Linear mixed model and Chi-square tests

From: Pytrik Reidsma (pytrik.reidsma_at_wur.nl)
Date: 02/02/05 

Date: Wed, 2 Feb 2005 12:59:04 +0000 (UTC)

Using the spss linear mixed model procedure, I applied some models on
my data, but have a few problems with interpretation. The first model
includes only fixed effects, the second also a random intercept and
the third also includes random slopes. I have data from 50.000 farms
on crop yields and farm characteristics and biophysical data. The
deviance of the models is very high (see appendix below). The
calculate Chi-square is also very high and therefore models including
random effects are highly significantly better. Are these good
conclusions? Or could you say too high deviances could mean the models
aren't good enough. The same data in a GLM with the region as a fixed
factor gives an r2 of 0.66. We can't calculate r2 linear mixed models,
but I've seen papers where they calculated the rho. In spss there are
no possibilities to include extra goodness of fit measures, or are
there?

Pytrik Reidsma

Appendix: some model results

 
Comparison of models for wheat yield
        Output Calculations
        -2LL parameters Chi-square df p-value
Fixed 64918.298 11
Rand i 57168.549 12 7749.749 1 0.000
Rand sl 56271.93 18 896.619 6 0.000