No significant differences among forests in regard to variation in tree 

 volumes vjere observed. Sampling techniques and rates can be uniform 

 throughout the region. A measure of the variation for each particular 

 sale that is to be sample tree measured should be obtained from cruise 

 data, from comparison with other similar sales, or from a small prelimi- 

 nary sample. Ordinarily, in straight randomized sampling, if no reliable 

 estimate of the variation can be obtained, then a sampling rate based 

 on a high coefficient of variation (about 80 percent) should be chosen 

 to be on the conservative side,. (See table in Appendix I.) 



The voluraes attained in scale book sampling in this study agreed well 

 with the actual scale book volumes. The accuracy percentages of the 

 sampling based on the spread of one standard error above or below the 

 sample mean, vjere as good or better than the actual accuracies of the 

 volume comparisons in two-thirds of the cases - just as the theoretical 

 probability would predict. 



Sample tree measurement first gives an estimate of tree measured volume. 

 That is, it is sure within knovjn accuracy limits of being as good as a 

 100-percent tree measured job, which in itself may or may not be suffi- 

 ciently accurate. In other words, a complete tree measured sale still 

 needs correcting to bring it up to the standards of log scaling. How- 

 ever, this correction or "adjustment" is sometimes made arbitrarily 

 from previous knowledge of defect conditions and tree measuring accu- 

 racies of similar timber. Especially on small sales, the practice of 

 using a "judgement" correction factor is followed by many forest officers 

 with satisfactory results. If the method is sufficiently accurate for 

 full tree measured sales, it is equally good for sample tree measured 

 sales. Figure 1 has been prepared to show the sample sizes needed for 

 a range of variations, sale sizes, and accuracies o The appropriate 

 sample sizes chosen with the aid of the graphs will give the indicated 

 accuracies in terms of tree measured volume. Methods will later be 

 discussed for determining sample sizes when check scaling is planned 

 and vjhen the chosen accuracy standards are based on scaled volumeo 



The sample sizes drawn from Figure 1 will give the indicated accuracies 

 with a probability of 68 percent. That is, they are based on a spread 

 of one standard error from the mean. If higher probabilities are 

 desired, larger samples will be needed. A sample size drawn from Figure 

 1 to give an accuracy of at least two percent would, for example, give 

 at least four percent accuracy vjith a 95 percent probability. 



-10- 



