The calculation provides estimates and confidence intervals
for the time since mtE and the population size at that time.
Our median-based estimate of the time since mtE is very closely
approximated by ˆs = log(1 + κ) in the scaled units, which translates
to ˆt = log(1 + (2m/σ 2) log λ)/ log(λ) generations in physical
units, where m is the ‘‘current’’ population. This formula, with
σ2 = 1, has also been obtained from a Wright–Fisher-based
coalescent model with a deterministically imposed population
growth (Chen and Chen, 2013). Furthermore, numerical simulations
by Cyran and Kimmel (2010) have shown close agreement between
coalescent times for a Wright–Fisher process and those for
a Galton–Watson process with a Poisson number of offspring per
individual, but poor agreement for other distributions of offspring
with σ2 not close to the slightly supercritical λ. For large populations,
the number of offspring in the Wright–Fisher model is
approximately Poisson. Given that our formalism has an explicit
dependence on the variance σ2 of the number of offspring per
individual, we would recommend instead comparison of a Galton–
Watson process with σ2 ̸≈ λ with a Cannings exchangeable
model with the variance of the number of offspring tuned to match
σ2.
Our estimate of the population size at the time of mtE is very
closely approximated by ˆκ0 = 1
4 θ2κ/(1+κ) in scaled units, where
θ ≈ 2.4868 is the positive solution to I1(θ) = θ. In physical units,
this translates to m0 = 14
mθ2 log λ/(1 + (2m/σ 2) log λ). This is
generally higher by a factor of about 1.5 than a value estimated
by naive extrapolation of a deterministic exponential growth. The
difference arises mainly from the fact that possible trajectories
corresponding to extinction of the entire population are included
in the calculation of our median-based estimate.