1. Accuracy desired 



When accuracy demands are lowered, a marked reduction 

 in sample size results. From the seller's point of 

 view, accuracy need not be high because (a) lovj accu- 

 racy means a saving in the volume determination job 

 and (b) errors tend to compensate over many sales. 

 From the purchaser's point of view, high accuracies 

 are desirable because he may not have the nmnber of 

 sales necessary to cause the possible errors to com- 

 pensate. The accuracj^ standard should not be the 

 same for every sale. The cost of measuring as vjell 

 as the total value of the sale should be considered, 

 because a rate can be reached in sampling where addi- 

 tional expense of measurements is not justified by 

 the possible gain in accuracy. 



2. Amount of variation in tree volumes 



Although the relative variation (coefficient of vari- 

 ation) is reasonably uniform for all species and size 

 classes, sufficient range in variation exists to 

 influence the sample size materially. An estimate 

 of the variation for a particular sale is needed before 

 determining the sample size and subsec^uent sampling 

 rate. Variation can be estimated from cruise data, 

 from a small preliminary samples or from a good knowl- 

 edge of the tim.ber concerned., (See table in Appendix !•) 



3 . Use of stratified sampling 



Stratified sampling, that is, sampling large and small 

 trees, or large, medium and small trees independently, 

 materially reduces the sample size needed fcr a given 

 accuracy. If the dividing point betvjeen the large and 

 small strata is chosen properly, all of the large trees 

 can be measured and the sample drawn from the remainder 

 with a net saving in number of trees measured. Further 

 reduction in sam.ple size can be made by increasing the 

 number of strata. 



Figure 1 gives sample sizes required on sales for a range of numbers 

 of trees, variations, and accuracies. Sample sizes read from this 

 chart vvfill give estimates of complete tree measured volume vjithin 

 the accuracies sho^m and for a probability of 68 percent. Estimating 

 log-scaled volume requires a correction based on a check scale. 



-27- 



