Poisson sampling and some alternatives in timber sales sampling

Sampling schemes are studied for timber sales where every tree in the population has to be visited. A strong linear relationship is known to hold between the variable of interest, volume in m³ (y), and tree basal area times height (x) (in cm²m). Purposive sampling of trees over the X-range and stratified sampling with regression estimators are recommended as worthwhile alternatives to Poisson sampling for efficiency and for yielding consistently good results where the model between the dependent and independent variables is known to be reasonable. Two other designs, selecting the n trees with largest X-values and the trees with 20% smallest and 80% largest X-values, are shown to be high-risk procedures.