An Economic Assessment of Banana Genetic Improvement and ...

DETAILS OF SAMPLE SURVEY DESIGN 179
Table D.2 Number of survey sites (primary sampling units), by elevation and diffusion
status, Uganda and Tanzania
Stratum
Low elevation
High elevation
Country
Not exposed Exposed Not exposed Exposed
Total
Uganda 14 8 4 1 27
Tanzania 1 9 1 2 13
Total 15 17 5 3 40
Selection of Secondary
Sampling Units
The SSU both in Uganda and Tanzania is a
village. In Uganda, in each LC3, there are
several parishes (LC2s), and each parish
consists of several villages (LC1s). One
SSU was selected per PSU. The probability
of selection (or sampling fraction) of a SSU
varies by PSU and is equal to 1/M p , where
M p represents the number of villages in the
pth PSU (p = 1, . . . , 40 in the sample). For
most exposed LC3s in Uganda, there is only
one exposed LC1 per PSU. Where there is
more than one exposed village per PSU, the
SSU was drawn with a random number
from the list of those villages with more
than 100 households according to the 1991
census. Whether a community selected in
the sample had been properly classified as
exposed or nonexposed was then verified at
the site. Two communities were replaced
because they were fishing villages with little
banana production.
Selection of Households
within a Community
A constant sample size was maintained for
two reasons. First, as mentioned above, there
was no information on variation in underlying
population parameters to guide optimal
sample allocation. The second reason was to
maintain comparable workloads among
enumerators, who resided in the communities
for 1 year to complete the survey. The
sample villages were visited, and a current
list of households in each village was requested
from the chairman of the LC1. The
total number of households selected per village
was 20. The probability of selection (or
a sampling fraction) of a household varies
by village and is equal to 20/H s , where H s
is the number of households in village s
(s = 1, . . . , 40 in the sample). If there was
an order in the list of households, random
numbers were used for selection. Otherwise,
a random start with systematic random
sampling from the list was employed.
Unit of Observation
The basic unit of observation for the sample
survey is the farm household. A farm household
is defined according to the culture of
which the household is a part, and includes
female-headed and child-headed (orphaned)
households, as well as male-headed households
with more than one wife. The total
sample size surveyed comprises 800 households
(540 in Uganda and 260 in Tanzania).
In some analyses conducted in this report,
the plot or the banana cultivar is the unit of
observation.
Weights
The overall probability of selecting a household
in the subsample (denoted as PSH) is a
unique number, defined as the product of
the sampling fractions at each level: PSH =