Lecture 7
• Sampling Concepts and Design
• Sample Size Determination
1
Sample or Census (Key terms)
An Element
An object/ person about which or from which the information is
desired
A Population
The total of all elements that share some common set of
characteristics
A Census
A complete count of each element in a population
A Sample
A subset/subgroup of the population
Go for a Sample or a Census?
Time and budget constraints (limited time?)
Attention on individual cases (deep investigation on each
individuals needed?)
Variance in the characteristics (big variance?)
Cost of errors (sampling/ non-sampling errors) (If sampling errors
could lead to serious consequences (e.g., in drug safety testing, a
misjudgment may put lives at risk)) 2
Match the terms and definition/examples
Definition: The total of all elements that share common
characteristics.
Example: All undergraduate students in a university.
Definition: A single object/person from whom information is desired
(the basic unit of a population).
Example: One undergraduate student in a university.
Definition: A complete count of every element in the population.
Example: Testing every batch of a new drug for safety (no element
is omitted).
Definition: A subset of the population selected for research.
Example: Surveying 30 students from a class of 150 to study study
habits.
Terms: element; population; census; sample 3
The Sampling Design Process
Define the population
Determine the Sampling frame
Select Sampling techniques
Determine the sample size
Execution
4
Step 1: Define the Target Population
Population of Interest
Specify the characteristics of individuals or things from whom
the information is needed.
For example: Geography, Demographics or usages.
A sampling unit
5
Step 2: Determine the sampling frame
The actual list of each element of the target population
from which a sample is drawn.
Seldom exists in a complete fashion
E.g. Telephone book, computer system for generating phone
numbers or company directories
Sampling frame error
Should seek to minimize
E.g. by verifying in data collection process or statistical
adjustment
6
Step 3: Select a sampling technique
Two broad types
Probability sampling Nonprobability sampling
Sampling elements Personal judgment by
are selected by researcher to select
chance/ by the sampling
random elements/ nonrandom
7
Probability Sampling
Pros
Be sure of obtaining cross-section of the population
Sampling error can be computed
Survey results can be projected to total population
Cons
More expensive
Time consuming
8
Nonprobability Sampling
Pros
Cheaper at cost e.g. Exploratory research
Data can be quickly obtained
Produced good estimates of population characteristics
Can be reasonably representative if collected in a careful
manner
Cons
Sampling error can’t be computed (i.e. unable to evaluate
precision)
Difficult to project to total population
Representative of the sample to the population
9
Nonprobability Sampling Methods
Nonprobability
Sampling Methods
Convenience Snowball
Judgment Quota
10
Convenience Sampling
Cheaper at cost and speed in collection
Low quality sample
Good in generate ideas, gain insights or develop
hypotheses
Commonly used in Focus Groups, pilot test and pretest
of questionnaire
11
Judgmental sampling
A form of convenience sampling technique
Choosing samples totally based on researchers’ judgments
Application: Test Marketing samples, samples drawn in
business-to-business research, retailing market research
Cheap in cost and convenient
But subjective leads to generalizations to a specific population
may not be possible
12
Quota sampling
An extension of judgmental sampling
E.g. Pre-specified the sample……….
20% male, 80% females
40% users, 60% non-users
May incur selection bias
No assurance of representativeness
May cross-referenced the ratio with accurate secondary data
But with close controls, quota sampling can obtain results close
to probability sampling
13
Snowball sampling
An initial group of respondents is selected (can be randomly
selected)
Then, they are asked to identify other respondents who belong
to the target population of interest (turns out to be nonrandom)
Used to study relatively rare or difficult to find characteristics in
the population
Application: Direct Selling salesforce customer satisfaction or
B-to-B buyer-seller relationship study
14
Scenario Solution Hint
Recruit students at the entrance
Collect data on "students’ milk tea
1 of teaching buildings (easy to
flavor preferences" within 1 hour.
access).
Ask club leaders to recommend
Research "consumption habits of
2 core members (researcher’s
members in high-end campus clubs".
judgment).
Survey 100 students, with a Recruit students in the
3 requirement of 50% male and 50% cafeteria, ensuring the gender
female. quota is met.
Research "students who love niche First find 2 hand account lovers,
4 hand account (stationery journal) then ask them to recommend
culture". peers.
15
Probability Sampling Techniques
Probability Sampling Techniques
Simple Random Systematic Stratified Cluster
Sampling Sampling Sampling Sampling
16
Probability Sampling Techniques
Simple Random Sample
Stratified Random Sample
Systematic Random Sample
Cluster Sampling
Simple random Stratified random
Systematic Cluster / Area
17
Simple Random Sampling
• A simple random sample of n measurements from a
population is a subset of the population selected.
• Every sample of size n from the population has an equal
chance of being selected.
18
Simple Random Sample:
Population Members
Population size of N = 30
Desired sample size of n = 6
19
How to draw a simple random sample?
Procedure:
1. Number all members of the population sequentially.
2. Use a table, calculator, or computer to select random
numbers from the numbers assigned to the population
members.
3. Create the sample by using population members with
numbers corresponding to those randomly selected.
20
Simple Random Sample:
Sample Members
The 6 random numbers are 03, 09, 08, 30, 23, 11
21
Random-number Table
22
Stratified Sampling
• Groups or classes inside a population that share a common
characteristic are called strata (plural of stratum).
• The groups or strata are often sampled in proportion to their
actual percentages of occurrence in the overall population.
Population
Strata Sample
23
How to draw a stratified sample?
Procedure:
1. Divide the entire population into distinct subgroups
called strata.
2. The strata are based on a specific characteristic, such as
age, income, education level, and so on.
3. All members of a stratum share the specific
characteristic.
4. Draw random samples from each stratum.
24
Systematic Sampling
• It is assumed that the elements of the population are
arranged in some natural sequential order.
• Then we select a random starting point and select every kth
element for our sample. (Eg. Start at 2, select every 3rd
element)
Population
Sample (every 3rd person selected) 25
How to draw a systematic sample?
Procedure:
1. Number all members of the population sequentially.
2. Then, from a starting point selected at random,
include every kth member of the population in the
sample.
26
Cluster Sampling
• It is a method used extensively by government agencies
and certain private research organizations.
• Population is divided by demographic area into sections.
• Then we randomly select sections or clusters.
• Every member of the cluster is included in the sample.
Randomly
Selected
Cluster
Population Cluster
27
Population
How to draw a cluster sample?
Procedure:
1. Divide the entire population into pre-existing
segments or clusters. The clusters are often
geographic.
2. Make a random selection of clusters. Include every
member of each selected cluster in the sample.
28
Clusters Example
Town
Company
Home
College
Area of a city
Geographic region
29
30
Sampling Error vs Nonsampling Error
A sampling error is the difference between
measurements from a sample and corresponding
measurements from the respective population. It is
caused by the fact that the sample does not perfectly
represent the population.
How to reduce the sampling error?
A nonsampling error is the result of poor sample
design, sloppy data collection, faulty measuring
instruments, bias in questionnaires, and so on.
31
Sampling or nonsampling error?
The sample size was too small, leading to a survey
result of "students’ monthly living expenses" that
was HK$200 higher than the actual value.
Ambiguous questionnaire questions caused respondents
to mistakenly fill in "weekly consumption" instead of
"monthly consumption".
The sampling frame omitted "new residents", resulting in
a lower-than-actual result in the "community satisfaction
survey".
During random sampling, the proportion of "high-income
groups" in the sample was excessively high.
Interviewers made recording mistakes, leading to
deviations in data entry. 32
So, probability or nonprobability ?
Depends on:
1. Research design
Exploratory research favors non-probability methods
2. Magnitude of sampling and nonsampling errors
Probability methods can compute sampling errors
3. Variability in the population
Probability methods are more appropriate in highly
heterogeneous population
4. Statistical considerations
Probability methods favors the use of statistical methods
5. Operational efficiency considerations
More expensive in Probability methods due to requirement
of statistically trained researchers
33
Scenario Questions
1. Research Design A startup wants to initially understand why
(Exploratory vs. customers dislike their new app (no clear
Conclusive) research hypothesis yet).
2. Magnitude of A drug company needs to test the safety of a
Sampling & new vaccine (a small sampling error could risk
Nonsampling Errors lives).
A brand wants to survey all users of its
3. Variability in the
product (users include teens, adults, seniors
Population
with very different preferences).
A researcher plans to use statistical tests
4. Statistical
(e.g., t-tests) to analyze if sales differ by
Considerations
region.
A small café wants to quickly find out if
5. Operational Efficiency
customers like its new cake (no budget for
(Cost & Expertise)
trained researchers).
34
Step 4. Sample Size Determination
Determining the final sample size depends on :
1. The nature of analysis
2. Resources constraints
3. Rules of thumb (or sample sizes used in similar studies)
4. The use of Statistical Methods
In general:
Exploratory research: smaller sample size
Conclusive research (descriptive/causal): larger sample size
35
Sampling Size Formula
Sample Size determination by Means:
2 2
n= Z σ
2
D
Z = The desired level of confidence that the sample result will fall within a certain
range (result +/- sampling error), in standard errors or Z values.
D = The specification of the precision level
σ = An estimate of the population standard deviation
36
Estimating σ
Use results from prior survey
Conduct a pilot survey
Use of Secondary Data
Z’ and ‘D’ are set by the researcher, based on statistical,
managerial, and financial criteria.
2 2
n= Z σ
2
D
37
2 2
n= Z σ
An example 2
D
Suppose a researcher wants to determine the sample size of
a survey of identifying the average $ spent of customers of
their visit to Park’N Shop
A similar survey which was taken 6 months ago showed the
σ = $85
The researcher wants to use the Confidence Level (CL) of
95% (i.e. z score = 1.96 from table)
The precision level will be set at +/- $10 (i.e. D= 10)
n= 1.962 x 852
2
10
Hence, the required sample will be 278
38
Step 5. Execution of sampling process
Specify operational instructions
Setting clear definitions of steps
Setting clear guidelines to interviewers
An example of a Sampling plan:
39
An example of Sampling Plan
1. Target population: Any female who is HK citizen and aged
between 18-45 who used makeup and skincare products during
the past 3 months
2. Sampling frame: Customers from three department stores located
in X, Y & Z
3. Sampling technique: Probability sampling (systematic sampling) –
For every 3 customers coming out of the skincare store, choose
the 3rd customer.
4. Sample Size: 300
5. Execution: Allocate 3 interviewers into stores A, B & C and survey
100 females by random systemically (every 3 walk by and choose
the last one)
40