LSMEANS

[mthorne]

Can someone point me to where I can see how differences of LSMEANS are calculated in GLIMMIX using the DIFF option?

[PaigeMiller]

LSMEANS calculations: https://documentation.sas.com/?docsetId=statug&docsetTarget=statug_glm_details41.htm&docsetVersion=1...

Essentially, Least Squares Means tries to adjust for different number of observations in cells, which can bias the calculation of regular old means.

[mthorne]

Hi Paige,
Thanks for replying! I am actually interested in the calculations that are used in the DIFF option, which compares one LSMEAN to another, because I am trying to understand how it compares to a Fisher's LSD test.

[PaigeMiller]

First you calculate the individual LSMEANS and then you do a subtraction to get a DIFF. 

The formula for Fisher's LSD is here: https://www.statisticshowto.com/how-to-calculate-the-least-significant-difference-lsd/

So I believe that FIsher's LSD uses the difference of actual factor means while LSMEANS uses the difference of least-squares means. So the difference between two means won't even agree.

[mthorne]

I'm not sure it is that simple. Just because one is not exactly equal to another doesn't equate to a difference in the analyses that I have run.

[PaigeMiller]

I don't understand what you are saying.

[mthorne]

I guess I don't understand what the calculation that indicates one LSMEAN is different from another. I have hand calculated Fisher's LSDs and I understand that it depends on the outcome of the LSD equation, but I can't find that similar equation for PDIFF. I have attached a DIFF plot from one of my analysis and it looks like if the variance of a LSMEAN doesn't intersect with a central value or vector, then that lsmean is different from zero or lsmeans that do intersect with that line. I have also attached a difference of lsmeans table that indicates which lsmeans are different from others. But, how is this determined is my question?
Mark

[PaigeMiller]

Here's a worked example: https://support.sas.com/kb/24/984.html

[mthorne]

Thank you, this is helpful! So then is TDIFF the same as PDIFF?
Mark

[PaigeMiller · Accepted solution]

TDIFF is the t-value of the test to see if the LSMEANS are different. PDIFF is the probability associated with the t-value.

[mthorne]

Okay, thank you very much!
Mark