Is inference still on the AP Statistics exam?

Yes, and here is the floor

Two of the five units in the Fall 2026 course are inference units:

• Unit 3: Inference for Categorical Data: Proportions, 15 topics, 15-25% of the multiple-choice section, roughly 30 class periods.
• Unit 4: Inference for Quantitative Data: Means, 10 topics, 10-20% of the multiple-choice section, roughly 18 class periods.

That is 25 of the course's 55 topics, about 45%. It is roughly 48 of the roughly 107 class periods, again about 45%. Unit 3 alone is the biggest unit in the course by both measures: no other unit has 15 topics and no other unit is allotted as many class periods. Only on multiple-choice weighting does another unit outrank it, Unit 1 at 20-30% against Unit 3's 15-25%, and those bands overlap.

On top of that, one of the four free-response questions is reserved for inference. In the exam format, question 3 is listed as Inference (Hypothesis Test or Confidence Interval), worth 10 points and 12.5% of the exam. The other three are named for course practices, not for content, so inference is the only content area with a question guaranteed to it in advance.

Put the guaranteed pieces together and at least a quarter of your score is inference. The next section shows that arithmetic, because the number is worth trusting rather than taking on faith.

Where the 25% floor comes from

Four sourced facts do all the work.

1. Section I, multiple-choice, is 50% of the exam score. Section II, free-response, is the other 50%.
2. Unit weightings are shares of Section I only, not of the whole exam. Unit 3 is 15-25% of it and Unit 4 is 10-20% of it.
3. Adding the endpoints correctly, not the midpoints, Units 3 and 4 are $15 + 10 = 25\%$ to $25 + 20 = 45\%$ of Section I.
4. Free-response question 3 is inference, worth 12.5% of the exam.

A share of Section I is worth half as much of the exam, so 25-45% of Section I is 12.5-22.5% of your score. Add the 12.5% that question 3 contributes and inference lands between 25% and 35% of the exam from those two sources alone.

The 25% end is a hard floor, and it is a pessimistic one. Reaching it needs Units 3 and 4 both sitting at the very bottom of their ranges, which leaves 75% of Section I for the other three units. Their upper bounds are 30%, 25% and 20%, adding to exactly 75%, so the floor is only attainable if Units 1, 2 and 5 all sit at their maximum on the same paper.

Say what the floor means precisely. It counts one guaranteed free-response question and the least Units 3 and 4 can be worth. It credits inference with nothing from the other three free-response questions and nothing from Unit 2, whose sampling distributions and central limit theorem work is what makes inference possible at all. So 25% is reachable, but only at that single corner of the published ranges. Anywhere else in them the share is higher, up to the 35% these two sources can reach.

Why the redesign coverage reads like a cut

Two true statements about the redesign get combined into a conclusion that does not follow.

The first is that the chi-square goodness-of-fit test was removed. The second is that the course went from nine units to five. Read together they suggest that inference was scaled back. Follow the removals through and a different picture appears.

Fewer unit numbers does not mean less material. In the current course, inference is the only content area that gets two units, one for categorical data and one for quantitative data, and between them they run 25 topics and about 48 class periods.

Inference did lose two things, and both are worth naming rather than glossing. College Board's future revisions page lists the chi-square goodness-of-fit test and the whole of the old Unit 9, Inference for Quantitative Data: Slopes, among the removals. So there is no inference for a regression slope on this exam, and Unit 5, Regression Analysis, is a descriptive and diagnostic unit of five topics.

One trap in that Unit 9 title. The removed unit was Inference for Quantitative Data: Slopes. The live Unit 4 is Inference for Quantitative Data: Means. Same first four words, different unit. If you saw the removal and concluded that inference for means went with it, that is the sentence to unlearn: intervals and tests for a population mean, a population mean difference, and the difference between two population means are all examinable, as topics 4.2, 4.4 and 4.5 for one mean or a mean difference and 4.7, 4.9 and 4.10 for two means.

What was removed from AP Statistics has the complete published list and what each entry covered.

Chi-square: what went, and what is still inference

This is where careless summaries go wrong, so the two halves need separating.

Removed: the chi-square goodness-of-fit test. One categorical variable, a single row of observed counts, and a claimed distribution to compare them against. It was old topics 8.2 and 8.3, and it is named in the removal list.

Live, and squarely inference: the chi-square tests for homogeneity and independence. They are the last two topics of Unit 3:

• 3.14 Setting Up a Chi-Square Test for Homogeneity or Independence
• 3.15 Carrying Out a Chi-Square Test for Homogeneity or Independence

Both work from a two-way table, both use $\chi^2 = \sum (O - E)^2 / E$, and both sit inside an inference unit rather than in a chi-square unit of their own. That placement is the point: in this course chi-square is inference, and it is examined inside the inference unit for categorical data.

So the chi-square removal is real, and it left the two tests that work from a two-way table inside Unit 3. Is chi-square goodness of fit on the AP exam? works through the distinction test by test, and chi-square two-way practice drills the version that is assessed.

One removal-list number points at a live inference topic

The removal list quotes topic codes from the nine-unit course. The current course reuses codes of the same shape for different content, and one collision lands directly on a live inference topic.

Read the list against a current syllabus and you would cross off two live topics, 2.9 and 4.9. 4.9 is the one that matters here: in the Fall 2026 course it is the setup for the test comparing two population means, and 4.10 carries it out.

Match on titles, never on codes. If a source tells you a topic number was removed, look up what that number is called in the Fall 2026 framework before you act on it. Every unit page on this site lists its current codes and titles.

What the 25 inference topics actually cover

Unit 3 handles inference for proportions, in 15 topics. It opens with estimators and the sampling distribution of $\hat{p}$ ("p-hat"), builds the one-proportion confidence interval and significance test, introduces p-values at topic 3.6 and Potential Errors When Performing Tests at topic 3.8, then repeats the whole cycle for the difference between two proportions, and finishes with the two chi-square topics.

Unit 4 handles inference for means, in 10 topics. Sampling distributions for $\bar{x}$ ("x-bar"), the confidence interval and test for a population mean or mean difference, which is where paired data lives, then the sampling distribution, interval and test for the difference between two means.

Two structural features are worth noticing before you revise.

First, the two units run the same five moves in the same order: sampling distribution, confidence interval, justify a claim from the interval, set up a test, carry out the test. Unit 3 threads three topics in around the cycles, estimators at 3.1, p-values at 3.6 and potential errors at 3.8, and runs the whole sequence twice, once for one proportion and once for two. Learn the shape once and Unit 4 is mostly Unit 3 with $t$ in place of $z$ and a different standard error.

Second, two-sample procedures are 10 of the 25 topics: 3.9 through 3.13 for proportions and 4.6 through 4.10 for means. Two-sample work is not an advanced extra here, it is 40% of the inference topics.

The conditions are shared across all of it, which is why the conditions for inference checklist is worth learning once rather than five times.

What this means for revision

If a quarter of the exam is guaranteed inference, and the two inference units plus question 3 can reach 35% between them, inference is where marginal study hours pay best. Three practical consequences.

Question 3 is known in advance. You know its content area before you open the paper, so a rehearsed structure is worth real points. For a test: state the hypotheses with symbols defined in context, name the procedure, check the conditions, compute the statistic and the p-value, then conclude in context against $\alpha$. For an interval: define the parameter, name the procedure, check the conditions, compute the interval, then interpret it in context and answer the question that was asked. Question 3 can be either. Drill it with mixed review inference practice, and see the AP Statistics FRQ guide for how the points are awarded.

Choosing the procedure is its own step. Deciding which procedure the situation calls for is separate from carrying it out, and an inference answer needs both. One proportion or two, one mean or two, paired or independent, or a two-way table. Which statistical test to use is the decision tree, and the which test interactive makes you commit to an answer before it tells you.

Do not revise the removed procedures. No goodness-of-fit test and no inference for a regression slope. If a practice set asks for a confidence interval for a slope, it was written for the nine-unit course; should I use old AP Statistics review books? covers how to audit older material without throwing it away.

For the exam as a whole, what is on the AP Statistics exam has both sections in full, and how many units are in AP Statistics has every weighting and class-period count in one table.