Skip to content

Research Design, Data Collection and Sampling

UGC NET Commerce: types of business research, research designs, primary and secondary data, classification, sampling methods, CLT, standard error, estimation.

  • 20 min read
  • Updated
On this page · 17 sections
  1. Prerequisites
  2. The Hook: 1,200 Customers Speak for Twelve Million
  3. 1. Research: Concept and Types
  4. 2. Research Designs
  5. 3. Data: Collection
  6. 4. Data: Classification and Tabulation
  7. 5. Sampling: Concepts
  8. 6. Methods of Sampling
  9. 7. Sampling Distribution, CLT and Standard Error
  10. 8. Statistical Estimation
  11. Memory Tricks
  12. Don’t Confuse
  13. Common Mistakes
  14. Practice
  15. Quick Revision Table
  16. Key Takeaways
  17. What’s Next

Prerequisites

The Hook: 1,200 Customers Speak for Twelve Million


1. Research: Concept and Types

Research is a systematic, objective and replicable search for knowledge, or for answers to a defined problem, through the collection and analysis of data. Business research applies it to managerial decisions: market potential, consumer behaviour, pricing, capital structure, HR practices.

BasisTypes
PurposeFundamental (basic, pure: theory-building) · Applied (solve a specific business problem) · Action (practitioner researches own setting to improve it)
ObjectiveExploratory · Descriptive · Diagnostic · Causal / explanatory (cause and effect) · Predictive
DataQuantitative · Qualitative · Mixed methods
LogicDeductive (theory → hypothesis → test) · Inductive (observations → generalisation)
TimeCross-sectional (one point in time) · Longitudinal (same units over time: panel or trend studies)
SettingLaboratory · Field · Simulation
NatureConceptual (abstract ideas, theory) · Empirical (observation and experiment)
OtherHistorical; ex post facto; case study; evaluation research; one-time vs longitudinal

2. Research Designs

A research design is the blueprint for collecting, measuring and analysing data: the arrangement of conditions that answers the research question with economy of effort. It settles what, where, when, how much and by what means.

A good design is flexible, appropriate, efficient and economical; it minimises bias and maximises the reliability of the data, and it gives the least experimental error.

Three broad types

DesignPurposeFeaturesBusiness example
Exploratory (formulative)Discover ideas and insights; define the problem; develop hypothesesFlexible; literature survey, experience survey (talking to knowledgeable people), analysis of insight-stimulating cases, focus groupsWhy are young customers leaving our brand?
Descriptive and diagnosticDescribe characteristics of a group (descriptive); find how often something occurs or with what it is associated (diagnostic)Rigid, pre-planned; must guard against bias; surveys, panelsProfile of credit-card users by age and income
Hypothesis-testing (experimental / causal)Test cause-and-effect relationshipsManipulation, control, randomisationDoes a 10% price cut raise sales of a product?

Exploratory designs are flexible; descriptive and experimental designs must be rigid so that the results are reliable and unbiased.

Experimental design principles

R.A. Fisher, who developed experimental design in agricultural research, set out three principles:

  1. Replication — repeat the treatment on several units, so that experimental error can be estimated and precision increased.
  2. Randomisation — allot treatments to units at random, to protect against the effect of extraneous factors.
  3. Local control — group similar units into blocks, so that variation between blocks is removed from experimental error.
Informal designsFormal designs
Before-and-after without controlCompletely randomised design (CRD): treatments allotted fully at random; analysed by one-way ANOVA
After-only with controlRandomised block design (RBD): units grouped into homogeneous blocks, treatments randomised within each block; two-way ANOVA
Before-and-after with controlLatin square design (LSD): controls two sources of variation (rows and columns); each treatment once in each row and column
Factorial designs: two or more factors at once, including their interaction

3. Data: Collection

Primary vs secondary data

BasisPrimary dataSecondary data
MeaningCollected first-hand for the present purposeCollected earlier by someone else, for another purpose
OriginalityOriginalAlready in existence, often processed
Cost and timeHighLow
SuitabilityExactly fits the purposeMay not fit; needs checking
ExamplesSurvey of your own customers, test-market salesCensus, RBI publications, company annual reports, industry databases

Before using secondary data, check its reliability (who collected it, how, with what bias), suitability (definitions, units, coverage and period match your purpose) and adequacy (enough detail and coverage).

Methods of collecting primary data

MethodFeaturesStrengthWeakness
ObservationRecord behaviour as it happens; structured or unstructured; participant or non-participant; disguised or undisguisedCaptures actual behaviour, not reportsCannot observe attitudes, past behaviour; observer bias
Personal interviewFace to face; structured or unstructured; depth interviewDepth, clarification, high responseCostly; interviewer bias
Telephone / online interviewRemoteQuick, cheapShort; limited to reachable respondents
Mailed or online questionnaireRespondent fillsCheap; wide coverage; no interviewer biasLow response; needs literate respondents
ScheduleEnumerator fills after asking the questionsWorks with non-literate respondentsExpensive; enumerator must be trained
Focus groupModerated discussion with a small groupRich, exploratory insightNot representative; dominant members
Panel / consumer diarySame respondents report repeatedlyShows change over timePanel conditioning, dropouts
Projective techniquesWord association, sentence completion, thematic apperception, role playReveal motives respondents will not stateHard to interpret

Questionnaire design: questions should be short, simple and unambiguous; avoid leading, double-barrelled (“Is the app fast and secure?”) and loaded questions; move from general to specific (funnel sequence); put sensitive questions late. A pilot study (pre-test) on a small group checks the questionnaire before the main survey.

Measurement and scaling

Scale (Stevens)PropertyPermissible averageExample
NominalCategories onlyModeGender, region, brand used
OrdinalOrderMedian (and mode)Rank of brands, income bracket
IntervalEqual intervals, arbitrary zeroMeanTemperature in °C, a rating scale treated as interval
RatioEqual intervals, true zeroAll, including GM, HM and CVSales, income, age

Attitude scales used in business research:

ScaleHow it works
Likert (summated)Agreement with statements on a 5- (or 7-) point scale; item scores summed
Semantic differential (Osgood)Bipolar adjectives at the ends of a 7-point scale (“modern … old-fashioned”)
StapelSingle adjective with a scale from −5 to +5, no neutral zero
Thurstone (equal-appearing intervals)Judges first rate many statements; the respondent agrees or disagrees with the selected ones
Guttman (cumulative, scalogram)Items arranged so that agreeing with a strong item implies agreeing with all weaker ones
Paired comparisonChoose between items two at a time; gives ranking
Constant sumAllocate a fixed total (say 100 points) among attributes

Reliability is consistency (test–retest, split-half, Cronbach’s alpha); validity is measuring what you intend (content, criterion, construct). A measure can be reliable without being valid, but not valid without being reliable.


4. Data: Classification and Tabulation

Classification is arranging data into groups or classes by common characteristics. It condenses data, brings out similarities and differences, and prepares data for tabulation and analysis.

Basis of classificationExample
Geographical (spatial)Sales by state
Chronological (temporal)Sales by year (a time series)
Qualitative (by attributes)By gender or occupation; simple / dichotomous (two classes: literate–illiterate) or manifold (many classes)
Quantitative (by variables)By income or age; discrete variables (number of children) or continuous variables (height)

Frequency distributions: vocabulary

TermMeaning
Class limitsLower and upper values of a class (20–30)
Class interval (width)Upper limit − lower limit
Class mark (mid-value)(Lower + upper)/2
Exclusive methodUpper limit of one class = lower limit of the next (10–20, 20–30); the upper limit is excluded. Required for continuous data
Inclusive methodBoth limits included (10–19, 20–29); convert to exclusive (9.5–19.5) before computing median or mode
Open-end class“Below 10” or “60 and above”
Frequency densityFrequency ÷ class width (for histograms with unequal classes)
Cumulative frequencyRunning total; “less than” and “more than” types; plotted as ogives

Number of classes: usually between 5 and 15. Sturges’ rule gives \( k = 1 + 3.322 \log_{10} N \). For N = 100: \( 1 + 3.322 \times 2 \approx 7.6 \), so about 8 classes.

Tabulation

A table presents classified data in rows and columns. Its parts:

  1. Table number
  2. Title (what, where, when)
  3. Head note (units, for example “₹ crore”)
  4. Captions (column headings) and stubs (row headings)
  5. Body (the figures)
  6. Footnote (explanations, exceptions)
  7. Source note (where the data came from)

Classification comes before tabulation: classification groups the data, tabulation displays the groups.


5. Sampling: Concepts

TermMeaning
Population (universe)All units under study; finite or infinite
Sampling frameThe list from which the sample is drawn (voters’ list, customer database)
Sampling unitThe unit selected (household, firm, invoice)
ParameterA measure of the population (\( \mu \), \( \sigma \), P)
StatisticA measure of the sample (\( \bar{x} \), s, p), used to estimate the parameter
CensusEvery unit studied
Sample surveyOnly a part studied

Why sample? Lower cost, faster results, greater scope, and sometimes greater accuracy (fewer, better-trained enumerators). Sampling is the only option when testing destroys the unit (life of bulbs, crash tests) or the population is infinite. A census is preferred when the population is small, heterogeneous, or every unit must be known (as in a national population census).

Two laws are cited as the basis of sampling: the law of statistical regularity (a reasonably large random sample tends to have the characteristics of the population) and the law of inertia of large numbers (large aggregates are more stable than small ones, because changes in parts tend to offset each other).

Sampling and non-sampling errors

Sampling errorNon-sampling error
Arises because only a part of the population is studiedArises in collection, processing and response
Present only in sample surveysPresent in both census and sample surveys (often larger in a census)
Decreases as sample size increasesMay increase with size
Measured by the standard errorNot easily measured
Examples: chance differences between sample and populationNon-response, wrong answers, faulty questionnaire, coding and tabulation errors, inadequate frame

6. Methods of Sampling

graph TD
    A["Sampling methods"] --> B["Probability (random)
known, non-zero chance"] A --> C["Non-probability (non-random)
chance unknown"] B --> B1["Simple random"] B --> B2["Systematic"] B --> B3["Stratified"] B --> B4["Cluster (area)"] B --> B5["Multistage"] C --> C1["Convenience"] C --> C2["Judgement (purposive)"] C --> C3["Quota"] C --> C4["Snowball"]

Probability methods

MethodHowBest when
Simple randomEvery unit (and every sample of size n) has an equal chance; lottery or random number tables; with or without replacementPopulation homogeneous and a frame exists
SystematicEvery kth unit after a random start; \( k = N/n \)Ordered list; quick. Risk: periodicity in the list matching k
StratifiedDivide into homogeneous strata (internally similar, different from each other); random sample from each stratum; proportional or disproportional allocationPopulation heterogeneous; improves precision
Cluster (area)Divide into clusters (each a mini-population, internally heterogeneous); select some whole clusters at randomNo full frame; population spread over a large area
MultistageSample in stages: states → districts → villages → householdsLarge national surveys

Non-probability methods

MethodHowNote
ConvenienceWhoever is easy to reachCheapest; most biased
Judgement (purposive)Researcher picks typical or expert unitsUseful in exploratory research and small samples
QuotaFixed numbers from each category (age, gender), chosen non-randomly by the interviewerLooks like stratified, but no randomness
SnowballRespondents refer further respondentsHidden or rare populations with no frame

Only probability samples allow the sampling error to be measured and confidence statements to be made.


7. Sampling Distribution, CLT and Standard Error

If we draw all possible samples of size n from a population and compute the mean of each, the distribution of those means is the sampling distribution of the mean.

PropertyValue
Mean of the sampling distribution\( E(\bar{x}) = \mu \) (the sample mean is unbiased)
Standard deviation of the sampling distributionStandard error \( SE_{\bar{x}} = \dfrac{\sigma}{\sqrt{n}} \)
Finite population, without replacement\( SE_{\bar{x}} = \dfrac{\sigma}{\sqrt{n}}\sqrt{\dfrac{N - n}{N - 1}} \) (the finite population correction, negligible when n/N is small)
SE of a proportion\( SE_p = \sqrt{\dfrac{PQ}{n}} \)
SE of difference of two means\( \sqrt{\dfrac{\sigma_1^2}{n_1} + \dfrac{\sigma_2^2}{n_2}} \)
SE of a standard deviation (large n)\( \dfrac{\sigma}{\sqrt{2n}} \)

Uses of the standard error: it measures the reliability of a sample estimate (smaller SE, more reliable), sets the confidence limits for the parameter, and is the denominator of every large-sample test statistic.

Because SE varies as \( 1/\sqrt{n} \), halving the SE needs four times the sample. With \( \sigma = 12 \): n = 36 gives SE = 2; n = 144 gives SE = 1.


8. Statistical Estimation

Estimation uses a sample statistic (the estimator) to infer the value of a population parameter. The value it gives for a particular sample is the estimate.

TypeMeaningExample
Point estimationA single value“Average satisfaction is 3.8”
Interval estimationA range with a stated confidence level“Between 3.74 and 3.86 with 95% confidence”

Properties of a good estimator (R.A. Fisher)

PropertyMeaningExample
UnbiasednessExpected value equals the parameter\( \bar{x} \) for \( \mu \); \( s^2 = \frac{\sum(x - \bar{x})^2}{n - 1} \) for \( \sigma^2 \) (dividing by n gives a biased estimator)
ConsistencyGets closer to the parameter as n increases\( \bar{x} \)
EfficiencySmallest variance among unbiased estimatorsFor a normal population, the mean is more efficient than the median
SufficiencyUses all the information in the sample about the parameter\( \bar{x} \) for \( \mu \) of a normal population

Mnemonic: CUES (Consistency, Unbiasedness, Efficiency, Sufficiency).

Confidence intervals

For a mean with known \( \sigma \) (or large n):

$$\bar{x} \pm z_{\alpha/2}\frac{\sigma}{\sqrt{n}}, \qquad z = 1.645\ (90\%),\ 1.96\ (95\%),\ 2.58\ (99\%)$$

For a proportion: \( p \pm z\sqrt{pq/n} \). For a small sample with unknown \( \sigma \): \( \bar{x} \pm t_{n-1}\,s/\sqrt{n} \).

A 95% confidence interval means: if we repeated the sampling many times, about 95% of the intervals built this way would contain the true parameter. It does not mean the parameter has a 95% chance of lying in this one interval; the parameter is fixed.

Sample size for estimating a mean within a margin of error E:

$$n = \left(\frac{z\,\sigma}{E}\right)^2$$

Memory Tricks

Don’t Confuse

ThisvsThat
Parameter: population (\( \mu \), \( \sigma \))vsStatistic: sample (\( \bar{x} \), s)
Standard deviation: spread of individual valuesvsStandard error: spread of a statistic across samples
Sampling error: only in samples; falls with nvsNon-sampling error: in census too; does not fall with n
Stratified: random within every stratumvsQuota: non-random within categories
Stratified: homogeneous strata, all sampledvsCluster: heterogeneous clusters, some taken whole
Questionnaire: respondent fillsvsSchedule: enumerator fills
Exploratory: flexiblevsDescriptive: rigid, pre-planned
Classification: groupingvsTabulation: presenting in rows and columns
Caption: column headingvsStub: row heading
Point estimatevsInterval estimate with confidence level

Common Mistakes


Practice

All questions below are original practice questions written in the UGC NET format. They are not previous-year questions.

Level 1: Foundation

  1. Name Fisher’s three principles of experimental design.
  2. A population has SD 16. What is the standard error of the mean for samples of 64?
  3. In a table, what are row headings called?
Answers
  1. Replication, randomisation and local control.
  2. \( 16/\sqrt{64} = \) 2.
  3. Stubs (column headings are captions).

Level 2: Exam-style MCQs

Q1. (Practice) A researcher divides a city’s retailers into kirana stores, supermarkets and online sellers, and draws a random sample from each group in proportion to its size. The method is:

  • (a) Cluster sampling
  • (b) Quota sampling
  • (c) Proportional stratified sampling
  • (d) Systematic sampling
Answer and explanation

(c). Homogeneous groups, random selection within each group, in proportion to size. Without the random selection it would be quota sampling.

Q2. (Practice) A sample of 64 has mean 80; the population SD is 16. The 99% confidence interval for the population mean is approximately:

  • (a) 76.08 to 83.92
  • (b) 74.84 to 85.16
  • (c) 78.00 to 82.00
  • (d) 38.72 to 121.28
Answer and explanation

(b). SE \( = 16/8 = 2 \); \( 80 \pm 2.58 \times 2 = 80 \pm 5.16 \). Option (a) is the 95% interval; (d) uses σ (16) instead of the SE (2).

Q3. (Practice) To estimate a mean within ±2 units at 99% confidence when the population SD is 10, the minimum sample size is:

  • (a) 97
  • (b) 166
  • (c) 167
  • (d) 129
Answer and explanation

(c) 167. \( n = (2.58 \times 10/2)^2 = 12.9^2 = 166.41 \), rounded up to 167.

Q4. (Practice) Consider the following statements:

  1. Sampling error decreases as the sample size increases.
  2. Non-sampling errors are absent in a census.
  3. The standard error of the mean is inversely proportional to the square root of the sample size.

Which of the statements given above is/are correct?

  • (a) 1 and 3 only
  • (b) 1 and 2 only
  • (c) 2 and 3 only
  • (d) 1, 2 and 3
Answer and explanation

(a) 1 and 3 only. A census has no sampling error but does have non-sampling errors (non-response, recording, processing).

Q5. (Practice) Which research design is most flexible?

  • (a) Exploratory
  • (b) Descriptive
  • (c) Diagnostic
  • (d) Experimental
Answer and explanation

(a) Exploratory. Its aim is to discover ideas and hypotheses, so the design adapts as insights emerge. The others must be rigid to avoid bias.

Level 3: Assertion–reason, match and sequence

For assertion–reason questions, use these options:

  • (a) Both A and R are true, and R is the correct explanation of A
  • (b) Both A and R are true, but R is not the correct explanation of A
  • (c) A is true, but R is false
  • (d) A is false, but R is true

Q1. (Practice) Assertion (A): A z-based confidence interval for the mean can be used for a large sample even when the population is skewed. Reason (R): By the central limit theorem, the sampling distribution of the mean approaches normality as the sample size increases.

Answer and explanation

(a). The CLT is precisely what justifies the normal approximation for large samples.

Q2. (Practice) Assertion (A): The sample variance computed with divisor (n − 1) is an unbiased estimator of the population variance. Reason (R): An unbiased estimator always has the smallest variance among all estimators.

Answer and explanation

(c). A is true. R is false: smallest variance is efficiency, a separate property. An estimator can be unbiased without being efficient.

Q3. (Practice) Match List I with List II.

List I (Sampling method)List II (Feature)
A. SystematicI. Respondents refer other respondents
B. ClusterII. Every kth unit after a random start
C. SnowballIII. Whole groups selected at random
D. JudgementIV. Researcher selects typical units
  • (1) A-II, B-III, C-I, D-IV
  • (2) A-III, B-II, C-I, D-IV
  • (3) A-II, B-III, C-IV, D-I
  • (4) A-II, B-I, C-III, D-IV
Answer and explanation

(1) A-II, B-III, C-I, D-IV.

Q4. (Practice) Match the property of an estimator with its meaning.

List IList II
A. UnbiasednessI. Minimum variance
B. ConsistencyII. Uses all information in the sample
C. EfficiencyIII. Expected value equals the parameter
D. SufficiencyIV. Approaches the parameter as n grows
  • (1) A-III, B-IV, C-I, D-II
  • (2) A-IV, B-III, C-I, D-II
  • (3) A-III, B-I, C-IV, D-II
  • (4) A-III, B-IV, C-II, D-I
Answer and explanation

(1) A-III, B-IV, C-I, D-II.

Q5. (Practice) Arrange the parts of a statistical table as they appear from top to bottom:

  • A. Source note

  • B. Title

  • C. Body

  • D. Table number

  • E. Captions

  • (1) D, B, E, C, A

  • (2) B, D, E, C, A

  • (3) D, E, B, C, A

  • (4) D, B, C, E, A

Answer and explanation

(1) D, B, E, C, A. Number, title, (head note), captions, body, (footnote), source.


Quick Revision Table

ItemKey point
Research designsExploratory (flexible) · descriptive/diagnostic (rigid) · experimental (causal)
Fisher’s principlesReplication, randomisation, local control
Formal designsCRD, RBD, Latin square, factorial
Primary / secondaryFirst-hand / already collected; check reliability, suitability, adequacy
Questionnaire / scheduleRespondent fills / enumerator fills
ScalesNominal, ordinal, interval, ratio; Likert, semantic differential, Stapel, Thurstone, Guttman
ClassificationGeographical, chronological, qualitative, quantitative
Sturges’ rule\( k = 1 + 3.322\log_{10}N \)
Table partsNumber, title, head note, captions, stubs, body, footnote, source
Probability samplingSimple random, systematic, stratified, cluster, multistage
Non-probabilityConvenience, judgement, quota, snowball
SE of mean\( \sigma/\sqrt{n} \); FPC \( \sqrt{(N - n)/(N - 1)} \)
SE of proportion\( \sqrt{PQ/n} \)
CLTSampling distribution of the mean → normal as n grows (n ≥ 30 working rule)
Estimator propertiesUnbiased, consistent, efficient, sufficient (CUES)
CI\( \bar{x} \pm z\sigma/\sqrt{n} \); 1.645, 1.96, 2.58
Sample size\( (z\sigma/E)^2 \), round up

Key Takeaways

What’s Next

Advertisement
Advertisement