Extended Data Fig. 3: Analysis of 200 simulated datasets with random varying sampling rates through time using a BBB model with constant sampling rate (a = 0).

The times of origin were accurately estimated (a); circles and bars indicate posterior estimates and 95% credible intervals. The relative errors on the time of origin were smaller in datasets with richer simulated fossil record (b). The size of the 95% credible intervals around the times of origin decreased with increasing numb er of fossils (c). The log variances were slightly underestimated (d), while the estimated sampling rates (e; the X-axis is log₁₀-transformed) cannot be plotted against true values because the underlying simulations were based on time-heterogeneous sampling with di?erent rates in each time bin. However, we plot for comparison the distribution from which sampling rates were sampled, randomly for each time bin (f; the X-axis is log 10 -transformed).