9 min readConcept sheetMCQ · classify the method, match and sequence

Data collection, questionnaire design and sampling

Primary data is collected first-hand; secondary data already exists. A good questionnaire avoids leading and double-barrelled questions. Probability sampling gives every unit a known, non-zero chance of selection.

Primary vs secondary dataPrimary: collected fresh for the study (survey, observation, experiment). Secondary: already published or recorded (Census, RBI bulletins, company records)

Questionnaire vs scheduleA questionnaire is filled in by the respondent; a schedule is filled in by the enumerator

Probability methodsSimple random, systematic, stratified, cluster, multistage

Non-probability methodsConvenience, judgement (purposive), quota, snowball

Methods of collecting primary data

Method	How it works	Strength	Limitation
Observation	The researcher watches behaviour, e.g. shoppers in a store	Records what people do, not what they say	Cannot capture motives; observer bias
Personal interview	Face-to-face questions, structured or unstructured	Deep answers; can probe and clarify	Costly and slow; interviewer bias
Telephone or online survey	Questions by phone, email or web form	Fast and cheap over wide areas	Low response; reaches only connected people
Mailed questionnaire	The respondent fills it in and sends it back	Cheapest; respondents can take their time	Low response rate; needs literate respondents
Schedule	An enumerator asks the questions and fills in the form	Works with illiterate respondents; high response	Needs trained enumerators; costly
Focus group	A small group (typically about 6 to 12 people; sources differ) discusses a topic with a moderator	Rich ideas for exploratory research	Small and not representative; a few may dominate
Experiment	Change one variable and measure the effect, e.g. a price test	Shows cause and effect	Artificial settings; costly

Questionnaire vs schedule

QuestionnaireFilled in by the respondent, usually sent by post or online. Cheap and covers a wide area, but response is low and respondents must be literate.

ScheduleFilled in by an enumerator, who asks the questions face to face. Costlier, but response is higher and it works with illiterate respondents, as in the Census.

NET often asks who fills in the form. That one fact separates the two.

Designing a questionnaire

Textbooks such as Malhotra give a sequence like this one.

1
Specify the information needed from the research objectives.
2
Decide how it will be administered: in person, by telephone, by mail or online.
3
Decide the content of each question: is it needed, and can the respondent answer it?
4★
Choose the question form: open-ended, or closed (dichotomous, multiple choice, scales).
5★
Word each question simply. Avoid leading, loaded, ambiguous and double-barrelled questions.
'Rate the price and quality' asks two things at once
6
Order the questions: easy openers, general before specific (the funnel), sensitive and classification questions last.
7
Design the layout, pre-test (pilot) the draft on a small sample, and revise it.

★ one-way, controls the pace

Scales of measurement (Stevens, 1946) and attitude scales

Scale	Key idea	Example or author
Nominal	Labels or categories only; the mode is the right average	Gender, region
Ordinal	Rank order; median and percentiles allowed	Ranking five brands
Interval	Equal intervals but no true zero; mean and SD allowed	Temperature in °C
Ratio	Equal intervals and a true zero; all statistics allowed	Income, sales, age
Likert	Agree-disagree statements, usually 5 points, summed into a score	Rensis Likert, 1932
Semantic differential	Bipolar adjectives at the two ends of a 7-point scale	Osgood, Suci and Tannenbaum, 1957
Thurstone	Statements placed at equal-appearing intervals by judges	L. L. Thurstone
Guttman	Cumulative: agreeing with a strong item implies agreeing with milder ones	Louis Guttman
Stapel	Unipolar, −5 to +5 with no zero, around a single adjective	Jan Stapel

The sampling process

The sampling frame is the list from which the sample is drawn, such as a voter list or a customer database.

Sampling techniques

Technique	Type	How it works	Example
Simple random	Probability	Every unit has an equal chance; lottery or random numbers	Drawing 50 roll numbers by lottery
Systematic	Probability	Random start, then every kth unit, where k = N ÷ n	Every 20th invoice
Stratified	Probability	Split into homogeneous strata; draw a random sample from each	Students by course, with proportional allocation
Cluster	Probability	Split into groups (often areas); pick whole clusters at random	Picking 10 wards of a city and surveying every home in them
Multistage	Probability	Sampling in stages, from larger to smaller units	State → district → village → household
Convenience	Non-probability	Whoever is easiest to reach	Mall-intercept survey
Judgement (purposive)	Non-probability	The researcher picks units thought typical or expert	Interviewing chosen industry experts
Quota	Non-probability	Fixed numbers per category, filled non-randomly	50 men and 50 women, chosen by the interviewer
Snowball	Non-probability	Each respondent refers others	Hard-to-reach groups, such as gig workers

Stratified vs cluster sampling

StratifiedGroups are homogeneous within and differ from each other. A random sample is drawn from every stratum. Improves precision.

ClusterGroups should be heterogeneous within, each like a mini-population. Some whole clusters are picked at random. Cuts cost, but is usually less precise.

Worked example: stratified allocation, systematic interval and standard error

Problem

A college has 1,200 students: 600 in BCom, 400 in BBA and 200 in BSc. (a) Allocate a sample of 60 proportionally. (b) Find the systematic sampling interval for 60 from the full list. (c) The SD of students' daily spending is ₹120. Find the standard error of the mean for n = 36.

Working

(a) Each stratum gets n × stratum size ÷ N. BCom = 60 × 600 ÷ 1,200 = 30; BBA = 60 × 400 ÷ 1,200 = 20; BSc = 60 × 200 ÷ 1,200 = 10. Total = 60.
(b) k = N ÷ n = 1,200 ÷ 60 = 20. With a random start of 7, pick 7, 27, 47, … up to 7 + 59 × 20 = 1,187.
(c) SE = σ ÷ √n = 120 ÷ √36 = 120 ÷ 6 = ₹20.
If n rises to 144, SE = 120 ÷ 12 = ₹10: a sample four times as large halves the error.

Answer

30, 20 and 10 students; k = 20; SE = ₹20 (₹10 with n = 144).

Sampling vs non-sampling error

Sampling errorArises because only part of the population is studied. It falls as n rises (SE = σ ÷ √n) and is zero in a census.

Non-sampling errorArises from faulty frames, non-response, wrong answers, interviewer bias or data entry. A bigger sample does not cure it, and it occurs in a census too.

How NET asks this

NET asks you to classify a method (probability or not, primary or secondary), match techniques with descriptions, sequence the sampling process, and spot faulty questions such as double-barrelled or leading ones.

Exam angle

Typical MCQ: “Is quota sampling a probability method?” No. It looks like stratified sampling, but the interviewer, not chance, picks the units within each quota. Trap: equating quota with stratified sampling.

Typical MCQ: “If the sample size is made four times larger, what happens to the standard error of the mean?” It halves, because SE = σ ÷ √n and √4 = 2. Trap: saying it falls to one-fourth.

Typical MCQ: “Who fills in a schedule?” The enumerator, who asks the questions. A questionnaire is filled in by the respondent. Trap: treating the two as the same.

Remember it

Probability sampling: 'Some Students Study Commerce Methodically': Simple random, Systematic, Stratified, Cluster, Multistage. Non-probability: 'Can't Judge Quality Samples': Convenience, Judgement, Quota, Snowball.