Module 3: Sampling — Notes
Lesson 1: Learning Objectives
Module Overview:
The module focuses on sampling theory, sampling distributions, and estimation, with the following key learning objectives:
Learning Objectives
- Understand Sampling, Sampling Distribution and Estimation, Types of Sampling
- Explain Central Limit Theorem, Sampling Distribution of the Mean and Sampling Distribution of Proportion
- Discuss Estimation Types and Statistic for Estimating Population Mean
- Describe Confidence Interval, Estimating Population Mean Using t Statistic and Confidence Interval Estimation for Population Proportion
- Learn Relevant Industry Example and Case Study
Structure: 21 lessons, covering 3.1_Sampling/Sampling Distribution/Estimation topics (3.1.1 through 3.1.10+): Introduction of Sampling, Types of Sampling, Non-Sampling Errors and Precautions, Central Limit Theorem, Sampling Distribution of the Mean, Sampling Distribution of Proportion, Estimation Introduction, Types of Estimation, Statistic for Estimating Population Mean, Confidence Interval.
Lesson 2: 3.1 Sampling, Sampling Distribution and Estimation
Context: Reading Time: 2 Minutes 50 seconds | SLM Page 74-75
How Does Sampling Work?
Sampling = the procedure by which researchers select a predefined number of observations from a larger population — enables using a tiny percentage of the population to study a big group. Depending on the study type, the sample strategy may involve systematic or basic random sampling. Frequently used in statistics, psychology, and finance.
What Is a Sampling Distribution?
A sampling distribution is a statistical probability distribution derived from a greater number of samples taken from a certain population — the distribution of frequencies of various possible outcomes for a population statistic. Enables organisations (governments, corporations) to base judgements on acquired information. One key technique: the sampling distribution of the mean.
Estimation
The population = all the study's relevant components. Collecting data from the whole population is often unfeasible (time/expense/constraints), so a sample (a subset) is used instead — statistical inference is the process of drawing conclusions about a population from a sample.
| Population (Parameters) | Sample (Statistics) |
|---|---|
| Population mean | Sample mean |
| Population variance | Sample variance |
| Population proportion | Sample proportion |
Two types of estimates: - Point estimate — a single sample-statistic value used as a lone estimate of a population parameter; says nothing about accuracy/quality. - Interval estimate — preferred by statisticians, since it comes with a stated confidence level that the interval contains the population parameter. Confidence intervals are population-parameter estimates expressed as intervals.
Pure reading lesson, no quiz/widgets.
Lesson 3: 3.1.1 Introduction of Sampling
Context: Lesson 3 of 21. Reading Time: 2 Minutes 58 Sec | SLM Page No. 75-77. Video tutorial (started only).
Meaning
- Sampling is the process of choosing a small number of components from a larger, well-defined group, expecting the data from the small group to allow conclusions about the larger group. The primary idea is to derive conclusions about the total population by choosing a subset of its members; it is used when a census is not feasible or reasonable.
- Sampling = choosing a portion of an aggregate (universe) to base a judgement or conclusion about the whole; inferring information about a population from a sample.
Purpose of sampling (5 numbered hotspots)
- Lower cost - a sample-based study costs far less than a census.
- More accurate results - it is commonly accepted (and supported by research) that sampling data improves study quality more than census data does.
- Accessibility - often not feasible to gather data from every element; some elements are hard or impossible to reach, so a representative subset makes otherwise inaccessible populations studiable.
- Faster data gathering - collection is quicker and shortens the time between needing information and getting it.
- Population element availability - sometimes sampling is necessary: e.g. testing breaking strength destroys the material, so a census is impossible; if the population is infinite, sampling is the only possible procedure.
Features of the sampling method (6 accordion items)
- Economy - cost control: far fewer physical resources and less time than a census.
- Reliability - with due diligence in choosing sample units and a homogeneous topic, a sample survey can have almost the same reliability as a census.
- Detailed study - few units allow intensive study, with multiple approaches applied.
- Scientific base - the underlying theory rests on statistical principles.
- Greater suitability in most situations - wide applicability since examining a few units normally suffices.
- Accuracy - the degree to which bias is reduced; when sample elements are generated correctly, some underestimate and others overstate population values.
Essentials of sampling
- Accurate (representative) - same properties as the universe it was drawn from.
- Homogeneity - identical in nature to the universe; no differences.
- Sufficient samples - enough items for a dependable, representative result.
- Optimisation - best efficiency at the lowest cost; larger samples give greater efficiency but cost more, so choose a suitable size.
Quizzes
- What is the primary reason managers use sampling in decision-making? -> To gain insights efficiently without studying the entire population
- Which of the following best defines a sampling error? -> The difference between sample characteristics and population characteristics due to chance
Lesson 4: 3.1.2 Types of Sampling
Context: Lesson 4 of 21. Reading Time: 18 Minutes | SLM Page No. 77-85. Video tutorial (started only). Longest lesson of the module (26 hotspots, 2 tab sets).
Two broad categories of sample design
- Element selection - each element is chosen separately and directly from the population (unrestricted sampling).
- Representation - elements are selected on a probability basis or by other means; restricted sampling covers all other types with added controls.
flowchart TD A[Sampling designs] --> B[Probability sampling] A --> C[Non-probability sampling] B --> B1[Simple random] B --> B2[Systematic random] B --> B3[Stratified random] B --> B4[Cluster] B --> B5[Sequential / multiphase] B --> B6[PPS] C --> C1[Convenience] C --> C2[Judgement] C --> C3[Quota] C --> C4[Snowball]
Probability sampling
- Every unit in the target population has a known, non-zero probability of selection (may or may not be equal depending on the design). Results can be extrapolated to the target population within a margin of error. Units are chosen at random; the sample frame and target population must be defined precisely.
- Sampling efficiency = trade-off between sampling cost and precision (precision = degree of uncertainty about the attributes measured). Cost rises with precision, sampling mistakes fall; researchers should create the most efficient design.
Simple random sampling (SRS)
- Every member has an equal chance, each choice independent; every possible sample combination is equally likely.
- Example: choose n = 10 of N = 30 employees -> each has 10/30 = 0.333 chance. Probability of selection = n / N. Random-number tables or software can draw the sample.
- Advantages: results generalise within a predetermined margin of error; objective estimates of population features; each unit has an equal and known chance regardless of sample size.
- Disadvantages: hard to obtain an accurate, up-to-date, complete list of the target population; for huge populations identifying every unit is laborious and costly - best suited to small populations.
Systematic random sampling
- Start at a random element between 1 and k, then draw every kth element - the skip interval (sampling ratio) = population size / sample size.
- Steps: determine population size; compute the sampling ratio; select every kth entry.
- Two cautions: the list's natural order must be unrelated to the trait studied; the skip interval must not coincide with a systematic pattern in the population.
- Advantages: versatility and simplicity - no need to number a large list; cheap and quick given available lists. Disadvantages: possible undetected patterns in the data (sample may not be representative); the researcher must know the exact population size, which is hard when it is very large/unknown.
Stratified random sampling
- Divide the target population into groups (strata) and sample from each. Useful when the population divisions are skewed or the distribution has extremes. Aim: reduce variability within each stratum, increase differences between strata; ideal stratification uses the main variable studied. Produces homogeneous smaller groups; a representative sample is drawn from each.
- Three reasons to use it: improve statistical efficiency; supply enough data to examine subpopulations; allow different research techniques/protocols in different strata.
Cluster sampling
- Divides units into mutually exclusive and collectively exhaustive subpopulations (clusters); each cluster is thought to represent the population's heterogeneity -> intergroup homogeneity and intragroup heterogeneity; randomly select clusters and survey every member of the chosen clusters.
- Single-stage: pick clusters at random and examine every element. Multistage: repeat cluster sampling over several phases (e.g. urban/semi-urban/rural areas -> sections within each -> banks within each area).
- Advantages: cost-effective and operationally feasible; a list of clusters (phone exchanges, blocks, regions) is easier to build than a list of all units. Disadvantages: clusters tend to be homogeneous, so estimates are less accurate when intra-/inter-cluster conditions aren't met; hence less common.
- Area sampling: clusters are geographic units (state, district, city, block); cheaper and frame-independent (a city map of blocks is enough).
Sequential / multiphase (double) sampling
Use a first sample to gather early data, then a subsample of it to investigate further; practical and cost-effective.
Sampling with Probability Proportional to Size (PPS)
When clusters have unequal numbers of elements, select clusters with probability proportional to their size; list cumulative element counts, pick numbers systematically - the numbers identify which cluster to take and how many members, drawn by systematic or simple random sampling. Gives results like SRS, cheaper and faster.
Non-probability sampling
No known chance of selection for population elements, so findings cannot be generalised; used for quick, cheap preliminary data or when population size is unknown. - Convenience: collect from easily accessible people; typical in the exploratory phase; assumes the population is homogeneous; representativeness cannot be measured; mind non-response/self-selection error. Pros: many respondents quickly. Cons: can damage validity/reliability of instruments during scale development; results can't be generalised; sampling error can't be estimated. - Judgement (purposive): participants chosen on an expert's opinion of who best meets the study criteria; suited to early exploratory research or when information sits with a specific group; better than convenience if the judgement is right, but representativeness still can't be gauged and data are rudimentary. - Quota: participants chosen to fill predetermined quotas by attitude, behaviour or demographics (gender, age, education, income, occupation); ensures defined subgroups are represented in desired proportions and can reduce selection bias, but results depend on the researcher's subjective judgement and true representativeness can't be measured. - Snowball (referral): early respondents help find others; first respondents chosen randomly, later ones non-randomly; used for small, distinct, hard-to-list populations (qualitative research). Cheaper, smaller samples, but higher risk of bias and no generalisation.
Stratified: strata are internally homogeneous, sample taken from every stratum. Cluster: clusters are internally heterogeneous, only some clusters are sampled (but surveyed fully).
Quizzes
- Which of the following is a characteristic of probability sampling? -> Each member of the population has a known chance of being selected
- Which sampling method involves dividing the population into subgroups and randomly sampling from each subgroup? -> Stratified Sampling
Lesson 5: 3.1.3 Types of Sampling, Non-Sampling Errors and Precautions
Context: Lesson 5 of 21. Reading Time: 1 minute 10 sec | SLM Page No. 85-86. Video tutorial (started only). 9 hotspots.
Sampling error (hotspots)
- Occurs when an analyst's sample does not accurately reflect the full population, so sample results differ from what the entire population would give.
- A sampling error is the difference between the sampled value and the true population value, whether or not the sample is representative or skewed.
- Even randomised samples have some sampling error because a sample is only an estimate of the population.
- Sampling errors can be reduced by increasing the sample size and ensuring the sample represents the whole population (course example: ABC Company streaming-subscription service).
Non-sampling error (hotspots)
- An error from data collection that makes the data differ from true values; different from sampling error.
- It can be random or systematic and is hard to spot in a survey, sample or census.
- Systematic non-sampling errors are worse than random ones - they can force a study/survey/census to be scrapped.
- More errors -> less reliable information; non-sampling errors increase bias.
Quizzes
- Which of the following is an example of a non-sampling error? -> Entering incorrect data during analysis
- What is the best way to reduce measurement error in a survey? -> Pilot-test the survey and train data collectors
Lesson 6: 3.1.4 Central Limit Theorem
Context: Lesson 6 of 21. Reading Time: 1 minute 02 sec | SLM Page No. 86. Video tutorial (started only).
- The Central Limit Theorem (CLT): the distribution of sample means approximates a normal distribution (bell curve) as the sample size gets larger, regardless of the shape of the population distribution.
- For a population with finite variance, the mean of all samples of the same size approximately equals the population mean; each sample mean distribution is roughly normal with variance about population variance / sample size.
- Sample sizes of 30 or more are considered sufficient for the theorem to hold.
- The average of the sample means (and their standard deviations) equals the population mean (and SD); a sufficiently large sample can predict population characteristics accurately.
CLT: sample means -> normal as n grows (n >= 30 rule of thumb); standard error = sigma / sqrt(n).
Quizzes
- What does the Central Limit Theorem (CLT) state? -> Sample means from any population will form a normal distribution if the sample size is large enough
- Which formula is used to calculate the standard error of the mean? -> SE = sigma / sqrt(n)
Lesson 7: 3.1.5 Sampling Distribution of the Mean
Context: Lesson 7 of 21. Reading Time: 1 Minute 15 Seconds | SLM Page 86-87. Video tutorial available (intro only per user policy).
Overview
A sampling distribution can produce unique values for different statistics (mean, standard deviation, etc.) when computing various statistical measures for each sample. Together with their respective frequencies, all such values of a given statistic (e.g., the mean) form the sampling distribution of the mean.
Example: Sampling in Manufacturing
Imagine a scenario where microchip testing is required in factories: - The universe (population) = all microchips produced - Testing each item is not feasible (time, cost, constraints) - Solution: Take a sample — a subset of items for testing - A sample = "small universe" — the quantity extracted from the population for testing
Key Concept: Sampling Distribution of the Mean
When you take a specific number of samples from a population and compute the mean for each sample: - Each sample provides a unique mean value - All these mean values, together with their frequencies, form the sampling distribution of the mean standard deviation
Quizzes
-
Q: What happens to the standard error of the mean as the sample size increases? - ✅ It decreases (standard error = σ / √n, so larger n → smaller SE) - Wrong options: It increases, It stays the same, It becomes zero
-
Q: Which of the following statements is true about the mean of the sampling distribution of the mean? - ✅ It is equal to the population mean (fundamental property of sampling distributions) - Wrong options: Always greater, Always less, Depends on sample size
Lesson 8: 3.1.6 Sampling Distribution of Proportion
Context: Lesson 8 of 21. Reading Time: 1 Minute 25 Seconds | SLM Page 87.
Overview
When taking repeated random samples of size n from a population where the proportion in the category of interest is p: - The mean of all sample proportions (p-hat) equals the population proportion (p) - This describes the sampling distribution of the sample proportion
Standard Deviation
The standard deviation of all sample proportions (spread of the distribution) is: $$σ_p = \sqrt{\frac{p(1-p)}{n}}$$
Key insight: Standard deviation falls with increasing sample size (n appears in denominator under square root).
Normality Condition
If the sample size n is large enough, the p-hat distribution becomes approximately normal, according to tradition: - np ≥ 10 AND n(1-p) ≥ 10
Quizzes
-
Q: What is the formula for the standard error of the sample proportion? - ✅ σp = √[p(1-p)/n] (standard error formula for proportions)
-
Q: According to CLT, when is the sampling distribution of proportion approximately normal? - ✅ When both np≥5 and n(1−p)≥5 (conditions for normal approximation)
Lesson 9: 3.1.7 Estimation: Introduction
Context: Lesson 9 of 21. Reading Time: 1 Minute 15 Seconds | SLM Page 87-88.
What is Estimation?
Estimation = the process of using sample data to estimate unknown population parameters (θ1, θ2, ... θk).
Example: Normal Distribution
If a random variable x follows a normal distribution with parameters: - x ~ N(μ, σ) — where μ is the mean and σ is the standard deviation - We may be interested in estimating μ, σ, or both from a random sample x1, x2, ... xn
Multiple Estimators Available
For any given parameter, multiple estimators exist. For example, to estimate the population mean μ, you could use: - Sample mean - Median - Mode - Geometric mean - Harmonic mean
Similarly, the sample standard deviation (s) can estimate the population standard deviation (σ).
Types of Estimation
Point Estimate = a single sample statistic used to estimate a population parameter (e.g., using the sample mean to estimate the population mean).
Quizzes
-
Q: Which of the following is a property of a good estimator? - ✅ It becomes more accurate as sample size increases (consistency property of estimators) - Wrong options: Always overestimates, Has largest variance, Avoids using data
-
Q: What does a 95% confidence interval mean? - ✅ 95% of such intervals from repeated samples will contain the true population parameter (correct interpretation of confidence intervals) - Wrong options: Parameter always in interval, 95% of population in interval, 95% chance sample mean is correct
Lesson 10: 3.1.8 Types of Estimation
Context: Lesson 10 of 21. Reading Time: 2 Minutes | SLM Page No. 88-89. Video tutorial (started only).
- Statisticians use sample statistics to estimate population parameters: sample proportions estimate population proportions; sample means estimate population means. An estimate can take two forms:
- Point estimate - a single statistic value representing a population parameter. E.g. the sample mean x-bar is a point estimate of the population mean mu; the sample proportion p-hat estimates the population proportion P. A population parameter theta is an unknown constant; from a random sample x1...xn we form an estimator theta-hat, which varies from sample to sample, so it is a random variable with its own sampling distribution.
- Interval estimate - states that a population parameter lies between two numbers, e.g. a < mu < b. Also called an estimate by a confidence interval: a range of values defined by two numbers that represent the expected values of the population parameter.
Quizzes
- Which of the following best describes a point estimate? -> A single value used to estimate a population parameter
- What happens to the width of a confidence interval when the sample size increases? -> It becomes narrower
Lesson 11: 3.1.9 Statistic for Estimating Population Mean
Context: Lesson 11 of 21. Reading Time: 1 Minute 40 Seconds | SLM Page No. 89. Video tutorial (started only).
- Estimating a population mean from a random sample is a very common task. If the population standard deviation sigma is known, the confidence interval for mu is based on the normally distributed sampling distribution of the sample mean (z-based).
- If sigma is unknown (and n small) the t-statistic is used - see Lessons 12-13.
Quizzes
- When should you use the t-distribution instead of the z-distribution for estimating the population mean? -> When the sample size is small and the population standard deviation is unknown
- What does the standard error of the mean (SEM) measure? -> The variability of sample means in the sampling distribution
Lesson 12: 3.1.10 Confidence Interval
Context: Lesson 12 of 21. Reading Time: 1 Minute | SLM Page No. 90. Video tutorial (started only, per policy). 6 numbered hotspots around a "CONFIDENCE INTERVAL" diagram.
The 6 hotspots (verbatim from the live course)
- Definition - A confidence interval gives an estimated range of values which is likely to include an unknown population parameter; the estimated range is calculated from a given set of sample data.
- Notation - The common notation for the parameter in question is θ. Often, this parameter is the population mean μ, which is estimated through the sample mean X̄.
- Confidence level C - The level C of a confidence interval gives the probability that the interval produced by the method employed includes the true value of the parameter θ.
- When σ is unknown - In many situations, the value of σ is unknown, thus it is estimated with the sample standard deviation, s; and/or the sample size is small (less than 30), and it is unsure whether data came from a normal distribution.
- Critical value - (In the latter case, the Central Limit Theorem can't be used.) In either situation, the z*-value cannot be used from the standard normal (Z-) distribution as a critical value anymore. It is essential to use a larger critical value than that, because of not knowing the data quantity.
- Formula - CI for one population mean: X̄ ± t*(n-1) · s/√n, where t(n-1) is the critical t-value from the t-distribution with n-1 degrees of freedom (n = sample size).
Use the t-distribution (not z) when the population SD is unknown and the sample is small; the CI is X̄ ± t* · s/√n with df = n − 1.
Quizzes
- Which factor leads to a narrower confidence interval? ✅ Larger sample size
- When should you use the t-distribution instead of the z-distribution for a CI for the mean? ✅ When the population standard deviation is unknown and sample size is small
Lesson 13: 3.1.11 Estimating Population Mean Using t-Statistic
Context: Lesson 13 of 21. Reading Time: 2 Minutes 5 Sec | SLM Page No. 90-92. Video tutorial (started only). 4 hotspots.
Reading
- Statistical analysis of two populations requires two samples. If two normal distributions have unknown variances and an experiment employs a small sample size, the t-test is performed to compare them: it determines whether there is a significant difference between the means of the two populations based on the samples drawn from each.
- Variables: n = sample size, x̄ = sample mean, μ0 (given value to be tested), s = sample standard deviation.
Hotspots on the t-statistic
- The statistic has a t-distribution instead of a normal distribution when the standard deviation of the sample is used in place of the standard deviation of the population.
- It is impractical to give a distinct curve table for each sample size because each has a unique t-distribution; essential t-values are given in a single table for a range of sample sizes at common alpha levels (0.10, 0.05, 0.01, ...).
- The t-distribution approximates the standard normal (z) distribution for very large data. In practice, whenever the population SD is unknown, t-distributions work best.
- Table values are arranged by degrees of freedom (df), not sample size. For a one-sample mean problem, df = n - 1.
Uses of the t-test
- One-sample location test - tests whether the mean of a normally distributed population has a value given in a null hypothesis.
- Two-sample location test - tests the equality of the means of two normally distributed populations.
- All such tests are generally called Student's t-tests; strictly, that term applies when the two populations' variances are assumed equal. Without that assumption it is Welch's t-test. Tests on non-overlapping units are called unpaired / independent-samples t-tests.
- Paired / repeated-measures t-test - tests whether the mean difference between two measurements on the same unit is zero (e.g. tumour size before vs after therapy). A related method tests whether a regression line's slope deviates noticeably from 0.
Quizzes
- When should the t-distribution be used instead of the z-distribution for estimating a population mean? ✅ When the population standard deviation is unknown and the sample size is small
- Which of the following is a key assumption for using the t-distribution? ✅ The population should be approximately normally distributed
Lesson 14: 3.1.12 Confidence Interval Estimation for Population Proportion
Context: Lesson 14 of 21. Reading Time: 1 Minute 10 sec | SLM Page No. 92. Video tutorial (started only). 3 flashcards.
- The confidence interval (CI) for a population proportion shows the statistical probability that a characteristic is likely to occur within the population.
- Video summary slide ("Factors affecting interval width"): larger n -> narrower interval (more precise); higher confidence level (1-α) -> wider interval (less precise); sample proportion p̂ near 0.5 -> interval is widest.
Flashcards (diabetes example)
- Estimating the proportion of people with diabetes: a diagnosis of diabetes is a "success" (outcome of interest), a lack of diagnosis is a "failure".
- The formula for the CI for a population proportion follows the same format as the estimate of a population mean. The standard deviation of the sampling distribution of the proportion is √(p̂(1−p̂)/n), so CI = p̂ ± Zα/2 · √(p̂(1−p̂)/n); Zα/2 is set by the desired confidence; the unknown p and q are replaced by the sample proportions p′ and q′.
- p̂ = x/n, where x is the number of successes in the sample and n is the sample size.
Quizzes
- Which condition must be met to use the normal approximation for a CI for a population proportion? ✅ n·p̂ ≥ 10 and n·(1−p̂) ≥ 10
- Which factor leads to a wider CI for a population proportion? ✅ Higher confidence level
Lesson 15: 3.1.13 Relevant Industry Example
Context: Lesson 15 of 21. Reading Time: 2 Minutes | SLM Page No. 93-94. Video: "Industry Examples" (agriculture - corn yield sampling: population = all 5,000 acres, sample = 100 acres measured; e-commerce - abandoned carts).
Worked example: smartphone ownership
A market research company surveys n = 500 randomly picked adults; x = 421 own smartphones. Find a 95% CI for the true proportion. - p′ = x/n = 421/500 = 0.842 (point estimate); q′ = 1 − 0.842 = 0.158 - CL = 0.95 -> α = 0.05 -> α/2 = 0.025 -> z0.025 = 1.96 - CI = p′ ± zα/2 · √(p′q′/n) -> 0.810 ≤ p ≤ 0.874 - Interpretation: 81% to 87.4% of adults in the city own a smartphone, at 95% confidence. - Rationale for 95%: the true population share would be contained in 95% of the confidence intervals created this way.
Quizzes
- Why is sampling used in the agricultural corn-yield example? ✅ Because measuring all 5,000 acres is too costly and time-consuming
- E-commerce example: standard error of the sample proportion of abandoned carts if p = 0.20 and n = 500? ✅ 0.0179
Lesson 16: 3.1.14 Case Study - Assurance of Quality in Pharmaceutical Production
Context: Lesson 16 of 21. Reading Time: 2 Minutes 25 seconds | SLM Page No. 94-95. Video started only.
- Context: a pharmaceutical firm tests and samples raw materials and finished goods in its quality-assurance process to meet legal requirements; sampling strategies grounded in statistical principles (e.g. the standard normal distribution) support well-informed quality decisions.
- Goal: demonstrate how industry sampling methods, such as the standard normal distribution, are used for quality control in pharmaceutical production.
Action plan (hotspots)
- Design of Sampling Plan - the QC group specifies how often and how to sample raw materials and final goods; the plan ensures enough representative samples from every batch.
- Sample Collection - samples of finished products (tablets, capsules) and raw materials (excipients, active pharmaceutical ingredients) are gathered per the plan; random sampling ensures they represent the whole batch.
Actions - Analytical statistics
- Quality tests (assays, purity, dissolution, microbiological limit tests) are run on the samples; results are analysed statistically against predefined acceptance criteria.
- For quantitative tests (e.g. assays) compute the mean and standard deviation of the results.
- If results are normally distributed, use the standard normal distribution: Z = (X − μ)/σ shows how many SDs each result lies from the mean.
- Acceptance standards: set from internal quality standards and regulation, e.g. ±2 SD from the mean; a Z-score inside that range is acceptable.
- Making choices: inside the range -> meets quality standard, can be distributed/processed; outside -> investigate (root-cause analysis, corrective action); the batch may be rejected, reworked or retested depending on the deviation.
- Conclusion: statistical sampling (e.g. the standard normal distribution) lets pharma firms monitor and maintain product quality through manufacturing, supporting informed acceptance decisions, regulatory compliance and patient safety.
Quizzes
- Education case study: why is the t-distribution used instead of z to estimate the population mean GPA? ✅ Because the population standard deviation is unknown and the sample size is small
- Healthcare case study: what is the sample proportion of patients satisfied with their care? ✅ 0.80
Lesson 17: Summary (module recap)
- Sampling is a fundamental idea used in all activities: choosing a small number of components from a larger, well-defined group, expecting the data from the small group to allow conclusions about the larger group.
- Sampling efficiency is the trade-off between sampling cost and precision, and it varies across probability sampling approaches. Precision = the degree of uncertainty about the attributes being measured. Cost and precision are strongly correlated, while sampling mistakes are negatively related to them.
- Systematic sampling: main advantages are versatility and simplicity - no need to number the entries in a large list before sampling; lists are available and the sample is drawn faster than with random sampling, so it is attractive and cost-effective.
- Single-stage cluster sampling: the population is split into manageable clusters and the required number of clusters is picked at random.
- A subsample can be chosen for additional research; sampling is a more practical and cost-effective way of gathering data.
Lesson 18: Glossary
| Term | Meaning (course wording) |
|---|---|
| Stratified random sampling | Divide the target population into distinct groups called strata and select the sample from each stratum |
| Cluster sampling | Probability technique that divides the sample units into mutually exclusive and collectively exhaustive subpopulations called clusters |
| Confidence interval | The CI for a population proportion shows the statistical probability that a characteristic is likely to exist within the population |
| Point estimate | A single statistic value representing a population parameter |
| Interval estimate | States that a population parameter lies between two numbers, which define an interval |
Lesson 19: Check Your Understanding (20 questions, passing mark 0%, scored 20/20)
See the Quiz Bank (Lesson 19 section) for all 20 questions with explanations. Themes: sample vs census (cost, efficiency), probability sampling, systematic (skip interval, hidden patterns), stratified (reduce variability within strata, represent subgroups), cluster/PPS, judgement sampling, simple random sampling (probability = n/N, needs a complete frame), sampling efficiency.
Lesson 20: Exercise (self-study, write in notebook)
- Discuss the value and significance of sampling in research and its key features; explain how sampling controls cost (fewer physical resources and less time than a census) with examples where the savings matter.
- Discuss advantages and disadvantages of simple random sampling.
- Explain stratified random sampling - key features, advantages, limitations.
- Discuss convenience sampling - characteristics, advantages, limitations.
- Discuss the Central Limit Theorem - significance, assumptions, practical implications.
Lesson 21: Learning Activities (self-study) and Further Readings & Bibliography
Learning Activities 1. For a research project collecting data from a population, research the types of sampling (random, stratified, cluster, convenience); in groups, study one technique in detail. 2. For a city household study using cluster sampling: discuss cluster sampling's key features and rationale; brainstorm potential clusters (neighbourhoods, districts, blocks) as sampling units.
Further readings / bibliography 1. Levin, Rubin, Rastogi, Siddiqui - Statistics for Management, Pearson, 7th ed., 2016. 2. Prem S. Mann - Introductory Statistics, 7th ed., Wiley India, 2016. 3. James, Witten, Hastie, Tibshirani - An Introduction to Statistical Learning with Applications in R, Springer, 2016. 4. Srivastava V.K. et al. - Quantitative Techniques for Managerial Decision Making, Wiley Eastern. 5. Levin & Kirkpatrick - Quantitative Approaches to Management, McGraw Hill Kogakusha. 6. (repeat of 2) Mann, Introductory Statistics. 7. Budnik, McLeavey, Mojena - Principles of Operations Research, AIT BS, New Delhi. 8. Sharma J.K. - Operations Research: Theory and Applications, Macmillan, New Delhi.
✨ Quick Revision Pack (Module 3)
60-second summary
- Sampling = studying a part to learn about the whole (population or universe). Needs a sampling frame (list of units).
- Probability sampling: simple random, systematic (every kth), stratified (proportional groups), cluster (whole groups), multistage, PPS. Non-probability: convenience (chunk), judgement (purposive), quota, snowball, consecutive.
- Errors: sampling error (chance difference, shrinks as n grows) vs non-sampling error (bias, non-response, measurement mistakes; does not shrink with n).
- Central Limit Theorem: sample means are approximately normal for large n (≥ 30) whatever the population shape; SE = σ/√n.
- Estimation: point estimate vs interval estimate (confidence interval = estimate ± margin of error).
Diagrams
Probability vs non-probability sampling
flowchart TB S["Sampling methods"] --> P["Probability (known chance)"] S --> N["Non-probability (judgement / convenience)"] P --> P1["Simple random"] P --> P2["Systematic (every kth)"] P --> P3["Stratified"] P --> P4["Cluster / multistage"] N --> N1["Convenience (chunk)"] N --> N2["Judgement (purposive)"] N --> N3["Quota"] N --> N4["Snowball"]
Stratified vs cluster
| Stratified | Cluster | |
|---|---|---|
| Groups are | Homogeneous inside, different between | Mini-versions of the population |
| You sample | Some units from every group | Whole groups chosen at random |
| Aim | Precision, represent all segments | Cost saving over wide areas |
Central Limit Theorem in one picture
flowchart LR A["Any population shape"] --> B["Take many samples of size n"] --> C["Plot the sample means"] --> D["Bell-shaped, mean = μ, SD = σ/√n"]
Memory tricks
Stratified = "layers of a cake" (take a slice from every layer). Cluster = "boxes of chocolates" (pick whole boxes). Systematic = "every kth person in the queue".
Quota sampling is non-random stratified sampling. Interviewers fill quotas by judgement.
Non-sampling errors are not reduced by a bigger sample; sampling error is. A large badly chosen sample can be worse than a small well-chosen one.
SE = σ/√n: to halve the standard error, quadruple the sample size.
Convenience sampling is quick and cheap but prone to bias (the sample may not represent the population). Combine with probability sampling to reduce bias.
Formula sheet
| Concept | Formula |
|---|---|
| Standard error of mean | σ/√n (use s if σ unknown) |
| SE of a proportion | √(p(1 - p)/n) |
| Confidence interval (mean) | x̄ ± z·σ/√n |
| Sample size (mean) | n = (zσ/E)² |
| Sample size (proportion) | n = z² p(1 - p)/E² |
| Systematic sampling interval | k = N/n |
| Proportional allocation | nᵢ = n x (Nᵢ/N) |
Practice problems (with answers)
- Standard error: σ = 20, n = 100 → SE = 2.
- Confidence interval: x̄ = 150, σ = 20, n = 100, 95% → 150 ± 1.96(2) = (146.08, 153.92).
- Sample size: σ = 15, margin of error E = 3, 95% (z = 1.96) → n = (1.96 x 15/3)² = 9.8² ≈ 97.
- Systematic sampling: N = 1,000, n = 50 → k = 20; pick a random start between 1 and 20, then every 20th unit.
- Proportional stratified: population 600 men, 400 women, sample 100 → 60 men, 40 women.
- Proportion SE: p = 0.4, n = 100 → √(0.4 x 0.6/100) = 0.049.
- Classify: surveying whoever passes the mall entrance → convenience sampling; picking 5 villages at random and surveying every household → cluster sampling; each respondent recommends the next → snowball sampling.
- Short answer (5 marks): Difference between sampling and non-sampling error. → Sampling error arises from observing only a sample and falls as n rises; non-sampling errors (bias, non-response, wrong measurement, coverage) occur even in a census and are not reduced by larger samples.