Imagine you are a doctor checking the health of 10 crore Indians. You cannot run tests on every person — you pick a manageable group, draw conclusions, and project them back to the full population. That is sampling in one sentence.
Formally, a population is the complete set of all units under study (denoted N). A sample is a subset of that population (denoted n). The numerical characteristic of the population is a parameter (fixed, usually unknown), and the corresponding measure computed from the sample is a statistic (varies from sample to sample).
The gap between what your sample tells you and what the true population parameter actually is — that gap is called sampling error. It is not a mistake; it is an unavoidable consequence of not studying the whole population. Minimising it is the whole art of sampling theory.
Why does this matter for SSC CGL? The Statistics paper tests you on identifying sampling methods from scenario descriptions, computing standard errors, allocating sample sizes under stratified sampling, and finding systematic sampling units. These are mechanical once you know the framework — but students who confuse the methods under exam pressure lose easy marks.
The analogy that locks this in: think of a population as a huge pot of rice on the stove. Simple random sampling is grabbing a fistful from anywhere. Stratified sampling is ladling from each layer (top, middle, bottom) proportionally. Systematic sampling is scooping every 20th grain in a row. Each method has tradeoffs in cost, precision, and practical feasibility — and the exam will give you a scenario and ask you to name the method or compute a number from it.
Every unit in the population has an equal and independent probability of selection. This is probability sampling at its purest.
Two variants:
Standard Error of the sample mean $\bar{x}$:
$SE(\bar{x}) = \dfrac{\sigma}{\sqrt{n}}$$SE(\bar{x}) = \dfrac{\sigma}{\sqrt{n}} \cdot \sqrt{\dfrac{N-n}{N-1}}$The term $\sqrt{\dfrac{N-n}{N-1}}$ is the Finite Population Correction (FPC) factor. When N is very large relative to n, FPC is close to 1 and can be ignored — which is why in SRSWR (where we treat N as effectively infinite), the FPC disappears entirely.
Look — in SRSWR the population size N drops out of the formula completely. The exam loves this distinction.
The population is divided into non-overlapping, exhaustive subgroups called strata. Random samples are then drawn independently from each stratum.
Why bother? Because if one stratum is very different from another (e.g., urban vs. rural students), a purely random sample might over-represent one group. Stratification forces representation.
Proportional Allocation: Sample size from stratum h is:
$n_h = n \cdot \dfrac{N_h}{N}$
where $N_h$ is the size of stratum h and N is total population size. Every stratum contributes to the sample in proportion to its share of the population.
Optimum (Neyman) Allocation: $n_h \propto N_h \sigma_h$ — strata with larger size or larger internal variation get more sample units. SSC CGL rarely tests Neyman allocation numerically but you should know it by name.
Line up the N population units. Compute the sampling interval $k = N/n$ (always an integer in exam problems). Randomly pick a starting unit r where $1 \leq r \leq k$. Then select units: r, r+k, r+2k, r+3k, ...
The i-th selected unit = $r + (i-1) \cdot k$
This is clean, operationally easy (no need for a random number table after the first pick), and the exam reliably tests "find the 10th selected unit" type problems.
One hidden trap: if the population has a periodic pattern with the same period as k, systematic sampling can be badly biased. The exam occasionally tests this as a disadvantage.
The population is divided into clusters (geographic areas, schools, etc.). Entire clusters are selected at random, and every unit within the selected cluster is included.
The key contrast with stratified sampling: in stratified sampling you sample from every stratum; in cluster sampling you sample only the selected clusters, not all clusters.
An estimator is a function of sample observations used to estimate a population parameter. Good estimators have these properties:
$E(\hat{\theta}) = \theta$. On average, the estimator hits the true parameter. The sample mean $\bar{x}$ is an unbiased estimator of the population mean $\mu$.| Sampling Error | Non-Sampling Error | |---|---| | Due to chance selection of units | Due to mistakes in measurement, recording, processing | | Decreases as n increases | Does not necessarily decrease with larger n | | Defined: Statistic − Parameter | Can occur in a census too |
When a question asks which sampling method gives every member an equal chance, check two things: (1) is selection truly random with no grouping? (2) is probability of selection the same for all units? If both yes, it is Simple Random Sampling. Eliminate Stratified (groups exist, unequal probabilities possible), Cluster (whole clusters selected, individuals in unselected clusters have zero chance), and Systematic (only the first unit is chosen truly randomly). This elimination cuts a 30-second reading question to 10 seconds.
Every systematic sampling problem follows one formula: i-th unit = $r + (i-1) \cdot k$, where $k = N/n$. Memorise this as "Random start plus (position minus one) times interval." For "10th unit, start 7, N=500, n=50": k = 10, answer = 7 + 9×10 = 97. No need to list all 9 previous units. Standard method (listing): 45 seconds. Formula method: 8 seconds.
Proportional allocation = one multiplication: $n_h = n \times \dfrac{N_h}{N}$. The ratio $N_h / N$ is the stratum's share of the population. Compute this share first, then multiply by total sample size n. For N=1000, strata of 400/300/200/100, total sample n=100: shares are 0.4, 0.3, 0.2, 0.1, so samples are 40, 30, 20, 10. Mental arithmetic once you see "proportional allocation" — standard calculation: 40 seconds, pattern recognition: 12 seconds.
SRSWR = no FPC. SRSWOR = multiply $\sigma/\sqrt{n}$ by $\sqrt{(N-n)/(N-1)}$. The exam always tells you which one it is. If it says "with replacement" or asks you to compute SE and gives you only $\sigma$ and n (no N), it is SRSWR — just compute $\sigma/\sqrt{n}$. If N appears in a without-replacement scenario, apply FPC. This saves you from hunting for N when it is irrelevant: SRSWR SE for σ=20, n=100 is 20/10 = 2 in 5 seconds flat.
Stratified: you go IN to every stratum and sample. Cluster: you pick clusters OUT and take everything inside. One word test — if the question says "divided into groups and then randomly selected FROM EACH group," it is Stratified. If it says "groups were randomly selected and all members of chosen groups were included," it is Cluster. This keyword scan takes 5 seconds and eliminates both wrong answers simultaneously.
When you see a sampling question in the exam hall, run this check in order:
Is a formula asked? (SE, sample size, unit number) → Identify SRSWR/SRSWOR/Systematic/Stratified, then plug directly into the relevant formula. No prose needed.
Is a method being identified from a scenario?
Is a property of estimators asked? Check: unbiased = $E(\hat\theta) = \theta$; efficient = minimum variance among unbiased estimators; consistent = converges to $\theta$ as n increases.
Is the question about errors? Statistic minus Parameter = Sampling Error. Anything else (recording errors, non-response) = Non-Sampling Error.
If you cannot place the question within 15 seconds, mark and skip. These questions reward preparation, not in-exam derivation.
Why this question: This is the fundamental definitional question. If you cannot answer it, every identification question in the paper is at risk.
Solving path: Apply the equal-probability test. Simple Random Sampling is the only method where selection probability is the same for every single unit and independent of grouping. Stratified involves groups, Systematic has a fixed pattern after the first pick, Cluster selects whole groups. Answer: Simple Random Sampling.
Why this question: Stratified vs. Cluster is the most commonly confused pair. This question nails the defining feature of stratification.
Solving path: Key phrase — "divided into 5 groups... and then 20 students are randomly selected FROM EACH group." Sampling from every subgroup = Stratified. If only some groups were selected and all members of those groups included, it would be Cluster. Answer: Stratified Sampling.
Why this question: Standard error computation under SRSWR is a direct formula application. The exam tests whether you know N is irrelevant here.
Solving path: SRSWR formula: $SE = \sigma / \sqrt{n} = 20 / \sqrt\{100\} = 20/10 = 2$. Population size N = 500 is a distractor — in SRSWR, N never appears. Answer: 2.
Why this question: Systematic sampling unit calculation. The exam gives you a random start and asks for a specific unit number.
Solving path: $k = N/n = 1000/50 = 20$. Second unit = first unit + k = 8 + 20 = 28. Answer: 28.
Why this question: The sampling error definition is a guaranteed 1-mark question. Do not confuse it with non-sampling error or standard error.
Solving path: Sampling error = difference between a population parameter and the corresponding sample statistic. It arises purely from chance selection. Non-sampling error arises from data collection/recording mistakes, and can occur even in a census. Standard error is a measure of variability of the statistic. Answer: Sampling Error.
Why this question: Proportional allocation with equal strata is the simplest case but still catches students who confuse the formula.
Solving path: Sampling fraction = $n/N = 50/500 = 1/10$. Each stratum has 100 units. Units from each stratum = $100 \times (1/10) = 10$. Or equivalently: $n_h = n \times (N_h/N) = 50 \times (100/500) = 10$. Answer: 10.
Why this question: Proportional allocation with unequal strata — the harder version. Tests whether you can apply the formula when strata sizes differ.
Solving path: $n_3 = n \times (N_3/N) = 100 \times (200/1000) = 100 \times 0.2 = 20$. Answer: 20.
Why this question: The "find the i-th unit" systematic sampling question. This is the harder variant — position 10, not position 2.
Solving path: $k = N/n = 500/50 = 10$. i-th unit = $r + (i-1) \times k$. 10th unit = $7 + (10-1) \times 10 = 7 + 90 = 97$. Answer: 97.
Confusing Stratified and Cluster sampling. The distinguishing rule is simple: stratified = you sample from every group; cluster = you select only some groups and take all their members. Students who rely on the word "divided into groups" alone pick the wrong answer half the time.
Including N in the SRSWR standard error formula. In SRSWR, the formula is $\sigma/\sqrt{n}$, full stop. The finite population correction only applies in SRSWOR. When the exam gives you N in a SRSWR question, it is a deliberate distractor.
Adding k repeatedly instead of using the position formula. For the 10th systematic unit, students list all 9 previous units. Use $r + (i-1) \times k$ directly. Listing wastes 30+ seconds and introduces arithmetic errors.
Treating sampling error as a mistake. Sampling error is the inherent difference between a statistic and the true parameter due to chance selection — it is not caused by errors or carelessness. Non-sampling error is caused by mistakes. The exam tests this distinction with definition questions.
Misapplying proportional allocation when strata are unequal. Students assume equal strata and compute $n/\text{number of strata}$. Always use $n_h = n \times (N_h/N)$. When strata are equal, this formula happens to give the same result — but applying it mechanically saves you when strata differ.
Confusing Standard Error with Sampling Error. Standard Error is the standard deviation of the sampling distribution of a statistic (a measure of precision). Sampling Error is the difference between the parameter and the statistic for a specific sample (a realised value). The exam has offered both as options in the same question.