Posting: # 19975
I have a funny situation with one of my datasets:
In a semireplicated trial I am using the lsmeans package and syntax like
to extract CIs from my model M where treatment is coded as "Trt" having two levels and it works just fine.
However, if I do something like
then I am getting
So, R is telling me that it can derive a difference of lsmeans but it can't derive the lsmeans individually. I find this quite annoying, possibly hard to believe. On processing the same data with WNL, it (WNL) seems to be able to derive LSmeans, and it is getting the same difference in LSmeans as R.
Did anyone encounter this situation? I am quite baffled myself . Google isn't my friend. My own understanding of the LSmean derivation is a bit lacking, otherwise I'd much prefer to just write code myself and not rely on the package.
I really wish I could publish the dataset for someone to reproduce the situation, but I can't offer to do so at present.
“(...) targeted cancer therapies will benefit fewer than 2 percent of the cancer patients they’re aimed at. That reality is often lost on consumers, who are being fed a steady diet of winning anecdotes about miracle cures.” New York Times (ed.), June 9, 2018.
(edited by mittyri on 2019-02-28 16:49)
Posting: # 19980
First of all I don't know the answer.
But I'm recalling some similar situation when
I mailed to the author (Russel Lenth) and we discussed the problem. Some weeks later he updated the package and the problem was resolved. But you should be ready to share some example.
Hope it helps.
Edit: Package linked. [Helmut]
Posting: # 19988
» So, R is telling me that it can derive a difference of lsmeans but it can't derive the lsmeans individually. I find this quite annoying, possibly hard to believe.
With the last sentence I agree totally.
Nevertheless such situations exist .
I have sometimes encounter this situation in SAS using the LSMeans statement in Proc GLM.
See the FAQs for emmeans for an R implementation. Look under point 8.
Why this is so precisely is beyond my intellectual reach. Had to tackle such things like "estimability theory". Oh my dear!
My own explanation, quick and dirty only for myself, goes like this: Some factors constellations used in LSMeans which lead to the NonEst result cancel out if you derive LSMeans for the difference and therefore the latter are estimable.