Study population and Sampling

Golam Rabbani
University of Dhaka

Study population and Sampling 1 / 25

Sampling in research

External validity involved - The degree conclusions hold for other

persons, places or times
Two approaches:
the sampling model (start with population and select and sample) and
the proxy model (generalize results to others that have a gradient of
similarity)
Can improve validity by good sampling or detailing similarity features

Study population and Sampling 2 / 25

Sampling terminology

Population: for which you want to generalize to

Theoretical population - group wished to be generalized to.
Accessible population - can’t always get a good sample
Sample: the group ‘selected’ to be in the study - need to avoid bias
whether selection or random assigned sample.
Statistical sampling terms:
response a measured value of a variable
statistic a function of sample observations
parameter a unknown value of the population
Probability sampling
Non-probability sampling

Study population and Sampling 3 / 25

Probability sampling methods

a sampling method that gives each unit in the population a known,

non-zero chance of being selected is called a probability sampling method.
In the analysis of the sample, weights appropriate to the probabilities are
utilized to increase validity of the findings. The most commonly used and
most familiar methods are:
Simple Random Sampling
Stratified Random Sampling
Systematic Sampling
Cluster Sampling or Area Sampling
Multistage Sampling

Study population and Sampling 4 / 25

Simple random sampling (SRS)

A simple random sample of size n is a sample of n units selected in

such a way that every possible sample of the given size has the same
chance of being selected. Samples of different sizes may have
different chances of being selected.
The SRS method can be
with replacement (WR)
without replacement (WOR).

Study population and Sampling 5 / 25

Selection procedure

To obtain a simple random sample of size n using a table of random

numbers,
1 Assign every element in the population a number, beginning with 1
and ending with N.
2 Go to the table and “randomly” select a starting point.
3 Record the numbers as they are read from the table; one records as
many digits as there are digits in N (e.g., if N is 79, record two-digit
numbers), but disregard any numbers chosen that exceed the value of
N.
When the sample is selected in this manner, every element in the
population has an equal chance of selection, and every combination of n
elements has an equal chance; i.e., a simple random sample is obtained.

Study population and Sampling 6 / 25

Features of SRS

Good for This sampling will give good sample (better representation)
if the population of interest is homogeneous with respect to the
characteristic of the interest.
Not good for The situations when SRS is not a good option are:
When the units of the population are heterogeneous with respect to
the characteristics of study
As every unit has equal chance of selection, in a large survey the units
selected may be widely scattered posing great operational difficulty and
thus incurring high cost.
Preparation of the frame (list of units for sampling) for a large
population may be impossible or at least not easy, time consuming and
cost-wise prohibitive.

Study population and Sampling 7 / 25

Systematic random sampling

For a 1-in-k systematic sample, you order the units of the population in
some way, and randomly select one of the first k units in the ordered list.
This selected unit is the first unit to be included in the sample. You
continue through the list selecting every k t h unit from then on.
Advantages The advantages of the systematic sampling are:
Sample is easy and fast to select.
Sample is evenly spread over the entire reference population.
Sampling is possible also in situations where frame is not possible to be
constructed for example in case of a hospital out-door clinic. In this
situation we find a suitable sampling interval and select patients/people
accordingly.
Disadvantages If the frame or the sampling units has some hidden
periodicity, which coincides with selections the results will be biased.

Study population and Sampling 8 / 25

Sampling procedure

Suppose we want to select a sample of size 20 from a population of

200.
We first find the sampling interval by dividing N = 200 by n = 20 i.e.
Sampling Interval = Nn = 200
20 = 10.
select a number at random which may be any number between 1 and
k
select a number at random which may be any number between 1 and
10. Suppose 3 is selected. So item number 3 in the frame is our first
unit. The second unit will be 3 + 10 = 13th unit. Thus the series of
20 items that will be selected are: 3, 13, 23 . . . , 193

Study population and Sampling 9 / 25

Stratified random sampling

The population under study is first divided into homogenous groups or

classes called strata. Sampling is carried out in each stratum following
exactly the simple random sampling procedure.
Good for It gives a better representation of population for a
heterogeneous population. As a result, the variation of estimates from
sample to sample i.e., sampling variation is minimized and precession
increased.
Not Necessary for It is not necessary for a homogeneous population
because it encurs more complexity of stratifying than SRS, which was
not needed.

Study population and Sampling 10 / 25

Sampling procedure

in this procedure of sampling, selection is done according to the following

steps
Define the population
Identify the strata
Determine the sub-sample sizes
Select items within each stratum by SRS.

Study population and Sampling 11 / 25

Stratified random sampling

Notations: The following symbols all refer to stratum h.

Nh ⇒ Total number of units
nh ⇒ Number of units in sample
yhi ⇒ Value obtained for the i th unit
Nh
Wh = N ⇒ Stratum weight
nh
fh = Nh ⇒ sampling fraction in the stratum
PNh
i=1 yhi
Y¯h = Nh ⇒ True mean
Pnh
i=1 yhi
y¯h = nh ⇒ Sample mean

Study population and Sampling 12 / 25

Allocation of samples among the strata

The sample sizes in different strata can be allocated in one of the different
principles:
Equal allocation: In this allocation the sample size is allocated as
nh = Nn , i.e. the sample size from each stratum will be equal
irrespective of the size of the stratum in the population
Proportional allocation: In this allocation the sample size is allocated
by satisfying the condition Nnhh = Nn . That is, if the elements of a
stratum comprise (for example) the 15% of the population then the
15% of the selected sample should also come from that stratum.

Study population and Sampling 13 / 25

Allocation of samples among the strata (cont.)

The other allocation principles available are

Neyman allocation: In this allocation the sample size is allocated by
satisfying the conditions
Nh
nh = constant
and Snhh = constant.
Nh Sh
Satisfying both, we need nh = constant
Optimum allocation: In this allocation the sample size is allocated by
satisfying the conditions
Nh
nh = constant
nh Ch = constant
and Snhh = constant
Nh Sh
Satisfying all, we need n h Ch = constant

Study population and Sampling 14 / 25

Cluster sampling

A cluster is group of elements and the sampling in which the sampling

units consist of a group or cluster of smaller units is known as Cluster
Sampling. Clusters in a sampling may be equal or unequal. Natural
clusters are usually unequal in size, for example, Villages, families, farms
etc.
Each cluster must be mutually exclusive and together the clusters must
include the entire population. After clusters are selected, then all units
within the clusters are selected. No units from non-selected clusters are
included in the sample.
In cluster sampling, the clusters are the primary sampling units (PSU’s)
and the units within the clusters are the secondary sampling units (SSU’s).
Often the SSUs are termed as Elements.

Study population and Sampling 15 / 25

Reasons for cluster sampling

There are two main reason for cluster sampling, they are:
1 A frame required for SRS is often not available (especially when the
population is large) but a list or frame of clusters is more easily
available.
2 Travelling and other costs become less in cluster sampling.

Disadvantages of cluster sampling

the main disadvantage of cluster sampling is that the estimators of the
population parameters using cluster sampling are less precise than those
using SRS.

Study population and Sampling 16 / 25

Types of cluster sampling

Numerous methods of sampl selection and estimation have been suggested

for cluster sampling. All these methods belong to the following four main
classes:
1 Selection of equal clusters with or without replacement.
2 Selection of unequal clusters with equal probability without
replacement.
3 Selection of unequal clusters with unequal probability with
replacement.
4 Selection of unequal clusters with unequal probability without
replacement.

Study population and Sampling 17 / 25

Multistage sampling

There are many ways to modify cluster sampling for more complex
sampling situations.
One common modification is to take a sample of secondary units from
within sampled clusters instead of inspecting every secondary unit
within each sampled cluster.
The process of drawing samples from the selected clusters is called
subsampling.
If the secondary units (Stage 2) are the individuals of the study
population, then sampling would stop.

Study population and Sampling 18 / 25

Multistage sampling (Cont.)

Suppose, however, the secondary units are also clusters of individuals

(tertiary units) in the study population. We would then subsample
tertiary units within the secondary units (Stage 3).
This sample of tertiary units from the study population is called a
three-stage cluster sample.
Any sampling design which requires multiple stages of subsampling is
called a multistage design.
Two-stage and three-stage cluster samples are examples of
multi-stage designs.

Study population and Sampling 19 / 25

Features of Multi-stage sampling

Multi-stage designs are used for a variety of practical reasons. One

common motivation is the same as it was for one-stage cluster
sampling- the unavailability of a sampling frame of units.
Other reasons for using a Multi-stage sampling includes lower cost
and administrative convenience.

Study population and Sampling 20 / 25

Drawback of Multi-stage sampling

The main drawback of the multi-stage sampling is that it is

complicated and difficult to draw and to analyse.
Multi-stage sampling is also less efficient than SRS.

Study population and Sampling 21

Measures of variations in Multi-stage sampling

In one-stage cluster sampling, the estimate varies due to one source:

different samples of primary units yield different estimates.
In two-stage cluster sampling, the estimate varies due of two sources:
different samples of primary units and then different samples of
secondary units within primary units.
In general, if there a k stages of subsampling, there will be k sources
of variability.
Thus, variances and variance estimators for Multi-stage sampling with
k-stages will contain the sum of k components of variability.

Study population and Sampling 22