I'm trying to implement Welch's ANOVA in SAS, and, according to the documentation at http://support.sas.com/kb/22/526.html, I'm supposed to be able to do it in two equivalent ways: PROC GLM; CLASS trt; MODEL y = trt; MEANS trt / WELCH; RUN; or     PROC MIXED; CLASS trt; MODEL y = trt / DDFM=SATTHERTH; REPEATED / GROUP=TRT; LSMEANS trt; RUN; However, I only get the same results (*F*-statistic and *p*-value) from both when I have two groups, but not more than two (six, specifically). Online, I've found examples of others also using both procedures for the "same" Welch's ANOVA, but their results also differ when I run them. Are the d.f. actually approximated slightly differently in both procedures? Here is a working example: It's the hermit crab data from Design of Experiments: Statistical Principles of Research Design and Analysis (2nd Ed.) by Kuehl. DATA     DATA hermits; DO coastal_site=1 to 6; DO rep=1 to 25; INPUT hermit_count @@; OUTPUT; END; END; DATALINES; 0 0 22 3 17 0 0 7 11 11 73 33 0 65 13 44 20 27 48 104 233 81 22 9 2 415 466 6 14 12 0 3 1 16 55 142 10 2 145 6 4 5 124 24 204 0 0 56 0 8 0 0 4 13 5 1 1 4 4 36 407 0 0 18 4 14 0 24 52 314 245 107 5 6 2 0 0 0 4 2 2 5 4 2 1 0 12 1 30 0 3 28 2 21 8 82 12 10 2 0 0 1 1 2 2 1 2 29 2 2 0 13 0 19 1 3 26 30 5 4 94 1 9 3 0 0 0 0 2 3 0 0 4 0 5 4 22 0 64 4 4 43 3 16 19 95 6 22 0 0 ; RUN; PROC GLM version of Welch's ANOVA PROC GLM DATA=hermits; CLASS coastal_site; MODEL hermit_count = coastal_site; MEANS coastal_site / WELCH; RUN; PROC GLM output   PROC MIXED version of Welch's ANOVA PROC MIXED DATA=hermits; CLASS coastal_site; MODEL hermit_count=coastal_site / DDFM = SATTERTHWAITE; REPEATED / GROUP = coastal_site; LSMEANS coastal_site; RUN; PROC MIXED output Can anyone tell me why these are not giving the same test statistics? If they, in general, don't give the same answers (beyond two groups), which one should be trusted (i.e., which one is implementing the test correctly)? Is it just a matter of slightly different   [1]: http://i.stack.imgur.com/bmDHI.jpg   [2]: http://i.stack.imgur.com/hchzJ.jpg   ... View more

I'm trying to use the Query Builder to write a percentile (PCTL) function pseudo-manually, where I want to use a column of data from my data set in the expression (since the PTCL function requires the raw data from which you want to calculate the percentile). [Note: I'm finding the 25th percentile in this example, which I know can be done in other ways in EG, but I am wondering about this in general for percentiles not "offered" by Summary Statistics in the Tasks > Describe menu. It also applies to using the raw data in the Query Builder in general.] To do this, I do the following: Start at the Input Data window Select Query Builder Drag my variable of interest (column called Length) over to to the Select Data pane in the Query Builder (shown below) Select the "Add a New Computed Column" button (that looks like a calculator) Select radio button next to "Advanced expression" and click Next I manually type in the "Enter an expression:" pane: PCTL(25, Now I need to get my raw data (n = 46) into the PCTL function, separated by commas, so in the lower left pane, I double-click "Selected Columns" to see my variable length drop down here (see image below) Then, I see this message in the pane to the right: "The maximum number of rows to process for retrieving distinct values may be limited" When I click "Get Values," sure enough - it only pulls the distinct values from my variable Length, so when I then "Select Values" to insert them in the PCTL function, I don't have all my data, only the distinct values (and no, I did not select the "Select distinct rows only" box in the original Query Builder box"). In this case, I have n = 40 distinct values, so those are the only values that get inserted into the PCTL function. I can't have this if I want to calculate something like a percentile - I need all of the data, not just distinct values! Why does it do this? Is there a way to change this? Any help would be greatly (greatly) appreciated. ... View more