Next: 12 Computing the effective
Up: Inferring patterns of migration
Previous: 10 Sampling error
  Contents
Typically, in most studies, sampling only take place in some
of the populations, whereas no information is available
about the gene frequencies in surrounding subpopulations. One
way of handling this is simply to ignore the problem. A
more satisfactory solution, perhaps somewhat experimental,
is to compute the covariances for the entire population
system, including unsampled populations, and then compute
the likelihood of the data based on the appropriate
covariance submatrix. Similarly, in the simulations, the
process of genetic drift in the entire system should be
simulated to produce bootstrap replicates of the data in the
sampled regions of the population.
To incorporate unsampled subpopulations, Ne should be
set up contain the effective size of all subpopulations. In
addition, the elements of the boolean vector sampled
should be set up to specify which subpopulations are sampled
and which are not, for example:
> sampled <- c(F,F,F,T,T,T,T,F,F)
Also note that the length of Ns and Nh should
match the number of true elements in sampled.
Jarle Tufto
2001-08-28