Sampling Distributions and Statistical Estimation

Introduction

Business analysts rarely have data on every customer, transaction, employee, or market participant. Instead, they collect a sample and use it to learn about a larger population. This creates an essential statistical problem: a sample statistic is not usually equal to the corresponding population parameter. Different samples produce different results, even when they are selected from the same population.

Sampling distributions provide the theoretical foundation for measuring this variation. They explain how sample means, sample proportions, and differences between sample statistics behave across repeated samples. Statistical estimation then uses this knowledge to quantify uncertainty through point estimates, confidence intervals, and margins of error.

These ideas are central to analytics because a reported value without an indication of its uncertainty can be misleading. For example, an estimated customer satisfaction rate of 72% is more useful when accompanied by information such as a 95% confidence interval from 69% to 75%.

Learning Objectives

By the end of this chapter, you should be able to:

• distinguish populations, samples, parameters, and statistics;
• explain sampling variability and sampling distributions;
• apply the central limit theorem and calculate standard errors;
• construct and interpret point estimates and confidence intervals;
• calculate margins of error for means, proportions, and differences;
• select appropriate normal or t-based procedures;
• evaluate assumptions, limitations, and ethical issues in statistical estimation.

1. Populations, Samples, Parameters, and Statistics

A population is the complete set of units relevant to a study. It may consist of all customers who purchased from a company during a year, all deliveries made by a logistics firm, or all employees in an organization. A sample is a subset selected from the population.

A numerical characteristic of a population is a parameter. Common parameters include the population mean μ, population proportion p, and population standard deviation σ. A numerical characteristic calculated from a sample is a statistic, such as the sample mean x̄, sample proportion p̂, or sample standard deviation s.

The parameter is generally unknown, whereas the statistic is observed. Estimation uses the statistic to make a reasoned statement about the parameter.

2. Sampling Variability and Sampling Distributions

If repeated random samples of the same size are drawn from a population, their statistics will usually differ. This is sampling variability. It is not necessarily caused by mistakes; it is an inherent consequence of observing only part of a population.

The sampling distribution of a statistic is the probability distribution of that statistic across all possible random samples of a specified size. For example, the sampling distribution of x̄ describes the values of the sample mean that would result from repeated samples of size n.

A statistic has three important features in a sampling distribution:

1. Centre: the expected or average value of the statistic.
2. Spread: the amount of sampling variability.
3. Shape: the distribution’s form, such as normal or skewed.

For a random sample from a population with mean μ and standard deviation σ, the sample mean has:

and

Here, E(x̄) is the expected value of the sample mean, SE(x̄) is its standard error, σ is the population standard deviation, and n is the sample size. The expression applies to independent observations with a common finite variance. In practice, σ is usually unknown and is replaced by s:

The units of the standard error are the same as the units of the original variable. If delivery time is measured in minutes, the standard error of its sample mean is also measured in minutes.

The formula demonstrates the effect of sample size. Increasing n reduces the standard error, but according to a square-root relationship. To halve the standard error, the sample size must be multiplied by four.

Standard error versus standard deviation

The standard deviation describes variation among individual observations. The standard error describes variation among estimates from repeated samples. A dataset may contain highly variable individual incomes but still produce a relatively precise estimate of the mean if the sample is sufficiently large.

3. The Central Limit Theorem

The central limit theorem (CLT) states that, under suitable conditions, the sampling distribution of the sample mean becomes approximately normal as the sample size increases, even if the population itself is not normally distributed.

For independent observations with population mean μ and finite standard deviation σ:

This standardized form is approximately a standard normal variable. The approximation is generally reliable when the sample is large enough and the population is not extremely skewed or dominated by outliers. If the population is itself normal, the sampling distribution of x̄ is normal for every sample size.

The CLT does not say that individual observations become normal. It says that the distribution of sample means becomes approximately normal. It also does not eliminate bias caused by poor sampling. A very large biased sample may provide a precise estimate of the wrong population quantity.

Conditions for using the CLT

Analysts should consider:

• Randomness: observations should arise from a probability sample or a defensible randomisation process.
• Independence: one observation should not unduly determine another. For sampling without replacement, a common practical rule is that the sample should be no more than about 10% of the population.
• Sample size and shape: larger samples are needed for strongly skewed populations or heavy-tailed data.
• Finite variance: the usual mean-based result requires a meaningful, finite population variance.

For proportions, the sample proportion p̂ is approximately normal when the expected numbers of successes and failures are sufficiently large. A common rule is:

Because p is unknown, analysts often check this condition using p̂ or a planning value.

4. Sampling Distributions of Common Statistics

4.1 Sample means

When σ is known, the standard error of x̄ is σ/√n. When σ is unknown, the estimated standard error is s/√n. For small samples from a normally distributed population, inference about μ uses the Student’s t distribution rather than the standard normal distribution.

The t distribution is centred at zero and has heavier tails than the standard normal distribution. Its shape depends on the degrees of freedom, usually n − 1 for a one-sample mean. As degrees of freedom increase, the t distribution approaches the standard normal distribution.

4.2 Sample proportions

If X represents the number of successes in n independent trials and each trial has probability p of success, then X follows a binomial model. The sample proportion is:

Its mean and standard error are:

Since p is unknown, a confidence interval commonly uses the estimated standard error:

The proportion must be expressed consistently. A proportion of 0.72 is equivalent to 72%, but a margin of error of 0.03 is equivalent to 3 percentage points, not 0.03 percentage points.

4.3 Differences between two means

For two independent samples, the estimator of μ₁ − μ₂ is x̄₁ − x̄₂. If the samples are independent, the estimated standard error is:

where s₁ and s₂ are sample standard deviations and n₁ and n₂ are sample sizes. This formula does not assume equal population variances. An unequal-variance t procedure, commonly called Welch’s method, is often the safer default.

For paired observations, such as the same employees measured before and after training, analyse the within-pair differences. If d̄ is the mean difference and s_d is the standard deviation of differences:

Treating paired data as independent can discard useful information and produce an incorrect standard error.

4.4 Differences between two proportions

For two independent groups, the estimator of p₁ − p₂ is p̂₁ − p̂₂. Its estimated standard error for a confidence interval is:

For hypothesis tests of equality, a pooled estimate may be used under the null hypothesis. Confidence intervals and hypothesis tests therefore do not always use exactly the same standard-error formula.

5. Point Estimation

A point estimate is a single statistic used to estimate a population parameter. The natural point estimate of μ is x̄; of p is p̂; and of μ₁ − μ₂ is x̄₁ − x̄₂.

A good estimator is often evaluated using:

• Unbiasedness: its expected value equals the parameter.
• Consistency: it approaches the parameter as sample size increases.
• Efficiency: among comparable unbiased estimators, it has relatively small variance.
• Robustness: it is not excessively affected by unusual observations or modest violations of assumptions.

Unbiasedness does not guarantee that one observed estimate is close to the parameter. It is a long-run property. Analysts should also report precision and discuss sampling design.

6. Confidence Intervals

A confidence interval combines a point estimate with a margin of error. Its general structure is:

A 95% confidence interval is constructed by a method that captures the fixed population parameter in approximately 95% of repeated samples, assuming the model and sampling procedure are appropriate. It is not technically correct to say that there is a 95% probability that the already fixed parameter lies in this particular interval under the frequentist interpretation.

6.1 Confidence interval for a mean with known σ

Here, α is the total probability outside the interval and z_{α/2} is the standard-normal critical value. For 95% confidence, α = 0.05 and z_{α/2} ≈ 1.96.

6.2 Confidence interval for a mean with unknown σ

The t critical value has n − 1 degrees of freedom. This method is appropriate when observations are independent and the population is approximately normal for small samples, or when the sample is sufficiently large for the CLT to provide a reasonable approximation.

6.3 Confidence interval for a proportion

The simple Wald interval is:

It can perform poorly for small samples or proportions near 0 or 1. The Wilson interval is generally preferable in such circumstances, and exact or alternative methods may be considered when counts are very small. Software output should not be accepted without checking the method used.

6.4 Confidence intervals for differences

For independent means:

For independent proportions:

The critical value and standard error must match the design and inferential method.

7. Margin of Error and Planning Sample Size

The margin of error (MOE) is the distance from the point estimate to either endpoint of a symmetric confidence interval:

A higher confidence level increases the critical value and therefore increases the margin of error, holding sample size constant. A larger sample reduces the standard error and therefore reduces the margin of error.

For estimating a mean with known or planned standard deviation σ and desired margin E:

The result must be rounded upward. For a proportion, if a planning value p is available:

If no planning value is available, p = 0.50 gives the largest required sample size because p(1 − p) is maximised at 0.25. These formulas assume a simple random sample and do not automatically account for nonresponse, clustering, stratification, finite-population corrections, or design effects.

8. Worked Examples

Example 1: Estimating an average delivery time

A random sample of 64 deliveries has mean time x̄ = 42 minutes and sample standard deviation s = 8 minutes. Construct an approximate 95% confidence interval for the population mean.

Because n = 64 is reasonably large, use a t interval with 63 degrees of freedom. The 95% critical value is approximately 2.00. The standard error is:

The margin of error is:

Therefore:

Interpretation: using this sampling method, the procedure produces intervals that capture the population mean in approximately 95% of repeated samples. The data support an estimated average delivery time between 40 and 44 minutes.

Example 2: Estimating a customer conversion rate

In a random sample of 400 website visitors, 120 made a purchase. Thus:

The estimated standard error is:

For a 95% interval, use z ≈ 1.96:

The approximate interval is 0.30 ± 0.0449, or (0.255, 0.345). In percentage terms, the estimated conversion rate is 30%, with a margin of error of about 4.5 percentage points and an interval of approximately 25.5% to 34.5%.

The success and failure counts are 120 and 280, both exceeding 10, so a normal approximation is reasonable. For greater accuracy, especially near a boundary or with smaller counts, a Wilson interval should be considered.

Example 3: Comparing two average processing times

Group 1 has n₁ = 36, x̄₁ = 18 minutes, s₁ = 4 minutes. Group 2 has n₂ = 49, x̄₂ = 20 minutes, s₂ = 5 minutes. The estimated difference is −2 minutes, meaning Group 1 is faster by 2 minutes.

The standard error is:

An approximate 95% interval using a critical value near 2 is:

Thus the interval is approximately (−3.95, −0.05) minutes. Since the entire interval is below zero, the evidence is consistent with Group 1 having a lower mean processing time, although the closeness of the upper endpoint means the practical conclusion should be reported cautiously.

9. Applied Case: Estimating the Effect of a Service Intervention

Background

A fictional online retailer introduces a redesigned checkout process. Management wants to know whether the redesign reduces checkout time and increases the purchase rate.

Problem

The analytics team observes 80 randomly selected customers using the old process and 100 using the new process. The old-process mean checkout time is 6.8 minutes with s = 2.4; the new-process mean is 5.9 minutes with s = 2.1. Purchases occur for 44 of the old-process customers and 65 of the new-process customers.

Analysis

The estimated reduction in mean time is:

Its estimated standard error is:

Using an approximate 95% critical value of 1.98, the margin of error is about 0.68 minutes. The interval for new minus old is approximately (−1.58, −0.22) minutes. The redesign is therefore associated with a reduction between about 0.22 and 1.58 minutes under the stated assumptions.

The conversion estimates are 44/80 = 0.55 and 65/100 = 0.65. The estimated difference is 0.10, or 10 percentage points. The confidence interval requires the standard error for two independent proportions:

Using 1.96, the margin of error is approximately 0.139. The interval is about −0.039 to 0.239, or −3.9 to 23.9 percentage points. The estimated increase is positive, but the interval includes zero; the data do not establish a clear increase at the 95% confidence level.

Approach and outcome

Management should distinguish the two findings. The time reduction appears reasonably precise and practically meaningful. The conversion result is less conclusive because the samples are modest and the estimated difference is uncertain. A larger experiment, ideally with random assignment and consistent exposure conditions, would provide stronger evidence.

Learning points

• Separate statistical significance from practical importance.
• Report estimates with intervals rather than isolated percentages.
• Analyse each outcome using its appropriate standard error.
• Random assignment supports causal interpretation; random sampling supports generalisation.
• A non-significant result does not prove that no effect exists.

10. Assumptions, Limitations, and Professional Practice

A narrow confidence interval does not compensate for biased sampling. Common threats include convenience sampling, nonresponse, undercoverage, duplicate records, measurement error, and dependence among observations. For example, 1,000 website visits from one campaign may not represent all customers.

Analysts should document the target population, sampling frame, selection method, response rate, exclusions, missing-data treatment, estimator, confidence level, and software procedure. Results should be presented in the units meaningful to decision-makers, such as minutes or percentage points.

Multiple confidence intervals create a further issue: if many intervals are examined, some may miss their parameters by chance. Analysts should pre-specify primary outcomes or use suitable multiplicity adjustments when many simultaneous claims are made.

Ethically, uncertainty should not be hidden to make a result appear more persuasive. A dashboard should not display a highly precise-looking estimate based on a non-random or tiny sample. Privacy-preserving aggregation is also important: estimation should not expose individual records, and analysts should assess whether small groups can be re-identified.

Technology packages can calculate t values, Wilson intervals, bootstrap intervals, and complex survey estimates efficiently. However, software cannot determine whether the sample is representative, whether observations are independent, or whether the chosen population is ethically and substantively appropriate. AI-assisted analytics should therefore be audited for unsupported assumptions, fabricated precision, and selective reporting.

11. Best-Practice Checklist

Before reporting an estimate, ask:

1. What population does the estimate describe?
2. How were observations selected?
3. Is the estimator appropriate for the variable and design?
4. Are independence, sample-size, and distributional conditions plausible?
5. Was the correct standard error used?
6. Does the interval use an appropriate critical value?
7. Are results reported in clear units and percentage points where relevant?
8. Is the distinction between association and causation clear?
9. Are practical consequences discussed alongside statistical uncertainty?
10. Have limitations and potential biases been disclosed?

Chapter Summary

Sampling distributions describe the behaviour of statistics over repeated samples. Sampling variability is unavoidable, but it can be quantified through the standard error. For a sample mean, the standard error is approximately s/√n; for a sample proportion, it is approximately √[p̂(1 − p̂)/n]. The central limit theorem explains why sample means and, under suitable conditions, sample proportions are often approximately normally distributed.

Point estimates provide single best estimates of population parameters. Confidence intervals add a margin of error and express the precision of an estimation procedure. Means with unknown population standard deviations generally use the t distribution, while large-sample proportion procedures often use normal critical values or more reliable alternatives such as Wilson intervals. Differences between groups require standard errors that reflect independent or paired designs.

Confidence level, sample size, variability, sampling design, and data quality all influence the usefulness of an estimate. A precise estimate from a biased sample can still be misleading. Professional analysts therefore report methods, assumptions, uncertainty, and limitations transparently.

Key Terms

Population: The complete set of units relevant to a study.

Sample: A subset of the population used for analysis.

Parameter: A numerical characteristic of a population.

Statistic: A numerical characteristic calculated from a sample.

Sampling variability: Natural variation in a statistic across samples.

Sampling distribution: The probability distribution of a statistic across repeated samples.

Standard error: The standard deviation of a statistic’s sampling distribution, estimated in practice.

Central limit theorem: A result stating that sample means tend toward a normal sampling distribution under suitable conditions.

Point estimate: A single statistic used to estimate a population parameter.

Confidence interval: An interval produced by a procedure designed to capture a parameter at a stated long-run rate.

Margin of error: The distance between an estimate and an endpoint of a symmetric confidence interval.

Critical value: A quantile used to determine the width of an interval or rejection region.

Student’s t distribution: A family of distributions used especially for means when population standard deviation is unknown.

Degrees of freedom: A quantity determining the relevant t distribution, often n − 1 for one mean.

Sample proportion: The number of observed successes divided by the sample size.

Independence: A condition in which observations do not systematically determine one another.

Bias: Systematic deviation of an estimator or sampling process from the target quantity.

Welch method: An unequal-variance procedure for comparing two independent means.

Wilson interval: A confidence-interval method for proportions that often performs better than the simple Wald interval.

Review Questions

5 Short Answer Questions

1. Distinguish between a population parameter and a sample statistic.
2. What does the standard error of a sample mean measure?
3. State the central limit theorem in practical terms.
4. Why is the t distribution used when σ is unknown?
5. What effect does quadrupling the sample size have on the standard error of a mean?

5 Long Answer Questions

1. Explain sampling distributions, including their centre, spread, and shape, and discuss why they matter in business analytics.
2. Describe the assumptions required for a confidence interval for a population mean and explain the consequences of violating them.
3. Compare confidence intervals for means, proportions, and differences between two groups.
4. Explain the difference between statistical significance and practical importance using an analytics example.
5. Discuss why a large sample does not necessarily eliminate bias.

3 Analytical / Application Questions

1. A sample of 100 customers has an average spending of $84 and s = $20. Construct an approximate 95% confidence interval using a t critical value of 1.984 and interpret it.
2. Of 500 surveyed users, 210 report being satisfied. Calculate the sample proportion, approximate standard error, and 95% margin of error using z = 1.96. State whether the normal approximation condition is satisfied.
3. An analyst compares two independent groups but treats matched before-and-after observations as independent. Explain how this mistake affects the standard error and propose a more appropriate analysis.

Multiple Choice Questions with Answers

Question 1

Which quantity is a population parameter?

A. The sample mean x̄

B. The sample proportion p̂

C. The population mean μ

D. The sample standard deviation s

Correct Answer: C. The population mean μ

Question 2

A sample mean has s = 12 and n = 36. What is its estimated standard error?

A. 0.33

B. 2

C. 6

D. 72

Correct Answer: B. 2

Question 3

What does the central limit theorem primarily describe?

A. The normality of every individual observation

B. The approximate normality of sample means under suitable conditions

C. The elimination of bias from any sample

D. The equality of all population variances

Correct Answer: B. The approximate normality of sample means under suitable conditions

Question 4

A sample contains 80 successes in 200 trials. What is the sample proportion?

A. 0.04

B. 0.20

C. 0.40

D. 2.50

Correct Answer: C. 0.40

Question 5

Which change generally reduces the margin of error, holding confidence level and variability constant?

A. Reducing the sample size

B. Increasing the sample size

C. Increasing the critical value

D. Increasing the population standard deviation

Correct Answer: B. Increasing the sample size

Question 6

For a small sample mean with unknown population standard deviation and an approximately normal population, which distribution is generally used?

A. Uniform distribution

B. Student’s t distribution

C. Binomial distribution

D. Poisson distribution

Correct Answer: B. Student’s t distribution

Question 7

A 95% confidence interval is best interpreted as follows:

A. There is a 95% probability that the fixed parameter is in this completed interval

B. 95% of the sample observations lie in the interval

C. The method captures the fixed parameter in approximately 95% of repeated samples

D. The sample statistic is correct with probability 0.95

Correct Answer: C. The method captures the fixed parameter in approximately 95% of repeated samples

Question 8

For two independent sample means, which expression estimates the standard error of their difference?

A. s₁/n₁ + s₂/n₂

B. √[(s₁²/n₁) + (s₂²/n₂)]

C. (s₁ + s₂)/(n₁ + n₂)

D. √[(s₁/n₁) − (s₂/n₂)]

Correct Answer: B. √[(s₁²/n₁) + (s₂²/n₂)]

Question 9

If the sample size is multiplied by four, what happens approximately to the standard error of a sample mean?

A. It doubles

B. It is halved

C. It remains unchanged

D. It becomes four times larger

Correct Answer: B. It is halved

Question 10

Which issue can remain even when a sample is very large?

A. Sampling variability is always zero

B. Bias caused by a nonrepresentative sampling method

C. The need to define a population

D. The existence of a sample statistic

Correct Answer: B. Bias caused by a nonrepresentative sampling method