Mixed model bimodal distribution

[SAS_User]

Yes! Chemistry or Biology - neither of these is familiar territory for me.
From what I understand the samples are just grouped this way by chance. And, a huge number of samples will probably solve that problem. There was nothing different in the sample extraction process to cause the bimodality either. Unless anybody has other ideas, I will go ahead and use the model suggested by you, Steve - sample as fixed effect and other random effects like chemical or run, etc. Thank you very much for thinking through this and sharing ideas.

[Dale]

If you have information on the exact dates that the different samples were processed, perhaps you could find information on atmospheric conditions that might account for the bimodality. I presume that temperature is pretty stable within the lab. However, the lab may not be controlled with regard to atmospheric pressure or humidity. Perhaps one or the other of those is associated with the results which you observe.

Just out of curiosity, what do the conditional residuals look like when you fit a model with random rather than fixed sample effects? Do those display the bimodality problem?

FYI, you can get the conditional residual plots using ODS graphics commands and the RESIDUAL option on the MODEL statement of the MIXED procedure.

[SAS_User]

The experiment was done on two days and although there is difference in mean measurements on the two days, the bimodality is not due to days. And, I tried putting in sample as a random effect and the residual plot is normal! Thanks Dale! I think I ran way too many models and forgot to try the one I really wanted. This random effects model (no fixed effects) will help answer all the questions. And the results are what was expected out of this experiement

[SAS_User]

Yes! Chemistry or Biology - neither of these is familiar territory for me.
From what I understand the samples are just grouped this way by chance. And, a huge number of samples will probably solve that problem. There was nothing different in the sample extraction process to cause the bimodality either. Unless anybody has other ideas, I will go ahead and use the model suggested by you, Steve - sample as fixed effect and other random effects like chemical or run, etc. Thank you very much for thinking through this and sharing ideas.