1. Presentation by: Dr. Bhushan Kamble
Moderator: Dr. Poornima Tiwari
Professor,
Department of Community Medicine,
VMMC & SJH
Sampling and sampling variability
2. Outline of presentation
• Definitions
• Need for sampling
• Types of sampling design
Probability sampling
Non probability sampling
• Factors affecting choice of sampling design
• Sample size
Factors affecting sample size
Calculation of sample size for
• Descriptive studies
• Comparison studies
• Sampling variability
• Sampling errors
• References
3. Definitions
Population: The target group to which the findings (of a study)
would ultimately apply is called population1
Or
Population is the term statisticians use to describe a large set or
collection of items that have something in common2.
Sample: is that part of the target population which is actually
enquired upon or investigated1.
Or
Sample is a subset of population, selected in such a way that it
is representative of the larger population2
1. Indrayan A., Satyanarayana L., Medical Biostatistics, third edition, 2009
2. Last JM. Dictionary of Epidemiology, 3rd
edition, 2000.
4. Definitions (cont.)
Sampling: is the process of selecting a small number of
elements from a larger defined target group of elements such
that the information gathered from the small group will allow
judgments to be made about the larger groups.
conclusions based on the sample results may be attributed only
to the population sampled*
.
.
*Dawson B., Trapp RG, Basic and Clinical Biostatistics, second edition, 1994
5. Definitions cont..
Sampling unit: is the unit of selection
Unit of study or element: is the subject on which information is
obtained.
Sampling frame: list of all sampling units in the target population
is called a sampling frame.
Sample size: the number of units or subjects sampled for inclusion
in the study is called sample size.
Sampling technique: Method of selecting sampling units from
sampling frame
6. Population Vs. Sample
Sample
Population Sample
Parameter Statistic
We measure the sample using statistics in order to draw
inferences about the population and its parameters.
Population of Interest
7. Target population
Sampling frame
Sample
Population you want to
generalize results to
Population you have access
to for your study
Study population
How can you get access to
study population?
Study actually done on?
1.
2.
3…..
8. Need for sampling
1. Complete enumeration may not be possible.
2. Resources: Lower cost, Lesser demand on personnel.
3. Speed: Faster results due to lesser coverage.
4. Reliable information: Due to small size - better trained
personnel, more accurate methods, better supervision.
To draw conclusions about population from sample, there are
two major requirements for a sample.
Firstly, the sample size should be large.
Secondly, the sample has to be selected appropriately so that
it is representative of the population. Sample should have all
the characteristics of the population.
9. Disadvantages of sampling
1. Sampling entails an argument from the fraction to the
whole. Validity depends on representativeness of the
sample.
2. Fails to provide precise information in case of small
segments containing few individuals.
3. Not necessary in studies where complete enumeration is
needed.
4. May cause a feeling of discrimination among the subjects
who are not included in the study.
10. Types of sampling
Probability sampling Non probability sampling
Probability of selection of
each individual is known
and pre determined
Simple random sampling
Systematic random
sampling
Stratified random
sampling
Cluster random sampling
Multistage random
sampling
Probability of selection of
each individual is not
known
Quota sampling
Purposive/ Judgmental
sampling
Snowball/ Network
sampling
Convenience/ Grab
sampling (man in the
street)
11. Simple random sampling
Equal probability of selection of units for inclusion in the study
Requires a list of all sampling units (sampling frame)
Each individual is chosen randomly.
Methods:
Lottery method (possible for finite population)
Random number tables
Software that generate random numbers
15. Simple random sampling (contd.)
Simple random method
With replacement
Without replacement
Advantage
Very scientific method
Equal chance of all subjects for selection
Disadvantage
Requires sampling frame
Example:
Blood sampling – TLC, Hb estimation
16. Stratified random sampling
Preferred method when the population is heterogeneous with
respect to characteristic under study.
Population is divided into groups or strata on the basis of
certain characteristics.
A simple random sample is selected from each strata.
Ensures representation of different strata/ groups in the study
population.
Can be done by selecting individuals from different strata in
certain fixed predetermined proportions.
Proportional stratified sampling
Dis-proportionate stratified sampling
17. Stratified random sampling(contd.)
For example, if we draw a simple random sample from a
population, a sample of 100 may contain
10 to 15 from high socioeconomic group
20 to 25 from middle socioeconomic group
70 to 75 from low socioeconomic group
To get adequately large representation for all the three socio
economic structures, we can stratify on socioeconomic class
and select simple random samples from each of the three
strata.
19. Stratified random sampling(contd.)
Advantage:
All groups, however small are equally represented.
When we want to highlight a specific subgroup within the
population. Ensures presence of the subgroup.
Observe existing relationships between two or more
subgroups.
Can representatively sample even the smallest and most
inaccessible subgroups in the population. To sample the rare
extremes of the given population.
Higher statistical precision compared to simple random
sampling. (d/t lesser variability). So less time and money.
Disadvantage:
Requires a sampling frame for each stratum separately.
Requires accurate information on proportions of each stratum
20. Systematic random sampling
Systematic sampling is a commonly employed technique,
when complete and up to date list of sampling units is
available.
A systematic random sample is obtained by
Selecting the first unit on a random basis
Then others are included on the basis of sampling interval
I = N/n.
21. For example, if there are 100 patients (N) in a hospital and to select a
sample of 20 patients (n) by systematic random sampling procedure,
Step 1: write the names of 100 patients in alphabetical order or their
roll numbers one below the other.
Step 2: sampling fraction: divide N by n to get the sampling fraction
(k).In the example k=100/20 = 5.
Step 3: randomly select any number between 1 to k i.e. between 1 to
5. Suppose the number we select is 4.
Step 4: patient number 4 is selected in the sample.
Step 5:Thereafter every 4+k th patient is selected in the sample until
we reach the last one.
Systematic random sampling(contd.)
23. Advantage:
easy to draw, simplicity.
assurance that the population will be evenly sampled.
Disadvantage:
Requires sampling frame.
Eg. Random blinded rechecking of slides under RNTCP. Slides
are drawn from the register by systematic random sampling.
Systematic random sampling(contd.)
24. Cluster sampling
The population is divided into subgroups (clusters) like
families. A simple random sample is taken of the subgroups
and then all members of the cluster selected are surveyed.
Cluster sampling is used when the population is
heterogeneous.
Clusters are formed by grouping units on the basis of their
geographical locations.
Cluster sampling is a very useful method for the field
epidemiological research and for health administrators.
26. Types:
One stage – when all units in the selected cluster are selected.
Two stage – only some units from a selected cluster are taken
using simple random or systematic random sampling.
Advantages
Simple as complete list of sampling units within population not
required
Low cost
Can estimate characteristics of both cluster and population
Less travel/resources required
Disadvantages
Potential problem is that cluster members are more likely to be
alike, than those in another cluster (homogenous).
Each stage in cluster sampling introduces sampling error—
the more stages there are, the more error there tends to be
Usually less expensive than SRS but not as accurate
Cluster sampling (contd.)
27. A special form of cluster sampling called the “30 X 7 cluster
sampling”, has been recommended by the WHO for field
studies in assessing vaccination coverage.
In this a list of all villages (clusters) for a given geographical
area is made.
30 clusters are selected using Probability Proportional to Size
(PPS).
From each of the selected clusters, 7 subjects are randomly
chosen.
Thus a total sample of 30 x 7 = 210 subjects is chosen.
The advantage of cluster sampling is that sampling frame is not
required
Cluster sampling (contd.)
28. Steps:
List of all clusters (villages and sectors/wards) is made.
Population of each cluster is written against them.
Cumulative population is then written in serial order.
Sampling interval is calculated = Total cumulative population/30
Choose a random number between 1 and the SI. This is the Random
Start (RS). The first cluster to be sampled contains this cumulative
population
Calculate the following series: RS; RS + SI; RS + 2SI; …. RS+(d-
1)*SI.
The clusters selected are those for which the cumulative population
contains one of the serial numbers.
Probability proportional to size (PPS)
29.
30. Multistage random sampling
Multistage sampling refers to sampling plans where the sampling is
carried out in stages
using smaller and smaller sampling units at each stage.
Not all Secondary Units Sampled normally used to overcome
problems associated with a geographically dispersed population
31. Multistage random sampling
In this method, the whole population is divided in first stage
sampling units from which a random sample is selected.
The selected first stage is then subdivided into second stage units
from which another sample is selected.
Third and fourth stage sampling is done in the same manner if
necessary.
Example:
NFHS data is collected by multistage sampling.
Rural areas – 2 stage sampling – Villages from list by PPS,
Households from village
Urban areas – Wards (PPS) – CEB (PPS) – 30 households
from each CEB
33. Non probability sampling
The probability of each case being selected from the total
population is not known
Units of the sample are chosen on the basis of personal judgment
or convenience
There are NO statistical techniques for measuring random
sampling error in a non-probability sample. Therefore,
generalizability is never statistically appropriate
34. • Involves non random methods in selection of sample
• All have not equal chance of being selected
• Selection depend upon situation
• Considerably less expensive
• Convenient
Non probability sampling
35. Types of Non probability sampling
Convenience/Grab/Availability
Judgment/Purposive sampling
Quota sampling
Snowball/Network
36. Convenience/Grab/Availability sampling
Subjects selected because it is easy to access them.
No Students in your class, people on Street, friends etc
Advantages:
In pilot studies, convenience sample is usually used to obtain basic
data and trends.
In documenting that a particular quality of a substance or
phenomenon occurs within a given sample.
Disadvantages:
Not representative of the entire population – skewed results.
Limitation in generalization and inference making about the entire
population – low external validity.
37. Snowball/Network sampling
If the sample for the study is very rare or is limited to a very
small subgroup of the population.
Works like a chain referral.
Initial subject helps identify people with a similar trait.
Advantages:
To reach rare and difficult to access populations.
Cheap, cost – efficient.
Lesser workforce, lesser planning.
Disadvantages:
Little control over sampling technique.
Representativeness is not guaranteed.
Sampling bias d/t people referring known people who are
more likely to be similar.
38.
39. Purposive or judgmental sampling
The specialty of an authority can select a more representative
sample. Knowledge of research question required.
Subjects selected for a good reason tied to purposes of research.
Advantages:
Hard-to-get populations that cannot be found through
screening general population.
Usually used when a limited number of individuals possess
the trait of interest.
Disadvantages:
No way to evaluate the reliability of the expert or the
authority.
Biased since no randomization was used in obtaining the
sample. So results cannot be generalised.
40. Quota sampling
• The population is divided into cells on the basis of relevant
control characteristics.
• A quota of sample units is established for each cell.
• A convenience sample is drawn for each cell until the quota is
met.
• Pre-plan number of subjects in specified categories(e.g. 100
men, 100 women).
• In uncontrolled quota sampling, the subjects chosen for those
categories are a convenience sample.
• In controlled quota sampling, restrictions are imposed to limit
interviewer’s choice.
41. •To sample a subgroup that is of great interest to the study.
•To observe relationships between subgroups.
•Example – an interviewer may be told to sample 50 males and
50 females.
Advantages:
•Used when research budget limited
•Introduces some elements of stratification
Disadvantages:
•Variability and bias can not be controlled or measured
•Time consuming
42. Factors affecting choice of sampling
designs
Heterogeneity: need larger sample to study more diverse
population
Desired precision: need larger sample to get smaller error
Nature of analysis: complex multivariate statistics need
larger samples
Accuracy of sample depends upon sample size, not ratio of
sample to population
44. Factors affecting sample size
1. Study design: descriptive or
comparison study
2. Sampling design: smaller if
stratified, larger if cluster
3. Type and number of
variables being studied.
4. Maximum tolerable
probability of type I error.
5. Required power for a
specified clinically important
difference.
6. Specification of the
magnitude of difference that
would be considered
significant.
7. The extent of variability
among measurements( S.D.)
8. Whether underlying
distribution is normal or
skewed
9. Heterogeneity of population:
need larger sample to study
more diverse population
10. Desired precision: need
larger sample to get smaller
error
11. Nature of analysis: complex
multivariate statistics need
larger samples
12. Resources and time at hand
46. SAMPLE SIZE FOR QUALITATIVE OUTCOME
VARIABLE
n=4𝑃𝑃/𝑃2
n= sample size
P= estimated prevalence
Q= 1-P
L= allowable error
A survey is to estimate prevalence of influenza virus infection in school kids.
Suppose the available evidence suggests that approximately 20% (P=20) of the
children will have antibodies to the virus. Assume the investigator wants to
estimate the prevalence within 6% of the true value (6% is called allowable error;
L)
The required sample size is :
n = (4 x 20 x 80) / (6 x 6) = 177.78
Thus approximately 180 kids would be needed for the survey
47. Sample size for estimation of mean
n= z2a/2s2
l2
Where, n= sample size
s= standard deviation
l= absolute precision
z= relative deviate
a= alpha error
Za/2 = 1.96 for a= 0.05
n = 4 s2
l2
48. Example
Suppose that it was required to estimate diastolic blood pressure in a
population to within ±2mmHg (using a 95% confidence interval)
and the standard deviation of diastolic blood pressure was known
to be 15mmHg.
S= 15 l= 2
n = 4 s2
l2
N=4×(225/4)=216.09
The next highest integer is taken, giving a requirement of 217 subjects
49. Sample size for estimation of proportion
n= z2a/2p(1-p)
l2
Where, n= sample size
p= anticipated value of proportion in population
l= absolute precision
z= relative deviate
a= alpha error
Za/2 = 1.96 for a= 0.05
n= 4 p(1-p)
l2
50. Example
Suppose it is thought that there are about 28% smokers in the
population and it is required to estimate the percentage of
smokers to within ±3% (in absolute terms), using a 95%
confidence interval.
p= 0.28 l= 0.03
n= 4 p(1-p) n= 4 ×0.28(1-0.28)
l2
(0.03)2
n= 860.5
so that a survey of 861 persons is required,
51. Sample size for estimation of rate
n= 4 r2
l2
where: r = estimated rate in the population
l = absolute precision
Suppose that a rate is expected to be around 25 per million (per
year) and it is required to estimate it with a 95% confidence interval
to within ± 5 per million. The number of cases required to achieve
this level of precision is
n= 4 (25)2
(5)2
n=96.04
which means that 97 cases would have to be observed
52. Sample size for estimation of difference
between two population means
n= z2a/2 (s1
2
+ s2
2
)
l2
Where, n= sample size
s= standard deviation ( subscript 1,2 refer to two
populations)
l= absolute precision
z= relative deviate
a= alpha error
Za/2 = 1.96 for a= 0.05
n= 4 (s1
2
+ s2
2
)
53. Sample size for estimation of difference
between two population proportion
n= z2a/2[ p1(1-p1) + p2 (1-p2) ]
l2
Where, n= sample size
p= anticipated value of proportion in population
( subscript 1,2 refer to two populations)
l= absolute precision
z= relative deviate
a= alpha error
Za/2 = 1.96 for a= 0.05
n= 4 [ p1(1-p1) + p2 (1-p2) ]
l2
54. Sampling variability refers to the different values which a given
function of the data takes when it is computed for two or more
samples drawn from the same population.
Factors affecting sampling variability:
1.Inherent variation in the population
2. Sample size
3.Sampling distribution of the mean
4.Sampling error and bias.
Sampling variability
55. Eg. Population of 7000 children and their birth weight. The mean and
standard deviation for this distribution are 3.36 and 0.56 respectively.
N Sample 1 Sample 2 Sample 3 Sample 4 Sample 5
1 3.09 4.28 4.09 2.34 4.29
2 3.74 2.82 2.96 3.06 2.87
3 2.56 3.80 3.09 3.35 3.43
4 3.63 1.89 3.14 3.30 3.40
5 2.96 4.04 3.14 4.36 3.58
6 2.76 2.39 4.38 3.99 3.96
7 3.98 3.41 3.87 4.62 3.18
8 3.76 3.95 4.34 3.18 3.07
9 2.66 5.83 3.81 2.80 2.70
10 3.16 3.30 4.16 3.14 3.21
N 10 10 10 10 10
Mean 3.23 3.57 3.70 3.41 3.37
SD 0.51 1.10 0.56 0.71 0.48
Minimum 2.56 1.89 2.96 2.34 2.70
maximum 3.98 5.83 4.38 4.62 4.28
56. Irrespective of sample size , the sample means are expected to
fluctuate evenly about the true population mean.
The variation in sample means exhibited in the table is an example
of sampling variation due to chance.
If we take 50 observations ,mean is 3.46 kg. sampling error 3.46-
3.36= 0.10
The means vary less(by chance) if the sample size is large; that is
sampling error is smaller,the larger is the sample.
57. The distribution more closely clustered around a middle value as the
sample size increases.
The mean do not systematically increase or decrease with increasing
sampling and have more variability(larger SD) when the sample size
is small.
The standard deviation of the means steadily decrease as sample size
increases, more quickly when the sample size is small.
59. A sampling experiment(based on the distribution of birth weights): what happens
to mean and variability of a sample mean when we keep doubling the sample size
N Mean of population values=3.36
Mean of sample means(kg)
SD of population values=0.56
SD of sample means
(observed SE OF Mean;kg)
2 3.50 0.40
4 3.51 0.28
8 3.46 0.19
16 3.45 0.11
32 3.44 0.080
64 3.46 0.06
60. Sampling error
• Types of sampling error: 1. sample error
• 2. non sample error
SAMPLE ERROR: is incurred when the statistical characteristics of a
population are estimated from a subset, or sample, of that population.
For example, if one measures the height of a thousand individuals
from a country of one million, the average height of the thousand
is typically not the same as the average height of all one million
people in the country.
61. Sample error (random error)
• Error caused by the act of taking a sample
• They cause sample results to be different from the results of
census
• Size of error can be measured in probability samples
• Expressed as “standard error”
• of mean, proportion…
• We have no control over
• Sample error depends upon:
• Size of the sample (larger size lesser error)
• Distribution of character of interest in population
62. Non sample error
Non response error: A non-response error occurs when
units selected as part of the sampling procedure do not
respond in whole or in part
Response error: A response or data error is any systematic
bias that occurs during data collection, analysis or
interpretation
• Respondent error (e.g., lying, forgetting, etc.)
• Interviewer bias
• Recording errors
• Poorly designed questionnaires
63. References
1. Indrayan A., Satyanarayana L., Medical Biostatistics, third edition,
2009
2. Last JM. Dictionary of Epidemiology, 3rd
edition, 2000.
3. Dawson B.,Trapp RG, Basic and Clinical Biostatistics, second
edition, 1994
4. Daly LE, Bourke GJ, Interpretation and uses of medical statistics,
fifth edition, 2003
5. Detels R., Beaglehole R., OxfordTextbook of public health, fifth
edition,2011.