Population And Sampling

Prof S Balkaran
1
2020
2 Introduction

 We collect data from objects of our  Population and sample

enquiry to solve problems  Definition
 Research design is critical element  Population is the study object consists of
 individuals , groups, organisations and
The plan according to which we obtain
events
research participants/subjects
 Encompasses the total collection of all
 Collect information from them
Units of Analysis
 Need to specify  Population is the full set of cases from
 Number of groups which the sample is taken
 How are they drawn – randomly??what  Population need not necessarily need be
should be done in experimental research people
 Indicate the size of population by N
3  UNITS OF ANALYSIS  REPRESENTATIVE
 Refer to the elements of the Population as the  In order for results to be generalizable, the
UNITS OF ANALYSIS – refer to humans sample must therefore be representative
/groups/organisations/human products etc.
 Representative - implies the sample has the
 It is impractical and uneconomical to involve exact properties in the exact proportions from
all members of the population in research which it was drawn, but in smaller numbers.
project as its too large??  A representative sample is a miniature image /
 Consequently we rely on data obtained from a likeness of the population.
Sample of the population
 We indicate Size by n
 Population is a group of potential participants
to whom you want to generalise the results of a
study
4 SAMPLING
 Distinguish between:  PROBABILITY
 Probability  Determine the probability that any
 Non Probability element / member of the population will
be included in the sample
 PROBABILITY
 Simple random sampling
 Stratified random sampling  NON PROBABILITY
 Systematic sampling  Cannot specify this probability
 Cluster sampling  Elements which have a chance of being
 NON PROBABILITY included have a probability that exceeds
zero.
 Accidental
 In some instances some elements have no
 Purposive
chance (zero) of being included
 Quota snowball self-selection
 Convenient
5
 ADVANTAGE OF PROBABILITY  SAMPLING FRAME
SAMPLING
 Researchers need clarity about
 It enables us to indicate the probability
with which sample results deviate in population / unit of analysis to
differing degrees from the corresponding apply hypothesis
population values
 Involves a Sampling Frame
 Enables us to estimate sampling error
 Complete list in which each unit
 Relates to unrepresentativeness of a
of analysis is mentioned only
sample
once
 NON-PROBABILITY SAMPLING
 Sample should be representative
 Frequently used for reasons of of the sampling frame – ideally
convenience and economy is the same as the population
BUT often differs due to
practical problems relating to
the availability of information
  RANDOM SAMPLING  Each sub-population is a Stratum
6
 it is the most attractive type of  The division into groups may be based
Probability Sampling on a single variable such as gender
 Distinguish between (which has two strata:
 Simple Random Sampling  Men
 Each member of the population has the  Women
same chance of being included in the
sample and each sample of a particular
 It may involve a combination of more
size has the same probability of being than one variable:
chosen [SEE EXAMPLE ON PAGE 59]
 Gender
 Stratified Random Sampling
 Age
 If population is composed of various
clearly recognisable, non-overlapping [SEE EXAMPLE ON PAGE 61]
subpopulations –[which we call STRATA
(singular is Stratum) that differ from
each other in terms of variables
7  ADVANTAGES OF  OTHER PROBABILITY
STRATIFIED RANDOM SAMPLING
SAMPLING  Systematic Sampling
 2 Important Advantages  We include every N/nth element (where
is an integer , that is, a whole number) –
 [see 4.3.2.4 – page 62]
[See Figure 4.9 – Page 64]
 With a Stratified Random Sample  Cluster Sampling
we are ensured of a sample's
 Used in Large Scale Sampling difficult to
representativeness, irrespective to
obtain to obtain lists of all
the sampling of ample size because
members/Units of Analysis
it has been built into the sampling
strategy from the very beginning.  Therefore we first draw pre-existing,
heterogeneous groups, called
 [see examples A to G on Pages 62- CLUSTERS [See example on page
63] 65/66]
 Because we do Cluster Sampling in
phases, we do not have to list all the
members of the population
8  Cluster sampling – fewer locations  Table of various factors on choice of
have to be visited probability sampling techniques [ see
Table 4, Page 67]
 Advantage - considerable saving in
time and costs compared to Simple
Random Sampling  NON PROBABILITY
 However when some Cluster  The probability that any element /unit of
analysis) will be included in a non-
Sampling are homogeneous probability sample CANNOT be
(similar) in terms of variables of specified.
interest, Cluster Sampling may lead  In some instances – certain members
to Biased Samples – samples that may have NO chance at all of being
tend to lead towards a particular included in such a sample.
factor of the research topic.
9  Accidental Sampling (Incidental  Purposive Sampling
Sampling)  Claimed most important type of Non
Probability Sampling
 Most convenient collection of members
of the population that are near and  Researchers rely on experience and
readily available for research purposes previous research findings to deliberately
[See Figure 4.11 – Page 68] obtain units of analysis as being
representative of the population
 Problem – different researchers may
 Quota Sampling proceed in different ways to obtain such a
sample
 Effort made to have same proportions of
units of analysis in important strata –  Snowball Sampling
e.g. gender, age etc. in the population  1st phase - approach few individuals from
 Segmentation population
 These become “informants” and identify
other members
 The latter may in turn identify a further set
of relevant individuals like a rolling
snowball – grows in size
  Self Selection Sampling  Impact of various factors on choice of
10 Non-probability Sample Techniques [see
 Occurs when we allow individuals to
Table 4.2 – Page 72]
identify their desire to take part in the
research
 SAMPLE SIZE
 Convenience Sampling (Haphazard  The larger the sample size, the lower the
likely error in generalising to the population
Sampling)
 Sample Size is governed
 Involves selecting haphazardly those
cases that are easiest to obtain  Confidence we need to have in our data -
level of certainty
 Continues until we reach the required
 Margin of error which can be tolerated -
sample rate accuracy as required
 Technique used widely  Types of analyses - categories subdivide
data
 PRONE - to bias and influence beyond
control of researcher due to fact that  Size of total population from which sample
is drawn
cases appear in the sample because they
were easier to obtain.  TO BE CONTINUED…………….
11 Sample Size
 The larger the sample size - the lower  When we determine Sample Size:
the likely error in generalising to the  We determine the size of the sample (n) –
Population bear in mind the size of the population
 Choice of Sample Size is governed by:  The smaller the total population, the
relatively larger the sample size should
 Confidence we need to have in our be to ensure satisfactory results
data - level of certainty
 The influence of absolute sample size (n)
 Margin of error – that which we can instead of proportion () to the size of the
tolerate in terms of accuracy we population (N) is related to the standard
require error of the mean
 Types of analyses to be undertaken:  An increase in sample size, in proportion
 Number of categories and subdivision
to the size of the population results in a
decrease in standard error.
of data
 Size of total population from which
sample is taken
12 Sample Size / Response Rate
 RESPONSE RATE
 The desired sample does not depend on  A perfectly representative sample is
size of the population only but on the one that exactly represents the
variance (heterogeneity) population from which it is taken.
 Each a stratum of a highly  NON-RESPONSES
HETEROGENEOUS population is
 Are likely to occur
relatively HOMOGENOUS, a relatively
smaller stratified sample than that  Refuse to be involved (various
required for a random sample may be reasons)
sufficient  REASONS
 The number of units of analysis from  Refusal to respond
which we eventually obtain usable data  Ineligibility to respond
may be much smaller than the number  Inability to locate respondent
that we drew originally  Respondent located but unable to
make contact
13 Response Rate
 TOTAL RESPONSE RATE  SAMPLING ERRORS
 Obtained by dividing the total number of  It is impossible to select a perfect sample
respondents by the total number of
 Sampling Error expresses the lack of fit
ineligible respondents in the sample
between the sample and the population .
 Active Response Rate – excludes
 Defined as “the difference between the
ineligible respondents as well as those
characteristics of the sample and the
who despite repeated attempts were
characteristics of the population from
unreachable.
which the sample is selected.
 Calculated by dividing total number of
responses by the sample size minus the  VARIABILITY
number of respondents who are ineligible  Amount of variability or the spread of
or unreachable. theses average values gives us some idea
of the amount of sampling error [see
page74]