2.7. Statistical analyses
All data are reported as mean ± standard error of the mean (SEM).
Immunological datawere analysedwith a permutation-based approach
[31]. The repeated measures nonparametric ANOVA was set as follows:
for repeatedmeasureswithin control and LPS-treated groups the significance
of the two comparisons post 4 h vs pre (ΔT (4 h–0 h)) and post
24 h vs pre (ΔT (24 h–0 h)) was computed through a permutationbased
paired t-test; the overall effect was computed through direct
combination of the two test statistics (i.e. sumof the two statistics divided
by square root of 2). The interaction effects at 4 h and 24 h were
estimated as the difference between mean values (i.e. ΔT (24 h–0 h)
in treated minus ΔT (24 h–0 h) in controls) and tested through a permutation
test for independent samples. The alternatives are assumed
one-tailed since we could predict that any eventual effect of immunechallenge
on immunological parameters would have resulted in an
increase of values. Behavioural data were analysed in a similar way
with a two factor full model (i.e. treatment and time) within each pair
of octopuses. Since animals were not randomly assigned to the control
or to the treated group (see Section 2.5 in Materials and methods) we
avoided to make a statistical comparison between groups (interaction
time × treatment). Tests on behavioural responses are two-tailed.
With our relatively small sample size, the permutation-based approach
ensured exact control of the type I error without assuming normality
of the error terms. For each test a t-value and a p-value are reported
and since the significance is computed through permutations, degrees
of freedom are not used. The software packages R [32] with package
flip [33] were used for statistical analyses.