Findings, implications and lessons learned

• This example has shown how mixed models can deal effectively with different layers in the data; this form of analysis gives more valid significance tests and provides appropriate and correct standard errors − something that conventional least squares analysis of variance methods cannot do except in one or two very specific circumstances.

• REML has the ability, particularly with unbalanced data structures, to combine information from the different data layers. This has the advantage of improving the precision of fixed effect comparisons.

• Before applying mixed models it may sometimes be helpful to evaluate some of the important fixed effects first (as was done in Case Study 3), and then to add the random terms later, as has been done here.

• This case study has also shown how to recognise the structures of the different layers in a data set, and has explained the understanding of variance components associated with random effects.