12 CIRCULAR 816, U. S. DEPARTMENT OF AGRICULTURE 
A rather systematic distribution of stops is better suited to many of the 
objectives of survey work. Three factors that must be provided for are 
as follows: (1) Representation of important sections and crop types, (2) 
plan of travel, (3) freedom from personal choice in selecting stops. In 
travel careful planning to obtain economy and coverage is required. In 
general, location of fields may be easily predetermined as at a given 
distance from some known point. 
A systematic sampling procedure, such as suggested, will tend to give 
more accurate results than random sampling. At the same time, if error 
is calculated as if the sample were fully random, the error estimate will 
tend to be too high. When the systematic sample gives results closely 
equivalent to results from a random sample, as often occurs in this work, 
there will not be much inaccuracy in treating it as a random sample, and 
what inaccuracy there is will be on the conservative side. Some random¬ 
ness may be arranged after the three essential factors have been provided 
for. 
A restricted random-sampling plan, or stratified sampling, is often 
followed, in order to ensure that some units fall in each subarea of crop 
type. In such a plan the variance between types or subareas should be 
removed in calculating the sampling error, because the precision of such 
a survey is properly determined from the variance within the types or 
subareas, rather than between them. In this kind of sampling it may be 
difficult to arrange satisfactory stratification when the number of units is 
small. It may not be important, however, if areas are rather homogeneous, 
if one crop type is of outstanding importance, and if eggs are widely dis¬ 
tributed. In some other grasshopper-survey problems it may be important. 
If a large number of fields (25 or more) can be sampled, it may be 
possible to represent each important subdivision by several fields, pre¬ 
serving essential randomness in the location of these fields within the 
subdivisions. Randomness in location is easily achieved by some system 
of drawing numbers. Use of a group of similar counties as a unit may aid 
in attaining this objective and may ensure greater precision. If the number 
of fields is limited, it seems best to distribute them rather widely, select 
them by some objective method, and analyze the data as if the fields had 
been taken at random. The procedure described is not entirely correct, 
but it does not lead to serious mistakes. 
Time Utilization of Sampling 
A final evaluation of sampling plans may rest on the total possible cost. 
A plan giving the desired precision may be too expensive. Although sam¬ 
pling within fields is less important in reducing error than that between 
fields, it is less costly, and may give the best result in limiting the total 
expense of doing considerable work in a field, once the field is reached. 
Biometric theory enables us to determine the best distribution of 
samples between and within fields if we have some preliminary estimates 
of limitations on total cost, cost of sampling fields and units within 
fields, and between-field and within-field variances. For purposes of 
analysis, these costs have been estimated in terms of time requirements. 
The total cost of planning, reaching a field, writing up results, and other 
necessary work has been estimated at 1 hour. The total allowable time 
