1
$\begingroup$

I could use some advice on how to handle this situation. I have a Randomized Complete Block Design (RCBD) ANOVA with 5 blocks and 8 treatments, the response is dry weight of some plant, there is 1 replication for each cell in the RCBD, except in 1 case where the measurement could not be taken due to the plant expiring.

When I perform ANOVA with a fixed block factor the residuals are approximate normal but not homogeneic.

I have tried some simple transformations but to no avail (I am not that interested in transformations either, I'd much rather use some robust method)

I have managed to run HC estimation BUT the methods used, Anova() from car package and robustreg in SAS, does not produce Residual SS, which I need to calculate simultaneus tests. Further, I am unsure if these methods compensate for the empty cells.

Could you guys point me in the right direction?

$\endgroup$
1
$\begingroup$

Plant expiring should be considered a 0, not a missing observation unless the plant was dropped etc. In a well designed experiment we would expect residuals to be homogeneic, if not a transformation or a robust method can be used as you have stated. I would suggest looking into the multcomp package and specifically the glht function, it is possible to use this package together with sandwich package.

So, if the plant died it is not a missing value, it should be 0, this should solve your first question since you will have no missing values. Secondly, use glht with sandwich package, there is a very similar example to what you are talking about in the manual!

Good luck!

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.