STAT 3011 Fall 2026: Hypothesis Testing For Proportions Part 2
Overview
Kazeem Adepoju uses Pfizer’s 2020 vaccine-trial example, with 8 symptomatic COVID-19 cases among 18,198 vaccinated participants, to teach one-proportion hypothesis tests. He derives the z statistic against a 0.1% benchmark, compares left-, right-, and two-tailed p-values using R’s `pnorm`, and connects hypothesis tests with confidence intervals, Type I and Type II errors, and statistical power.
Key takeaways
- With 8 cases among 18,198 vaccine recipients, p-hat is approximately 0.00044; testing against p-zero = 0.001 gives z ≈ −2.392.
- The alternative hypothesis determines the p-value calculation: the same z gives about 0.0084 for a left-tailed test, 0.9916 for a right-tailed test, and 0.0168 for a two-tailed test.
- At α = 0.05, the Pfizer example rejects p = 0.001 under the left- and two-tailed alternatives, but fails to reject it under the right-tailed alternative.
- A hypothesis test evaluates evidence against a null value, while a confidence interval estimates plausible effect sizes; statistical significance alone does not establish clinical importance.
- Failing to reject a null hypothesis is not proof that it is true or that a benchmark has been met; it means the data did not provide sufficient evidence to reject it under the chosen test.
- Alpha controls the maximum Type I error probability, beta represents Type II error probability, and statistical power is 1 − beta.
Chapters
- Kazeem Adepoju outlines left-tailed, right-tailed, and two-tailed tests for a population proportion.
- He also plans to explain how confidence intervals relate to hypothesis testing.
- The example uses 18,198 vaccine recipients, of whom 8 developed confirmed symptomatic COVID-19.
- The sample proportion is calculated as p-hat = 8 / 18,198, approximately 0.00044, or 0.044%.
- A hypothesis test evaluates statistical evidence for an effect, while a confidence interval estimates its size.
- Adepoju distinguishes statistical evidence that a vaccine works from whether its effect is large enough to be clinically meaningful.
- The example treats 0.1%, or p-zero = 0.001, as the FDA benchmark and sets the alternative to p < 0.001.
- Adepoju identifies random sampling and sufficiently large expected counts as conditions, checking n × p-zero = 18.198 against a threshold of 15.
- The equality belongs in the null hypothesis; a less-than alternative makes this a left-tailed test.
- The test statistic uses (p-hat − p-zero) divided by the null standard error, √[p-zero(1 − p-zero) / n].
- For the Pfizer numbers, the z statistic is approximately −2.392 and the left-tailed p-value is approximately 0.0084.
- In R, `pnorm(-2.392)` returns the lower-tail probability; because 0.0084 is below α = 0.05, the null is rejected.
- Changing the alternative to p > 0.001 makes the test right-tailed; the same z statistic gives a p-value near 0.9916, calculated as `1 - pnorm(-2.392)`.
- A two-tailed alternative, p ≠ 0.001, represents a difference in either direction and uses twice the upper-tail probability beyond |z|.
- For |z| = 2.392, the two-tailed p-value is approximately 0.0168, calculated as `2 * (1 - pnorm(2.392))`.
- The same sample and z statistic can yield different p-values because each alternative specifies a different tail or set of tails.
- At α = 0.05, the example rejects the null for the left- and two-tailed tests but fails to reject it for the right-tailed test.
- A small p-value indicates that the observed statistic is unusual under the null; failing to reject does not prove the null or establish that a benchmark is met.
- Adepoju emphasizes choosing the alternative before observing the data and uses normal-curve sketches to show the relevant tail areas.
- A Type I error rejects a true null hypothesis; Adepoju compares it to convicting an innocent person when the null represents innocence.
- A Type II error fails to reject a false null; the courtroom analogy is acquitting someone who is guilty.
- The analogy links statistical decisions to the distinction between what the evidence suggests and what is true in reality.
- Alpha is the maximum tolerated probability of a Type I error; the lecture uses α = 0.05.
- Beta denotes the probability of a Type II error, and power is 1 − β, the probability of correctly rejecting a false null.
- Adepoju invokes George Box’s statement that all models are wrong but some are useful to explain why statistical models approximate reality.
- Adepoju says graded exam booklets will be returned in class on Friday and that he will explain retake options.
- The makeup exam is scheduled to open Friday at 3:00 p.m. and close Sunday at 11:59 p.m.; the 90-minute timer starts when a student clicks Start.
- He closes by saying he expects to return to in-person teaching on Friday after recovering from the flu.
Transcript excerpt
0:00 Okay. Uh, confirm, if you are hearing me clearly. Is everyone hearing me clearly? >> Sounds good. >> Okay. Uh, thank you so much. Uh, I want to welcome you to my lecture today. Of course, uh, we're meeting remotely today. Uh, because, uh, I pick a I pick a flu and I think I'm I'm going to be fine. Okay. I'm going to fine. I will return to class on Friday. Uh, going to be uh, returning in person. Uh, I'm going to say this on Friday. Uh, I'm going to come into that class with your exam booklets. The graded uh, booklets.
0:46 So, you're going to have the graded booklets. So, be ready uh, to meet uh, uh, to come to class on Friday. We started, uh, hypothesis testing uh, on Monday. Uh, you know, we've been focusing on hypothesis testing uh, on, you know, hypothesis testing on on on our our proportion. And you know what I'm going to do uh, today I'm actually, um, going to use example uh, from uh, Pfizer vaccine uh, clinical trial. I remember I've been using this example in class. Um, I'm going to do that today.
1:34 Now, I also I I wanted to pay attention. Uh, today I'm going to I'm going to walk us through three different kind of tests that we can have you know, in this area. I'm going to walk you through uh, one tail test. The one tail test is divided into two. We have right tail and left tail. I'm going to walk you through that today using this example. I'm also going to walk you through two-sided test, two-sided that we call two tails. Okay, then after that I'm also going to uh give, you know, uh differentiate uh
2:24 you know, what is the connection
This is the start of the transcript. Save this video free for the whole transcript, searchable and time-coded.
Related notes
Summary, takeaways, and chapters were generated by AI from the video's transcript and may contain errors. The video belongs to its creator, Prof Kazeem Adepoju.