Table 2.

Results of the linear mixed effects model analysis—models were fitted using the lme function in the nlme package for R. The random effects structure is given for the model. A first-order autoregressive function (corAR1) was used to model the temporal autocorrelation in the data that arose from making repeated measurements on replicates. The results of the model selection procedure on the fixed effect terms are also given and the most parsimonious models are highlighted in bold. Analysis revealed that the metabolic balance (R/P) varied significantly between treatments and months, though the interaction between treatment and month was not significant.

model	d.f.	AIC	logLik	L-ratio	p
random effects structure
random = ∼ 1|mesocosm,
corrected structure = corAR1 (month|mesocosm)
fixed effects structure
1. ln(R/P) ∼1 + month × treat	17	−28.5	31.3
2. ln(R/P) ∼1 + month + treat	11	−38.8	30.4	1.7	0.95
3. ln(R/P) ∼1 + treat	5	−9.3	9.7	41.5	<0.0001