[R] question in using nlme and lme4 for unbalanced data

[Ben Bolker]

On 10-10-25 04:59 PM, Bert Gunter wrote:
>  ...ignore the block variation entirely, "
> 
> If the between block variability is large, this will lose precision;
> with imbalance, it could also result in bias (prhaps not in this
> study...). The mixed or fixed effects choice is arbitrary; this is not
> -- blocks should be modeled, not ignored.
> 
> -- Bert

   I agree that in general blocking factors should not be ignored!  I
don't even think they should be discarded if they are
small/non-significant (in the ecological literature this is called
'sacrificial pseudoreplication' sensu Hurlbert (1984)).

 But the point is that for this particular case the *estimated* block
variation is *exactly* zero for one function and trivially small for the
other (I think it's exactly zero for lmer, but not sure), so in fact the
original poster will (I think?) get exactly (or almost exactly) the same
answer if they simply use lm() and ignore blocks entirely ...

  Or am I missing something?

  (Taking the liberty of cc'ing back to the list so I can look like an
idiot in public if necessary ...)

  cheers
    Ben