Sampling Distributions and Statistical Estimation

Introduction

Human resource decisions are often based on information collected from only part of a workforce. An organisation may survey 200 employees rather than all 20,000 employees, interview a sample of applicants, or examine a sample of absence records. The central statistical problem is to use sample evidence to make defensible statements about a larger population.

Sampling distributions and statistical estimation provide the theoretical foundation for doing this. A sampling distribution describes how a statistic would vary across repeated samples from the same population. Estimation uses a sample statistic to infer an unknown population parameter. Together, these ideas explain sampling error, confidence intervals, margin of error, and the limits of quantitative evidence in HR and business.

Learning objectives

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

1. Distinguish populations, samples, parameters, statistics, and sampling error.
2. Explain the concept and importance of a sampling distribution.
3. Derive and apply the sampling distributions of means and proportions in common conditions.
4. Use the central limit theorem appropriately.
5. Construct and interpret point estimates and confidence intervals.
6. Select suitable confidence-interval procedures for means and proportions.
7. Identify assumptions, sources of bias, ethical issues, and limitations in estimation.
8. Apply estimation to evidence-based HR decisions.

1. Populations, samples, parameters, and statistics

A population is the complete set of units relevant to a study. It could be all employees in a company, all job applicants during a year, or all monthly absenteeism rates for a division. A sample is a subset selected from that population.

A numerical characteristic of a population is a parameter. Common parameters include:

A statistic is calculated from sample observations. For a sample of n observations x₁, x₂, …, xₙ, the sample mean is:

Here, ∑xᵢ is the sum of the observations and n is the sample size. The statistic x̄ is used to estimate μ. A statistic is a random variable before the sample is observed because its value depends on which units happen to be selected.

Sampling error and bias

Sampling error is the random difference between a sample statistic and its population parameter. It is expected whenever a sample, rather than a census, is used. Sampling error generally decreases as sample size increases.

Bias is systematic error caused by the design or implementation of a study. Examples include:

• Undercoverage: some population members have little or no chance of selection.
• Nonresponse bias: selected people do not respond and differ meaningfully from respondents.
• Voluntary-response bias: people with unusually strong opinions are more likely to participate.
• Measurement bias: questions or instruments produce systematically inaccurate responses.

A larger sample reduces random sampling error but does not automatically remove bias. A very large, poorly designed survey can be less credible than a smaller probability sample.

2. What is a sampling distribution?

A sampling distribution is the probability distribution of a statistic calculated from all possible samples of a specified size drawn under a specified sampling method. For example, the sampling distribution of x̄ consists of the means of every possible sample of size n, together with their probabilities.

It is not the same as:

1. the distribution of individual population values;
2. the distribution of observations in one sample; or
3. the distribution of several different variables.

The sampling distribution is a theoretical model of repeated sampling. It allows analysts to quantify how much a statistic is expected to vary and therefore how precisely it estimates a parameter.

Three properties are especially important:

• Centre: the expected value of the statistic.
• Spread: the standard deviation of the statistic, called its standard error.
• Shape: whether the distribution is approximately normal, skewed, or follows another form.

The standard error

The standard error (SE) is the standard deviation of a sampling distribution. It measures the typical sampling-to-sampling variability of a statistic, not the variability of individual observations.

For a sample mean when the population standard deviation is known:

σ is the population standard deviation and n is the sample size. The units of SE are the same as those of the original variable—for example, hours, dollars, or points. As n increases, √n increases and the standard error decreases.

For sampling without replacement from a finite population of size N, when the sampling fraction is substantial, a finite population correction may be used:

The correction is relevant when sampling without replacement and n is commonly greater than about 5% of N. It should not be applied automatically to independent or effectively unlimited populations.

3. Sampling distribution of the sample mean

Suppose observations are independently selected from a population with mean μ and standard deviation σ. Then:

and

Therefore, the standard error is σ ÷ √n. The sample mean is an unbiased estimator of μ because its expected value equals μ.

Normal populations

If the population itself is normally distributed, the sampling distribution of x̄ is normal for every sample size n:

The notation indicates a normal distribution centred at μ with standard deviation σ ÷ √n. If σ is known, a standardised value is:

This z-score expresses the distance between a sample mean and the population mean in standard-error units.

The central limit theorem

The central limit theorem (CLT) states that, under suitable independence and finite-variance conditions, the sampling distribution of the sample mean becomes approximately normal as sample size increases, even when the population distribution is not normal.

The CLT does not say that individual observations become normal. It concerns the distribution of sample means. The required sample size depends on population shape: a roughly symmetric population may require a modest sample, while a highly skewed population or one containing extreme outliers may require a much larger sample.

The CLT supports approximate inference but does not correct poor sampling, dependence between observations, severe measurement error, or a biased sampling frame.

Worked example 1: standard error of an HR measure

An organisation knows from historical records that the standard deviation of monthly training hours is 12 hours. A simple random sample of 36 employees is selected. The standard error of the sample mean is:

Thus, repeated samples of 36 employees would produce sample means that typically vary by about 2 hours around the population mean. This is not a claim that every sample mean is exactly 2 hours from μ.

4. Sampling distribution of a sample proportion

A categorical variable can be represented as a success or failure. Let p be the population proportion of successes and let p̂ be the sample proportion. If X successes are observed in a sample of n, then:

Under independent Bernoulli trials:

and

The standard error has units of a proportion, although proportions are dimensionless. Because p is usually unknown, it is often replaced by p̂ when estimating the standard error:

The normal approximation is generally appropriate when both np ≥ 10 and n(1 − p) ≥ 10. When p is unknown, analysts often check np̂ and n(1 − p̂), while recognising that exact or alternative methods may be preferable for small samples or extreme proportions.

Worked example 2: employee engagement proportion

In a random sample of 400 employees, 280 report that they are satisfied with career-development opportunities. The sample proportion is:

Its estimated standard error is:

The sample proportion is therefore estimated with a standard error of approximately 0.0229, or 2.29 percentage points.

5. Point estimation

A point estimate is one numerical value used to estimate a population parameter. Examples include x̄ as an estimate of μ and p̂ as an estimate of p.

A useful estimator should be evaluated using several criteria:

• Unbiasedness: its expected value equals the parameter.
• Consistency: it approaches the parameter as sample size increases.
• Efficiency: among competing unbiased estimators, it has relatively small variance.
• Sufficiency: it uses the information in the sample effectively for the parameter being estimated.

These properties describe long-run performance, not a guarantee that a particular estimate is close to the parameter. A particular sample can produce an estimate above or below the true value.

6. Confidence intervals

A confidence interval (CI) combines a point estimate with a margin of error. Its general form is:

A 95% confidence interval is constructed by a method that, in repeated use under the stated assumptions, captures the fixed population parameter in approximately 95% of intervals. It is not technically correct to say that there is a 95% probability that the already-computed fixed parameter lies in this particular interval. In practical reporting, however, it is common to say that we are 95% confident because the method has 95% long-run coverage.

The confidence level, margin of error, sample size, and variability are related. Higher confidence requires a larger critical value and therefore a wider interval. Larger samples generally produce narrower intervals. Greater population variability produces wider intervals.

6.1 Confidence interval for a mean when σ is known

If the population standard deviation σ is known and the sampling distribution is normal or approximately normal, a confidence interval for μ is:

Here, α = 1 − confidence level, and z_(α÷2) is the standard-normal critical value leaving probability α÷2 in each tail. Common values are 1.645 for 90%, 1.960 for 95%, and 2.576 for 99% confidence.

Worked example 3: average recruitment time

A random sample of 64 recruitment cases has a mean completion time of 18 days. Historical evidence supports σ = 8 days. Construct a 95% confidence interval for the population mean.

The standard error is:

Using z = 1.96, the margin of error is:

Therefore:

The organisation can report that the interval-estimation method gives 95% confidence that the population mean recruitment time is between 16.04 and 19.96 days, subject to the sampling and measurement assumptions.

6.2 Confidence interval for a mean when σ is unknown

In most practical situations σ is unknown and is estimated by s. If the population is approximately normal, or the sample is sufficiently large, use the t-distribution:

The t-distribution has heavier tails than the standard normal distribution and depends on degrees of freedom, df = n − 1. As n increases, the t-distribution approaches the standard normal distribution.

Conditions include a random or representative sample, independence, and an approximately normal population for small samples. Strong skewness or outliers can make a small-sample t interval unreliable.

Worked example 4: training assessment score

A sample of 25 employees has a mean assessment score of 72 and a sample standard deviation of 10. A 95% interval is required. With df = 24, the approximate critical value is t = 2.064.

Thus, the interval is:

Rounded sensibly, the estimated population mean score is between 67.9 and 76.1 points at the 95% confidence level.

6.3 Confidence interval for a population proportion

For a sufficiently large random sample, the commonly used Wald interval is:

It should be used cautiously with small samples or proportions near 0 or 1. The Wilson interval or an exact binomial interval often provides better coverage in those circumstances.

Worked example 5: proportion satisfied with flexible work

Suppose 280 of 400 employees are satisfied with flexible working arrangements, so p̂ = 0.70. The estimated standard error from Example 2 is 0.0229. A 95% Wald interval is:

The result may be reported as approximately 65.5% to 74.5%, subject to the assumptions and limitations of the method.

7. Sample-size planning and margin of error

For a mean with known σ, the required sample size for a desired margin of error E is:

Because n must be a whole number, round upward. If σ is unknown, use a prior study, a pilot estimate, or a conservative planning value.

For a proportion, the planning formula is:

If no prior estimate of p is available, p = 0.50 is conservative because p(1 − p) is maximised at 0.25.

Worked example 6: planning an employee survey

An HR department wants a 95% confidence interval for a proportion with margin of error no greater than 0.05. Without prior information, use p = 0.50:

The required sample size is 385, before allowing for nonresponse. If an expected response rate is 70%, the number invited should be approximately:

This adjustment addresses expected participation, not selection bias among those who respond.

8. Sampling methods and their effect on estimation

Probability sampling gives each population unit a known, non-zero chance of selection and supports formal sampling-error calculations.

• Simple random sampling: every sample of a given size has equal selection probability.
• Systematic sampling: select every kth unit after a random starting point; periodic patterns must be considered.
• Stratified sampling: divide the population into relevant strata, such as departments, and sample within each. It can improve representation and precision.
• Cluster sampling: select groups such as branches or teams and survey units within them. It can reduce cost but often increases sampling variance because cluster members may be similar.

Convenience sampling and many online volunteer surveys do not provide a defensible probability basis for generalising to the entire population. Weighting can adjust known differences between sample and population, but weighting cannot fully repair unknown bias or poor measurement.

9. Detailed applied case: estimating employee burnout

Background

This illustrative case concerns Meridian Services, a fictional organisation with 2,400 employees. Senior management wants to estimate the proportion experiencing high burnout before allocating wellbeing resources.

Problem

An initial proposal is to post an open survey link on the intranet. HR analysts identify two risks: employees with strong burnout may be more motivated to respond, and employees without regular intranet access may be undercovered. The proposal could generate a precise-looking percentage that is not representative.

Analysis and approach

The analysts create a sampling frame containing all employees and stratify by business unit. They randomly select employees within each stratum, issue confidential invitations, send two reminders, and record response rates by stratum. The burnout measure uses a documented scale with a pre-specified threshold. The survey explains the purpose, voluntary nature, confidentiality arrangements, and use of data.

Suppose 450 of 600 invited employees respond, and 135 exceed the burnout threshold. The response proportion is 75%, while the estimated burnout proportion is:

The estimated standard error is:

A simple 95% interval is:

The analysts report an estimated burnout proportion of 30%, with an approximate 95% interval from 25.8% to 34.2%, while clearly stating the response rate, sampling design, and measurement limitations. If response patterns differ substantially by business unit, design weights may be applied and a survey-statistics procedure used.

Outcome and learning points

Management uses the estimate to prioritise workload review and manager training, but does not treat the interval as a diagnosis of individual employees. The case demonstrates that estimation is inseparable from study design, measurement quality, confidentiality, and responsible interpretation.

10. Interpretation, limitations, and ethical governance

A confidence interval addresses uncertainty from sampling under a specified model. It does not automatically include nonresponse bias, coverage error, poor wording, deliberate misreporting, data-processing mistakes, or changes over time.

Statistical significance and practical importance are also different. A very large sample may produce a narrow interval around a trivial difference, whereas a meaningful organisational difference may remain uncertain in a small sample. HR professionals should report effect sizes, intervals, sample sizes, response rates, and relevant subgroup information rather than relying only on a yes/no conclusion.

Ethical practice requires:

• informed and voluntary participation;
• protection of confidential and sensitive employee information;
• data minimisation and secure storage;
• transparency about purposes, sponsors, and reporting;
• avoidance of re-identification in small subgroups;
• fair treatment of employees who decline participation;
• careful communication that uncertainty is not concealed.

Automated and AI-supported survey analysis can classify comments or identify patterns, but it can reproduce language, demographic, or sampling biases. Models should be validated, documented, monitored, and used as decision support rather than as an unchallengeable authority. Human review is essential when results may affect employment conditions.

11. Best practices for professional estimation

1. Define the target population and parameter before collecting data.
2. Use a probability-based sampling design whenever generalisation is required.
3. Establish a reliable sampling frame and assess coverage.
4. Pre-test instruments and define variables and thresholds in advance.
5. Monitor nonresponse and compare respondents with known population characteristics.
6. Select an interval method that matches the variable, sample size, and assumptions.
7. Round results consistently and avoid false precision.
8. Report uncertainty, limitations, response rates, and weighting procedures.
9. Protect confidentiality, especially for small departments.
10. Distinguish descriptive evidence from causal claims.

Chapter Summary

Sampling distributions describe the behaviour of statistics across repeated samples. They provide the basis for understanding sampling error and standard error. The sample mean is an unbiased estimator of the population mean, with standard error σ ÷ √n when observations are independent. The central limit theorem explains why sample means are often approximately normal for sufficiently large samples. Sample proportions have mean p and standard error √[p(1 − p) ÷ n] under appropriate independent sampling conditions.

Point estimates provide single best estimates, while confidence intervals combine estimates with a quantified margin of error. The z-distribution is appropriate for a mean when σ is known; the t-distribution is generally used when σ is unknown. Proportion intervals require attention to sample size and extreme proportions. Sample-size planning links confidence level, variability, and desired precision. Finally, sound estimation depends not only on formulas but also on representative sampling, valid measurement, ethical governance, and transparent interpretation.

Key Terms

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

Sample: A subset selected from a population.

Parameter: A numerical characteristic of a population.

Statistic: A numerical quantity calculated from sample data.

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

Sampling error: Random difference between a statistic and its population parameter.

Bias: Systematic error that causes estimates to differ from the target parameter.

Standard error: The standard deviation of a sampling distribution.

Central limit theorem: Result stating that sample means tend toward a normal distribution under suitable conditions as sample size increases.

Point estimate: A single sample-based value used to estimate a parameter.

Unbiased estimator: An estimator whose expected value equals the parameter.

Confidence interval: An interval produced by a method with a stated long-run coverage rate.

Margin of error: The distance from a point estimate to an interval endpoint.

Critical value: A distribution-based multiplier used to construct an interval.

Degrees of freedom: The number of independent pieces of information used to estimate variability; for a one-sample mean, n − 1.

Finite population correction: A factor reducing standard error when a substantial fraction of a finite population is sampled without replacement.

Coverage error: Error caused when the sampling frame excludes relevant population members.

Nonresponse bias: Bias caused when nonrespondents differ from respondents.

Stratified sampling: Sampling separately within defined population subgroups.

Cluster sampling: Sampling groups or clusters rather than individual units directly.

Review Questions

Short Answer Questions

1. Distinguish a population parameter from a sample statistic.
2. What does the standard error of a statistic measure?
3. State the central limit theorem in the context of sample means.
4. Why does increasing sample size usually reduce the standard error of a mean?
5. Explain why a large sample cannot, by itself, eliminate sampling bias.

Long Answer Questions

1. Explain the sampling distribution of the sample mean, including its centre, spread, shape, and assumptions.
2. Compare z-based and t-based confidence intervals for a population mean.
3. Explain how confidence level, sample size, variability, and margin of error are related.
4. Discuss probability and non-probability sampling methods in an HR survey context.
5. Explain the ethical responsibilities involved in estimating sensitive employee characteristics.

Analytical / Application Questions

1. A random sample of 49 employees has a mean absence rate of 6.2 days, and the population standard deviation is known to be 2.8 days. Construct and interpret a 95% confidence interval for the population mean absence rate.
2. In a random sample of 500 employees, 325 support a proposed flexible-hours policy. Calculate the sample proportion and an approximate 95% confidence interval. State the conditions supporting the approximation.
3. An HR team receives 1,200 voluntary responses to an open employee survey and reports the result as representative of all 8,000 employees. Critically evaluate the claim and recommend a more defensible design.

Multiple Choice Questions with Answers

Question 1

Which term refers to a numerical characteristic of an entire population?

A. Statistic

B. Parameter

C. Sample

D. Sampling error

Correct Answer: B. Parameter

Question 2

If the population standard deviation is 15 and n = 25, what is the standard error of the sample mean?

A. 0.6

B. 3

C. 15

D. 75

Correct Answer: B. 3

Question 3

What does the central limit theorem primarily describe?

A. The distribution of individual observations becoming uniform

B. The sampling distribution of means becoming approximately normal under suitable conditions

C. The elimination of nonresponse bias as sample size increases

D. The population mean becoming equal to zero

Correct Answer: B. The sampling distribution of means becoming approximately normal under suitable conditions

Question 4

A sample contains 80 successes out of 200 observations. What is the sample proportion?

A. 0.20

B. 0.40

C. 0.60

D. 0.80

Correct Answer: B. 0.40

Question 5

Which distribution is generally used for a confidence interval for a mean when the population standard deviation is unknown?

A. Uniform distribution

B. Binomial distribution

C. t-distribution

D. Poisson distribution

Correct Answer: C. t-distribution

Question 6

For a 95% confidence interval using the standard normal distribution, the usual critical value is approximately:

A. 1.28

B. 1.645

C. 1.96

D. 2.576

Correct Answer: C. 1.96

Question 7

Which change will generally make a confidence interval narrower, all else equal?

A. Reducing the sample size

B. Increasing the confidence level

C. Increasing the sample size

D. Increasing population variability

Correct Answer: C. Increasing the sample size

Question 8

Which sampling method divides a population into subgroups and samples within each subgroup?

A. Cluster sampling

B. Stratified sampling

C. Convenience sampling

D. Voluntary-response sampling

Correct Answer: B. Stratified sampling

Question 9

A 95% confidence interval is best interpreted as a method that:

A. Gives a 95% probability that the fixed parameter lies in the computed interval

B. Captures the parameter in approximately 95% of repeated intervals under the assumptions

C. Guarantees that 95% of observations lie inside the interval

D. Eliminates all forms of survey bias

Correct Answer: B. Captures the parameter in approximately 95% of repeated intervals under the assumptions

Question 10

If 280 of 400 employees are satisfied with a policy, what is the estimated standard error of the sample proportion?

A. Approximately 0.00229

B. Approximately 0.0229

C. Approximately 0.0700

D. Approximately 0.3000

Correct Answer: B. Approximately 0.0229