THE IMPORTANCE OF SELECTING A SAMPLING MODEL BEFORE DATA-COLLECTION -AN EXAMPLE USING THE ENDANGERED HUMBOLDT MILK-VETCH (ASTRAGALUS-AGNICIDUS BARNEBY)

Numerous sampling models exist as alternatives to simple random or sys tematic sampling. The appropriate application of one of these models m ay increase efficiency and decrease the cost of sampling. However, unl ess a model is chosen and a sampling methodology appropriate to the mo del is developed prior to data collection, the estimator is likely to be biased. As an example, we present a case in which a single methodol ogy was used to sample pod production of Humboldt milk-vetch (Astragal us agnicidus Barneby) using three sampling models after the fact to ob tain estimates. The results of the three analyses - cluster, two-stage , and double sampling - are contrasted with respect to estimated preci sion and bias. The two-stage estimator yielded the least precise but t he only unbiased estimator. This example illustrates some serious nega tive consequences of the relatively uncommon practice of collecting da ta before a sampling model has been identified.