Inference for Quantitative Data: Means

What this unit covers

The topics below follow the published Statistics course framework for Unit 4. This unit is worth 10–20% of the exam, so budget your time against that rather than against how long the unit takes to teach.

Lessons in this unit

• Probability Rules14 min · 3 objectivesApply the complement, addition, and multiplication rules · Distinguish mutually exclusive events from independent events · Use the general addition rule to avoid double-counting
• Random Variables: Mean & Variance14 min · 3 objectivesCompute the expected value (mean) of a discrete random variable · Compute and interpret the variance and standard deviation of a random variable · Apply rules for the mean and variance of transformed and combined variables
• The Binomial Distribution15 min · 3 objectivesVerify the four conditions (BINS) for a binomial setting · Compute binomial probabilities and the mean and standard deviation · Distinguish "exactly," "at most," and "at least" probability requests
• The Geometric Distribution13 min · 3 objectivesRecognize a geometric setting and contrast it with a binomial setting · Compute the probability that the first success occurs on trial k · Find the expected number of trials until the first success
• Conditional Probability, Trees & the Rare-Disease Trap14 min · 3 objectivesCompute conditional probabilities from a tree diagram or two-way table · Distinguish P(A | B) from P(B | A) · Explain why a highly accurate test for a rare condition still yields many false positives
• Combining Random Variables: Means Always Add, Variances Do Not14 min · 3 objectivesCompute the mean and variance of a sum or difference of random variables · Explain why variances add for a difference as well as a sum · Identify when the independence condition fails

Formulas in Unit 4

Every term in Unit 4

All 36 terms we publish for Inference for Quantitative Data: Means, with definitions. Reading them through is the fastest way to find the ones you cannot define — then drill those in cram mode until you can produce them without the prompt.

Why t rather than z

The population standard deviation is unknown and estimated by s, which adds variability. The t distribution has heavier tails to account for it.

Degrees of freedom for one sample

df = n − 1. As df grows, the t distribution approaches the standard normal.

Conditions for t procedures

Random sample, 10% condition, and Normal/Large Sample — the population is normal, n ≥ 30, or a graph of the data shows no strong skew or outliers.

One-sample t-interval

x̄ ± t*(s/√n), interpreted as capturing the true population mean in context.

One-sample t-test statistic

t = (x̄ − μ₀)/(s/√n), compared to a t distribution with n − 1 degrees of freedom.

Matched pairs t-test

Compute the difference for each pair and run a one-sample t-test on those differences. Using a two-sample test here is a standard and costly error.

Two-sample t-interval

(x̄₁ − x̄₂) ± t*√(s₁²/n₁ + s₂²/n₂), with technology supplying the degrees of freedom.

Deciding one-sample or two-sample

Two independent groups means two-sample; two measurements on the same or paired subjects means matched pairs.

Robustness of t procedures

They perform well even when the population is not exactly normal, provided the sample is large or the data show no strong skew or outliers.

Choosing a significance level

Use a smaller α when a false positive is costly, and a larger one when missing a real effect is worse.

Reading a t-table

Find the row for your degrees of freedom, then the column for your confidence level or tail probability. Between-row values are rounded conservatively downward.

Checking normality from a sample

Graph the data. A dotplot or boxplot showing no strong skew and no outliers satisfies the condition for a moderate sample.

Why the 10% condition exists

Sampling without replacement makes observations slightly dependent; keeping the sample under 10% of the population makes the effect negligible.

Interpreting a t-interval for a difference

If the interval contains zero, there is no convincing evidence the two means differ. Say this in context rather than stating it abstractly.

Paired data recognition

Two measurements on the same subject, or naturally matched subjects. If you can subtract within a pair meaningfully, the design is paired.

Shape of the t distribution

Symmetric and bell-shaped like the Normal but with heavier tails, reflecting the extra uncertainty from estimating σ with s. Heavier tails mean larger critical values and wider intervals.

t approaches z

As degrees of freedom increase, s estimates σ more precisely and the t distribution converges to the standard Normal. By df of a few hundred the difference is negligible.

Degrees of freedom, two samples

The exact formula is messy, so software reports a non-integer value. The conservative hand method uses the smaller of n₁ − 1 and n₂ − 1, which gives a slightly wider interval.

Standard error of the mean

s/√n — an estimate of how much x̄ varies from sample to sample, not how much the data vary. Confusing it with s is a frequent conceptual error.

Standard error of a difference of means

√(s₁²/n₁ + s₂²/n₂). Variances add even though the parameter is a difference, so the standard error of a difference exceeds either individual standard error.

Pooled versus unpooled t

Pooling assumes the two population standard deviations are equal, which is rarely justified. The unpooled procedure is the default and is what calculators report by default.

Conditions for two-sample t

Two independent random samples or two randomly assigned groups; the 10% condition for each if sampling without replacement; and Normality checked separately in each group.

Paired or two-sample?

Ask whether each observation in one group is naturally linked to one specific observation in the other. If yes, reduce to differences and use a one-sample procedure with df = (number of pairs) − 1.

A two-mean interval containing zero

No convincing evidence of a difference between the two means. Note this is not evidence that they are equal — the interval may simply be wide.

One-sample t statistic

t = (x̄ − μ₀)/(s/√n) with df = n − 1. The numerator is the observed departure from the null; the denominator converts it into standard errors.

How robust t procedures actually are

t procedures remain approximately valid when the Normality condition is mildly violated. They tolerate more skew as n grows, which is why the condition is stated in terms of both shape and sample size.

Outliers and t procedures

Outliers are more damaging than skew, because they inflate s and shift x̄ simultaneously. An outlier in a small sample is grounds for not using a t procedure at all.

Checking Normality from the sample graph

Graph the data: a dotplot, boxplot or Normal probability plot. State what the graph shows rather than asserting the condition is met, since the graph is the evidence.

Reporting degrees of freedom

Always state df with a t procedure, because the critical value depends on it. An interval quoted without df cannot be checked by a reader.

Confidence interval for a difference of means

(x̄₁ − x̄₂) ± t*√(s₁²/n₁ + s₂²/n₂). Interpret it as an interval for the difference, not for either mean separately.

Margin of error for a mean

t*(s/√n). It shrinks with √n and grows with the confidence level and with the variability of the data.

Sample size for a mean

n = (z*σ/ME)², which requires an estimate of σ from a pilot study or prior data. This is why sample-size planning for means is harder than for proportions.

Why σ is almost never known

Knowing the population standard deviation while not knowing its mean is an unusual situation. This is why t procedures, not z procedures, are the practical default for means.

Interpreting the magnitude of t

A t of 2 means the sample mean sits two standard errors from the null value. Large |t| indicates a departure unlikely under the null; the p-value converts it to a probability.

One-sided p-value from two-sided output

Software usually reports a two-sided p-value. If the alternative is one-sided and the data fall in the predicted direction, halve it; if they fall the other way, the one-sided p-value exceeds 0.5.

Why paired designs need fewer subjects

Pairing removes between-subject variation, so the standard error is computed from within-subject differences only. The same effect is detectable with far fewer individuals.

What examiners penalize here

• When combining independent random variables, **variances add even when you subtract the variables**. To find the standard deviation of a difference X − Y, compute σ²_X + σ²_Y first, then take the square root. Never subtract standard deviations, and never add standard deviations directly.
• The quickest way to tell binomial from geometric: is the *number of trials fixed*? Fixed n, count successes → **binomial**. Keep going until the first success, count the trials → **geometric**. The phrase "until" almost always means geometric.

Practice Statistics

Our practice bank is drawn from across the whole course rather than filtered to one unit, which is closer to how the exam asks anyway — it will not tell you which unit a question is testing.

Questions about this unit

All 5 units of AP Statistics

1. Unit 1 · Exploring One-Variable Data and Collecting Data
2. Unit 2 · Probability, Random Variables, and Probability Distributions
3. Unit 3 · Inference for Categorical Data: Proportions
4. Unit 4 · Inference for Quantitative Data: Means
5. Unit 5 · Regression Analysis