The thing to remember is that the following two statements are equivalent as far as MM reporting:   random intercept/subject=block; random block;   It is just that the first statement results in faster calculations, and is required if method=quad   But it still boils down to "Block was fit as a random effect."   And random intercept/subject=block(location) is "Location within block was fit as a random effect."   So boiling this all down, would this make sense?:  "Treatments were randomly assigned within blocks, and multiple locations within each block were measured. Treatment was fit as a fixed effect, while block and location within block were fit as random effects."   You could probably look through the ecology or psychology literature to find examples using the phrase 'random intercept', especially where R packages were used to analyze the results.  Those might be enough to get you started.   SteveDenham ... View more

If you came to me for an analysis, I would consider this to be a linear mixed model problem, with repeated measures.   To analyze this you would need to get your data into a long format, with a single response value for each record, along with the design factors: ID, treatment (compressed/uncompressed) and site.  Possible code would look something like (data is simulated to look something like what you have in Table 1):   data one; call streaminit(12345); do i=1 to 2; do j=1 to 5; do k=1 to 6; if i=1 then do; trt='Uncompressed'; site=j; result=7 + rand('normal',0,2); end; else do; trt='Compressed'; site=j+5; result=9 + rand('normal',0,1); end; id=k; output; end; end; end; run; proc glimmix data=one; class trt id site; model result=trt site*trt/solution ddfm=bw; random site/residual subject=id(trt) type=cs; lsmeans trt/diff; lsmeans trt*site; run;   The method you propose will certainly work: Firstly, we calculated the average for compressed area (site1~site5) and the average for uncompressed area (site1~site5) for each patient, respectively. So, we compared the average between two groups (compressed area vs. uncompressed area) by paired t tests. However, is it correct way?    suspect that the p value for the F test for treatment will be very nearly the same as the p value for the paired t.   One issue is that the sites within each treatment are not truly repeated (that is, site 1 in the compressed area is not site 1 in the uncompressed area).  I would recommend renumbering the sites in one of the areas as 6 to 10.  Consequently, the model should not contain a main effect for site, in order to prevent non-estimability for the treatment least squares means.   Regarding the missing values question, the mixed model procedures are robust to these so long as the data are missing at random, which is a reasonable assumption.   Here is code in case you want to try a generalized estimating approach.  Note that the standard error is smaller.   proc genmod data=one; class trt id site; model result=trt trt*site/type3; repeated subject=id*site/type=exch; lsmeans trt/diff; lsmeans trt*site; run; Treatment means are the same.     SteveDenham   ... View more

While I like @StatDave 's response to look at HPGENSELECT, I would suggest a couple of things before you start doing variable selection.  Season, treatment, parity and body condition score (BCS) seem to me to be 3 design factors and a continuous covariate, and that covariate (BCS) is well known to have an effect on pregnancy rate in mammals.  So in truth you have just four variables, with possible interaction, and no real need to employ variable selection.  Try the following MODEL statement:   model result(event='pregnant')=season*treatment*parity BCS BCS*season BCS*treatment BCS*parity; This fits a fully saturated model for the design factors, with possible different slopes for the BCS relationship.  Work through this to eliminate the interaction terms where the slopes do not differ.  Once you have stabilized your selection of appropriate slope terms, you could then fit an effects model, with the relevant covariate/covariate by effect interaction terms in the model.  This approach is covered in Milliken and Johnson's Analysis of Messy Data, vol.3: Analysis of Covariance, or in SAS for Mixed Models (any of the editions 1 to 3) in the chapter on analysis of covariance.   Also, look at the following crosstabulation:   PROC FREQ data=maanshan320; tables parity*season*treatment*result; run; That should give 8 tables that are Nx2, where N is the number of treatments and 2 is the number of levels for result.  From those 8 tables, you should readily be able to identify where the separation is occurring, if anywhere, for the design factors.  Also, you may want to look at the results of PROC GLM, with BCS as the dependent variable, and the design factors crossed with the result variable as the independent variables.  In the LSMEANS statement, see how BCS separates as a result of the factors.   I think the root cause of the separation issue is the inclusion of high-order interactions with BCS.  For some combination or combinations of the design factors and the response, there are likely to be full separations of the covariate.  Additionally, fourth order interactions do a great job of modeling noise, especially when one is a continuous variable, which brings us back to @StatDave 's comment regarding fitting the data perfectly.  So think carefully about the biological question at hand (which looks like it might be related to feeding dairy cows and seeing what the resulting pregnancy rate is) and formulate a model that addresses those questions.   SteveDenham   ... View more

Well, I didn't go through all 300+ pages of data, but I did see a lot of what I referred to as "identicalness" in just the first subject.  There are levels of subscale and item2 which are identical once you drill down to the levels specified by your BY statement.  For instance, for the first subject, all values of RATING are 1, except for subscale n1, item 4.  Similar patterns are seen for the second data point for the first subject, where all values of RATING are 1 - no variability.  Given this, try METHOD=TYPE1 which should fit the terms sequentially, or loosen your restriction on the BY variables, so that some variability can be seen at the level you are examining.   SteveDenham ... View more