AP Statistics inference procedures cheat sheet

By Jude Wallis · Published

Choose an inference procedure from the data type and the number of samples. Proportions use z, means use t, and a two-way table of counts uses chi-square. Every interval is statistic plus or minus critical value times standard error.

This sheet covers the inference procedures in Units 3 and 4 of the Fall 2026 AP Statistics course. The chi-square goodness-of-fit test and inference for a regression slope were removed in the redesign and are listed here only as outside the framework.

Which procedure for which situation

Two facts about the data pick the procedure: whether the response is categorical or quantitative, and how many samples or groups there are.

DataSamples or groupsInference is aboutIntervalTest
Categorical, two outcomesOne sampleppOne-sample z-interval for a proportionOne-sample z-test for a proportion
Categorical, two outcomesTwo independent samples or two treatment groupsp1−p2p_1 - p_2Two-sample z-interval for a difference in proportionsTwo-sample z-test for a difference in proportions
Categorical, two-way table of countsIndependent samples or treatment groupsThe distribution of one categorical variable in each population or treatmentNoneChi-square test for homogeneity
Categorical, two-way table of countsOne sample, two variables recorded on each individualThe association between two categorical variablesNoneChi-square test for independence
QuantitativeOne sampleμ\muOne-sample t-interval for a meanOne-sample t-test for a mean
QuantitativePaired data, two measurements per pairμd\mu_dOne-sample t-interval on the differencesOne-sample t-test on the differences (paired t-test)
QuantitativeTwo independent samples or two treatment groupsμ1−μ2\mu_1 - \mu_2Two-sample t-interval for a difference in meansTwo-sample t-test for a difference in means

Here pp is a population proportion, μ\mu ("mu") is a population mean, and μd\mu_d is the population mean of the paired differences. For a longer walk through the choice, see which statistical test to use or the which test tool.

Confidence intervals: statistic, standard error, critical value

Every interval in the course has the same form:

statistic±(critical value)(standard error)\text{statistic} \pm (\text{critical value})(\text{standard error})
IntervalStatisticStandard errorCritical value
One proportionp^\hat{p}p^(1−p^)n\sqrt{\frac{\hat{p}(1-\hat{p})}{n}}z∗z^*
Difference in proportionsp^1−p^2\hat{p}_1 - \hat{p}_2p^1(1−p^1)n1+p^2(1−p^2)n2\sqrt{\frac{\hat{p}_1(1-\hat{p}_1)}{n_1} + \frac{\hat{p}_2(1-\hat{p}_2)}{n_2}}z∗z^*
One meanxˉ\bar{x}sn\frac{s}{\sqrt{n}}t∗t^* with df=n−1df = n - 1
Mean difference (paired)xˉd\bar{x}_dsdn\frac{s_d}{\sqrt{n}}, where nn is the number of pairst∗t^* with df=n−1df = n - 1
Difference in meansxˉ1−xˉ2\bar{x}_1 - \bar{x}_2s12n1+s22n2\sqrt{\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}}t∗t^* with dfdf from technology

p^\hat{p} ("p-hat") is a sample proportion, xˉ\bar{x} ("x-bar") is a sample mean, ss is a sample standard deviation, and xˉd\bar{x}_d and sds_d are the mean and standard deviation of the paired differences.

The common z∗z^* values are 1.645 for 90%, 1.960 for 95%, and 2.576 for 99%. Read t∗t^* from the t-table. For two independent samples, the smaller of n1−1n_1 - 1 and n2−1n_2 - 1 is a conservative choice of dfdf when working by hand.

Significance tests: null hypothesis, test statistic, distribution

Every z and t test statistic has the form statistic−null valuestandard error\frac{\text{statistic} - \text{null value}}{\text{standard error}}.

TestH0H_0Test statisticDistribution when H0H_0 is true
One proportionp=p0p = p_0z=p^−p0p0(1−p0)nz = \frac{\hat{p} - p_0}{\sqrt{\frac{p_0(1-p_0)}{n}}}Standard normal
Difference in proportionsp1=p2p_1 = p_2z=p^1−p^2p^c(1−p^c)(1n1+1n2)z = \frac{\hat{p}_1 - \hat{p}_2}{\sqrt{\hat{p}_c(1-\hat{p}_c)\left(\frac{1}{n_1} + \frac{1}{n_2}\right)}}Standard normal
One meanμ=μ0\mu = \mu_0t=xˉ−μ0s/nt = \frac{\bar{x} - \mu_0}{s/\sqrt{n}}t with df=n−1df = n - 1
Mean difference (paired)μd=0\mu_d = 0t=xˉd−0sd/nt = \frac{\bar{x}_d - 0}{s_d/\sqrt{n}}t with df=n−1df = n - 1
Difference in meansμ1=μ2\mu_1 = \mu_2t=xˉ1−xˉ2s12n1+s22n2t = \frac{\bar{x}_1 - \bar{x}_2}{\sqrt{\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}}}t with dfdf from technology
Chi-square, homogeneity or independenceSee the chi-square sectionχ2=∑(observed−expected)2expected\chi^2 = \sum \frac{(\text{observed} - \text{expected})^2}{\text{expected}}Chi-square with df=(rows−1)(columns−1)df = (\text{rows} - 1)(\text{columns} - 1)

Two differences from the interval table matter. A one-proportion test puts the null value p0p_0 in the standard error, because the calculation assumes H0H_0 is true. A two-proportion test uses the combined proportion p^c=x1+x2n1+n2\hat{p}_c = \frac{x_1 + x_2}{n_1 + n_2}, where x1x_1 and x2x_2 are the two success counts, because H0H_0 says the two populations share one proportion. See pooled vs unpooled standard error.

Conditions for each procedure

ProcedureRandom10% (sampling without replacement)Shape
One-proportion intervalRandom samplen≤0.10Nn \leq 0.10Nnp^≥10n\hat{p} \geq 10 and n(1−p^)≥10n(1-\hat{p}) \geq 10
One-proportion testRandom samplen≤0.10Nn \leq 0.10Nnp0≥10np_0 \geq 10 and n(1−p0)≥10n(1-p_0) \geq 10
Two-proportion intervalTwo independent random samples, or random assignmentEach sample at most 10% of its populationObserved successes and failures in both groups all at least 10
Two-proportion testTwo independent random samples, or random assignmentEach sample at most 10% of its populationn1p^cn_1\hat{p}_c, n1(1−p^c)n_1(1-\hat{p}_c), n2p^cn_2\hat{p}_c, n2(1−p^c)n_2(1-\hat{p}_c) all at least 10
One-sample t (mean or paired differences)Random sample or randomized experimentn≤0.10Nn \leq 0.10NPopulation approximately normal, or n≥30n \geq 30, or sample (of differences, if paired) with no strong skewness or outliers
Two-sample tTwo independent random samples, or random assignmentEach sample at most 10% of its populationBoth populations approximately normal, or both n≥30n \geq 30; any sample with n<30n < 30 shows no strong skewness or outliers
Chi-squareIndependence: one random sample. Homogeneity: independent random samples or random assignmentSample at most 10% of the populationAll expected counts greater than 5

NN is the population size. The 10% check applies only when sampling without replacement from a population. For any procedure, it is not needed when the data come from a randomized experiment. The reasoning behind each check is in the conditions for inference checklist.

Chi-square tests for two-way tables

Both chi-square tests in the course use the same arithmetic on a two-way table of counts. They differ in how the data were collected and in the wording of the hypotheses.

TestData collectionH0H_0HaH_a
HomogeneityIndependent random samples from two or more populations, or randomly assigned treatment groupsThe distribution of the categorical variable is the same in every population or treatmentThe distributions are not all the same
IndependenceOne random sample, two categorical variables recordedThere is no association between the two variables in the populationThere is an association between the two variables

The expected count for each cell is

expected count=row total×column totaltable total\text{expected count} = \frac{\text{row total} \times \text{column total}}{\text{table total}}

The p-value is the area to the right of the χ2\chi^2 ("chi-square") statistic under the chi-square curve with df=(rows−1)(columns−1)df = (\text{rows} - 1)(\text{columns} - 1). Use the chi-square table or the chi-square calculator. The two tests are compared in chi-square homogeneity vs independence.

Procedures outside the Fall 2026 framework

The College Board's list of topics removed for the Fall 2026 course includes two inference procedures that older materials still teach:

Regression is still in the course as Unit 5, where the line, the correlation, and the residuals are described and interpreted without an interval or a test. Details are in what was removed from AP Statistics.

Two-sample z-interval for a difference in proportions

A large school district takes independent random samples of 200 seniors and 200 juniors. In the samples, 84 seniors and 60 juniors have a part-time job. The district has more than 2,000 students in each grade. Construct a 95% confidence interval for the difference in the proportions with a part-time job (seniors minus juniors).

  1. Procedure: two independent samples and a categorical response with two outcomes, so use a two-sample z-interval for p1−p2p_1 - p_2.

  2. Conditions: the samples are independent random samples. Each sample of 200 is at most 10% of more than 2,000 students. The observed counts are 84 and 116 for seniors and 60 and 140 for juniors, all at least 10.

  3. Sample proportions: p^1=84200=0.42\hat{p}_1 = \frac{84}{200} = 0.42 and p^2=60200=0.30\hat{p}_2 = \frac{60}{200} = 0.30, so p^1−p^2=0.12\hat{p}_1 - \hat{p}_2 = 0.12.

  4. Standard error: 0.42(0.58)200+0.30(0.70)200=0.001218+0.001050=0.002268=0.0476\sqrt{\frac{0.42(0.58)}{200} + \frac{0.30(0.70)}{200}} = \sqrt{0.001218 + 0.001050} = \sqrt{0.002268} = 0.0476.

  5. Margin of error: 1.96×0.0476=0.09331.96 \times 0.0476 = 0.0933.

  6. Interval: 0.12±0.09330.12 \pm 0.0933, which is (0.0267,0.2133)(0.0267, 0.2133).

The interval is about (0.027,0.213)(0.027, 0.213). We are 95% confident that the proportion of all seniors in the district with a part-time job is between 0.027 and 0.213 higher than the proportion of all juniors with one.

Paired t-test on differences

Ten runners each run the same course twice, once in an old shoe model and once in a new one, with the order decided by a coin flip. For each runner the difference in time (old minus new) is recorded in seconds. The ten differences have mean 2.4 and standard deviation 3.1, and a dotplot of them shows no strong skewness and no outliers. Test at α=0.05\alpha = 0.05 whether the mean difference is greater than 0.

  1. Procedure: each runner gives two measurements, so the data are paired. Use a one-sample t-test on the differences.

  2. Hypotheses: H0:μd=0H_0: \mu_d = 0 and Ha:μd>0H_a: \mu_d > 0, where μd\mu_d is the true mean difference in time (old minus new) for runners like these.

  3. Conditions: the order of the shoes was randomly assigned. With n=10<30n = 10 < 30, the dotplot of differences shows no strong skewness and no outliers.

  4. Standard error: sdn=3.110=3.13.1623=0.9803\frac{s_d}{\sqrt{n}} = \frac{3.1}{\sqrt{10}} = \frac{3.1}{3.1623} = 0.9803.

  5. Test statistic: t=2.4−00.9803=2.448t = \frac{2.4 - 0}{0.9803} = 2.448 with df=10−1=9df = 10 - 1 = 9.

  6. P-value: the area to the right of 2.448 under the t curve with 9 degrees of freedom is about 0.018.

  7. Decision: 0.018≤0.050.018 \leq 0.05, so reject H0H_0.

t=2.448t = 2.448, df=9df = 9, p-value about 0.018. Because the p-value is less than 0.05, there is convincing evidence that the true mean difference in time (old minus new) is greater than 0 for runners like these.

Frequently asked questions

How do I tell paired data from two independent samples?

Data are paired when each value in one group has a specific partner in the other: the same subject measured twice, or two subjects matched on a trait with one given each treatment. Paired data use a one-sample t procedure on the differences. Two groups with no such link use a two-sample t procedure. See paired vs two-sample t-test.

Why does the two-proportion test pool the samples when the interval does not?

The test statistic is computed as though H0:p1=p2H_0: p_1 = p_2 were true. If the two populations share one proportion, the best estimate of it combines both samples, which gives p^c\hat{p}_c. An interval makes no such assumption, so each sample keeps its own p^\hat{p} in the standard error.

Which procedures use z and which use t?

Procedures for proportions use z. Procedures for means use t, because the population standard deviation is estimated with the sample standard deviation ss. The comparison is in t-test vs z-test.