Figure 2.

Finding suitable prior parameters for relaxed-clock models. a) Relationship between branch lengths estimated under a strict-clock and under a nonclock model. The outlier corresponds to the branch leading to Xyelidae. b) Increase in variance (squared deviation from the expectation) of nonclock branch lengths over time. The slope of the regression line was used directly to center the prior for the variance increase parameter of the IGR model, and via simulations for the TK02 (Thorne–Kishino 2002) anc CPP models. c) Outcome of the simulations under the CPP model showing success of different combinations of the variance of the rate multiplier and the rate of the Poisson process to approximate the observed variance in branch lengths. Colors (in the online version of this figure) represent the relative deviation from the observed variance. The results were used to choose appropriate priors. d) The rate ratios observed in random pairs of branches tend to be higher than the ratios between parent and offspring pairs, demonstrating slight but significant rate autocorrelation in the data set (P=0.0055; Kolmogorov–Smirnov test of equality of distributions). The bars on the random pairs represent SDs obtained from 1000 replicates of random pair sets of the same size as the parent–offspring pair set.