i.e., > "number" in the above code) that installed that particular category. > > Thanks! >