MARKETING RESEARCH

An Applied Orientation | Naresh K. Malhotra | 6th Edition

CHAPTER 11 DETAILED NOTES

Sampling: Design and Procedures

Topics Covered:

1. The Sampling Design Process (5 Steps)

2. Non-Probability Sampling Techniques

3. Probability Sampling Techniques

4. Choosing Between Non-Probability and Probability Sampling

PGDM (Insurance) — IIRM Hyderabad | Trimester III | 2024–26

 PART 1 — THE SAMPLING DESIGN PROCESS

The sampling design process includes five sequential steps that are closely interrelated and relevant to all
aspects of the marketing research project. Sample design decisions should be integrated with all other decisions
in a research project.

Step Action Key Concept

1 Define the Target Population Specify element, sampling unit, extent, and time

2 Determine the Sampling Frame A list or directions for identifying the target population

3 Select a Sampling Technique Probability vs non-probability; with/without replacement

4 Determine the Sample Size Qualitative + quantitative factors; resource constraints

5 Execute the Sampling Process Detailed operational specifications for each decision

Step 1 — Define the Target Population

Definition: Target Population

"The collection of elements or objects that possess the information sought by the researcher and about
which inferences are to be made." — Malhotra Must be defined in terms of: (1) Element, (2) Sampling Unit,
(3) Extent, (4) Time.

Definition: Element

"An object that possesses the information sought by the researcher. In survey research, the element is
usually the respondent." — Malhotra

Definition: Sampling Unit

"The basic unit containing the elements of the population to be sampled." — Malhotra The sampling unit
may be the same as the element (e.g., individual respondent) or different (e.g., household containing the
element).

Why definition matters:

Imprecise definition of the target population will result in research that is ineffective at best and misleading at
worst. For example, in assessing consumer response to a new men's cologne: Should the population include
only men? Women who buy cologne for partners? Men over 17 only? Each choice leads to a different sample
and different conclusions.

Example — Insurance Survey — IRDAI Customer Satisfaction Study

 IRDAI wants to conduct a customer satisfaction survey for policyholders of private non-life insurers after the
motor insurance tariff liberalisation. Element: Individual policyholder who holds an active motor OD policy
with a private insurer. Sampling Unit: The household (since often more than one family member holds a
policy and the decision-maker may differ from the named insured). Extent: Tier-1 and Tier-2 cities in
Maharashtra and Gujarat — chosen because these states have the highest private motor insurance
penetration. Time: Policy holders with policies renewing between April 2025 and March 2026. Without this
precision, a researcher might accidentally survey fleet owners (commercial policies) instead of individual
retail policyholders, completely distorting the satisfaction scores. The definition step prevents this costly
error.

Step 2 — Determine the Sampling Frame

Definition: Sampling Frame

"A representation of the elements of the target population. It consists of a list or set of directions for
identifying the target population." — Malhotra Examples: telephone directory, association membership list,
mailing list, city directory, map, random digit dialing program.

A key issue is sampling frame error — the discrepancy between the frame and the actual target population. A
frame may omit some elements (under-coverage) or include elements that don't belong (over-coverage). Three
ways to handle sampling frame error:

• Redefine the population in terms of the sampling frame (e.g., 'households with a correct telephone listing').
• Screen respondents during data collection to filter out ineligible elements.
• Adjust the data by a weighting scheme to counterbalance the sampling frame error.

Example — Group Health Insurance — Sampling Frame for SME Survey

A health insurer wants to survey HR managers at small and medium enterprises (SMEs) in Pune about
group health policy satisfaction. The insurer decides to use the MCCIA (Mahratta Chamber of Commerce)
member directory as its sampling frame. Sampling Frame Error: The MCCIA directory covers only
formal-sector, chamber-affiliated SMEs. It misses unregistered businesses and firms that are chamber
members but have lapsed. As a result, the frame over-represents well-organised, larger SMEs and
under-represents micro-enterprises. How the insurer handles this: During telephonic screening, every
contacted HR manager is asked 'Does your firm have between 10 and 200 employees?' to filter out firms
that fall outside the SME definition. Additionally, respondents from service-sector firms are weighted
downward because the MCCIA directory is skewed toward manufacturing. This corrects the composition
bias before analysis.

Step 3 — Select a Sampling Technique

Selecting a sampling technique involves three decisions: (1) Bayesian vs Traditional approach, (2) Sampling
with vs without replacement, (3) Nonprobability vs Probability sampling. The last decision is the most critical.

Definition: Sampling With Replacement

"A sampling technique in which an element can be included in the sample more than once." — Malhotra

Definition: Sampling Without Replacement

"A sampling technique in which an element cannot be included in the sample more than once." — Malhotra
 Definition: Nonprobability Sampling

"Sampling techniques that do not use chance selection procedures. Rather, they rely on the personal
judgment of the researcher." — Malhotra

Definition: Probability Sampling

"A sampling procedure in which each element of the population has a fixed probabilistic chance of being
selected for the sample." — Malhotra

Step 4 — Determine the Sample Size

Definition: Sample Size

"The number of elements to be included in a study." — Malhotra Determining sample size is complex and
involves qualitative and quantitative considerations. Quantitative determination (using confidence intervals)
is covered in Chapter 12.

Qualitative Factors in Sample Size Determination:

• Importance of the decision: More important decisions require larger, more precise samples.
• Nature of the research: Exploratory research uses small samples; conclusive/descriptive research requires
larger ones.
• Number of variables: More variables being measured require larger samples to control cumulative sampling
error.
• Nature of analysis: Multivariate or subgroup analysis demands larger samples.
• Sample sizes in similar studies: Industry norms provide benchmarks (see Table 11.2).
• Incidence rates: Lower proportion of eligible respondents in population → initial sample must be much larger.
• Completion rates: Anticipated refusals increase the required initial contacts.
• Resource constraints: Time, money, and availability of qualified interviewers set practical limits.

Typical Sample Sizes in Marketing Research (Table 11.2)

Type of Study Min. Size Typical Range

Problem identification research (e.g., market potential) 500 1,000–2,500

Problem-solving research (e.g., pricing) 200 300–500

Product tests 200 300–500

Test-marketing studies 200 300–500

TV/radio/print advertising (per ad tested) 150 200–300

Test-market audits 10 stores 10–20 stores

Focus groups 2 groups 6–15 groups

Step 5 — Execute the Sampling Process

Execution requires a detailed specification of how all sampling design decisions are to be implemented —
population definition, sampling frame, sampling unit, technique, and sample size. For example, if households
are the sampling unit, an operational definition of 'household' is needed; procedures for vacant units and
callbacks must be specified; the next-birthday method for within-household selection must be documented.

Example — Reinsurance Broker Survey — Full Sampling Design Execution

An IRDAI-commissioned study on cedant satisfaction with reinsurance brokers in India. Full sampling
design: Target Population: Risk managers or finance heads at Indian corporates with annual turnover
above ■500 crore that purchase reinsurance through a registered broker (element); their company
(sampling unit); Pan-India (extent); FY 2025-26 (time). Sampling Frame: IRDAI's registered corporate
policyholder database, supplemented by the FICCI risk management members list. Frame error is expected
because some companies place reinsurance directly — screened out during data collection. Technique:
Stratified sampling — strata defined by industry sector (manufacturing, energy, infrastructure, services) and
broker relationship (single-broker vs multi-broker). Sample Size: 400 respondents across strata. Incidence
rate is estimated at 60% (not all contacted risk managers deal with reinsurance broking). Completion rate
estimated at 70%. Initial contacts needed: 400 / (0.60 × 0.70) = approximately 952 contacts. Execution:
CATI (Computer-Assisted Telephone Interviewing) used. Callback schedule: maximum 3 attempts,
morning/afternoon/evening rotation. Within-company, the next-seniority-level person is contacted if the risk
manager is unavailable.
 PART 2 — NON-PROBABILITY SAMPLING TECHNIQUES

Definition: Nonprobability Sampling

"Sampling techniques that do not use chance selection procedures. Rather, they rely on the personal
judgment of the researcher. They do not permit an objective evaluation of the precision of the sample
results, and the estimates obtained are not statistically projectable to the population." — Malhotra

Overview of Non-Probability Techniques

Technique Description

Convenience Sampling Samples most accessible, convenient respondents. Least expensive. Not representative.
For exploratory use only.

Judgmental Sampling Researcher's expertise selects 'typical' or 'representative' elements. Subjective but useful
for commercial research.

Quota Sampling Two-stage: first set quotas to match population composition; then select elements by
convenience within each quota.

Snowball Sampling Initial random group refers further respondents. Used for rare characteristics or
hard-to-reach populations.

A — Convenience Sampling

Definition: Convenience Sampling

"A nonprobability sampling technique that attempts to obtain a sample of convenient elements. The
selection of sampling units is left primarily to the interviewer." — Malhotra Examples: mall-intercept
interviews without qualifying respondents, students/church groups, tear-out magazine questionnaires,
"people on the street" interviews.

Advantages: Least expensive and least time-consuming; units are accessible, easy to measure, cooperative.

Disadvantages: Serious selection bias; not representative of any definable population; not appropriate for
descriptive or causal research; cannot generalize to any population.

Best used for: Exploratory research — focus groups, questionnaire pretesting, pilot studies.

Example — Crop Insurance Awareness — Mall Intercept Study

A general insurance company is planning to launch a new standalone crop insurance product for small and
marginal farmers. Before designing the product, the marketing team wants a quick read on urban
consumers' awareness of crop insurance — not because urban consumers are the target, but because the
team needs to pilot-test the questionnaire before deploying it in rural agrarian districts. A convenience
sample of 80 shoppers at a Phoenix Market City in Pune is intercepted on a Saturday afternoon. The
questionnaire is administered and the responses reveal that several questions are ambiguous — for
example, 'What is the sum insured under PMFBY?' confuses respondents who interpret 'sum insured' as
'premium.' What this sample is good for: Identifying badly worded questions, testing sequencing, and
estimating interview length (~12 minutes). The sample is NOT used to make any market-size claims. Once
the questionnaire is refined, a probability sample of actual farmers in Vidarbha and Marathwada is drawn
for the real study.
B — Judgmental Sampling

Definition: Judgmental Sampling

"A form of convenience sampling in which the population elements are selected based on the judgment of
the researcher. The researcher, exercising judgment or expertise, chooses the elements to be included in
the sample, because he or she believes that they are representative of the population." — Malhotra
Common examples: test markets, purchase engineers as company representatives, bellwether precincts in
voting research, expert witnesses, selected department stores for new display tests.

Advantages: Low cost, convenient, quick. Useful when broad population inferences are not required and the
researcher has deep domain expertise.

Disadvantages: Subjective. Does not allow direct generalization to a specific population. Quality entirely
depends on researcher's judgment and expertise.

Example — Marine Insurance — Judgmental Selection of Survey Respondents

A reinsurance broker wants to understand the underwriting appetite of Indian marine cargo underwriters for
container-ship risks post the Red Sea disruptions. The research manager uses judgmental sampling —
selecting 15 senior marine underwriters from New India Assurance, Oriental Insurance, and Bajaj Allianz
who are known experts in hull and cargo underwriting with 15+ years of experience. The researcher
deliberately excludes junior underwriters because they lack the authority to accept or decline large
facultative risks. The researcher also excludes underwriters from companies that do not write marine
business. This judgment is based on the researcher's knowledge of the market structure. Why this works
here: The goal is to understand expert opinion on niche underwriting appetite — not to generalize across all
insurance employees. The small, purposively selected sample delivers richer, more relevant insights than a
random sample of 200 insurance employees would.

C — Quota Sampling

Definition: Quota Sampling

"A nonprobability sampling technique that is a two-stage restricted judgmental sampling. The first stage
consists of developing control categories, or quotas, of population elements. In the second stage, sample
elements are selected based on convenience or judgment." — Malhotra The quotas ensure that the
composition of the sample is the same as the composition of the population with respect to characteristics
of interest (e.g., gender, age, income, region).

Advantages: Lower cost, greater interviewer convenience, sample can be controlled for certain characteristics.
Under certain conditions obtains results close to conventional probability sampling.

Disadvantages: Selection bias within each quota (interviewer may avoid unfriendly respondents, undesirable
locations). Does not permit assessment of sampling error. Cannot project to population statistically.

Example — Life Insurance Awareness Study — Quota Sample Design

 An insurer wants to survey 1,000 adults about awareness of term life insurance across India. A random
probability sample would be expensive and time-consuming. Instead, a quota sample is designed with three
control characteristics: gender, age group, and income tier — matching the LIC policyholder database
composition. Control quotas set: • Gender: 52% Male (520), 48% Female (480) • Age: 18–30 (30%), 31–45
(40%), 46–60 (20%), 60+ (10%) • Income: Below ■5 lakh p.a. (45%), ■5–15 lakh (35%), Above ■15 lakh
(20%) Field interviewers at train stations, markets, and malls are given these quotas. They fill each quota
by convenience — approaching suitable individuals until all 1,000 slots are filled. Within the ■5–15 lakh
income bracket, 350 interviews must be completed by the time fieldwork closes. Key limitation exposed:
An interviewer working at a high-end mall fills the ■5–15 lakh bracket quickly by interviewing well-dressed
shoppers. This introduces a subtle upward bias — respondents in this bracket skew toward higher financial
literacy than the true population. This is why quota sampling results cannot be statistically projected.

D — Snowball Sampling

Definition: Snowball Sampling

"A nonprobability sampling technique in which an initial group of respondents is selected randomly.
Subsequent respondents are selected based on the referrals or information provided by the initial
respondents. This process may be carried out in waves by obtaining referrals from referrals." — Malhotra A
major objective of snowball sampling is to estimate characteristics that are RARE in the population.

Advantages: Substantially increases the likelihood of locating desired rare characteristics. Low sampling
variance and cost.

Disadvantages: Final sample is a nonprobability sample despite random initial selection. Referrals have similar
demographic/psychographic characteristics to referring respondents — introduces homogeneity bias.
Time-consuming.

Example — Cyber Insurance — Snowball Sampling for Early Adopters

An insurer wants to understand why a small group of Indian tech-sector CFOs are early adopters of
standalone cyber insurance (a very rare characteristic — penetration is below 3% in the SME segment). A
random sample of 10,000 SMEs would yield only ~300 cyber insurance buyers — extremely expensive
screening. Snowball approach: The insurer's broker relationships are used to randomly identify 10 CFOs
known to have purchased cyber insurance. Each is interviewed and asked: 'Do you know of other CFOs at
peer companies who have also purchased cyber insurance?' Each refers 2–3 colleagues, generating 25
referrals. These 25 are interviewed and asked the same question, producing another 40 names. By the
third wave, the sample reaches 80 cyber insurance buyers — enough for a meaningful
qualitative-quantitative analysis. The sample is NOT projected to all Indian SMEs. It is used specifically to
understand the motivations of early adopters — useful for designing targeted marketing campaigns and
refining product features for the next wave of buyers.
 PART 3 — PROBABILITY SAMPLING TECHNIQUES

Definition: Probability Sampling

"A sampling procedure in which each element of the population has a fixed probabilistic chance of being
selected for the sample. It is possible to prespecify every potential sample of a given size that could be
drawn from the population, as well as the probability of selecting each sample. Confidence intervals can be
calculated; the researcher can make inferences or projections about the target population." — Malhotra

Sampling efficiency reflects a trade-off between cost and precision. The four main probability techniques are:
Simple Random Sampling, Systematic Sampling, Stratified Sampling, and Cluster Sampling. Each is assessed
relative to SRS as the baseline.

1 — Simple Random Sampling (SRS)

Definition: Simple Random Sampling (SRS)

"A probability sampling technique in which each element in the population has a known and equal
probability of selection. Every element is selected independently of every other element and the sample is
drawn by a random procedure from a sampling frame." — Malhotra Equivalent to a lottery system. Requires
assigning unique IDs to all elements and using random number tables or computer routines.

Advantages:
• Easily understood and explained to clients.
• Results can be projected to the target population.
• Most statistical inference methods assume SRS data.

Disadvantages:
• Difficult to construct a complete sampling frame for most populations.
• Can produce very large or geographically dispersed samples — high cost.
• Lower precision (larger standard errors) than stratified sampling.
• May produce unrepresentative samples, especially with small n.

Example — Motor Insurance Claim Audit — SRS

An insurer has a database of 50,000 motor insurance claims settled in FY 2025-26 and wants to audit a
sample of 500 claims for documentation accuracy. Each claim is assigned a number from 1 to 50,000 in the
system. Using a random number generator, 500 unique numbers between 1 and 50,000 are produced: say
4,832; 17,219; 29,004… and so on. The corresponding 500 claim files are pulled and audited. Every claim
has a 500/50,000 = 1% probability of selection, independent of every other claim. The audit results can be
projected: if 6% of the sampled claims had documentation errors, the insurer can say with 95% confidence
that 5%–7% of all 50,000 claims have documentation errors — a statistically valid inference. This is the key
advantage of SRS over convenience or judgmental sampling.

2 — Systematic Sampling

Definition: Systematic Sampling

 "A probability sampling technique in which the sample is chosen by selecting a random starting point and
then picking every ith element in succession from the sampling frame. The sampling interval i = N/n
(population size / sample size), rounded to nearest integer." — Malhotra

Similar to SRS in equal probability, but only one random number is needed — the starting point r. The sample
then consists of elements r, r+i, r+2i, r+3i, … up to the required sample size. Can be used without knowing the
full sampling frame in advance (e.g., intercept every 10th customer leaving a bank branch).

Key caution: If the sampling frame has a cyclical pattern whose period matches the sampling interval i,
systematic sampling will be biased (e.g., always selecting the same day of the week in weekly data).

Example — Fire Insurance Renewal Follow-Up — Systematic Sampling

An insurer has a list of 12,000 commercial fire insurance policies due for renewal in Q1 FY26, sorted by
renewal date. The operations team wants to call 600 policyholders to offer a loyalty discount — but calling
all 12,000 is not feasible. Sampling interval: i = 12,000 / 600 = 20 Random start: A number between 1
and 20 is randomly selected — say 7. Sample: Policy nos. 7, 27, 47, 67, 87, 107... (every 20th policy on the
renewal list) Because the list is sorted by renewal date, this ensures the 600 selected policies are spread
evenly across all renewal months. A pure SRS might by chance over-select March renewals. Systematic
sampling avoids this without requiring complex stratification. The only risk: if the list alternates between
corporate and retail customers every 10 policies, an interval of 20 would always land on the same type —
the insurer checks for this cyclical pattern before proceeding.

3 — Stratified Sampling

Definition: Stratified Sampling

"A probability sampling technique that uses a two-step process to partition the population into
subpopulations, or strata. Elements are selected from each stratum by a random procedure." — Malhotra
Strata must be mutually exclusive and collectively exhaustive. The major objective is to INCREASE
PRECISION without increasing cost.

Criteria for selecting stratification variables:

• Homogeneity within strata: Elements within each stratum should be as similar as possible on the
characteristic of interest.
• Heterogeneity across strata: Elements in different strata should differ as much as possible.
• Relatedness: Stratification variable must be closely related to the characteristic being measured.
• Cost: Variable should be easy and inexpensive to measure and apply.

Proportionate vs Disproportionate Stratified Sampling:

In proportionate stratified sampling, the sample drawn from each stratum is proportional to that stratum's
share of the total population. In disproportionate stratified sampling, strata with higher variability are
over-sampled to improve precision. When the researcher is primarily interested in comparing strata against each
other, equal sample sizes from each stratum are often used.

Example — Property Insurance Premium Survey — Stratified by Business Size

 An insurer wants to estimate the average annual property insurance premium paid by Indian corporates, but
the premium distribution is highly skewed — micro-enterprises pay ■5,000–■50,000 while large corporates
pay ■50 lakh+. A SRS of 400 companies would give very imprecise estimates because the high-premium
large corporates would rarely be sampled. Stratification by annual turnover: • Stratum 1: Micro (turnover
< ■1 cr) — 60% of population → 120 sampled (proportionate) • Stratum 2: Small (■1–25 cr) — 25% → 100
sampled (proportionate) • Stratum 3: Medium (■25–250 cr) — 10% → 40 sampled (proportionate) • Stratum
4: Large (>■250 cr) — 5% → 140 sampled (DISPROPORTIONATE — over-sampled because variance in
premiums is very high) Total sample: 400. The large-corporate stratum is given extra weight because the
insurer's revenue is disproportionately driven by these 5%. Over-sampling them gives a precise estimate for
that critical stratum. The final population mean is derived by weighting each stratum's estimate by its true
population proportion.

4 — Cluster Sampling

Definition: Cluster Sampling

"First, the target population is divided into mutually exclusive and collectively exhaustive subpopulations
called clusters. Then, a random sample of clusters is selected based on a probability sampling technique
such as simple random sampling. For each selected cluster, either all the elements are included in the
sample or a sample of elements is drawn probabilistically." — Malhotra Objective: DECREASE COST (not
increase precision — the opposite of stratified sampling).

Key distinctions — Cluster vs Stratified Sampling:

Factor Cluster Sampling Stratified Sampling

Objective Decrease cost Increase precision

Subpopulations A sample of clusters chosen ALL strata included

Within subpopulation Each cluster should be Each stratum should be

HETEROGENEOUS HOMOGENEOUS

Across subpopulations Clusters should be HOMOGENEOUS Strata should be HETEROGENEOUS

Sampling frame needed Only for selected clusters For entire population

Types of Cluster Sampling:

• One-Stage Cluster Sampling: All elements within selected clusters are included. Fast and simple but less
precise.
• Two-Stage Cluster Sampling (Simple): Random sample of clusters selected first; then a random sample of
elements within each selected cluster.
• Multistage Cluster Sampling: Multiple stages of random sampling (e.g., geographic area → district → block
→ household).
• Area Sampling: Clusters are geographic areas (counties, housing tracts, blocks). Most common form in
consumer research.
• Probability Proportionate to Size (PPS) Sampling: Clusters selected with probability proportional to their
size; second-stage probability varies inversely. Net result: equal probability for every element.

Example — Health Insurance Survey — Cluster Sampling Across Districts

 IRDAI wants to assess awareness of the Pradhan Mantri Jan Arogya Yojana (PM-JAY) scheme among
Below Poverty Line (BPL) households in rural India. There is no master list of all BPL households
(impossible to compile for 300 million+ beneficiaries). Cluster sampling is ideal because geographic clusters
(districts/blocks) do exist. Stage 1: India's 750+ districts are listed. 50 districts are randomly selected using
SRS. Stage 2: Within each selected district, 10 Gram Panchayats (GPs) are randomly selected. Stage 3:
Within each selected GP, 20 BPL households are randomly selected from the beneficiary database. Total
sample: 50 × 10 × 20 = 10,000 households. This is multistage cluster sampling. The sampling frame is
needed only for the 50 selected districts, not for all of India — a huge cost saving. The trade-off:
households within the same GP tend to be similar (all rural, same socioeconomic level), so the sample is
less precise per unit cost than if the same 10,000 households were selected by SRS across all of India. But
SRS across all rural India would cost 10x more to execute.

5 — Other Probability Techniques: Sequential & Double Sampling

Definition: Sequential Sampling

"A probability sampling technique in which the population elements are sampled sequentially, data
collection and analysis are done at each stage, and a decision is made as to whether additional population
elements should be sampled. The sample size is not known in advance." — Malhotra

Definition: Double Sampling (Two-Phase Sampling)

"A sampling technique in which certain population elements are sampled twice. In the first phase, a sample
is selected and some information is collected. In the second phase, a subsample is drawn from the original
sample and additional information is obtained." — Malhotra

Example — Sequential Sampling — Testing IRDAI Draft Regulation

IRDAI drafts a new circular allowing insurers to use Aadhaar-based e-KYC for all policy issuances. The
regulator wants to know whether insurance agents (intermediaries) prefer e-KYC over physical KYC.
Sequential sampling is used: Agents are interviewed one at a time, and after each group of 20, a running
proportion is calculated. If 70%+ consistently prefer e-KYC, sampling stops — sufficient evidence exists. If
preference is below 50%, sampling also stops — strong evidence against. Only if results are borderline
(50–70%) does sampling continue. This avoids collecting 500 responses when 80 are sufficient.

Summary: Strengths and Weaknesses of All Sampling Techniques (Table 11.4)

Technique Type Strengths Weaknesses

Convenience Non-P Lowest cost, fastest, most accessible Severe selection bias; not representative;
cannot project

Judgmental Non-P Low cost; useful when researcher has deep Subjective; no generalization possible
expertise

Quota Non-P Controls sample composition; lower cost Selection bias within quotas; no error
than probability estimation

Snowball Non-P Best for rare characteristics; low cost per Referral homogeneity bias; time-consuming
locate

SRS Prob Easy to understand; projectable; basis of Hard to build frame; costly if geographically
statistical inference dispersed; lower precision
Systematic Prob Easier/cheaper than SRS; can use without Cyclical pattern bias possible
complete frame

Stratified Prob Highest precision; ensures all subgroups Hard to choose stratification variables;
represented expensive

Cluster Prob Most cost-effective; feasible when only Lower precision; difficult to form
clusters are known heterogeneous clusters

PART 4 — CHOOSING BETWEEN NON-PROBABILITY AND

PROBABILITY SAMPLING

Factor Favours Non-Probability Favours Probability

Nature of research Exploratory Conclusive/Descriptive

Nonsampling vs Sampling errors Nonsampling errors dominate → non-prob Sampling errors dominate → probability
gives more control needed

Population variability Homogeneous (low variance) Heterogeneous (high variance)

Statistical considerations Not required Required (confidence intervals, projections)

Time available Short Longer lead time acceptable

Cost budget Low Higher budget available

Need to project to population Not needed Essential

Example — When to Use What — Insurance Research Decision Guide

Scenario 1 — Use Non-Probability (Convenience): An insurer wants to pretest a 20-question customer

satisfaction questionnaire before deploying it nationwide. 30 employees at the head office are asked to fill
the form and flag confusing questions. No statistical projection needed. Convenience sampling is perfect.
Scenario 2 — Use Non-Probability (Quota): A broker wants a quick read on agent sentiment about a new
digital policy issuance portal. Time pressure is high (decision in 2 weeks). A quota sample of 150 agents
stratified by region (North/South/East/West) and tier (corporate/retail) is used. Results inform internal
decisions but are not published. Scenario 3 — Use Probability (Stratified): IRDAI wants to publish the
Annual Policyholder Satisfaction Index. Results will be publicly released and used to rank insurers.
Stratified probability sampling (by insurer, line of business, and geography) is mandatory — the findings
must be defensible, with confidence intervals reported. Scenario 4 — Use Probability (Cluster): A
microinsurance NGO wants to assess awareness of the Micro-insurance Regulation (2005) among
policyholders in rural Odisha. No master list of rural microinsurance policyholders exists — only village-level
records. Multistage cluster sampling (district → block → village → household) is the only feasible probability
approach.

CHAPTER SUMMARY — KEY TAKEAWAYS

• The sampling design process has 5 steps: Define population → Determine frame → Select technique →
Determine size → Execute. All steps are interrelated.
• Target population must be defined in terms of element, sampling unit, extent, and time — imprecise
definition is the #1 cause of flawed research.
• Sampling frame error is the gap between the frame and the actual population — can be handled by
redefining the population, screening respondents, or weighting data.
• Non-probability sampling (convenience, judgmental, quota, snowball) relies on researcher judgment, is
cheaper and faster, but cannot be statistically projected to the population.
• Probability sampling (SRS, systematic, stratified, cluster) uses chance selection — allows confidence
intervals, statistical inference, and valid generalisation.
• Stratified sampling increases precision by ensuring all subgroups are represented. Clusters within strata
should be homogeneous; strata themselves should be heterogeneous.
• Cluster sampling decreases cost. The logic is the reverse — clusters should be heterogeneous internally;
clusters themselves should be homogeneous to each other.
• The choice between techniques depends on: nature of research (exploratory vs conclusive), need to project
results, population variability, time, cost, and statistical requirements.

Chapter 11 Notes | Marketing Research: An Applied Orientation, 6th Ed. | Malhotra | PGDM (Insurance) — IIRM Hyderabad |
Trimester III 2024–26