AP Statistics Unit 6: Inference for Proportions - Confidence Intervals

Overview of Unit 6: Inference for Categorical Data

• Unit 6 Scope: This unit focuses on inference for categorical data, specifically emphasizing population proportions ($p$) and sample proportions ($\hat{p}$).
• Definition of Inference: Inference is the process of using sample statistics to make judgments or predictions about a larger population parameter.     * Example: If a sample of high school students shows that $78\%$ did their homework, inference helps determine what that means for the entire population of students (e.g., is it exactly $78\%$, slightly more, or slightly less?).
• Core Inference Procedures:     * Confidence Intervals: Used to estimate or make a prediction for what the population parameter might be based on a sample statistic.     * Significance Tests: Used to determine if a specific claim made about a population parameter is potentially true or false (addressed in Part 2 of the unit review).
• Types of Analysis:     * Single Sample: Analyzing one proportion from one population.     * Two Samples: Analyzing the difference between two sample proportions taken from two distinct populations.

Conditions for Inference

• Before conducting inference procedures (intervals or tests), specific conditions must be met to ensure the results are valid:
• Randomness: The sample must be collected randomly to avoid bias.
• Independence (10% Rule): The sample size ($n$) must be less than $10\%$ of the total population ($N$). This allows for the assumption of independence when sampling without replacement.
• Large Counts (Success/Failure Condition): The sample size must be large enough to assume a normal distribution for the sampling distribution. The sample must contain:     * 10 or more observed successes ($n\hat{p} \ge 10$).     * 10 or more observed failures ($n(1-\hat{p}) \ge 10$).     * Success Definition: The specific characteristic or outcome being measured.

Sampling Distributions and Confidence Intervals

• Point Estimate: A single value (the sample proportion, $\hat{p}$) used to estimate the population parameter. While it is unlikely to perfectly match the population proportion, it serves as the center for the interval.
• Sampling Distribution Characteristics:     * Center: The mean of all possible sample proportions is the true population proportion ($\mu_{\hat{p}} = p$).     * Spread: Measured by the standard deviation formula: $\sigma_{\hat{p}} = \sqrt{\frac{p(1-p)}{n}}$.     * Shape: Approximately normal if the Large Counts condition is met.
• The Logic of the Interval: In a normal distribution, approximately $95\%$ of all possible sample proportions fall within roughly $2$ (more accurately, $1.96$) standard deviations of the true population proportion. Therefore, if we take a sample proportion and add/subtract $1.96$ standard deviations, we have a $95\%$ probability of capturing the true population parameter within that span.

The Mechanics of a One-Sample Z-Interval

• Standard Error ($SE$): Because the true population proportion ($p$) is unknown (which is why the interval is being built), we cannot calculate the true standard deviation. Instead, we use the sample proportion ($\hat{p}$) to calculate the Standard Error:     * $SE_{\hat{p}} = \sqrt{\frac{\hat{p}(1-\hat{p})}{n}}$
• Critical Value ($Z^<em>$): Represents the number of standard deviations one must reach from the center to capture the desired percentage of the distribution. It is determined by the confidence level ($C\%$).      Alpha (\alpha): The area in the tails, calculated as $\alpha = 1 - C$.     * Finding $Z^<em>$: The area in each tail is $\frac{\alpha}{2}$. Use invNorm on a calculator or a Z-table to find the z-score corresponding to the area at the bottom ($\frac{\alpha}{2}$).      Example for $95\%$ Confidence: $\alpha = 0.05$, tail area = $0.025$, yielding $Z^* = 1.96$.     * Example for $90\%$ Confidence: $\alpha = 0.10$, tail area = $0.05$, yielding $Z^* = 1.645$.     * Example for $99\%$ Confidence: $\alpha = 0.01$, tail area = $0.005$, yielding $Z^* = 2.576$.
• Margin of Error ($MoE$): The total amount added to and subtracted from the point estimate:     * $MoE = Z^* \times \sqrt{\frac{\hat{p}(1-\hat{p})}{n}}$
• General Formula for One-Sample Z-Interval:     * $\hat{p} \pm Z^* \sqrt{\frac{\hat{p}(1-\hat{p})}{n}}$

The Four-Step Process for Confidence Intervals

• Step 1: Name the Procedure: Identify the test as a "One-sample Z-interval for a population proportion" and state the parameter being estimated in context.
• Step 2: Check Conditions: Verify Random, Independence ($10\%$ rule), and Large Counts ($n\hat{p} \ge 10$ and $n(1-\hat{p}) \ge 10$).
• Step 3: Perform Calculations: Identify $\hat{p}$, calculate $Z^*$, find the Standard Error, and build the interval.
• Step 4: Interpret Results: State the conclusion in context: "I am $C\%$ confident that the true proportion of [population] that [context] is between [lower bound] and [upper bound]."

Example Study: Teacher College Loan Debt

• Scenario: A random sample of $780$ US teachers found that $82\%$ have college loan debt. Construct a $98\%$ confidence interval.
• Step 1: One-sample Z-interval for the proportion of all US teachers with college loan debt.
• Step 2:     * Random: Random sample stated.     * Independence: $780$ is less than $10\%$ of all US teachers.     * Large Counts: Successes = $780 \times 0.82 = 640$; Failures = $780 \times 0.18 = 140$. Both are $\ge 10$.
• Step 3:     * $\hat{p} = 0.82$     * $Z^<em>$ for $98\%$: Tail area $1\%$ ($0.01$), $Z^</em> = 2.326$.     * $SE = \sqrt{\frac{0.82 \times 0.18}{780}} \approx 0.01377$     * $MoE = 2.326 \times 0.01377 \approx 0.032$     * Interval: $0.82 \pm 0.032 \rightarrow (0.788, 0.852)$.
• Step 4: "I am $98\%$ confident that the true proportion of all teachers in the United States that have student loan debt is somewhere between $78.8\%$ and $85.2\%$."

Advanced Interpretations and Applications

• Interpretation of Confidence Level ($C\%$): This refers to the success rate of the method. If we were to take many samples of the same size and build intervals for each, about $C\%$ of those intervals would capture the true population proportion.     * It is not a probability that the truth is in a specific interval; the truth is either in it or it is not.
• Using Intervals for Claims: If an entire interval is above or below a claimed value, it provide evidence for that claim.     * In the teacher example: Does it provide evidence that the proportion is less than $84\%$? No, because the interval goes up to $85.2\%$. Does it provide evidence the proportion is over $70\%$? Yes, because the entire interval ($78.8\%$ to $85.2\%$)) is above $70\%$.
• Finding Sample Size for a Desired Margin of Error:     * Formula: $MoE = Z^* \sqrt{\frac{\hat{p}(1-\hat{p})}{n}}$     * If $\hat{p}$ is unknown, use a conservative estimate of $0.5$ to maximize the required sample size.     * Example: For $95\%$ confidence ($Z^* = 1.96$) and $MoE = 0.02$:         * $0.02 = 1.96 \sqrt{\frac{0.5 \times 0.5}{n}}$         * $( \frac{0.02}{1.96} )^2 = \frac{0.25}{n}$         * $n = \frac{0.25}{( \frac{0.02}{1.96} )^2} \approx 2401$

Difference Between Two Proportions ($p_1 - p_2$)

• Procedure: Two-Sample Z-Interval for the difference of two population proportions.
• Formula:     * $(\hat{p}_1 - \hat{p}_2) \pm Z^* \sqrt{\frac{\hat{p}_1(1-\hat{p}_1)}{n_1} + \frac{\hat{p}_2(1-\hat{p}_2)}{n_2}}$
• Example Comparison: Teachers vs. Nurses debt.     * Teachers: $n_1 = 780$, $\hat{p}_1 = 0.82$.     * Nurses: $n_2 = 550$, $\hat{p}_2 = 0.70$.     * Observed Difference: $0.82 - 0.70 = 0.12$.     * $95\%$ Interval: $0.12 \pm 1.96 \sqrt{\frac{0.82 \times 0.18}{780} + \frac{0.70 \times 0.30}{550}} \rightarrow 0.12 \pm 0.0468 \rightarrow (0.0732, 0.1668)$.
• Interpretation of Signs in Two-Sample Intervals:     * Entirely Positive: If the result is (Positive, Positive) and the order was $\hat{p}_1 - \hat{p}_2$, then $p_1$ is higher than $p_2$.     * Entirely Negative: If the result is (Negative, Negative), then $p_2$ is higher than $p_1$.     * Contains Zero: If the interval is (Negative, Positive), there is no evidence of a difference. The difference could be negative, positive, or zero.

Using the AP Statistics Formula Sheet

• The formula sheet provides a generic confidence interval structure: $\text{statistic} \pm (\text{critical value})(\text{standard error of statistic})$.
• Standard Error Formulas provided on the sheet:     * For one proportion: $\sqrt{\frac{p(1-p)}{n}}$     * For the difference of two proportions: $\sqrt{\frac{p_1(1-p_1)}{n_1} + \frac{p_2(1-p_2)}{n_2}}$     * Note: When $p$ is unknown for intervals, substitute $\hat{p}$.