229
authority to replace employees who do not display a keen interest in maintaining the highest quality
standards.
A portion of a data set's accuracy can be measured by calculating the sampling errors. This
allows a statistician to make statements of the likelihood of a population parameter falling within a
certain range of the survey statistic. These calculations are valid only if the survey was designed and
carried out as a probability survey—meaning that each sampling unit has a known probability of
selection. To conduct a probability survey, it is necessary to construct a sampling frame and to follow
a scientific sample design.
Accurate information is derived from data that have both low sampling errors and low non-
sampling errors. As mentioned above, these errors are impossible to eliminate and difficult to
minimize. Data managers should be free to discuss the various sources of errors in a data set and
should publish sampling errors and data collection procedures in technical notes.
E. Types of surveys
While agricultural censuses were once believed to be the only certain way of obtaining
accurate agricultural information, a wide variety of innovative methods of collecting agricultural data
have been developed during the past several decades. Survey organizations have accepted the validity
of smaller, more specialized probability surveys and no longer see the need to do a complete
enumeration of all farming units. The proliferation of survey types has placed an added responsibility
on the data manager and statisticians to choose the most effective survey designs. Careful survey
selection is crucial if the four criteria listed above are to be met. For example, it is not possible to
meet the need for current food security information with a national census that takes two years to plan
and implement. On the other hand, a rapid rural reconnaissance survey will not suffice if detailed
structural information on the number of farmers operating under various land tenure arrangements is
needed. At times, these decisions may mean abandoning a time-honored survey design in favor of a
more efficient, cost-effective alternative.
The most important factor in maintaining efficient surveys is to limit the use of each survey
to only the most necessary information. Planners must remain vigilant in the effort to limit data
collection to only necessary items. A survey overloaded with data requests slows the process, makes
the data less accurate, and wastes time and money; the data provider must endure lengthy interviews
and the policymaker must search through extraneous data for the information needed to reach
meaningful decisions.
Table 8.1 provides additional information on the interactions and responsibilities necessary in
coordinating an agricultural data system. It should be recognized that all the criteria listed have very
likely been present at one time or another in the Zambian agricultural data system. The available data
series show that donors and the Zambian government have supported data collection efforts rather well
for over twenty years.