[R-sig-ME] RE : Unbalanced design mixed models

[S Ellison]

>> But statistically speaking, including the plots with only one individual in this context
> is incorrect right?

Why would that be incorrect? Each isolated point will still contribute to the fixed effect coefficient(s). And a likelihood for that observation still exists, based on a distribution which includes  the single 'residual' variance and the between-plot variance. As long as you have at least one plot with replication you'll have separable estimates of both variances. 

A practical problem is that with the plot variance estimate coming from very few plots it might not be a very good estimate. I'd also worry that you may have heteroscedasticity from plot to plot, and if so you'll not be able to build plot-specific variances  into the model without replication in every plot. However, if your simple model is sufficient to explain the error structure, that isn't a problem.

S Ellison



*******************************************************************
This email and any attachments are confidential. Any use...{{dropped:8}}