VALIDITY OF THE SURVEY METHOD OF RESEAECH. 9 
making questions submitted to farmers conform to the terms in 
which the farmer's knowledge exists. When this is done a proper 
study of data furnished by farmers may reveal numerous important 
facts never suspected either by the farmer or his questioner. For 
instance, if in the farm-management survey made some years ago 
in Lenawee County, Mich., the farmer had been asked directly what 
the manure of a horse or cow was worth to him, he probably would 
not have hazarded a reply. If he had it would have been little more 
than a guess, not an estimate. But when the question was broken 
up into its elements and he was asked to state the acreage and yields of 
his various crops, the prices at which his products were sold, the 
number and kinds of animals kept on the place, he answered readily 
enough. By taking these data from many farms and comparing 
those having relatively little stock with those having many, the 
actual increment in crop values due to the manure of a single animal 
was easily calculated. 1 
LAW OF ERROR. 
The law of error, frequently called the law of averages, may be 
stated in many different ways. Perhaps as comprehensive a state- 
ment of it as any is this : " Errors of measurement or observation 
tend, in the absence of ' bias,' to group themselves about the true 
value of the quantity measured in such manner as to eliminate each 
other in the final average." 
The manner in which such errors group themselves about the true 
average will be discussed in some detail a little later. 
Absolute accuracy is not obtainable in any kind of measurements. 
In any case it is merely a question of degree of accuracy. 
The accuracy of any average depends on three things. First, and 
most important of all, is freedom from " bias " ; that is, entire absence 
of any tendency to make each measurement too high or too low. In 
general, we have found bias singularly absent in practically all our 
field studies of farm practice. It is true that some farmers deliber- 
ately overestimate, but fortunately there seem to be about as many 
who deliberately underestimate. These over and under estimates tend 
to cancel each other and thus to reduce their effect on the resulting 
averages. 
Second in importance is the number of items on which an average 
depends. The larger the number the more reliable the average. The 
reason for this lies in the fact that when a number of items is aver- 
aged the larger the number the better the chance that any error will 
be canceled by a similar error in the opposite direction. 
Since no measurement of any kind is absolutely accurate, every 
measurement represents an error of greater or less magnitude. 
1 See Dept. Agr. Bui. 341, Table LX, p. 98. 
