Behavioural Surveillance Surveys - The Wisdom of Whores
Behavioural Surveillance Surveys - The Wisdom of Whores
Behavioural Surveillance Surveys - The Wisdom of Whores
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Under the heading <strong>of</strong> non-probability<br />
sampling methods are a variety <strong>of</strong> approaches<br />
that are not based upon the statistical<br />
principles which govern probability samples.<br />
<strong>The</strong>re are various reasons for using<br />
non-probability methods. Some methods<br />
(e.g. snowball or network sampling) are<br />
designed for use when probability sampling<br />
is not feasible. In snowball sampling, key<br />
informants in a sub-population identify other<br />
members <strong>of</strong> their community. <strong>The</strong>se people<br />
are contacted, and they in turn identify<br />
further contacts. <strong>The</strong> process goes on until an<br />
adequate sample is achieved. Other methods<br />
(e.g. purposive sampling) are designed to<br />
provide the maximum amount <strong>of</strong> information<br />
possible for key groups <strong>of</strong> study subjects in<br />
order to develop and/or test social theories.<br />
Yet others (convenience sampling) are<br />
designed to obtain a sample <strong>of</strong> subjects<br />
at the least possible cost. In general,<br />
non-probability sampling methods are not<br />
intended to produce “representative” data<br />
for larger populations, although they are<br />
sometimes (incorrectly) used to try to do so.<br />
Probability sampling has two major<br />
advantages. Firstly, it is less prone to bias<br />
than non-probability methods and secondly,<br />
it permits the application <strong>of</strong> statistical theory to<br />
estimate sampling error from the survey data<br />
themselves. Consistent use <strong>of</strong> probability<br />
sampling methods in the context <strong>of</strong> BSS has<br />
the critical advantage <strong>of</strong> producing data which<br />
are comparable from one survey to the next,<br />
and which can therefore be used to measure<br />
statistically significant changes in risk behavior<br />
over time. <strong>The</strong>refore probability sampling<br />
methods are the preferred choice for BSS<br />
whenever feasible. <strong>The</strong> major disadvantage <strong>of</strong><br />
probability sampling is that a list or sampling<br />
frame is needed, and this can take time and<br />
resources to produce. While there are ways to<br />
make the task <strong>of</strong> developing sampling frames<br />
less costly and time consuming, the use <strong>of</strong><br />
probability sampling methods will nevertheless<br />
involve greater time and expense than<br />
sampling approaches that do not require a list<br />
or sampling frame.<br />
While they are generally cheaper and easier<br />
to use, non-probability sampling methods have<br />
several important drawbacks. <strong>The</strong> first is the<br />
risk <strong>of</strong> sampling bias resulting from the<br />
subjectivity that <strong>of</strong>ten enters into the sample<br />
selection process. Where a list <strong>of</strong> sampling<br />
units is not available from which to select a<br />
sample following fixed rules, there is the<br />
danger that certain types <strong>of</strong> subjects will be<br />
disproportionately included in and others<br />
disproportionately excluded from the sample.<br />
Secondly, there is the issue <strong>of</strong> replicability,<br />
which is <strong>of</strong> key importance for surveys<br />
intended to monitor behavioral trends over<br />
time. Where sample selection criteria are not<br />
defined in operationally precise terms so that<br />
they can be replicated in subsequent survey<br />
rounds, there is a danger that observed<br />
changes will be due to changes in sampling<br />
rather than real changes in behavior. Finally,<br />
non-probability methods provide no statistical<br />
basis for assessing the precision or reliability<br />
<strong>of</strong> survey estimates. In fact, conventional<br />
statistical tests cannot reliably be used with<br />
non-probability samples, although in practice<br />
this limitation is <strong>of</strong>ten overlooked.<br />
B EHAV I OR A L S U R V EI L L A NC E SURV EY S CHAPTER 4<br />
31