AP Statistics Inference (Units 6 & 7): The Conditions-Checking Step Students Skip and Lose Points On
- Edu Shaale
- 5 days ago
- 16 min read
Serious About Your AP Scores? Let’s Get You There
From understanding concepts to scoring 4s and 5s, EduShaale’s AP coaching is built for results — with personalised learning, small batches, and exam-focused strategy.
22-33% combined MCQ exam weight of Units 6 and 7 | 3 conditions verified on every inference FRQ | ≥10 the Large Counts / expected-count threshold | 2.92 the 2025 AP Statistics mean score, out of 5 |
60.3% of students scored 3 or higher in 2025 | 4 distinct inference procedures across Units 6-7 | 0 credit typically given for “conditions met” with no numbers shown | n>30 the sample-size line that changes which Normal check applies |
Table of Contents
The Sentence That Reads Fine and Still Scores Zero
Ask an AP Statistics student to name the conditions for a two-sample t-test and most will answer instantly: “Random, Independent, Normal.” Ask them to verify those conditions using the actual numbers in the problem in front of them, and the confidence usually drops. This is not a knowledge gap in the way most students think about knowledge gaps. It is a specific, well-documented, and very fixable habit -- and the College Board's own course materials name it directly. Teachers are told, in the framework document itself, that students “often check conditions superficially -- e.g., just listing ‘SRS’ -- without explicitly connecting them to the problem.”
That single sentence describes the exact pattern behind a large share of the lost points in Units 6 and 7 -- Inference for Categorical Data: Proportions and Inference for Quantitative Data: Means -- which together make up somewhere between 22% and 33% of the multiple-choice exam and underpin the majority of the free-response section. Conditions are not a formality tacked onto the front of a calculation. They are a scored component in their own right, checked against the specific numbers in the problem, and a response that skips straight to the test statistic without doing this is treated the same as a response that never checked at all.
This guide is not a general overview of statistical inference. It is a close, numbers-first look at exactly what the Random, Normal, and Independent conditions require for each of the four inference procedures taught across Units 6 and 7 -- one-proportion, two-proportion, one-mean, and two-mean or matched-pairs -- including the specific places where the requirement quietly changes: which proportion to plug into the Large Counts check, why a paired t-test is checked on the differences rather than the raw data, and why “n is large” means something different for a mean than it does for a proportion. It ends with a condition-by-condition checklist built to be used the way it is meant to be used: during a timed free-response question, not after one.
Worth knowing: This piece assumes familiarity with the four-step State-Plan-Do-Conclude structure that underlies every inference FRQ. If that structure itself is new, our AP Statistics FRQ Tips guide covers it end to end before this piece goes deeper into the Plan step specifically. |
1. Why Conditions Exist -- and Why “SRS” Alone Is Not Enough
Every inference procedure in Units 6 and 7 -- every z-interval, z-test, t-interval, and t-test -- rests on a single mathematical assumption: that the sampling distribution of the statistic being used (a sample proportion or a sample mean) is approximately normal. That assumption is what allows a z-score or t-score to be looked up against a known distribution and converted into a p-value or a margin of error in the first place. Conditions exist to check whether that assumption actually holds for the data in front of you. They are not bureaucratic box-ticking; they are the justification for using the formula at all.
This is why the College Board treats condition-verification as its own line on the scoring guideline, filed under the official skill category Statistical Argumentation, specifically Learning Objective 4.C: verify that inference procedures apply in a given situation. A response that calculates a flawless test statistic without this step has produced a number without justifying that the number means anything -- and the rubric is built to catch precisely that.
The three questions every condition actually answers: Random asks: can this data be generalised to a population at all? Independence asks: does knowing one observation tell you anything about another? Normal / Large asks: is the sampling distribution shaped closely enough to a normal curve for a z- or t-procedure to be valid? Every condition check on the exam is really one of these three questions, answered with the specific numbers in the problem. |
2. The Three Conditions Every Inference Procedure Shares
Before looking at how Units 6 and 7 differ, it helps to see what every single inference procedure in both units has in common. Every procedure -- proportions or means, one sample or two -- is built from the same three-condition skeleton:
Condition | What It Checks | How It's Verified |
Random | Whether the data can be generalised beyond the sample | Data collected via a random sample or a randomised experiment -- stated in the problem, not assumed |
Independent | Whether one observation affects the probability of another | Random assignment/selection, plus the 10% condition when sampling without replacement: n ≤ 10% of the population |
Normal / Large | Whether the sampling distribution is close enough to normal to trust a z- or t-procedure | Differs by procedure -- this is where proportions and means genuinely diverge (Sections 3 and 4) |
The first two conditions barely change across the whole course. The third one -- Normal / Large -- is where most of the real technical variation lives, and it is also where most of the avoidable point loss happens, because the exact numeric check depends on which parameter is being estimated and which specific procedure is being used. Sections 3 and 4 walk through that variation directly.
3. AP Statistics Conditions for Inference: Unit 6, Proportions
Unit 6 covers inference for one population proportion and for the difference between two population proportions. The Random and Independent conditions work exactly as described above. The Large Counts condition is where the details matter -- because which value gets plugged in changes depending on whether the procedure is a confidence interval or a significance test, and whether it involves one sample or two.
Procedure | Large Counts Check Uses | Condition, Written Out |
One-proportion confidence interval | Sample proportion, p̂ | np̂ ≥ 10 and n(1-p̂) ≥ 10 |
One-proportion significance test | Hypothesised value, p₀ | np₀ ≥ 10 and n(1-p₀) ≥ 10 |
Two-proportion confidence interval | Each sample's own p̂₁ and p̂₂ | n₁p̂₁, n₁(1-p̂₁), n₂p̂₂, n₂(1-p̂₂) all ≥ 10 |
Two-proportion significance test | Pooled/combined proportion, p̂c | n₁p̂c, n₁(1-p̂c), n₂p̂c, n₂(1-p̂c) all ≥ 10 |
Table built from the AP Statistics CED's official Essential Knowledge statements for Learning Objectives UNC-4.B and VAR-6.J (Effective Fall 2020). The pooled proportion is calculated as p̂c = (n₁p̂₁ + n₂p̂₂) / (n₁ + n₂).
The logic behind the switch is straightforward once it is stated plainly: a confidence interval makes no assumption about what the true proportion is, so it uses the best estimate available -- the sample proportion. A significance test, by contrast, starts by assuming the null hypothesis is true, so the Large Counts check has to use whatever value the null hypothesis assumes -- p₀ for one sample, or the pooled estimate under “p₁ = p₂” for two samples. Using p̂ where p₀ belongs (or vice versa) does not usually change the final decision, but it is checking the wrong thing, and a rubric built around verifying the correct condition will not award the point for checking a different one.
The 10% Condition for Two Samples: Check Both, Separately
One detail that is easy to lose when moving from one-sample to two-sample procedures: the 10% condition applies to each sample independently, against its own population. If a study compares 40 students from School A to 35 students from School B, the condition is not a single combined check -- it is n₁ ≤ 10% of School A's population AND n₂ ≤ 10% of School B's population, verified and written as two separate statements.
4. AP Statistics Conditions for Inference: Unit 7, Means
Unit 7 covers inference for one population mean, the difference between two population means, and the special case of matched pairs. Random and Independent again work as described in Section 2. The Normal / Large condition, however, is built on a genuinely different logic than the Large Counts condition in Unit 6 -- and the CED is explicit that students are expected to understand both how and why the two differ.
Situation | How the Normal Condition Is Checked |
Population is stated to be normally distributed | Condition is automatically satisfied, regardless of sample size |
Sample size n > 30 | The Central Limit Theorem applies -- treat the sampling distribution as approximately normal even if the population shape is unknown |
Sample size n ≤ 30 | The condition depends on the data itself: the sample must be reasonably free of strong skewness and outliers, checked using a dotplot, boxplot, or stemplot |
Table built from the AP Statistics CED's Essential Knowledge statement UNC-4.P.1 (confidence intervals for a mean) and the parallel statement for significance tests.
There is no single numeric threshold here the way there is with np ≥ 10 for proportions -- and that is precisely the point students trained mainly on proportions tend to miss. A response that writes “n = 45 > 30, so Normal condition met” for a mean is using the right logic. A response that tries to apply np ≥ 10 to a means question has imported the wrong condition from the wrong unit entirely.
Matched Pairs: Check the Differences, Not the Two Original Samples
Matched-pairs data -- the same subjects measured twice, such as a class's scores before and after an intervention -- is treated as a single sample of differences, not two separate samples. This has a direct consequence for condition-checking that a meaningful share of students get backwards: the Normal / Large condition is checked on the distribution of the differences, not on the two original sets of scores. Graphing the “before” scores and the “after” scores separately and checking each for skewness answers the wrong question. The one distribution that actually matters is the set of before-minus-after differences for each pair.
Quick self-check: If a problem gives you two sets of numbers for the same group of people or objects, measured under two conditions, subtract first. Whatever distribution you are asked to describe, check, or graph next should be the distribution of those differences -- not the two original columns. |
5. Proportions vs Means: The Same Three Conditions, Different Numbers
Units 6 and 7 are taught back-to-back for a reason: the underlying logic -- state a parameter, verify conditions, calculate, conclude -- never changes. What changes is the specific arithmetic behind the Normal / Large condition, and mixing the two up is one of the most common cross-unit errors on the exam.
| Unit 6: Proportions | Unit 7: Means |
Parameter | p (population proportion) | μ (population mean) |
Distribution used | Normal (z) | t, with n-1 degrees of freedom |
Normal / Large condition | np ≥ 10 and n(1-p) ≥ 10, using p̂, p₀, or pooled p̂c depending on procedure | Population stated normal, OR n > 30 (CLT), OR n ≤ 30 with no strong skew/outliers |
Formula sheet provided? | Yes | Yes |
Common cross-unit error | Applying the mean's “n > 30” rule instead of checking np and n(1-p) | Applying np ≥ 10 to a mean, where it has no meaning |
The practical habit worth building here is simple: before checking any condition, name the parameter first. If it is p, the Large Counts arithmetic in Section 3 applies. If it is μ, the population-shape logic in Section 4 applies. Confusing the two costs the condition-verification point even when every other part of the response is correct.
6. The Four Condition-Checking Mistakes That Cost the Most Points
Naming conditions without verifying them. Writing “Random, Normal, Independent -- all conditions met” states the checklist but performs none of the checking. A response built around verification, not recitation, treats an unverified list the same as a skipped step -- the specific numbers from the problem have to appear on the page.
Using p̂ where p₀ belongs, or vice versa. The single most common Unit 6 slip: plugging the sample proportion into a significance test's Large Counts check instead of the hypothesised value, or the reverse on a confidence interval.
Checking the raw data instead of the differences on matched pairs. Graphing or describing the two original sets of scores instead of the paired differences answers a question the procedure never asked.
Treating the 10% condition as a single combined check on two-sample designs. Each sample needs its own 10% verification against its own population -- one combined statement is not equivalent to two separate ones.
7. Same Data, Different Score: A Worked Comparison
The scenario below is original, built to mirror how the College Board's own scoring guidelines treat condition verification, not a reproduction of any specific released question.
The scenario: A university admissions office wants to know whether a new optional interview process changes the proportion of applicants who accept their offer of admission. In a randomised trial, 150 applicants are randomly assigned to receive the new interview process; historically, 62% of all admitted applicants accept their offer. Of the 150, 102 accept. Does this provide convincing evidence, at α = 0.05, that the interview process changes the acceptance rate? |
Response A -- Conditions Named, Not Verified
“Conditions: Random ✓, Normal ✓, Independent ✓. z = 1.47, p = 0.14. Fail to reject.”
Where this lands on the rubric: The calculation may well be correct, but every condition is asserted rather than checked. There is no evidence the writer confirmed np₀ and n(1-p₀) individually, no population size for the 10% condition, and no indication random assignment was connected to this specific scenario. On a rubric that scores verification as its own component, this pattern is treated as equivalent to not checking conditions at all -- a Partially Correct or Incorrect classification for that part, regardless of how correct the final numbers are. |
Response B -- Conditions Verified With the Actual Numbers
Condition | What Response B Wrote |
Random | Applicants were randomly assigned to the new interview process, as stated in the problem. |
Independent | 150 is almost certainly less than 10% of all applicants nationally to this type of programme, so the 10% condition is reasonably met. |
Large Counts | Assuming H₀ is true (p = 0.62): np₀ = 150(0.62) = 93 ≥ 10, and n(1-p₀) = 150(0.38) = 57 ≥ 10. Both are satisfied. |
Same eventual z-statistic, same p-value, same final decision. The difference is entirely in whether the numbers behind each condition were actually shown. Response B's Plan step lands in the Essentially Correct category for condition verification; Response A's does not -- and that single component is frequently the difference between a 3 and a 4 on a question that is otherwise identical.
8. A Step-by-Step Conditions-Checking Framework
A practical sequence for verifying conditions on any Unit 6 or 7 free-response question, in the order it should be written:
Name the parameter first. p or μ? This single decision determines every check that follows.
State Random explicitly, tied to the problem. Not “SRS ✓” but a sentence naming how this specific sample or experiment was actually generated, per the problem text.
Check Independence with a number, not a word. Write the population size (or note it is a randomised experiment, where independence follows from random assignment) and, for two samples, check both separately.
Identify which value belongs in the Normal / Large check. For proportions: p̂, p₀, or the pooled p̂c, per Section 3's table. For means: population shape, n > 30, or a graph-based skew/outlier check, per Section 4.
Show the arithmetic, not just the conclusion. np₀ = [number] ≥ 10, not “Large Counts met.” A grader awarding this component is checking for the presence of the calculation, not trusting the checkmark.
Pro tip: Write this sequence in the same order every time, on every practice FRQ, until it takes under 90 seconds. The Do and Conclude steps get faster once Plan stops requiring a decision about what to check. |
9. Myths About Checking Conditions
Myth: “Listing the three condition names earns the same credit as verifying them.”
Reality: The CED itself flags this as a common and costly shortcut. Verification means connecting each condition to the specific numbers in the problem, not reciting a checklist.
Myth: “The Large Counts condition always uses the sample proportion.”
Reality: Only for confidence intervals. Significance tests use the hypothesised value p₀ (or the pooled estimate for two-sample tests), because the test proceeds by assuming the null hypothesis is true.
Myth: “If n is at least 30, the Normal condition is automatically fine for any procedure.”
Reality: That threshold applies to means, via the Central Limit Theorem. For proportions, the relevant check is np ≥ 10 and n(1-p) ≥ 10 -- a different calculation with a different threshold.
Myth: “For matched pairs, check both sets of raw scores for Normality.”
Reality: Matched pairs are analysed as a single sample of differences. The Normal / Large condition is checked on those differences, not on the two original datasets.
10. Beyond Units 6 and 7: Why This Skill Doesn't Stop Here
The habit of verifying conditions with actual numbers, rather than naming them from memory, carries directly into Units 8 and 9 -- Chi-Square inference and inference for the slope of a regression line -- where a parallel set of conditions (Random, Independence, and an Expected Counts or Linearity check) is verified the same way and scored the same way. Students who build the numbers-first habit here in Units 6 and 7 are, in practice, also preparing for roughly the final 4% to 10% of the exam's content weight in one pass rather than two.
Ready to Start Your AP Journey?
EduShaale’s AP Coaching Program is designed for students aiming for top scores (4s & 5s). With expert faculty, small batch sizes, personalized mentorship, and a curriculum aligned to the latest AP format, we help you build deep conceptual clarity and exam confidence.
Subjects Covered: AP Calculus, AP Physics, AP Chemistry, AP Biology,
AP Economics & more
📞 Book a Free Demo Class: +91 90195 25923
🌐 www.edushaale.com/ap-coaching
Free Diagnostic Test: testprep.edushaale.com
11. Frequently Asked Questions
Q: What are the three conditions for inference in AP Statistics?
A: Random, Independent, and Normal (or Large Counts, for proportions). Random checks whether the data can be generalised to a population; Independent checks whether observations affect one another's probability, verified through the 10% condition when sampling without replacement; Normal or Large Counts checks whether the sampling distribution is close enough to normal to justify using a z- or t-procedure. All three apply to every inference procedure in Units 6 and 7, though the exact numeric check for the third condition differs by procedure.
Q: What is the Large Counts condition in AP Statistics?
A: It is the condition verifying that a sampling distribution of a proportion is approximately normal, checked by confirming that both np and n(1-p) are at least 10. Which value is used for p depends on the procedure: the sample proportion for a confidence interval, the hypothesised value for a one-sample significance test, and a pooled estimate for a two-sample significance test.
Q: Do you use p-hat or p-naught for the Large Counts condition?
A: It depends on whether the procedure is a confidence interval or a significance test. Confidence intervals use the sample proportion, p-hat, because no assumption about the true proportion has been made. Significance tests use the hypothesised value, p-naught (or a pooled proportion for two-sample tests), because the test proceeds by assuming the null hypothesis is true.
Q: What is the 10% condition and when does it apply?
A: The 10% condition verifies independence when sampling without replacement from a finite population: the sample size should be no more than 10% of the population size. It applies to every inference procedure that uses a sample drawn without replacement, and for two-sample procedures it must be checked separately for each sample against its own population.
Q: How is the Normal condition checked for a mean, compared to a proportion?
A: For a mean, the Normal condition is satisfied if the population is stated to be normally distributed, if the sample size exceeds 30 (via the Central Limit Theorem), or, for smaller samples, if the sample data show no strong skewness or outliers. This is a different check from the Large Counts condition used for proportions, which relies on np and n(1-p) both being at least 10 rather than on sample size alone.
Q: Why do I check conditions on the differences for a matched-pairs test, not the original data?
A: Matched-pairs data is treated as a single sample made up of the differences between paired measurements, not as two independent samples. Because the inference procedure itself operates on those differences, the Normal / Large condition must be verified on the distribution of the differences, not on the two original sets of scores.
Q: What happens if I name the conditions but don't check them with numbers?
A: A response that lists condition names without connecting them to the specific numbers in the problem is generally treated the same as a response that never checked conditions at all, since the scoring criteria require verification, not recitation. This is one of the most common and most avoidable sources of lost points on inference free-response questions.
Q: Is checking conditions really worth separate points on the AP Statistics exam?
A: Yes. Verifying that an inference procedure applies to a given situation is an official, separately assessed skill in the AP Statistics course framework, and free-response scoring guidelines typically treat condition verification as its own component within a question, distinct from the calculation and the conclusion.
Q: What's the difference between checking conditions for one sample versus two samples?
A: The underlying three conditions are identical, but two-sample procedures require each condition to be checked for both samples rather than once. The 10% condition needs a separate population comparison for each sample, and the Large Counts or Normal condition for two-sample procedures typically involves either four separate counts (proportions) or a check across both samples (means).
Q: Do these same conditions apply to Chi-Square tests and slope inference in Units 8 and 9?
A: The same underlying logic -- Random, Independence, and a distribution-shape check -- carries forward, though the specific mechanics differ: Chi-Square procedures check expected counts rather than np and n(1-p), and slope inference for a regression line adds a linearity condition. Building a numbers-first condition-checking habit in Units 6 and 7 transfers directly to these later units.
12. EduShaale -- Expert AP Statistics Coaching
EduShaale's AP Statistics coaching is built around exactly the gap this guide describes: strong students who understand inference conceptually but lose points because condition-verification never became an automatic, numbers-first habit.
Condition-by-Condition Drilling: Every practice free-response question is scored specifically on whether conditions were verified with real numbers, not just named -- the exact distinction this guide is built around.
Units 6 & 7 Intensive: A dedicated block covering every procedure in this guide -- one-proportion, two-proportion, one-mean, two-mean, and matched pairs -- with condition-checking rehearsed as its own timed skill, separate from calculation.
Rubric-Based Mock Scoring: After every practice exam, students see exactly which condition-verification component cost them credit, using the same Essentially/Partially/Incorrect classification the real exam uses.
Diagnostic-First Coaching: A graded diagnostic free-response set at the start of coaching identifies whether a student's gap is conceptual, computational, or -- as it most often is -- a missing verification habit.
Free 60-Minute Strategy Session -- book here
Live Online 1-on-1 AP Statistics Coaching
WhatsApp +91 9019525923 | edushaale.com | info@edushaale.com
EduShaale's most important observation: Students rarely lose the conditions point because they don't understand statistics. They lose it because “verify conditions” quietly became “name conditions” somewhere in their practice, and nobody caught the substitution until the actual exam did. |
13. References & Resources
Official College Board Resources
EduShaale AP Resources
AP and Advanced Placement are registered trademarks of the College Board, which was not involved in the production of this guide. Condition definitions and formulas are drawn from the official AP Statistics Course and Exam Description; score data reflects the 2025 exam administration and is updated periodically -- verify current figures at apcentral.collegeboard.org and apstudents.collegeboard.org before use in a final study plan. This guide is for educational purposes only.
© 2026 EduShaale | edushaale.com | info@edushaale.com | +91 9019525923
Comments