Behavioural Surveillance Surveys - The Wisdom of Whores

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