Phylogenetic or family trees reconstructed from a sample of individual
s belonging to a population or species can be used to infer population
dynamic history. A method for making such inferences involves visual
inspection of the graphs of the numbers of lineages in the phylogeny p
lotted against the times when they occur. One transformation of the li
neages axis often produces a linear lineages-through-time plot if the
population has been growing exponentially. However, that transformatio
n is known to fail under certain circumstances. We report a series of
simulation studies designed to determine the conditions under which th
e transformation fails. As long as the sample represents less than 0.5
% of individuals in the total population, the Type 1 error rate is acc
eptable. This means that studies of virus populations are likely not t
o suffer from transformation failure, so long as a random sample from
the population is represented. When the transformation does fail, the
resultant lineages-through-time plot is very similar to a two-stage ep
idemic process. (C) 1996 Academic Press Limited