240
4. CSO sample surveys, 1982 to 1985
Following a hiatus in surveys from 1979 to 1982, annual surveys were conducted between
1982 and 1985 with assistance from the FAO. In 1981/82, a pilot census of agriculture was carried
out in preparation for the census of agriculture that was scheduled to be conducted during the 1982/83
crop season. The actual census was never conducted, and, as a result, the FAO and the CSO agreed
to do a sample survey of agriculture on a rotating basis. The four major agricultural
provinces—Central, Eastern, Southern, and Lusaka—were actually surveyed in 1982/83. Recall data
were collected during the following year to complete the 1982/83 survey in the remaining provinces.
It is not entirely clear how the 1984/85 data were collected. While the authors of the survey
publication state that a two-stage sample was used with survey enumeration areas (SEAs) serving as
the primary sampling units, they do not provide information on sample size at either the primary or
secondary levels. Again, no estimates of sampling errors were calculated or published for the data.
5. CSO sample surveys, 1986 to 1993
Since 1986, the CSO has conducted annual surveys of crop forecasting on the noncommercial
sector, initially with technical assistance provided from BUCEN and independently since 1990. Results
have been less than spectacular, at least with respect to the publication of the annual surveys. The
results of only the 1986 survey have been published. According to CSO officials, crop forecasting data
are regarded as confidential until the Minister of Agriculture certifies them acceptable. Publication of
such data has been the prerogative of that ministry because the final announcement was a blend of two
forecasts collected separately, one by the CSO, and the other by the Ministry of Agriculture. The CSO
is thus unable to publish the data themselves without upstaging or contradicting the minister's role.
For these surveys, a new sample design was used which comprised a three-stage sampling
system. The sample is used for both the crop forecasting exercise and the annual survey with the goal
of publishing data at the district, provincial, and national levels. The first stage of sampling is done
from the census supervisory areas (CSAs). Within each CSA, SEAs have been identified based on
readily identifiable boundaries. From one to four SEAs are located in each CSA. Sample CSAs are
first selected with probability proportional to the number of households located in each based on the
1990 census (1980 census in earlier data-gathering efforts). Once a CSA is selected, a sample SEA
is then selected, again based on probability proportional to the number of households. During the
earlier years of the survey, 800 SEAs were selected, however the number has been reduced to 400
for recent surveys. Households in each sample SEA are listed and farm households identified.
Following the listing process, farm households are stratified based on area planted to crops.
Three strata are identified: small (less than 5 hectares), medium or emerging (5-20 hectares), and
large (greater than 20 hectares). Fifteen sample households are systematically selected from each of
the small and medium groups. Normally, only a few households are identified as medium-sized
farmers in any given sample SEA. If fewer than 15 households qualify as medium-sized, all of these
households are selected and the remainder of the intended 15 sample households are reassigned to the
small-farm stratum. For example, if a sample SEA contained 100 small farmers and 3 medium
farmers, the final sample would include the 3 medium farmers plus 27 of the small farmers. Only if
the total number of farmers—small and medium—were less than 30 would the sample size be less than
30. This results in a sample size that nearly always approaches the maximum of 30 per SEA or a total