Appendix C. Observation model.
We use negative binomial models for measurements of aphid and mummy abundance. These can be motivated heuristically by assuming that stem counts in each quadrant are iid Poisson random variates conditional on a per-quadrant mean, and that the per-quadrant mean is a Gamma random variate with mean equal to the field-level mean (from the process model) and shape parameter k. Different shape parameters are used for aphids (kA) and mummies (kM).
For stem counts, the total number
of aphids counted on all stems in a quadrant has mean nst
UJ+PJ+UA+PA), where nst is the
number of stems counted. The total number of mummies counted during stem counts
has mean nst (M+H). The number of aphids counted
in tensweeps in a quadrant has mean nsw
s(UA+PA
+ 0.75(UJ+PJ)), where the conversion factor
s
is the effective number of hundreds of stems swept in a single tensweep, and
nsw is the number of tensweeps in a given quadrant. Using
data from a previous field season, we estimated that 25% of juveniles are not
counted in a ten-sweep. The number of mummies observed during time counts in
a quadrant has mean ntcktc (M+H),
where the conversion parameter
tc
is the effective number of hundreds of stems surveyed in a three-minute time
count and ntc is the number of three-minute time counts.
For the dissection data, we model
the number of parasitized adult aphids in each quadrant as a beta-binomial with
mean PA/(UA+PA) and overdispersion parameter
W. For
the hyperparasitism data, not all mummies emerged, so we fit two parameters
to model the fraction of (unhyperparasitized) wasps and hyperparasitoids that
emerge
W
and
H,
respectively. We model the total of emerged parasitoids, emerged hyperparasitoids,
and unemerged mummies as a Dirichlet-trinomial (the extension of a beta-binomial)
with overdispersion parameter
H.
The expected fraction of emerged parasitoids is
WM
/(M+H), and the expected fraction of emerged hyperparasitoids
is
HH
/(M+H).