Experimental design question

[belench19]

I am struggling with the complexity of my experimental design... I apologize in advance for any incorrect terminology or grammar.

I know that I have a 2x2x2 factorial design with 2 different dam diets (during pregnancy), 2 different offspring diets, and genders (obviously 2 levels). Females will be fed one of two diets during pregnancy. Upon weaning males and female offspring will be randomized to receive one of two offspring diets (giving me 8 treatment combinations). In the offspring I will be looking at mRNA expression in brain regions and some behaviors (Barnes maze, elevated T, etc.).

I am concerned that there may be a litter effect (specific dam) that influences gene expression. I do not know if I need to do a split-plot or if something needs to be nested. My concern in this is because not all litters provide equal number of males and females (sometimes all one gender), so blocks will not be all even. Should I just control for effect of litter/dam in some other way?

 

[hlsmith]

Are these rats? What is dam? I am guessing there are quite a few things you need to control for if they can be clustered in litters. Might also need a multilevel model. I believe there might be a good example of a multilevel model out there on salamanders.

 

[Miner]

Dam is the female parent of an animal.

This sounds like a split plot design. Offspring are nested within dam restricting full randomization resulting in different error structures between dams and offspring. This case assumes a single litter from each dam.

This could also be a split-split plot design. Offspring are nested within litters nested within dam. This case would arise if you have multiple litters from each dam.

 

[belench19]

Thank you both for the help. As stated, dam is the mother (mice in this case). Right now I have not decided if I would like to use multiple litters from each dam. It will cut down on cost of experiment considerably because you always have some who you just will not get pregnant and others who are more fertile.

This means I will likely go split-split plot design. Will the fact that not every litter will produce each combination of sex*offspring diet invalidate results because of error structure? Does this need to be considered in sample size calculation?

 

[Miner]

It would be best to consider this in the sample size. Split-split-plot designs are difficult enough to analyze without adding missing experimental runs. Missing runs will not prevent you from assessing main effects, but probably will prevent you from assessing some interactions.

 

[belench19]

Would you expect as much of an issue if just split plot (as opposed to split-split plot) were used?

 

[Miner]

Assuming that you have a single litter per dam, no. Would you expect litter to litter difference to be a major source of variation, or interact with the treatment?

 

[belench19]

There is a possibility that we will look into some epigenetic markers/modifications, which means there is a possibility that dam or dam+sire (father) combination will be a source of significant variation. I really cannot thank you enough for your help. I am struggling with the model but will figure it out.

Yijk=u+