STAT 3011 Fall 2026: Confidence Interval of difference of Proportions
Overview
Prof Kazeem Adepoju introduces confidence intervals for the difference between two independent population proportions, explaining how sample proportions estimate the difference and how the standard error and confidence-level-specific z critical value determine the interval. He applies the method to female and male student-athlete graduation rates, finding an estimated difference of 24.3 percentage points and a 99% confidence interval of approximately 5.5 to 43.1 percentage points.
Key takeaways
- A two-proportion confidence interval estimates the absolute population difference p1-p2 using p̂1-p̂2, not a pooled proportion.
- For independent groups, the standard error is the square root of p̂1(1-p̂1)/n1 plus p̂2(1-p̂2)/n2, and the margin of error is z* times that standard error.
- Check all four counts before using the normal approximation: successes and failures must each number at least 10 in both groups.
- Higher confidence requires a larger critical z value and therefore a wider interval; common two-sided values are 1.645, 1.96, and 2.576 for 90%, 95%, and 99%.
- In the graduation example, 43 of 53 female athletes and 58 of 102 male athletes graduated, yielding an estimated difference of 24.3 percentage points and a 99% interval of roughly 5.5 to 43.1 points.
- A confidence interval describes uncertainty in the estimated effect size, while a hypothesis test addresses evidence against a specified no-difference claim.
Chapters
- Prof Kazeem Adepoju reports a Midterm 1 mean of 80.51 and median of 83, with extra credit bringing the highest score to 105.
- He says makeup and retake scores were compared and the higher score substituted.
- Exam 2 will cover sampling distributions of means, confidence intervals, and hypothesis tests for one proportion or one mean; two-group comparisons are excluded.
- Students are directed to practice with the old Exam 2 posted on Canvas and expect an exam guide before the end of the week.
- The lesson shifts from one-group inference to comparing the proportion of an outcome across two groups.
- Examples include comparing political-party vote shares and symptomatic COVID-19 rates among vaccinated and placebo groups.
- For the vaccine example, the population quantities are the risks in the vaccinated and placebo groups, denoted by separate proportions.
- Subtracting two population proportions measures their absolute difference; a difference of zero indicates equal proportions.
- A hypothesis test evaluates evidence that a difference exists, while a confidence interval estimates its plausible size.
- A ratio compares groups on a relative scale; vaccine efficacy can be expressed as one minus the vaccinated-to-placebo risk ratio.
- Population proportions are generally unknown because studying every person is impractical, so inference uses samples from both groups.
- For each group, the sample proportion is the number of observed successes divided by that group's sample size.
- In the COVID-19 example, successes are participants who develop symptomatic COVID-19, counted separately for vaccinated and placebo samples.
- Subtracting the two sample proportions produces a point estimate of the population difference, but one sample can yield an estimate that is too high or too low.
- Repeated random samples can produce different estimates, which is why a point estimate alone does not describe uncertainty.
- Repeated investigations generate a sampling distribution of differences between sample proportions; its center corresponds to the population difference.
- The standard error estimates sampling variability using both groups: the square root of p̂1(1-p̂1)/n1 plus p̂2(1-p̂2)/n2.
- The standard deviation describes the theoretical sampling distribution using population proportions, while the standard error substitutes sample proportions.
- The normal approximation supports a z-based confidence interval when the required success and failure counts are sufficiently large.
- The interval is the estimated difference plus or minus a margin of error, calculated as the critical z value multiplied by the standard error.
- Common two-sided critical values are approximately 1.645 for 90%, 1.96 for 95%, and 2.576 for 99% confidence.
- Increasing the confidence level increases the critical value and margin of error, producing a wider interval.
- The standard 95% confidence level corresponds to a 5% significance level in the related hypothesis-testing framework.
- The two groups must be independent for the method presented in class.
- Each group must have at least 10 observed successes and at least 10 observed failures: n1p̂1, n1(1-p̂1), n2p̂2, and n2(1-p̂2) must each be at least 10.
- These counts support using an approximately normal sampling distribution for the difference in sample proportions.
- For a one-proportion confidence interval, the analogous checks use the sample proportion; hypothesis-test conditions instead use the null-hypothesis proportion.
- Education example: compare graduation proportions for female and male student athletes.
- Public-opinion example: compare support for a proposed policy in Minnesota and California.
- Healthcare example: compare side-effect proportions between two treatments; manufacturing example: compare defect rates under new and old processes.
- Each scenario has a binary outcome and two independent groups, making a difference between proportions an appropriate target.
- The sample includes 53 female athletes, of whom 43 graduated, and 102 male athletes, of whom 58 graduated.
- The success and failure counts in both groups meet the minimum of 10, so the normal-approximation condition is satisfied.
- The estimated graduation proportions are 43/53 and 58/102; their difference is approximately 0.24269, or 24.3 percentage points.
- Using the 99% critical value and the two-group standard error gives an interval of about 0.055 to 0.431.
- The 99% interval is interpreted as an estimated female-athlete graduation advantage between about 5.5 and 43.1 percentage points.
- For confidence levels beyond the common 90%, 95%, and 99% values, compute z* as the standard normal quantile at 1-alpha/2.
- For a 96% interval, alpha is 0.04, so the quantile is evaluated at 0.98; the same procedure gives approximately 1.96 for 95%.
- The class concludes with the z-quantile procedure and a brief closing.
Transcript excerpt
0:01 Yeah, because everyone know I'm Okay. Um, good morning everyone. All those of you at the back, uh, can
0:49 you confirm if you can hear me? Okay. Thank you so much. Thank you for confirming that. Uh before I get started today. Um I think we're done with mamm one. I provided an opportunity uh for makeup and retake and I'm I'm happy that uh some of you uh actually u made use of the opportunity. So um this is what I've done. You're going to see that your your grade changed this morning. I've already uh substituted on you know if you if you wrote a me and one and the makeup of course I choose uh
1:39 the higher score. Okay, if you notice that. So this is the analysis. Okay. uh the mean uh stand at 80.51 that is the mean performance in meter one and the median is 83 uh 0 you know if you want to the minimum the maximum of course uh the maximum is 105 you know we added some extra credit so so take a look at This is a distribution of scores. 17 students score between 0 and 59. 16 students score between 60 and 69. And of course 29 score between 70 and
2:31 79. 31 score between 80 and 89. And of course 53 students score 90 plus. So that is what I I'm hoping that the exam 2 will have a higher performance compared to this you know um probably one of the reason I I I actually want to believe that this is not your best actually or maybe what happen is you know it it takes uh some time to get settled you know when in the in the new semester four semester but now you are used to you understand
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.