Participant-Based Survey Sampling Guide for Feed the Future Annual Monitoring Indicators 52
Note that Feed the Future requires that IPs produce estimates for the indicators above according to
specified disaggregates. The “Number of Hectares under Improved Management Practices” indicator
requires first-level disaggregated estimates by type of hectare (i.e., cropland, cultivated pasture,
aquaculture, rangeland, conservation area, freshwater or marine ecosystem, or other). The “Number of
Individuals Using Improved Management Practices” indicator requires a first-level disaggregated
estimate by value chain actor type (i.e., smallholder producer, non-smallholder producer, people in
government, people in private sector firms, people in civil society, and others). The “Value of Sales”
indicator requires a first-level disaggregated estimate by type of product or service. There are also
varying requirements for second-level disaggregated estimates (e.g., sex and age) that are nested within
first-level disaggregated estimates for all three indicators above. The “Value of Sales” indicator has an
additional requirement for third-level disaggregated estimates nested within the first two levels. The
“Yield of Agricultural Commodities” indicator requires reporting by commodity and requires second-
level disaggregated estimates by farm size (i.e., smallholder, non-smallholder), as well as third-level
disaggregated estimates by sex and age (i.e., 15–29 and 30+).
However, survey implementers should not produce estimates for required disaggregates based on
separate sample size calculations for those disaggregates. While it would be ideal to ensure precision for
the indicator estimate for a disaggregate at the same level as that for the overall estimate for the
indicator, this would entail a separate sample size calculation at the disaggregate level. Such a
calculation would mean taking into account the input parameters specific to disaggregates (e.g., the
total number of participants for the disaggregates and the target value of the indicator at the level of
the disaggregate population of participants).
More importantly, a sample size calculation for disaggregates separately could substantially increase the
overall sample size for the survey, making estimates at such levels of precision costly. For instance, in
the example above, for the “Number of Individuals Using Improved Management Practices” indicator,
assume there is only one first-level disaggregate by value chain actor type, say, smallholder producer.
Then, assuming the same input parameters as those used at the overall level, at the second level of
disaggregation, a separate sample size calculation for males and females would result in a requirement
of 809 sampled participants for each category of males and females, for a total sample size of 1,618
across both sexes. Increasing the overall sample size twofold (in this example) would drive up the cost of
the survey substantially. Therefore, survey implementers should not produce estimates of indicators by
their required disaggregates based on separate sample size calculations. Instead, they should compute
the estimates of disaggregated indicators based on the portion of the overall sample size that happens
to fall into the category of disaggregate, and accept the loss in precision. For instance, In the above
example, if we assume that the overall sample size for the survey is 809 of whom 500 respondents are
males and the remainder are females, then the disaggregated estimates by male and female should be
based on the reduced sample sizes of 500 and 309, respectively, even though there will be a loss in
precision for these estimates. However, IPs may decide that it is worth the additional investment of
resources to collect more precise sex-disaggregated data at some points in the project.
Finally, in light of the discussion above, Feed the Future recommends that IPs adopt a minimum overall
sample size for the survey of 525 participants. That is to say
should be 525 or more after taking
into account the three adjustments to
and assuming an anticipated response rate of 95%. If the
actual response rate encountered in the field is 95%, there will be completed interviews for 500 sampled