AP Statistics hypothesis tests cheat sheet

By Jude Wallis · Published

A significance test states hypotheses about a parameter, computes a test statistic, and finds a p-value assuming the null hypothesis is true. If the p-value is at most alpha, reject the null; otherwise fail to reject it. The conclusion is written in context in terms of the alternative.

Significance tests are developed in Unit 3 of the Fall 2026 AP Statistics course (topics 3.5 through 3.8 and 3.12 through 3.15) and applied to means in Unit 4 (topics 4.4, 4.5, 4.9, and 4.10). The course has no test for a regression slope and no chi-square goodness-of-fit test.

The steps of a test

StepWhat to write
HypothesesH0H_0 and HaH_a in symbols, with the parameter defined in context
Procedure and conditionsThe name of the test, and each condition checked with numbers
Test statistic and p-valueThe value of zz, tt, or χ2\chi^2, the degrees of freedom if any, and the p-value
DecisionThe p-value compared with α\alpha, and "reject H0H_0" or "fail to reject H0H_0"
ConclusionA sentence in context about the evidence for HaH_a

H0H_0 ("H-naught") is the null hypothesis, HaH_a is the alternative hypothesis, and α\alpha ("alpha") is the significance level, chosen before looking at the data. The procedures and their conditions are in which statistical test to use and the conditions for inference checklist.

Stating hypotheses

TestH0H_0HaH_a is one of
One proportionp=p0p = p_0p<p0p < p_0, p>p0p > p_0, p≠p0p \neq p_0
Two proportionsp1=p2p_1 = p_2p1<p2p_1 < p_2, p1>p2p_1 > p_2, p1≠p2p_1 \neq p_2
One meanμ=μ0\mu = \mu_0μ<μ0\mu < \mu_0, μ>μ0\mu > \mu_0, μ≠μ0\mu \neq \mu_0
Mean difference (paired)μd=0\mu_d = 0μd<0\mu_d < 0, μd>0\mu_d > 0, μd≠0\mu_d \neq 0
Two meansμ1=μ2\mu_1 = \mu_2μ1<μ2\mu_1 < \mu_2, μ1>μ2\mu_1 > \mu_2, μ1≠μ2\mu_1 \neq \mu_2
Chi-square for homogeneityThe distributions of the variable are the same for all populations or treatmentsThe distributions are not all the same
Chi-square for independenceThere is no association between the two variablesThere is an association between the two variables

Rules that apply to every row:

  • Hypotheses are about population parameters such as pp and μ\mu. They are never about sample statistics such as p^\hat{p} and xˉ\bar{x}.
  • H0H_0 contains the equality. HaH_a is the claim the study is looking for evidence of.
  • The direction of HaH_a comes from the question asked, before the data are examined.
  • Define each parameter in words, including the population.

See null vs alternative hypothesis and one-tailed vs two-tailed tests.

Test statistic and p-value

For z and t tests, the test statistic measures how many standard errors the sample statistic is from the null value:

test statistic=statistic−null valuestandard error\text{test statistic} = \frac{\text{statistic} - \text{null value}}{\text{standard error}}

The p-value is the probability, computed assuming H0H_0 is true, of getting a test statistic as extreme as or more extreme than the one observed, in the direction of HaH_a.

HaH_aP-value
Parameter >> null valueArea to the right of the test statistic
Parameter << null valueArea to the left of the test statistic
Parameter ≠\neq null valueTwice the area in the tail beyond the test statistic
Chi-square testArea to the right of χ2\chi^2

A small p-value means the observed result would be unusual if H0H_0 were true, which is evidence for HaH_a. The p-value is not the probability that H0H_0 is true. See what a p-value means and the p-value calculator.

Decision and conclusion sentences

ResultDecisionConclusion sentence
p-value ≤α\leq \alphaReject H0H_0"Because the p-value of [value] is less than or equal to α=\alpha = [value], we reject H0H_0. There is convincing evidence that [HaH_a in context]."
p-value >α> \alphaFail to reject H0H_0"Because the p-value of [value] is greater than α=\alpha = [value], we fail to reject H0H_0. There is not convincing evidence that [HaH_a in context]."

Both conclusions are written about HaH_a. Only the word "not" changes.

Wording to avoidUse instead
"Accept H0H_0""Fail to reject H0H_0"
"This proves that...""There is convincing evidence that..."
"H0H_0 is true""There is not convincing evidence that [HaH_a]"
A conclusion with no contextName the parameter and the population

Failing to reject H0H_0 does not show that H0H_0 is true. It means the data are consistent with H0H_0. For a randomized experiment, the conclusion may describe a cause-and-effect relationship. See how to write a hypothesis test conclusion.

Type I error, Type II error, and power

DecisionH0H_0 is trueH0H_0 is false
Reject H0H_0Type I errorCorrect decision
Fail to reject H0H_0Correct decisionType II error
QuantityDefinitionProbability
Type I errorRejecting H0H_0 when H0H_0 is trueα\alpha
Type II errorFailing to reject H0H_0 when H0H_0 is false1−power1 - \text{power}
PowerRejecting H0H_0 when H0H_0 is false1−P(Type II error)1 - P(\text{Type II error})
Change, with everything else held fixedPowerProbability of a Type II error
Larger sample sizeIncreasesDecreases
Larger α\alphaIncreasesDecreases
True parameter farther from the null valueIncreasesDecreases
Smaller standard errorIncreasesDecreases

Raising α\alpha increases power and also increases the probability of a Type I error. Describe each error in context: what was decided, what was actually true, and what the consequence would be. See type 1 vs type 2 errors and the power and errors interactive.

Tests and confidence intervals together

A two-sided test and a confidence interval answer related questions.

ProcedureRelationship
One-sample t-test of μ=μ0\mu = \mu_0, two-sided, level α\alphaRejects H0H_0 when μ0\mu_0 is outside the (1−α)(1 - \alpha) confidence interval for μ\mu, and fails to reject when μ0\mu_0 is inside it
Tests for proportionsUsually agree with the matching interval, but can differ near the boundary because the test and the interval use different standard errors

A 95% confidence level matches α=0.05\alpha = 0.05 for a two-sided test. A test gives the strength of evidence against one null value. An interval gives the full set of plausible values. See confidence interval vs hypothesis test and when a confidence interval and p-value disagree.

One-sample z-test for a proportion, with the errors in context

A company reports that 25% of its roughly 5,000 customers use its app. A manager believes the proportion is now higher. In a random sample of 160 customers, 52 use the app. Carry out a test at α=0.05\alpha = 0.05, and describe a Type I error in context.

  1. Hypotheses: H0:p=0.25H_0: p = 0.25 and Ha:p>0.25H_a: p > 0.25, where pp is the proportion of all the company's customers who use the app.

  2. Conditions: the sample is random. 0.10(5,000)=5000.10(5{,}000) = 500 and 160≤500160 \leq 500. np0=160(0.25)=40np_0 = 160(0.25) = 40 and n(1−p0)=160(0.75)=120n(1-p_0) = 160(0.75) = 120 are both at least 10.

  3. Sample proportion: p^=52160=0.325\hat{p} = \frac{52}{160} = 0.325.

  4. Standard error using p0p_0: 0.25(0.75)160=0.001172=0.0342\sqrt{\frac{0.25(0.75)}{160}} = \sqrt{0.001172} = 0.0342.

  5. Test statistic: z=0.325−0.250.0342=2.19z = \frac{0.325 - 0.25}{0.0342} = 2.19.

  6. P-value: the area to the right of z=2.19z = 2.19 is 1−0.9857=0.01431 - 0.9857 = 0.0143.

  7. Decision: 0.0143≤0.050.0143 \leq 0.05, so reject H0H_0.

  8. Type I error in context: concluding that more than 25% of customers use the app when the proportion is actually 25%.

z=2.19z = 2.19, p-value about 0.014. Because the p-value is less than 0.05, we reject H0H_0. There is convincing evidence that more than 25% of the company's customers use the app.

Two-sided one-sample t-test that fails to reject

A machine is set to fill bags with 100 grams of coffee. A random sample of 20 bags from a day's production of several thousand has mean 103.5 grams and standard deviation 8.0 grams, and a dotplot shows no strong skewness and no outliers. Test at α=0.05\alpha = 0.05 whether the mean fill differs from 100 grams.

  1. Hypotheses: H0:μ=100H_0: \mu = 100 and Ha:μ≠100H_a: \mu \neq 100, where μ\mu is the mean fill, in grams, of all bags produced that day.

  2. Conditions: the sample is random, 20 bags are less than 10% of several thousand, and with n=20<30n = 20 < 30 the dotplot shows no strong skewness and no outliers.

  3. Standard error: 8.020=8.04.4721=1.7889\frac{8.0}{\sqrt{20}} = \frac{8.0}{4.4721} = 1.7889.

  4. Test statistic: t=103.5−1001.7889=1.957t = \frac{103.5 - 100}{1.7889} = 1.957 with df=20−1=19df = 20 - 1 = 19.

  5. P-value: the area to the right of 1.957 with 19 degrees of freedom is about 0.0326, and the test is two-sided, so the p-value is 2(0.0326)=0.0652(0.0326) = 0.065.

  6. Decision: 0.065>0.050.065 > 0.05, so fail to reject H0H_0.

t=1.957t = 1.957, df=19df = 19, p-value about 0.065. Because the p-value is greater than 0.05, we fail to reject H0H_0. There is not convincing evidence that the mean fill of all bags produced that day differs from 100 grams.

Frequently asked questions

What does a p-value mean?

It is the probability of getting a result at least as extreme as the one observed, in the direction of HaH_a, if H0H_0 is true. It is computed assuming H0H_0, so it cannot be the probability that H0H_0 is true.

Why do we say fail to reject instead of accept?

A large p-value means the data are consistent with H0H_0. They are also consistent with many other parameter values close to the null value, so the test has not shown that H0H_0 is true.

How are the significance level and a Type I error related?

The significance level α\alpha is the probability of a Type I error: rejecting H0H_0 when it is true. Choosing a smaller α\alpha makes a Type I error less likely and, with everything else fixed, makes a Type II error more likely.