248
and realistic budgets and human resource support must be provided. Without the insistence of top
government officials on timely publications, the timing will inevitably slip. Data managers must see
their mission as providing high-quality information that is available for current decisions.
D. Improving accuracy
It is difficult to judge the accuracy of the agricultural estimates because there are few
alternative measures for comparison. The data series reviewed show variations that may have been
caused by either natural variations in agricultural production or data collection errors. However, the
magnitude and rapidity of fluctuations suggest significant collection errors. Unless a thorough review
of a survey's accuracy is completed during the survey period, much of the information that would
reveal the true accuracy of the data is lost.
For example, it is not difficult to understand why errors exist in the CSO data series when
surveys take several years to be published. It is difficult to accurately track data for the three-year
waiting period now occurring between data collection and data entry. Data have to be tracked at every
point in the data entry/editing process—upon receipt, during data entry, and at each point in the
computer editing process. Steps need to be taken to ensure that manual edits performed in the field
and after receipt at the central office are performed correctly. This requires both training and good
manuals which define the editing required.
In the future, it is hoped that both the MAFF and the CSO survey units will publish technical
notes that attempt to explain data fluctuations. A mature and confident survey unit displays an open
interest in eliminating survey shortcomings. All surveys, even in experienced hands, have points of
difficulty. For the Zambian units to ignore or minimize them is not helpful to the data user. When the
survey managers begin publishing comments on the possible biases in the survey, the data users will
gain confidence in both the data and the survey unit. Again, this type of information can be easily lost
in the delay between collection and analysis/publication.
The survey units should make every effort to employ improved statistical techniques designed
to minimize sampling and non-sampling errors. Employing "total survey design" techniques would be
a helpful approach. This methodology allows the survey designers to identify the various error-causing
components of a survey and to minimize both sampling and non-sampling errors.
Training is often a key component in minimizing survey errors. 19 All survey personnel should
be provided with comprehensive training so they fully understand and appreciate the impact their
actions have on data quality. For example, field personnel are often expected to spend long days
following intricate procedures in search of data that have little meaning to them. This situation often
leads to fabrication of the data by the field enumerators. However, when the staff is given information
about the survey sample design and insights into the purpose of the survey, the quality of their
performance is higher.
19
The CSO is presently engaging its statisticians in on-the-job training of manual editing using 1992/93 post-harvest data.
The statisticians, after the program, are then expected to supervise junior officers in future surveys. Based on early feedback,
the statisticians are gaining a lot of experience with common data errors generated by enumerators in the field and data entry
personnel in Lusaka (CSO, personal communications).