We based an evaluation of adaptive cluster sampling (Thompson, 1990, J
ournal of the American Statistical Association 85, 1050-1059) on a sim
ulation experiment where samples were drawn from an enumeration of thr
ee species of waterfowl wintering in central Florida, U.S.A. The initi
al samples were taken either by simple random sampling or with probabi
lity proportional to available habitat. Efficiency of adaptive cluster
sampling relative to simple random sampling was highest when 1) the w
ithin-network variance was close to the population variance, and 2) th
e final sampling fraction was close to the initial sampling fraction.
The within-network variance is determined by the spatial distribution
of the population, quadrat size, and the condition that determined whe
n to adapt sampling. The final sampling fraction depends on the previo
us factors as well as the size and selection of the initial sample. So
me combinations of these factors led to increased precision compared t
o simple random sampling and some did not. Implications to design of w
ildlife surveys are discussed.