Transcript Overview of Sampling Methods II

Overview of
Sampling Methods II
(Session 04)
SADC Course in Statistics
Learning Objectives
By the end of this session, you will be able to
• describe accessibility sampling, quota
samples, purposive sampling
• explain what is meant by a systematic
sample, cluster sample, a multistage
sample
• take a sample according to one of the
above sampling schemes
• explain the difference between probability
and non-probability samples.
To put your footer here go to View > Header and Footer
2
Pre-statistical sampling
• Accessibility sampling – sample only the
most convenient sampling units –
sometimes called convenience sampling
(not advised)
• Purposive sampling – sampling a given
number of ‘typical’ or ‘representative’
sampling units
• Quota sampling - a particular form of
purposing sampling where choice of actual
sample is left to the enumerator’s discretion
– enumerator asked to fill a pre-specified quota (a
fixed sample size for each sample segment )
To put your footer here go to View > Header and Footer
3
Difficulties with above schemes
• Accessibility samples will usually be highly
biased – not an advisable approach
• Purposive sampling often done at initial stages
of sampling to ensure good coverage – with
good reason sometimes – more on this later
• Quota sampling (often done in opinion polls,
market surveys, etc) has the advantage of
being cheap and quick and not requiring the
existence of a sampling frame. However, it can
lead to an very biased sample if interviewer
convenience has a big effect (often NOT the
case for telephone polling)
To put your footer here go to View > Header and Footer
4
Probability sampling
• These are samples where every individual
in the population has a known non-zero
probability of entering the sample.
• Such schemes allow the sampling error to
be quantified and the chance of bias
reduced.
• Simple and stratified random sampling
discussed in the previous session are
examples of probability based sampling
procedures – others outlined below.
• In practice, partial deviations from probability sampling occur with good reason.
To put your footer here go to View > Header and Footer
5
Systematic sampling
This method requires a well-established
sampling frame, i.e. list of all population
members.
The procedure involves selecting one element
at random from the first k elements in the
list, then selecting every kth unit thereafter,
progressing through the list in a systematic
way.
This leads to approximately (1/k)*100% of
the population entering the sample
To put your footer here go to View > Header and Footer
6
Remarks about systematic sampling
• The process is simple, and is useful where
a list of units already exists, e.g. telephone
directory, list of customers in a bank
• It can also be useful in studies requiring a
good geographical spread, by imposing a
grid on a map of the region.
• It assumes that the original list from which
the sample is drawn is itself organised in a
“random” manner which is independent of
the key variables of interest in the study.
To put your footer here go to View > Header and Footer
7
Limitations of systematic sampling
• The assumption that the original list is
random may not be true.
• The theory is less well developed. Hence
analysis of the data relies on assuming
that the sample is like a simple random
sample.
• Requires the availability of a good
sampling frame and knowledge of the
size of the target population.
To put your footer here go to View > Header and Footer
8
Cluster sampling
• Taking a simple random sample can be
administratively difficult.
• More convenient to divide the population
into non-overlapping groups (clusters)
• Then sample a few clusters at random
• Then enumerate all members in the chosen
clusters
This process is referred to as cluster sampling.
More discussion on this will follow in sessions
13 and 14.
To put your footer here go to View > Header and Footer
9
Cluster sampling – further notes
• In the initial division of the population, aim
to make each cluster as heterogeneous as
possible.
• The sampling frame is required only for the
chosen clusters, so useful when a sampling
frame does not exist for the whole
population
• The division of the population into clusters is
different from that used in identifying strata.
Here, the aim is to have high within- cluster
variation.
To put your footer here go to View > Header and Footer
10
Multi-stage sampling
• Consider again the population divided into a
number of clusters.
• But now, instead of including all units in the
cluster, take a random sample of units
within each cluster.
• Above would be called a two-stage
sampling design
• This may be extended to more than twostages
– e.g. may select districts, then enumerations
areas within districts, then household within
enumeration areas, to give a 3-stage design.
To put your footer here go to View > Header and Footer
11
Multi-stage sampling
• Most large-scale surveys are conducted
using a multi-stage sampling procedure.
• Can be used in combination with
stratification, e.g.
– first divide population into strata
– continue the sampling within each
stratum according to a multi-stage
sampling procedure
• There will be more discussion concerning
multi-stage sampling procedures in sessions
13 and 14.
To put your footer here go to View > Header and Footer
12
References
• Moser, C.A. and Kalton, G. (1971) Survey
Methods in Social Investigations. Gower
Publishing Company Limited.
• Scheaffer, R.L., Mendenhall, W., Ott, L.
(1990) Elementary survey sampling, (4th
Edition). PWS-Kent Publishing Company,
pp. 390.
• Woodward, M. and Francis, L.M.A. (1988)
Statistics for Health Management and
Research (see Chapter 10 for an
overview). Edward Arnold, London. ISBN
0-340-42009-X
To put your footer here go to View > Header and Footer
13
Some practical work follows …
To put your footer here go to View > Header and Footer
14