Chapter 12 — Front Matter
Purpose and Overview
Every statistical study begins with a question about a population. But populations are usually too large, too expensive, or too spread out to measure completely. Instead, researchers collect a sample, compute a statistic from that sample, and use the result to draw a conclusion about the whole population. This process works — but only if we understand one essential fact: statistics computed from samples are not fixed values. They vary from sample to sample.
Chapter 12 introduces the logic of statistical inference by studying that variability. Students will learn why repeated samples produce different results, how the pattern of those results can be described mathematically, what the Central Limit Theorem says about sample means, and how standard error measures typical sampling variability. From that foundation, students will learn to construct and interpret confidence intervals — structured statements that estimate population values while honestly acknowledging uncertainty.
This chapter does not ask students to accept inference on faith. It builds inference step by step, starting with simulation and intuition, moving through sampling distributions, and arriving at the confidence interval as a principled, honest tool for communicating what data can and cannot tell us.
Opening Question
A polling organization surveys 1,200 randomly selected registered voters and finds that 54% support a proposed education funding measure. The organization reports: "54% of voters support the measure, with a margin of error of ±3 percentage points." What does that number — 54% — actually represent? Is it the true population percentage, or is it an estimate? Why does the margin of error exist? If a different organization surveyed a different 1,200 voters the same week, would they also get 54%? And why should anyone trust a poll of 1,200 people to represent millions of voters? These questions cannot be answered by arithmetic alone. They require understanding how sample statistics behave when sampling is repeated — which is exactly what this chapter studies.
14Sections
10Objectives
13Key Terms
12+Worked Examples
Learning Objectives
- 1Explain why sample statistics vary from sample to sample.
- 2Define a sampling distribution and describe its shape, center, and spread.
- 3Describe the sampling distribution of a sample proportion and a sample mean.
- 4State the Central Limit Theorem in plain language and explain what it means for sample means.
- 5Define standard error and distinguish it from standard deviation.
- 6Compute and interpret a point estimate.
- 7Construct a confidence interval for a proportion using the margin of error formula.
- 8Interpret a confidence interval correctly in context.
- 9Explain what "confidence level" means and what it does not mean.
- 10Identify and explain common mistakes in estimation and inference.
Sections 12.1 – 12.3
Why Samples Vary and How to Study That Variation
Sampling Variability · Notation · Simulation
Section 12.1
Why Samples Vary
When a company surveys 400 customers about satisfaction with a new product, the results depend partly on which 400 customers were selected. If the survey were repeated with a different group of 400 customers drawn from the same population, the results would almost certainly differ — not because the surveys were done incorrectly, but because different people were asked.
This is the fundamental reality of sampling: every random sample produces a different result. This is not a flaw in statistics. It is an honest acknowledgment of how data work. Even with perfect sampling technique, two polls of the same population on the same day can produce different percentages, two quality control samples from the same factory line can produce different average measurements, and two surveys of the same students about study habits can produce different means.
Variability Is Not Error
Students often confuse sampling variability with mistakes. The distinction is essential:
| Concept | What It Means | Is It a Problem? |
|---|
| Variability | Statistics differ from sample to sample because different random individuals were selected | No — expected, predictable, and can be modeled |
| Bias | The sampling method systematically leans in one direction, misrepresenting the population | Yes — a flaw in design |
| Error | Mistakes in recording, measuring, or calculating | Yes — should be corrected |
Key Idea
If you drew a random sample of 20 beads from a jar of 100 red and 100 blue beads, you might get p̂ = 0.55 red. A second sample might give p̂ = 0.45. A third might give p̂ = 0.60. None of these results is wrong — each is a legitimate estimate. The question is: how much do they differ, and does the pattern of their differences have a predictable shape? That predictable shape is called a sampling distribution.
Section 12.2
Sampling Variability
Two main factors affect how much statistics vary across samples. Sample size is the first: larger samples tend to produce statistics that are closer to the true population value. A sample of 1,000 voters tells us more than a sample of 10 voters about the full electorate. As sample size increases, sampling variability decreases.
Population variability is the second: if the population has very little spread — nearly everyone gives the same response — samples will be similar to each other. If the population has a lot of spread, samples can differ substantially.
Key Idea — Statistics Are Random Variables
Before studying any individual sample, a statistician can think of the sample statistic as a random variable — a quantity whose value depends on chance, specifically on which individuals happen to be selected. Like any random variable, a sample statistic has a distribution. That distribution is what Chapter 12 studies.
Notation Review
| Symbol | Name | Type |
|---|
p | Population proportion | Parameter — fixed, describes the full population |
p̂ | Sample proportion | Statistic — varies from sample to sample |
μ | Population mean | Parameter — fixed, describes the full population |
x̄ | Sample mean | Statistic — varies from sample to sample |
σ | Population standard deviation | Parameter — describes spread in the full population |
s | Sample standard deviation | Statistic — estimates σ from the sample |
n | Sample size | Known — the number of individuals sampled |
Section 12.3
Simulation of Repeated Samples
One of the best ways to understand sampling variability is to simulate it. A simulation mimics random sampling many times so that the pattern of results can be observed directly.
Example 12.3.1
Describing a simulated sampling distribution
- ProblemA student simulates drawing 200 random samples of size n = 40 from a population where the true proportion is p = 0.45. She records p̂ from each simulated sample and creates a histogram. The histogram is approximately bell-shaped, centered near 0.45, with most values falling between 0.30 and 0.60. Describe what the histogram tells her.
- ThinkThe histogram displays the simulated sampling distribution of p̂ — how sample proportions vary across repeated samples of size 40.
- RepresentEach bar in the histogram represents a range of p̂ values. The height shows how many of the 200 simulated samples produced a p̂ in that range.
- AnalyzeThe histogram is centered near 0.45, matching the true population proportion. The bell shape suggests the sampling distribution is approximately normal. The spread (0.30 to 0.60) shows how much sample proportions can vary when n = 40.
- InterpretSample proportions from random samples of size 40 will typically be close to 0.45, but they vary. The distribution is approximately normal, centered at the true proportion, with predictable spread.
Key finding: Sample proportions cluster near the true p = 0.45, vary predictably, and form an approximately normal distribution.
What Simulations Reveal
- Sample statistics are not all the same — they vary from sample to sample
- They cluster around the true population parameter
- Their spread depends on sample size (larger n → less spread)
- Their shape approaches a normal distribution under the right conditions
Sections 12.4 – 12.8
Sampling Distributions, the CLT, and Standard Error
Distribution of p̂ · Distribution of x̄ · Shape, Center, Spread · CLT · Standard Error
Section 12.4
Sampling Distribution of a Proportion
When a categorical variable is studied, the statistic of interest is usually the sample proportion p̂: the fraction of the sample with a specific characteristic. If repeated random samples of the same size n are taken from a population with true proportion p, and p̂ is computed from each sample, the collection of all possible p̂ values forms a sampling distribution of the sample proportion.
Properties of the Sampling Distribution of p̂
Shape: Approximately normal when both np ≥ 10 and n(1–p) ≥ 10.
Center: The mean equals the population proportion: μ(p̂) = p
This confirms that p̂ is an unbiased estimator — on average across all possible samples, p̂ equals p.
Spread: The standard deviation is: σ(p̂) = √[ p(1–p) / n ]
This decreases as n increases — larger samples produce more consistent estimates.
Conditions for Using This Result
Three conditions must be satisfied: Random sampling — the sample must be drawn randomly. Large enough sample — both np ≥ 10 and n(1–p) ≥ 10 must hold. Independence — individual observations should be independent; if sampling without replacement from a finite population, the sample size should be no more than 10% of the population.
Example 12.4.1
Describing the sampling distribution of a proportion
- ProblemA large university has found that 35% of all students have used the campus tutoring center in the past semester. A researcher plans to randomly select n = 120 students. Describe the sampling distribution of p̂.
- ThinkPopulation proportion p = 0.35, sample size n = 120. Verify conditions, then state shape, center, and spread.
- Check Conditions
np = 120(0.35) = 42 ≥ 10 ✓
n(1-p) = 120(0.65) = 78 ≥ 10 ✓
Independence: 120 is well under 10% of a large university population ✓
- Calculate
μ(p̂) = p = 0.35
σ(p̂) = √[ 0.35 × 0.65 / 120 ]
= √[ 0.2275 / 120 ]
= √0.001896
≈ 0.0435
- InterpretThe sampling distribution of p̂ is approximately normal, centered at 0.35, with a standard deviation of about 0.044. In repeated random samples of 120 students, the sample proportion who used the tutoring center will typically fall within about 0.044 of the true proportion 0.35.
Shape: approximately normal. Center: 0.35. Spread: σ(p̂) ≈ 0.044.
Display Instruction
Draw a normal curve centered at 0.35. Mark one standard deviation on each side: 0.306 and 0.394. Add marks at two and three standard deviations. Label the curve with the mean and the value σ(p̂) ≈ 0.044. Shade the region within one standard deviation and label its approximate probability (68%).
Student Note — Using the Formula
The formula for σ(p̂) uses the population proportion p. In practice, we usually do not know p — that is why we are taking a sample. In Section 12.10, we will substitute the sample proportion p̂ when constructing a confidence interval. For now, we are studying the mathematical theory of how samples behave.
Section 12.5
Sampling Distribution of a Mean
When a quantitative variable is studied, the statistic of interest is usually the sample mean x̄. The sampling distribution of the sample mean describes how x̄ varies across all possible random samples of size n from a given population.
Properties of the Sampling Distribution of x̄
Center: μ(x̄) = μ — the sample mean is an unbiased estimator of the population mean.
Spread (Standard Error of the Mean): σ(x̄) = σ / √n
Shape: Exactly normal if the population is normal (any n). Approximately normal for large n by the Central Limit Theorem, even when the population is not normal.
Example 12.5.1
Sampling distribution of a sample mean
- ProblemThe time to complete a standardized reading test is normally distributed with μ = 42 minutes and σ = 8 minutes. A researcher selects random samples of n = 25 students. Describe the sampling distribution of x̄.
- ThinkBecause the population is normally distributed, the sampling distribution of x̄ is also normally distributed for any sample size. Compute mean and standard deviation.
- Calculate
μ(x̄) = μ = 42 minutes
σ(x̄) = σ / √n = 8 / √25 = 8 / 5 = 1.6 minutes
- InterpretThe sampling distribution of x̄ is normal with mean 42 minutes and standard deviation 1.6 minutes. Sample means will typically fall within about 1.6 minutes of the true mean. The individual test times vary much more (σ = 8 min), but averaging 25 values reduces that variability substantially — this is the averaging effect of samples.
Shape: normal. Center: 42 min. Standard error: 1.6 min.
Section 12.6
Shape, Center, and Spread of Sampling Distributions
| Sampling Distribution of p̂ | Sampling Distribution of x̄ |
|---|
| Center | μ(p̂) = p | μ(x̄) = μ |
| Spread | σ(p̂) = √[ p(1–p) / n ] | σ(x̄) = σ / √n |
| Shape | Approximately normal when np ≥ 10 and n(1–p) ≥ 10 | Exactly normal if population is normal; approximately normal for large n by CLT |
| Estimating | Population proportion p | Population mean μ |
Example 12.6.1
Comparing sampling distributions for two sample sizes
- ProblemTwo researchers each study the same population where p = 0.70. Researcher A uses samples of n = 30. Researcher B uses samples of n = 120. Compare the sampling distributions of their sample proportions.
- Calculate
Researcher A (n = 30):
μ(p̂) = 0.70
σ(p̂) = √[ 0.70 × 0.30 / 30 ] = √0.007 ≈ 0.0837
Researcher B (n = 120):
μ(p̂) = 0.70
σ(p̂) = √[ 0.70 × 0.30 / 120 ] = √0.00175 ≈ 0.0418
- InterpretBoth sampling distributions are centered at 0.70 — neither method is biased. However, Researcher B's estimates are much more consistent: typical sampling error of about 0.042, compared to about 0.084 for Researcher A. Quadrupling the sample size (30 to 120) halved the standard error.
Both unbiased; Researcher B's standard error (≈ 0.042) is half that of Researcher A (≈ 0.084).
Section 12.7
Central Limit Theorem as an Introduction
Central Limit Theorem
When random samples of size n are drawn from any population with mean μ and finite standard deviation σ, the sampling distribution of the sample mean x̄ is approximately normal when n is sufficiently large, regardless of the shape of the population distribution.
A common guideline: n ≥ 30 is usually sufficient. For heavily skewed populations, larger samples may be needed.
The CLT is remarkable because real-world populations are rarely normal: incomes are right-skewed, response times are skewed, counts of events are discrete. Yet when researchers take large samples from these populations and compute sample means, those means follow an approximately normal distribution — which is why normal-distribution methods apply so broadly in inference.
Example 12.7.1
Applying the Central Limit Theorem
- ProblemDaily customer wait times at a service center are strongly right-skewed, with μ = 7.2 minutes and σ = 4.5 minutes. A quality analyst takes random samples of n = 50 customers. What is the approximate distribution of x̄?
- ThinkThe population is not normal, but n = 50 ≥ 30. The CLT applies, so the sampling distribution of x̄ is approximately normal.
- Calculate
μ(x̄) = 7.2 minutes
σ(x̄) = 4.5 / √50 = 4.5 / 7.071 ≈ 0.636 minutes
- InterpretEven though individual wait times are strongly right-skewed, the distribution of sample means from samples of 50 customers is approximately normal with mean 7.2 min and standard deviation 0.636 min. The analyst can use normal distribution methods to make probability statements about x̄.
By the CLT: x̄ ~ N(7.2, 0.636). Normal methods apply despite the skewed population.
MGU Library Connection
- The Central Limit Theorem is proved rigorously in college-level Probability and Statistics courses
- In this chapter it is treated intuitively and used as a practical tool for inference
- See MGU Probability and Statistics for the formal mathematical treatment
Section 12.8
Standard Error
The standard error of a statistic is the standard deviation of its sampling distribution. It answers the question: if we drew many random samples of this size, how much would the statistic typically vary from sample to sample?
| Term | Measures | Applied To |
|---|
| Standard deviation (s or σ) | Variability of individual values in a distribution | A single dataset or population |
| Standard error | Variability of a statistic across repeated samples | A sampling distribution |
Standard Error Formulas
Standard error of p̂ (when p is unknown, substitute p̂):
SE(p̂) = √[ p̂(1–p̂) / n ]
Standard error of x̄ (when σ is unknown, substitute s):
SE(x̄) = s / √n
Example 12.8.1
Computing standard error of the mean
- ProblemA random sample of n = 64 batteries is tested. The sample mean battery life is x̄ = 18.3 hours with a sample standard deviation of s = 3.2 hours. Compute the standard error of the sample mean.
- Calculate
SE(x̄) = s / √n = 3.2 / √64 = 3.2 / 8 = 0.40 hours
- InterpretThe standard error is 0.40 hours. In repeated random samples of 64 batteries, sample means will typically vary from the true population mean by about 0.40 hours. Individual battery lives vary much more (s = 3.2 hours), but averaging 64 values reduces that variability substantially.
SE(x̄) = 0.40 hours — the averaging effect of n = 64 reduces variability from 3.2 to 0.40 hours.
Sections 12.9 – 12.14
From Sample to Population: Point Estimates and Confidence Intervals
Point Estimates · Confidence Intervals · Margin of Error · Interpretation · Applications
Section 12.9
Point Estimates
A point estimate is a single value computed from a sample and used to estimate an unknown population parameter. Point estimates are useful because they provide a specific answer — in a news report or a management meeting, a single number is easy to communicate and act on.
However, point estimates have an important limitation: they carry no information about their own uncertainty. If a poll reports p̂ = 0.54, you know what the sample found, but you do not know how close 0.54 is likely to be to the true population value. This is why confidence intervals are generally preferred in serious statistical reporting.
| Parameter | Point Estimate | Formula |
|---|
| Population proportion p | Sample proportion p̂ | p̂ = (count with characteristic) / n |
| Population mean μ | Sample mean x̄ | x̄ = Σx / n |
Example 12.9.1
Computing a point estimate
- ProblemA researcher surveys 250 randomly selected adults and finds that 182 report exercising at least three times per week. What is the point estimate of the population proportion?
- Calculate
p̂ = 182 / 250 = 0.728
- InterpretThe point estimate is p̂ = 0.728, or 72.8%. Based on this sample, about 72.8% of adults in the population exercise at least three times per week. This is the best single estimate from the sample, but it does not tell us how close it is to the true population proportion — for that, we need a confidence interval.
p̂ = 0.728. A point estimate without a margin of error does not fully communicate uncertainty.
Section 12.10
Confidence Intervals
A confidence interval is an interval of values, constructed from sample data, designed to capture the true population parameter with a stated level of confidence. It takes the form:
Confidence Interval for a Proportion
p̂ ± z* · √[ p̂(1–p̂) / n ]
| Confidence Level | Critical Value z* |
|---|
| 90% | 1.645 |
| 95% | 1.960 |
| 99% | 2.576 |
The formula has three parts: p̂ is the center of the interval; z* is the critical value controlling width based on the confidence level; and √[ p̂(1–p̂)/n ] is the standard error of p̂. Together they produce an interval wide enough to capture the true proportion with the chosen probability across all possible samples.
Conditions for a Valid Confidence Interval for a Proportion
Three conditions must be met: Random sample — data must come from a random sampling process. Large enough sample — n·p̂ ≥ 10 and n·(1–p̂) ≥ 10. Independence — each observation is independent; sample is no more than 10% of the population if sampling without replacement. If these conditions fail, the confidence interval cannot be trusted even if the arithmetic is correct.
Example 12.10.1
Constructing a 95% confidence interval for a proportion
- ProblemA random sample of 400 high school students is asked whether they have a paid part-time job. Of the 400 students, 148 say yes. Construct a 95% confidence interval for the true population proportion.
- StateLet p = the true proportion of all high school students with a paid part-time job.
- Check Conditions
p̂ = 148 / 400 = 0.37
n·p̂ = 400(0.37) = 148 ≥ 10 ✓
n(1–p̂) = 400(0.63) = 252 ≥ 10 ✓
Random sample: stated in the problem ✓
Independence: 400 is well under 10% of all high school students ✓
- Calculate
SE(p̂) = √[ 0.37 × 0.63 / 400 ]
= √[ 0.2331 / 400 ]
= √0.000583
≈ 0.02414
Confidence interval = 0.37 ± 1.96 × 0.02414
= 0.37 ± 0.0473
= (0.323, 0.417)
- CheckThe interval (0.323, 0.417) is centered at p̂ = 0.37 and lies entirely within [0, 1], as any proportion must. The width of ±0.047 is reasonable for n = 400. ✓
- InterpretWe are 95% confident that the true proportion of all high school students with a paid part-time job is between 32.3% and 41.7%.
95% CI: (0.323, 0.417) — between 32.3% and 41.7% of high school students have a paid part-time job.
Display Instruction
Draw a horizontal number line from 0.28 to 0.46. Mark the sample proportion p̂ = 0.37 with a solid dot. Draw a horizontal segment from 0.323 to 0.417 to represent the confidence interval. Mark the endpoints with short vertical tick marks. Label the confidence level (95%) above the interval and the margin of error (±0.047) below it.
Section 12.11
Margin of Error
The margin of error is the distance from the point estimate to either endpoint of the confidence interval:
Margin of Error Formula
ME = z* · SE(p̂) = z* · √[ p̂(1–p̂) / n ]
What Determines the Margin of Error
| Factor | Effect on Margin of Error |
|---|
| Higher confidence level | Larger z* → wider interval → larger margin of error |
| Larger sample size n | Smaller SE → narrower interval → smaller margin of error |
| p̂ near 0.50 | Maximum p̂(1–p̂) → largest SE and largest margin of error for that n |
Example 12.11.1
Effect of sample size on margin of error
- ProblemA researcher expects p̂ ≈ 0.50. Compare the 95% margins of error for samples of n = 100, n = 400, and n = 1600.
- Calculate
n = 100: ME = 1.96 × √[0.50×0.50/100] = 1.96 × 0.050 = 0.098 (≈ ±9.8%)
n = 400: ME = 1.96 × √[0.50×0.50/400] = 1.96 × 0.025 = 0.049 (≈ ±4.9%)
n = 1600: ME = 1.96 × √[0.50×0.50/1600] = 1.96 × 0.0125 = 0.0245 (≈ ±2.5%)
- InterpretQuadrupling the sample size (100→400 or 400→1600) cuts the margin of error in half. This follows from the formula: the standard error decreases with 1/√n, so to halve the margin of error, the sample size must be multiplied by four. This is why high-precision national polls often use samples of 1,000–1,500 respondents.
To halve the margin of error, quadruple the sample size. Going from n = 100 to n = 1600 reduces ME from ±9.8% to ±2.5%.
Data Ethics Note
The margin of error in published polls reflects only sampling variability from random sampling. It does not account for bias from poor question wording, nonresponse, voluntary response, or coverage error (the sampling frame not matching the actual population). A poll with a margin of error of ±3% but serious design flaws can be far more misleading than a poll with a margin of error of ±6% that was carefully designed. Always ask: was the sample truly random?
Section 12.12
Interpreting Confidence Intervals
Confidence intervals are frequently misinterpreted — even by people who report them in news media and research papers. The key is understanding what "95% confident" actually means.
The true population proportion p is a fixed number. It either is or is not inside any specific computed interval. There is no probability involved once the interval is computed. The correct interpretation refers to the method, not the specific interval.
The Correct Interpretation
"If this confidence interval procedure were repeated many times with different random samples of the same size from the same population, approximately 95% of the resulting intervals would contain the true population proportion."
Or, more practically: "We are 95% confident that the true population proportion falls between [lower bound] and [upper bound]."
| Incorrect Statement | Correct Statement |
|---|
| "There is a 95% probability that p falls in this interval." | "We are 95% confident that p falls in this interval." |
| "95% of all individuals in the population have values in this interval." | The interval estimates the population parameter, not individual values. |
| "95% of all possible samples will produce a p̂ in this interval." | The interval captures the population parameter, not future sample statistics. |
Example 12.12.1
Writing a correct confidence interval interpretation
- ProblemA 95% confidence interval for the proportion of adults who read at least one book per month is (0.31, 0.43). Write a correct interpretation.
- ThinkThe interval was constructed from sample data. The confidence level refers to the reliability of the method, not the probability for this specific interval.
- InterpretWe are 95% confident that the true proportion of all adults who read at least one book per month is between 0.31 and 0.43, or between 31% and 43%. If this estimation procedure were repeated with many different random samples of the same size, approximately 95% of the resulting confidence intervals would contain the true population proportion.
We are 95% confident the true proportion is between 31% and 43%. This refers to the method's long-run reliability, not to the probability for this single interval.
Display Instruction
Draw a vertical line representing the true (unknown) population proportion p. Draw 20 horizontal line segments representing 20 different confidence intervals, each centered at a different sample proportion. Show 19 of the 20 intervals crossing the vertical line (capturing p) and 1 interval that falls entirely to one side (missing p). Label this display: "Approximately 95% of all confidence intervals constructed this way will capture the true parameter."
Student Warning
It is a very common mistake to say "there is a 95% probability that p is in this interval." Once the interval is computed, the true value either is or is not inside it — we simply do not know which. The 95% confidence refers to the reliability of the method, not the probability for any specific computed interval.
Section 12.13
Estimation in Polling, Science, and Quality Control
Polling and Public Opinion
Political and opinion polls routinely report margins of error. A poll might report: "48% of likely voters support Candidate A, with a margin of error of ±3 percentage points." This means the 95% confidence interval is (0.45, 0.51). Because this interval includes 0.50, the data do not provide enough evidence to confidently say which candidate leads — the candidates are said to be in a "statistical tie."
Scientific Research
A drug study might find that a treatment reduced blood pressure by an average of 8 mmHg in a sample of 80 patients. A 95% confidence interval for the true mean reduction might be (5.2, 10.8) mmHg. This interval tells the scientific community not just the observed effect, but how precisely it was measured. A narrow interval means the effect is estimated reliably. A wide interval means more research may be needed.
Example 12.13.1
Interpreting a polling confidence interval
- ProblemA news organization reports: "In a poll of 1,050 randomly selected adults, 61% said they are satisfied with their local public schools. The margin of error is ±3 percentage points at the 95% confidence level." (a) State the 95% confidence interval. (b) Interpret in context. (c) Can we conclude that more than half of all adults are satisfied?
- Solve (a)
CI = 0.61 ± 0.03 = (0.58, 0.64)
- Solve (b)We are 95% confident that the true proportion of all adults satisfied with local public schools is between 58% and 64%.
- Solve (c)Yes. The entire confidence interval (0.58 to 0.64) lies above 0.50. Even the lower bound is above 50%. This provides reasonable evidence that more than half of all adults are satisfied.
CI: (0.58, 0.64). Since the entire interval exceeds 0.50, the data support the conclusion that a majority are satisfied.
Section 12.14
Common Estimation Mistakes
Mistake 1 — Treating the confidence interval as a probability statement about the parameter
"There is a 95% probability that the true proportion is in my interval."
Correction: The parameter is fixed. The 95% refers to the method's long-run reliability. Say: "We are 95% confident the true proportion falls in this interval."
Mistake 2 — Thinking a larger confidence level always gives a "better" interval
"A 99% confidence interval is better than a 95% interval because it's more reliable."
Correction: A 99% interval is wider, meaning less precise. Greater confidence comes at the cost of precision. If precision matters, the answer is a larger sample size, not just a lower confidence level.
Mistake 3 — Confusing standard deviation and standard error
Using s = 3.2 hours directly in a confidence interval formula instead of SE = s/√n.
Correction: Standard deviation measures variability among individual data values. Standard error measures variability among sample statistics. Use standard error in a confidence interval formula.
Mistake 4 — Assuming margin of error accounts for all sources of error
"A poll with a ±3% margin of error is reliable, so we can trust its results completely."
Correction: The margin of error reflects only sampling variability from random sampling. It does not account for nonresponse bias, poorly worded questions, a bad sampling frame, or response bias. A technically correct margin of error does not validate flawed data collection.
Mistake 5 — Thinking the interval describes the distribution of individual responses
"The 95% CI (0.52, 0.68) means 95% of people surveyed own pets."
Correction: The interval estimates the population proportion parameter, not the range of individual responses. The 95% refers to the reliability of the method across repeated samples.
Mistake 6 — Concluding two groups are "the same" because intervals overlap
"Both groups have overlapping confidence intervals, so they must have the same population proportion."
Correction: Overlapping confidence intervals do not necessarily mean population parameters are identical. A formal hypothesis test (Chapter 13) is needed to properly compare two groups.
Mistake 7 — Computing a confidence interval without checking conditions
Constructing a 95% CI from a voluntary-response web poll of n = 15 respondents.
Correction: Always verify the three conditions first — random sampling, large enough sample (n·p̂ ≥ 10 and n·(1–p̂) ≥ 10), and independence. If conditions fail, the formula does not apply.
Data Ethics Note
Misrepresenting the margin of error is one of the most common ways that polling results are used deceptively in public discourse. When a journalist ignores the margin of error and declares a race "won" when results are within that margin, they are misusing statistical tools. Students of statistics have a responsibility to read margins of error and confidence intervals carefully — and to push back when they are reported incorrectly.
Chapter 12 — Back Matter
Chapter Summary
- Sec 12.1Every random sample produces a different result — this is sampling variability. It is expected and predictable, not a mistake or bias. The variation itself follows a mathematical pattern called a sampling distribution.
- Sec 12.2Sampling variability is driven by sample size and population variability. Larger samples produce statistics that cluster more tightly around the true parameter. Population and sample notation (p, p̂, μ, x̄, σ, s, n) should be used precisely and consistently.
- Sec 12.3Simulations reveal the pattern of sampling variability directly: sample statistics cluster near the true parameter, form an approximately bell-shaped distribution, and vary in predictable amounts that decrease as sample size increases.
- Sec 12.4The sampling distribution of p̂ is approximately normal when np ≥ 10 and n(1–p) ≥ 10, with mean p and standard deviation √[p(1–p)/n]. Sample proportion is an unbiased estimator of the population proportion.
- Sec 12.5The sampling distribution of x̄ has mean μ and standard deviation σ/√n (the standard error of the mean). It is exactly normal when the population is normal; otherwise approximately normal by the CLT for large n.
- Sec 12.6Both p̂ and x̄ are unbiased estimators. Their standard errors decrease as n increases. Shape approaches normal under the conditions described for each type. These three properties — shape, center, spread — fully characterize a sampling distribution.
- Sec 12.7The Central Limit Theorem: for large enough n, the sampling distribution of x̄ is approximately normal regardless of the population's shape. A common guideline is n ≥ 30. The CLT is why normal methods apply so broadly in inference.
- Sec 12.8Standard error is the standard deviation of a sampling distribution. For p̂: SE = √[p̂(1–p̂)/n]. For x̄: SE = s/√n. Standard error is not the same as standard deviation — it measures variation among sample statistics, not individual values.
- Sec 12.9A point estimate is a single sample value used to estimate a population parameter. It is useful but carries no information about its own uncertainty. p̂ estimates p; x̄ estimates μ.
- Sec 12.10A confidence interval is a point estimate ± margin of error. For proportions: p̂ ± z* · √[p̂(1–p̂)/n]. Conditions must be verified before use. The three common critical values are z* = 1.645 (90%), 1.960 (95%), 2.576 (99%).
- Sec 12.11Margin of error = z* · SE. Larger n reduces the margin of error; higher confidence increases it. The margin of error reflects only sampling variability — not bias or measurement error from flawed data collection.
- Sec 12.12A 95% confidence interval means the method captures the true parameter 95% of the time in the long run. It is incorrect to say "there is a 95% probability the true value is in this interval." The confidence level refers to the method, not to any single computed interval.
- Sec 12.13Confidence intervals appear in polling, scientific research, and quality control. In polling, a result "within the margin of error" is a statistical tie. Margins of error are almost always reported at the 95% confidence level.
- Sec 12.14Seven recurring errors: treating CI as probability, preferring wider CIs, confusing SD with SE, overstating what margin of error covers, misapplying the interval to individuals, misinterpreting overlapping intervals, and skipping condition checks.
Key Terms Review
sampling variability
Natural variation among sample statistics across repeated random samples from the same population.
sampling distribution
Distribution of a statistic across all possible samples of a given size from the same population.
unbiased estimator
A statistic whose mean over all possible samples equals the true population parameter.
standard error
The standard deviation of a sampling distribution; measures typical variation among sample statistics.
Central Limit Theorem
For large enough n, the sampling distribution of x̄ is approximately normal regardless of population shape.
point estimate
A single sample value (p̂ or x̄) used to estimate a population parameter.
confidence interval
An interval of values designed to capture the true population parameter with a stated level of confidence.
confidence level
The long-run percentage of confidence intervals, built the same way, that capture the true parameter.
margin of error
The distance from the point estimate to either endpoint of the confidence interval; = z* · SE.
critical value (z*)
The multiplier in the confidence interval formula that corresponds to the chosen confidence level.
sample proportion (p̂)
The proportion with a characteristic in the sample; an unbiased estimator of the population proportion p.
sample mean (x̄)
The arithmetic average of sample values; an unbiased estimator of the population mean μ.
Concept Review Questions
- In your own words, explain why different random samples from the same population produce different statistics.
- What is a sampling distribution? How does it differ from the distribution of a single dataset?
- What does it mean for a statistic to be an unbiased estimator? Give one example.
- What happens to the standard error of a sample proportion as sample size increases? Explain why this makes sense.
- State the Central Limit Theorem in plain language. Why is it remarkable?
- Explain the difference between standard deviation and standard error. When do you use each?
- A researcher reports a 95% confidence interval of (0.41, 0.57). What does this interval mean in context?
- Why is it incorrect to say "there is a 95% probability that the true proportion falls in this interval"?
- What does the margin of error measure? Name two things it does not measure.
- List all three conditions that must be satisfied before a confidence interval for a proportion is valid.
Skill Practice
- p = 0.40, n = 80. Describe shape, mean, and standard deviation. Verify conditions.
- p = 0.25. Find n such that σ(p̂) = 0.03.
- p = 0.60. Compare σ(p̂) for n = 25, n = 100, n = 400.
- p = 0.10, n = 40. Check the conditions. Can the normal approximation be used?
- p = 0.75, n = 200. Find the mean and standard deviation of the sampling distribution.
- μ = 55, σ = 12, n = 36. Describe the sampling distribution of x̄.
- Right-skewed population, μ = 14.5, σ = 5.0, n = 50. Describe the sampling distribution. What theorem applies?
- Normal population, μ = 100, σ = 15, n = 9. Describe the sampling distribution.
- What sample size is needed to reduce SE(x̄) from 4.0 to 2.0 when σ = 20?
- Two samples from the same population: n = 30 and n = 270, σ = 15. Compute SE(x̄) for each. What is their ratio?
- n = 150, p̂ = 0.64. Compute SE(p̂).
- n = 48, x̄ = 182 g, s = 24 g. Compute SE(x̄).
- n = 200, p̂ = 0.50. Find SE(p̂). Then find SE(p̂) if n = 800.
- Survey: 182 of 250 adults exercise 3x/week. Compute p̂.
- Sample: 36 items, x̄ = 74.2, s = 9.0. Compute the point estimate and SE(x̄).
- n = 300, 186 use social media daily. Construct a 95% CI. Verify conditions.
- n = 500 voters, p̂ = 0.51. Construct a 90% CI.
- n = 200, p̂ = 0.15. Construct a 99% CI. Verify conditions.
- Target ME ≤ 0.04 at 95%, p̂ ≈ 0.50. Find minimum n.
- n = 120, p̂ = 0.07 defective. Construct a 95% CI. Interpret in context.
- p̂ = 0.52, n = 900. Compute ME at 95% confidence.
- Find ME for n = 100, 400, 1600 with p̂ = 0.50 at 95%.
- A poll reports ME = ±4%. Find the approximate sample size used (assuming p̂ ≈ 0.50, 95%).
- How does ME change when confidence level increases from 90% to 99%? Compute using p̂ = 0.40, n = 250.
Applications
- A national health organization surveys 1,000 randomly selected adults and finds that 380 report meeting recommended weekly exercise guidelines. (a) Compute p̂. (b) Compute SE(p̂). (c) Construct a 95% CI. (d) Interpret in context. (e) The organization claims "fewer than 40% of adults meet the guidelines." Does the confidence interval support this claim?
- A seed company claims 90% germination. A botanist randomly plants 250 seeds under controlled conditions; 214 germinate. (a) Compute p̂. (b) Construct a 95% CI. (c) Does the interval support the company's claim? Explain.
- A city surveys 600 residents; 54% support a new transit plan. (a) Compute the 95% ME. (b) Construct and interpret the CI. (c) The plan requires more than 50% support to pass. What does the CI suggest?
- A quality control engineer samples 100 circuit boards and finds p̂ = 0.06 are defective. Construct a 95% CI. If the company's standard requires fewer than 10% defects in the population, does the CI suggest compliance?
- A student surveys 80 classmates and finds that 52 prefer hybrid learning. Construct a 90% CI. Verify all conditions. Interpret in context.
Interpretation Practice
- A 95% confidence interval for the proportion of adults who own a pet is (0.52, 0.68). Determine which of the following interpretations are correct and which are incorrect. Explain each. (a) "There is a 95% probability that the true proportion is between 0.52 and 0.68." (b) "We are 95% confident that the true proportion of all adults who own a pet is between 52% and 68%." (c) "95% of all adults surveyed own pets." (d) "If this survey were repeated many times, approximately 95% of the resulting confidence intervals would contain the true proportion."
- A 95% CI for a school survey proportion is (0.61, 0.79). A researcher claims this proves that more than half of all students prefer the new schedule. Is this claim supported? Explain.
- One survey produces a 95% CI of (0.44, 0.56). Another produces a 95% CI of (0.48, 0.52). Both are based on the same population. What explains the difference in width?
Error Analysis
Problem 1
"My 95% confidence interval is (0.42, 0.56), so there is a 95% chance the true proportion is in this range."
Explain the error. Write a correct interpretation of the same interval.
Problem 2
A researcher says "I switched from a 95% to a 99% CI to get a more useful, more precise estimate."
Identify the error in reasoning. What actually changes when you increase the confidence level?
Problem 3
A student computes SE = s = 3.2 hours and uses it directly in a confidence interval formula, skipping the √n step.
Identify the error. What should have been computed, and how?
Problem 4
"A poll of 50 voters found 52% prefer Candidate A, with a margin of error of ±14 points at 95% confidence. This is a statistical tie." Verify the margin of error using the formula. Is the "statistical tie" conclusion justified?
Compute the margin of error from scratch using p̂ = 0.52, n = 50, z* = 1.96. Is the article's conclusion valid?
Problem 5
A student collects data by standing at the cafeteria exit and asking students about lunch preferences. She computes a 95% CI and her classmate says: "The margin of error covers any problems with how you collected the data."
Is the classmate correct? What does the margin of error actually cover, and what does it not cover?
Chapter 12 Checkpoint
Chapter 12 Checkpoint
Complete all parts without assistance — assesses all key skills and concepts from Chapter 12
Checkpoint 12.1 — Sampling Distribution of a Proportion
A large school district reports that 45% of all students participate in at least one extracurricular activity. A researcher plans to survey a random sample of 100 students.
- Check the conditions necessary to use the normal approximation for p̂.
- Describe the shape, center, and standard deviation of the sampling distribution of p̂.
- What is the probability that the sample proportion falls between 0.40 and 0.50? (Use z-scores and the properties of the normal distribution.)
Checkpoint 12.2 — Central Limit Theorem
The number of minutes students spend on homework each night has mean μ = 45 min and standard deviation σ = 18 min. The distribution is moderately right-skewed.
- A researcher takes random samples of size n = 36. Describe the sampling distribution of x̄.
- What theorem justifies the shape you described?
- Would the same conclusion hold for samples of n = 5? Explain.
Checkpoint 12.3 — Standard Error
A random sample of 180 voters found that p̂ = 0.62 support a ballot measure.
- Compute the standard error of p̂.
- Interpret the standard error in context.
- How would the standard error change if the sample size increased to 720?
Checkpoint 12.4 — Confidence Interval
A random sample of 350 adults found that 203 had visited a dentist in the past year.
- Compute the point estimate p̂.
- Check the conditions for a valid confidence interval.
- Compute the 95% confidence interval.
- Interpret the confidence interval in context.
- A dental association claims that more than 60% of adults visit a dentist annually. Does the confidence interval support this claim? Explain carefully.
Checkpoint 12.5 — Interpretation
A political analyst reports a 95% confidence interval of (0.48, 0.56) for the proportion of voters who support a ballot measure.
- What is the point estimate?
- What is the margin of error?
- Can the analyst conclude that the majority of voters support the measure? Explain.
- The analyst says: "I am 95% sure the true proportion is in this range." Is this a correct interpretation? If not, rewrite it correctly.
- Would a 99% confidence interval be narrower or wider? Explain without calculating.
Answers may be placed in the answer key.
Looking Ahead
Chapter 13: Hypothesis Testing
Chapter 12 introduced estimation — one of the two main branches of statistical inference. Chapter 13 introduces the other: hypothesis testing. Where a confidence interval asks "What is the plausible range for the population parameter?", a hypothesis test asks "Is there enough evidence to reject a specific claim about the population?" You will learn to write null and alternative hypotheses, compute test statistics, interpret p-values, make decisions using significance levels, and understand Type I and Type II errors. Every skill from this chapter — standard error, sampling distributions, normal approximations, and the logic of long-run probability — carries forward directly into hypothesis testing.
Bridge to College Statistics and Data Science
Chapter 12 introduces estimation — one of the two main branches of statistical inference. In college statistics courses and beyond, the ideas developed here extend in several important directions.
Confidence intervals for means use the t-distribution rather than the standard normal, because the population standard deviation is typically unknown and must be estimated from the sample. The t-distribution, introduced by William Gosset in 1908, accounts for the extra uncertainty that introduces.
Confidence intervals for differences allow researchers to compare two populations — for example, the difference in means between a treatment group and a control group, or the difference in proportions between two demographics.
Sample size planning uses the margin of error formula solved for n, so researchers can determine how large a sample is needed before data collection begins.
Bootstrap methods use repeated resampling from the original data to estimate the sampling distribution empirically, without relying on mathematical formulas. These computational approaches are covered in Data Science Foundations and modern college statistics courses.
MGU Library Connection
- Probability and Statistics: formal derivations of the Central Limit Theorem and t-distributions
- Data Science Foundations: simulation-based inference, bootstrapping, and resampling methods
- Appendix K: Normal Distribution Reference — z-tables and area calculations
- Appendix L: Sampling Distribution Reference — formulas for p̂ and x̄
- Appendix M: Confidence Interval Reference — formulas, conditions, critical values, and interpretation guide
- Appendix O: Calculator and Spreadsheet Skills — confidence interval computations with technology