Simulation results
In simulations using with complete sampling of candidate fathers, true
fathers had mean posterior probabilities of 0.993, with a minimum in any
simulation of 0.984, indicating that the marker set used here has high statistical power
to resolve relationships. However, when no information about dispersal
is included in paternity analysis, the remaining uncertainty about
paternity causes apparent distance between mates to be inflated by
approximately 40m, regardless of the true dispersal distance
(figure \ref{173111}A, left-most panel). This bias increases to several
hundred metres when sampling of fathers is incomplete as paternity is
incorrectly assigned to unrelated individuals further from the mother,
inflating the tail of the dispersal kernel (figure \ref{173111}A).
In contrast, when dispersal is inferred jointly with paternity, the
inferred shape parameter values of the dispersal kernel are close to the
true value of one (figure \ref{173111}B). In fact, when sampling of father
is complete, the shape parameter is somewhat inflated (i.e. the pollen
dispersal kernel is inferred to decay more rapidly than the exponential
distribution). This inflation decreases as the proportion of missing
fathers increases, and begins to underestimate the shape parameter of
the dispersal kernel when many fathers are unsampled.