241
of 12,000 from 400 sample SEAs. Usually, no large-farm households are identified. If they are
identified, data are collected from these households via the commercial farm survey.
After the sample farmers are identified, each is visited three times. During the first visit,
which is conducted in January/February, the listing process is completed, sample farmers are
identified, and preliminary crop forecast data are collected. During the second visit, which is
conducted in March/April, data are collected for the final crop forecast. During the third visit, which
is conducted in October/November, a post-harvest survey is completed. This last data set forms the
basis for the annual agricultural survey report.
Based on a preliminary analysis of the sample size, the CSO's noncommercial survey is both
too small and too large. While a thorough review of the data is necessary to determine the optimum
number of samples to be selected at each stage of the survey, experience suggests that a sample size
of 400 PSUs would tend to have sampling errors that are too high for data presentation at the district
level. At the same time, a secondary sample size of 30 farmers is clearly larger than required at the
final stage of sampling. More likely, 5 to 10 small farmers and a 100 percent sample of medium
farmers selected from each sample SEA would provide the basis for acceptably accurate provincial
and national estimates. This would reduce the data handling load by 75 percent which would greatly
facilitate the speed of collection, summarization, and publication. While the data should probably only
be published at the provincial and national levels, it may be feasible to provide district estimates for
only the larger districts that exhibit reasonable sampling errors. 14
A review of the "Phase 3 Final Crop Forecasting Questionnaire " shows the survey instrument
to be of reasonable length. However, the post-harvest survey is extremely detailed with 33 pages.
Thus, three sets of data are to be collected from 24,000 (presently 12,000) farmers each year. This
means the CSO staff is faced with a total of 72,000 (or 36,000) questionnaires to be entered, edited,
and summarized annually. The post-harvest survey questionnaire is far too long and is also not
designed for efficient data entry. It should be reduced so only the essential data are being gathered and
simplified to allow easier data entry.' 5
The large size of the questionnaire is also problematic during the summary phase.
Conversations with CSO personnel indicate that corners are cut to speed the summarization process.
For example, as far as can be discerned, no computer edits of the data are conducted. Given the
pressures of trying to push large amounts of data through a summary system without adequate edit
"According to officials in the CSO, the government's policy is to encourage development planning in all sectors at least
down to the district level. The CSO thus feels compelled to collect and publish data at the district level to help facilitate
government planning. Rather than abandoning the current practice, the CSO instead recommends improving the quality and
lowering the cost of district-level data collection.
While the management of CSO also feels that the questionnaire is probably too long, it stresses that length alone is
not a hindrance. Space is provided for capturing all the possible production activities of households, but rarely does a
household have more than 4-5 activities (out of 20 in section 2) to report, leaving substantial empty space. Also, few
households report information on growing fruit trees in section 3 and vegetables in section 4. Thus, the majority of the
questionnaires returned to the CSO have substantial space that has been left blank. Moreover, the CSO management feels
that the current data being collected are essential for monitoring purposes. Before discarding items for sake of increasing
efficiency, the CSO management advises caution against loss of important information, a point of view shared by the authors.