1. Plain-English Intuition: Tasting the Soup and Gathering the Evidence
Imagine a chef cooking a massive 50-liter cauldron of vegetable soup for a banquet:
- To find out if the soup has enough salt, the chef does not drink all fifty liters.
- Instead, the chef thoroughly stirs the pot so that ingredients are evenly dispersed, dips in a single tablespoon, and tastes it. If the pot was properly stirred, that single spoonful accurately represents the entire 50-liter pot.
- In social research, Sampling is that spoonful.
- If a sociologist wants to know the voting preferences or fertility attitudes of 100 million citizens, surveying all 100 million (a complete Census) would cost billions of rupees and take five years. By drawing a scientifically randomized sample of 2,000 citizens, the researcher can estimate the attitudes of the entire nation with a margin of error of less than three percent.
- But once you choose whom to study, how do you gather the truth? Do you mail them a written Questionnaire, sit down for an intimate In-depth Interview, immerse yourself in their daily life through Participant Observation, or combine multiple methods (Triangulation)?
This module provides the comprehensive toolkit of Sampling Designs, Data Collection Techniques, and Measurement Reliability and Validity that every empirical sociologist relies upon.
2. Sampling Foundations: Logic, Terminology, and Errors
CANONICAL SAMPLING VOCABULARY
| Term | Methodological Definition & Concrete Example |
|---|---|
| Universe / Population | The total aggregate of all elements, persons, or units conforming to a set of specifications (e.g., all registered voters in Uttar Pradesh). |
| Sample | A representative subset drawn from the target population to estimate characteristics of the whole. |
| Sampling Unit | The elementary single unit selected for study (e.g., individual, household, village, school, or factory). |
| Sampling Frame | The actual physical or electronic master list of all units in the target population from which the sample is drawn (e.g., electoral roll, voter list telephone directory, university admission register). |
| Sampling Error | The mathematical discrepancy between sample statistic and true population parameter; decreases as sample size (n) increases. |
| Non-Sampling Error | Human errors (poor questions, non-response, recording bias, cheating); increases with the overall size of the survey (highest in full censuses). |
The Iron Law of Errors:
- Sampling Error: Occurs solely due to examining a sample rather than the entire population. It is inversely proportional to sample size: as sample size (n) increases, sampling error decreases.
- Non-Sampling Error: Occurs due to human blunders: ambiguous questions, poorly trained interviewers, dishonest responses, data-entry typos, and non-response. Crucially, non-sampling error increases as survey size expands because managing a massive team across millions of households multiplies operational chaos.
3. Probability vs. Non-Probability Sampling
A. Probability Sampling Techniques
1. Simple Random Sampling (SRS)
Every single unit in the population has a known, equal, and independent non-zero probability of being included in the sample:
- Methods of Selection:
- Lottery Method: Placing identical numbered slips into a container, shuffling thoroughly, and drawing blindfolded.
- Random Number Tables: Using standardized mathematical tables of random digits, notably L.H.C. Tippett’s Table of Random Numbers (1927) (10,400 four-figure numbers), or tables by Fisher & Yates, and Kendall & Babington Smith.
- Limitation: Requires a complete, up-to-date sampling frame of every individual unit, which is rarely available for large national populations.
2. Systematic Random Sampling
Units are selected at regular numerical intervals from an ordered list:
- Sampling Interval ((k)): Calculated as the population size ((N)) divided by the desired sample size ((n)):
- Procedure: A Random Start is selected between (1) and (k) (e.g., using a random number table). If the random start is (4) and (k = 10), the sample consists of elements: (4, 14, 24, 34, 44, \dots)
- The Danger of Periodicity: If the sampling frame contains a recurring cyclical pattern that coincides with the interval (k) (e.g., if every 10th house on a street is a corner house with higher wealth), severe systematic bias is introduced.
3. Stratified Random Sampling
Used when the population is heterogeneous across key social subgroups:
- Procedure: The population is divided into mutually exclusive, internally homogeneous subpopulations called Strata (e.g., stratifying by Caste: General, OBC, SC, ST; or by Income tiers). A simple random sample is then drawn independently from within each stratum.
- Proportionate Stratified Sampling: The sample size drawn from each stratum is directly proportional to that stratum's share in the total population.
- Disproportionate Stratified Sampling: Used when a critical minority group is so small in the population (e.g., Parsis or transgender individuals forming 0.1%) that a proportionate sample would yield too few cases for statistical analysis. The researcher over-samples this group, applying statistical weighting later.
4. Cluster Sampling
Used when a complete sampling frame of individuals does not exist, but the population is naturally grouped into geographical or institutional Clusters (e.g., city blocks, villages, school classrooms):
- Procedure: The researcher selects a random sample of clusters (e.g., 20 random villages in a district), and then interviews all individuals or a random sample within the selected clusters.
- Contrast with Stratified Sampling:
- Stratification: Internally homogeneous strata; samples are drawn from all strata.
- Cluster: Internally heterogeneous clusters; only a sample of clusters is selected.
5. Multi-Stage Sampling
Extends cluster sampling across several geographical tiers for large-scale national surveys (e.g., National Family Health Survey / NFHS):
- Stage 1: Random selection of States.
- Stage 2: Random selection of Districts within selected States.
- Stage 3: Random selection of Villages / Urban Wards within selected Districts.
- Stage 4: Random selection of Households within selected Villages.
B. Non-Probability Sampling Techniques
1. Purposive (Judgmental) Sampling
The researcher uses their professional sociological judgment to deliberately select cases that are exceptionally typical, expert, or illustrative of the phenomenon (e.g., selecting senior trade union leaders to understand union history).
2. Quota Sampling
The non-probability equivalent of stratified sampling:
- The population is divided into categories (e.g., 50% men, 50% women; 30% youth).
- Interviewers are assigned a Quota (e.g., "Interview 20 women under age 25"). Unlike stratified sampling where subjects are chosen by random tables, here the interviewer selects anyone who meets the criteria until the quota is filled. Common in commercial market research and exit polls.
3. Snowball Sampling (Network / Chain-Referral)
Designed specifically for hidden, deviant, stigmatized, or hard-to-reach populations where no sampling frame exists:
- The researcher identifies one or two initial informants (e.g., an active intravenous drug user, an undocumented refugee, or an elite corporate lobbyist).
- After gaining their trust, the researcher asks them to introduce or refer other members of their underground network.
- The sample expands through referrals like a rolling snowball.
4. Convenience (Accidental) Sampling
Selecting cases simply because they are easily accessible to the researcher (e.g., a professor surveying students sitting in their own classroom, or a journalist interviewing people walking past a railway station). It carries near-zero external validity.
4. Data Collection Tools: Observation, Survey, and Interview
COMPARATIVE MATRIX OF PRIMARY DATA COLLECTION TOOLS