Empirical comparison of uniform and non-uniform probability sampling for estimating numbers of red-cockaded woodpecker colonies
Four sampling and estimation methods for estimating the number of red-cockaded woodpecker colonies on National Forests in the Southeast were compared, using samples chosen from simulated populations based on the observed sample. The methods included (1) simple random sampling without replacement using a mean per sampling unit estimator, (2) simple random sampling without replacement with a ratio per pine area estimator, (3) probability proportional to 'size' sampling with replacement, and (4) probability proportional to 'size' without replacement using Murthy's estimator. The survey sample of 274 National Forest compartments (1000 acres each) constituted a superpopulation from which simulated stratum populations were selected with probability inversely proportional to the original probability of selection. Compartments were originally sampled with probabilities proportional to the probabilities that the compartments contained woodpeckers ('size'). These probabilities were estimated with a discriminant analysis based on tree species and tree age. The ratio estimator would have been the best estimator for this survey based on the mean square error. However, if more accurate predictions of woodpecker presence had been available, Murthy's estimator would have been the best. A subroutine to calculate Murthy's estimates is included; it is computationally feasible to analyze up to 10 samples per stratum.