238
at the provincial and national levels. Table 8.5 lists the published time series data for maize production
on noncommercial farms, as an example.
While the annual survey of noncommercial farmers began in 1971/72, the crop forecasting
portion of the survey was first initiated with technical and equipment support being provided by the
ZATPID II Project (equipment and long-term advisory support) and the US Bureau of Census
(BUCEN—training and short-term technical support) in 1985/86. As can be seen in table 8.2, this
effort has been less than successful in producing publishable data sets. Despite continuous collection
of detailed annual survey data since 1986, only the results from that year have been published."
Crop forecasts, however, have been published for the period 1986 through 1993. 12 The following
is a description of the evolution of the CSO sample survey in its three iterations.
3. CSO sample surveys, 1972 to 1978
Designed as a probability survey of the entire country, the sampling frame was built by first
stratifying the country into urban, peri-urban, and rural areas. The urban stratum contained all
principal towns with populations of 30,000 or more; the peri-urban stratum comprised all urban
centers with populations less than 30,000 plus all provincial and district administrative centers; the
rural stratum comprised the remainder of the country.
Once the rural stratum was identified, a 10 percent sample of basic areas (BAs—the primary
sampling unit [PSU] for the survey) was selected. A total of 134 BAs were selected in 1972 out of
a total rural population of 1,145 BAs. 13 Field enumerators listed all of the noncommercial farming
households in the sample BAs, and a sample of 10 percent of the households was randomly selected
for interviewing. The number of selected BAs ranged from 3 to 28 per province. Fortunately, the
amount of data collected under these surveys was reasonable—especially when compared to the
magnitude of the present efforts. Copies of the questionnaires used can be found in the back of each
publication.
Despite the probability design of the survey which would have permitted sampling errors to
be calculated, the publications do not discuss any sampling or non-sampling errors. Given the large
size of the PSUs, it is difficult to say what the sampling errors might have been. Normally, a sample
size of 10 percent in the second stage of sampling is more than adequate for agricultural surveys. As
with most agricultural surveys, the non-sampling errors are a very important part of the estimation
procedure and are largely immeasurable. If the survey managers had noted the major data collection
problems and how they were handled, it would have provided future data users with some hints
concerning the quality of the time series.
11 Summary tables for both the 1986/87 and 1987/88 surveys have been completed but not published. (Disks containing
the 1986/87 summary data have been misplaced.) The 1988/89 raw data have been entered on the computer but have not
been summarized because the expansion factors (weights) needed for summarization have not been entered.
12 Although these data have been published, a complete data set is not available. Some data were able to be recreated
from disk backups.
13 The size of these BAs must have been huge. If all Zambia was divided into 1,145 BAs, the average size of a BA
would have been approximately 65 square kilometers. At present, 9,682 rural survey enumeration areas (SEAs) are used
as PSUs. The average size of a SEA is approximately 8 square kilometers.