LS Means Lines model

[hende333]

Hello -

I am analyzing my data with ProcGlimmix and the LSMeans statement, with the Lines command for letter separation. I am writing about the statistical analysis I am doing, and realized I don't know what model that the Lines command uses to determine separation (Fishers, Bonferroni, Tukey, etc.) I've included the code, although I don't think it's necessary.

Does anyone know what test is used, and what source I can use to cite that? I have been looking all over but haven't been able to find a straight answer. Thank you so much!

proc glimmix data=Bruise nobound plots=studentpanel;
class cls Inoc bruise Rep ; 
model wcorr = cls|bruise|inoc /dist=n ddfm=kr;
random rep;
lsmeans cls|bruise /pdiff lines cl;
run;

[PaigeMiller]

I did not know off the top of my head. So did I do? I read the documentation.

It says: 

Yes, that's kind of small, but if you click on it, it becomes readable.

[hende333]

Thanks Paige, but I am specifically asking about the Lines letters separation within LS Means. I could not find that in the documentation at all. 

[PaigeMiller]

---

@hende333 wrote:

Thanks Paige, but I am specifically asking about the Lines letters separation within LS Means. I could not find that in the documentation at all. 

---

Ok, with that in mind, I call upon the stats wizard @Rick_SAS to see if he has additional insight.

[Rick_SAS]

I'm not great with PROC GLIMMIX. Someone like @SteveDenham is probably a better choice. I assume you've read the doc for the LINES option.

If you do not use the ADJUST= option, you get all pairwise least squares mean differences and their significance. The unadjusted p-values are from a statistic that is assumed to have a t distribution. Since you specified DDFM=KR, the test statistic is the one specified in the doc. If you have access to Westfall, Tobias, and Wolfinger (2011, 2nd  Ed), they discuss the LINES option and p-values on pp 280-281. (If you specify an ADJUST= option, the p-values are adjusted.) 

I'd like to suggest that the table that the LINES option creates is a poor way to visualize the pairwise significance of the differences. For a discussion, see "Graphs for multiple comparisons of means: The lines plot." You can use the PLOTS= option on the LSMEANS statement to request a better visualization. My favorite visualization is PLOTS=DIFFOGRAM, which creates a faithful representation of the significant differences.