Data analysis and REML
We filtered period estimates from Biodare2 to exclude rhythms with
RAE>0.65 and period >28h as we found that
rhythms beneath these cut-offs gave the most reliable period estimates.
We next used restricted maximum likelihood (REML) to fit a linear mixed
model to the 191 accession dataset and thus obtain accession means for
period and RAE which were adjusted for the effects of cabinet and
experimental run (see details in Extended Methods). An additional step
was required to calculate phase means for each accession as the raw
phase data is circular and relative to dawn (0) with a full circle
representing 24 hours. Phase data was analysed with a circular
regression model using the Genstat RCIRCULAR procedure to obtain
accession mean phases adjusted for cabinet and run effects (Fisher &
Lee, 1992).
For temperature experiments, no period or RAE cut-offs were imposed as
we predicted an increase in both these variables with lower temperatures
(Dodd et al., 2014; Rees et al., 2019). We fitted linear models to the
period and RAE data and identified the contribution of each component by
analysis of variance (Supplementary Tables 12-14). The above data
analysis was done using Genstat 18th edition
(RRID:SCR_014595).
Map figures (Figure 2 and 4) were created using the ggmaps package in R
using Google maps (2018) (Kahle & Wickham, 2013).