Table 2. Correlations between residual RMRt and DEE from various models.

Predictors in models*	No. of parameters	AIC response = log10 (RMRt)	AIC response = log10 (DEE)	Correlation between residuals (95% C.I.)†	t H0: zero correlation	P value
Intercept only	1	+25.4	+7.4	0.34 [0.12, 0.59]	3.06	0.0031
BM‡	2	+0.2	+2.2	0.26 [0.03, 0.51]	2.28	0.026
Site	4	+30.6	+9.9	0.37 [0.15, 0.62]	3.31	0.0014
Site + BM	5	+3.6	+3.0	0.26 [0.03, 0.51]	2.29	0.025
Site × date§	8	+28.1	+7.1	0.28 [0.06, 0.53]	2.48	0.015
Site × date + BM¶	9	0.0	0.0	0.16 [-0.07, 0.40]	1.37	0.18

Given are Pearson correlation coefficients between residual RMR and residual DEE from models with the same predictors. Also given are the models' AIC values for each of the two response variables. The AIC model selection criterion ranks the models according to their predictive ability (models with lower values give better predictions; see Materials and Methods). The values given are the differences from the model with the lowest value (i.e., site × date + BM for both response variables). A difference in AIC of less than one unit is considered unimportant. [We have added 95% confidence interval (C.I.) because this is relevant for the discussion on insufficient power to detect within-site correlations.]

^*All models include intercept.

^†Confidence intervals calculated by using Fisher's transformation (46).

^‡BM is body mass.

^§Site-specific intercept and slope on date effect.

^¶Interaction effect on log_e(RMR): F_{3, 145} = 3.66, P = 0.01. Interaction effect on log_e(DEE): F_3,74 = 3.33, P = 0.02.