ACCURACY OF ESTIMATES 



AREA 



In determining the extent of various cover types and stand- 

 condition classes there are two possible sources of error (1) 

 errors in classifying the cover of the field samples and in 

 compiling the field data, and (2) sampling errors. The former 

 result from mistakes of judgment or technique and the complexity 

 of the cover which not infrequently grades from one class into 

 another with no clearly defined boundaries „ These errors were 

 minimized by the exercise of care and skill, but it is seldom 

 possible to evaluate them. An effort was made to maintain a 

 high order of accuracy and uniformity of standards in the 

 classification, collection, and compilation of sample data by 

 field checks, by a continuing program of training, and by 

 cross checks in the office. 



Sampling errors (standard errors of estimate), on the other 

 hand, do not involve human errors, but rather are theoretical 

 measures of the reliability of estimates based on the variabil- 

 ity exhibited by sample measurements. They generally vary 

 inversely with the square root of the number of samples and 

 directly with the square root of the unsampled proportion of 

 the total population. Hence, they can be controlled by alter- 

 ing either the number of samples, the size of individual 

 samples, or both. 



Area estimates for the portion covered by complete survey have 

 no sampling errors. Such technique errors that may have 

 occurred in spite of all reasonable precautionary measures are 

 small or negligible. Analysis of sample variations for the 

 portion covered by sampling indicates that the standard errors 

 of estimate are t 2.0 percent for total, ±2.$ percent for 

 commercial, and ± 4-4 percent for noncommercial forest land 

 areas. Accordingly, the probabilities are 2 out of 3 that for 

 the areas sampled the actual total, commercial, and noncommercial 

 forest land areas are, respectively, within t 146,000, ± 152,000, 

 and 1 63,000 acres of the estimates, if measurements and comput- 

 ing errors introduced no bias. 



-39- 



