[R-sig-ME] Are mixed models a cure for pseudoreplication?

[Christoph Scherber]

Dear all,

I have sampled grasshoppers from a set of 40 sites on 10 transects per site.

Now if I analyse the pooled data (N=40), the minimal adequate model is the null model.

However, if I account for non-independence and include site as a random effect ("pseudoreplicated"
N=400), I can see much clearer relationships, and the minimal adequate model contains three
explanatory variables plus a two-way interaction.

Which model is correct? The pooled (non-lme) one, or the lme?

I have the feeling that the pooled model gets it wrong in making me believe that "nothing happens"
in the data, while the mixed model is nicer but obviously inflates standard errors.

Any suggestions?

Best wishes,
Christoph



-- 
PD Dr Christoph Scherber
Georg-August University Goettingen
Department of Crop Science
Agroecology
Grisebachstrasse 6
D-37077 Goettingen
Germany
phone 0049 (0)551 39 8807
fax 0049 (0)551 39 8806
http://www.gwdg.de/~cscherb1