Generalized nonlinear models relating measured <i>E. propertius</i> CO₂ production rate to measured body temperature.

<p>All models provided a better fit than a linear model (<i>CO2 = S×T</i>; log-likelihood test, p<0.05 in all cases). Model 5 was selected because there was no significant difference in explanatory power between Models 4 and 5 (likelihood ratio test, L = 3.06, p = 0.08) and Model 5 has fewer terms. A power of the mean model for variance was used in all models. Empirical input: CO₂: CO₂ production rate, ml·min⁻¹; T: Temperature, °C; M: body mass, mg. Coefficients estimated by maximum likelihood: S: an overall scaling factor; M_s: a mass scaling factor; T_s: a temperature scaling factor (i.e. thermal sensitivity). L_met: theoretical minimum life-supporting metabolic rate. AIC: Akaike's Information Criterion; df: degrees of freedom.