2.3.4. Statistical analysis
The full models for ion leakage and compound content included
the predictors: temperature, day length, and cold acclimatisation
and their second- and third-order interactions. For statistical inferences,
we employed model selection using Akaike’s information
criterion corrected for small sample sizes, (AICc), a likelihoodbased
measure of model plausibility that penalises models with a
higher number of parameters. In cases with DAIC < 1, the model
with the fewer parameters was assumed better (Burnham &
Anderson, 2002). For freeze tolerance, the response variable ion
leakage is a proportion, hence a general linear model with normal
distribution and a logit link was used. For other responses, a general
linear model with normal distribution and identity link was
assumed. Particular variables were considered significant if their
95% confidence interval did not overlap zero. Differences among
levels of a variable were identified using the Tukey test at the
95% significance level. Data were log transformed when necessary
to ensure homogeneity of the variance. Data presented here are
back-transformed for clarity. The statistical package R (version
2.15.0, R Development Core Team 2010; R Foundation for Statistical