237
the data, uses the information provided by the voluntary respondents as indicators of what all farmers
are doing. Using this method, the average of the data from the mail responses is calculated and
multiplied by the estimated number of commercial farms. Since the CSO commercial farm data were
published by farm size within provinces, the procedure would be somewhat more complicated. The
statistician would first have to estimate the number of farms in each size group in each province and
then apply an appropriate average value to each.
In the 1977/78 report, the data user is warned that the estimation procedure is not totally
accurate and that the "maize and other major crops would be the most accurate." This is because there
is usually less variability in area, yield, and production data when nearly all the farmers are actively
producing the crop. Experiences from other African countries suggest that a direct expansion of data
from voluntary respondents may be flawed for a number of reasons. Some amount of overestimation
or upward bias was introduced when the smaller commercial farms did not receive a questionnaire and
thus were not given a chance to respond. On the other end of the scale, large farmers are notorious
for either refusing to report or grossly underreporting their agricultural holdings. Since the commercial
farms survey was totally voluntary with no response validation process, it is reasonable to expect that
the holdings of the largest farmers are underrepresented. This would cause a downward bias in the
estimates derived by expanding the data. While it is impossible to speculate on the impact of the
adjustments on the crop estimates, it is fair to say that the low response rate from an unknown and
possibly underrepresented portion of the population leaves serious questions about the level of
estimated crop production reported. For example, the data series in table 8.4 listing the estimated
number of farms shows much variability—changing radically from one year to the next, especially for
the three years beginning in 1982/83 and then again in 1989/90. If it is difficult to have confidence
in an estimate as basic as the number of farms, it is even more difficult to imagine how the more
detailed estimates of crop area and production might be accurate. 10
Another anomaly in the commercial farm survey is the timing of publications. Although the
questionnaires continued to be mailed each year, the results are often not published until several years
later. In the 1987/88 report (published in September 1990), the director thanks the staff for catching
up with 12 years of unpublished data. Evidently the demand for detailed commercial farm data was
strong enough to cause the CSO to resurrect and publish data that had not been given a high priority
at the time it was collected.
2. CSO noncommercial crop surveys and crop forecasts
To complement the commercial farm data series, the CSO also collects data from the
noncommercial farmers (all farmers not falling under the definition of commercial farmers). Attempts
to measure the agricultural production of the noncommercial farmers is done via a series of sample
surveys. At present, the CSO collects data on the noncommercial farm sector for the purpose of
providing both annual crop forecasts as well as more detailed statistics. The crop forecasts focus on
the number of farms growing the major crops, the area planted, and expected production. These
estimates are published midway through the crop year as part of the early warning process. In
contrast, the annual surveys are designed to provide more detailed information on land use, crop
production, livestock numbers, and prices received for agricultural produce. Estimates are published
Additional remarks on estimating procedures may be found in the section titled "Setting and Publishing Agricultural
Estimates" (see p. 245).