Fig. 3

Conditional inference trees for the first-to-break measure by each combinatorial mode of inheritance (N = nuclear and M = mitochondrial, giving combinations for A as NN, B as MN and C as MM). The simulation data on the first-to-break for each strategy is classified into a categorical variable to describe whether one or multiple strategies have > 10% difference (in either direction) in their first-to-break measure, including the ‘sequences’ lower benchmark but excluding the ‘maximum’ upper benchmark (because this would mask meaningful comparisons). Non-measured data types (see Fig. 1) are given nominal values that ensure their hierarchical interpretation: where ‘Toward Threshold’ is set to 1000, ‘Away from Threshold’ is set to 1500 and ‘Extinction’ is set to 2000. Data classifications are given for all strategies and their combinations, but only five classifications are needed to describe the data: ‘ = ’ where all strategies have < 10% difference, ‘CX’ where mosaics (C) and mixtures (X) have < 10% difference but > 10% difference than rotations (R) or sequences, ‘CXR’ where mosaics (C), mixtures (X) and rotations (R) have < 10% difference but > 10% difference than sequences, ‘R’ where rotations have > 10% difference than all other strategies and ‘X’ where mixtures have > 10% difference than all other strategies. The conditional inference tree is used to partition the data classification output based on the parameter space inputs based on the 1 million randomly sampled parameter combinations for the 17 parameters (see Table 4). Trees are built and drawn using R:ctree, which uses permutation tests to iterate an algorithm that tests the independence between the inputs and output variables and makes a binary split in the variable with the strongest differentiation of output distributions. The parameter of each split is given in the nodes within the tree, which reports the parameter (as per Table 4) and the p-value of the independence test; the quantitative place of the split in the parameter itself is recorded in the line between nodes. The iterations that form the tree stop when algorithm can no longer make a split into terminal nodes with > 5% of the data, which is a control applied for the visualization of the tree to ensure a manageable number of terminal nodes. The distributions of data classification are given in the terminal nodes as a bar chart, where the y-axis describes the proportion of data points