Behavioural Surveillance Surveys - The Wisdom of Whores

Selecting primary sampling
units (clusters) for “fixed”
populations
When conventional PSUs are used for “fixed”
populations, the procedure recommended for
choosing first stage sampling of PSUs (clusters)
will depend upon whether information on the
size of PSUs (i.e. the number of sub-population
members associated with each site or PSU)
is available prior to sample selection. Below,
two scenarios are considered - the first where
the number of sub-population members
associated with each PSU is known or can be
estimated at reasonable cost, and the second
where such information cannot be obtained
at an acceptable cost.
when measures of PSU size are available
Statistically, the most efficient two-stage
sample design is one in which PSUs are
selected with probability-proportional-to-size
at the first stage of sample selection and a
fixed number of sub-population members is
chosen from each PSU at the second-stage.
The term probability-proportional-tosize
(PPS) simply means that the chance of
a PSU being selected depends upon its size.
The larger the PSU, the higher the likelihood
that it will get selected into the sample. This
compensates for the fact that an individual in
a large PSUs will be less likely to be selected
from that PSU into the final sample than an
individual in a small PSU (because there
will be more individuals competing to be
selected.) Using PPS, a brothel that employed
100 women would be twice as likely to be
selected as a primary sampling unit as a brothel
that employed 50 women. But as long as the
number of women selected from each brothel
was the same at the second stage of sampling
every individual woman in both brothels
would have the same probability of being
selected into the final respondent sample.
To use PPS selection procedures, the
number of individuals associated with each
PSU must be known in advance. This number,
known as a measure of size, does not need
to be an exact count. A rough approximation
for each site is good enough.
As the number of PSUs listed in a sampling
frame is often large, the recommended
procedure for choosing sample clusters is
through systematic sampling, where one
PSU or sample cluster is chosen at random
and every ith cluster thereafter is automatically
chosen for the sample, based on a calculated
sampling interval.
If sample PSUs are selected with a
probability weighted according to their size,
as described below, and an equal number of
individuals is chosen per PSU at the second
stage of sample selection, the end result is a
self-weighted sample. This means that
every person in the universe described by the
sampling frame has the same overall
probability of being selected into the final
sample. In addition to being relatively
efficient in terms of sampling precision, this
design eliminates the need to weight the data
during analysis.
The steps involved in selecting a sample
of clusters using systematic sampling with
probability-proportional-to-size are described
in Figure 1, and an example of the selection
procedure is provided in Table 3. Form 1 in
Appendix 4 can be used in the field to assist
with the selection of clusters by probability
proportional to size.
38
C H A PTER 4 B EHAV I OR A L S U R V EI L L A NC E S U R V EY S