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