AP Statistics Unit 9: Comprehensive Study Guide on Inference for Slope

Overview of Unit 9: Inference for Slope

• Unit 9 is the final unit of content in the AP Statistics curriculum.
• The primary focus is extending the concepts of bivariate quantitative data from Unit 2 to make judgments about entire populations using sample data.
• Bivariate Quantitative Data Review: This data consists of two quantitative variables that can be visualized on a scatter plot to create a Least Squares Regression Line (LSRL).
• The Concept of Inference for Slope: The slope calculated from a sample ($b$) is just one estimate. Inference allows us to use this sample slope to make a judgment about the true population slope ($\beta$).

The Sampling Distribution of Sample Slopes

• To perform inference, one must understand the behavior of all possible sample slopes from samples of the same size taken from the same population.
• Center: The mean of the sampling distribution of sample slopes is the true population slope ($\beta$), provided the samples are drawn randomly to avoid bias.
• Spread (Standard Deviation): The formula for the standard deviation of the sampling distribution is:   - $\sigma_b = \frac{\sigma}{\sigma_x \sqrt{n}}$   - In this formula, $\sigma$ is the standard deviation of the true regression line, $\sigma_x$ is the standard deviation of the population X-values, and $n$ is the sample size.
• Standard Error ($SE_b$): Because population parameters ($\sigma$ and $\sigma_x$) are typically unknown, we use the Standard Error of the sample slopes as an estimate:   - $SE_b = \frac{s}{s_x \sqrt{n-1}}$   - Here, $s$ is the standard deviation of the sample residuals, $s_x$ is the standard deviation of the X-values in the sample, and $n-1$ is used in the denominator (the reasons for using $n-1$ are beyond the scope of AP Statistics).   - Note: This value is almost always provided in a computer regression analysis output table.
• Shape: The distribution of sample slopes is approximately normal provided certain conditions (specifically regarding residuals) are met.
• T-Model Application: When building the sampling distribution using standard error, a T-model is used with degrees of freedom ($df$) calculated as:   - $df = n - 2$   - The subtraction of 2 reflects the use of two variables in bivariate data.

Constructing a T-Interval for Slope

• A T-interval is used to estimate the true population slope ($\beta$).
• Four-Step Process:   1. Name the Interval: Explicitly state it is a "T-interval for the slope of [context]."   2. Check Conditions: Verify the assumptions necessary for the sampling distribution.   3. Build the Interval: Use the formula $b \pm t^* \times SE_b$.   4. Interpret the Interval: State the confidence level and what the interval represents in terms of the population relationship.

Necessary Conditions for Inference

• Linearity: The scatter plot of the sample data must appear somewhat linear. A regression line should not be applied to non-linear data.
• Randomness: Observations must be selected randomly to avoid bias and allow for generalization.
• Independence (10% Condition): The sample size $n$ must be less than $10\%$ of the population size to assume independence.
• Normality of Residuals: The residuals must be approximately normal with no major outliers or extreme skewness. This is verified by checking a histogram or dot plot of the residuals.
• Constant Variance (Equal SD of Residuals): The standard deviation of the Y-values should not vary with X. When looking at a residual plot, there should be similar variability (vertical spread) across all X-values, rather than a "fan pattern" (small residuals at one end and large ones at the other).

Example Case Study: Car Weight and Gas Mileage

• Scenario: A dealership analyzes the relationship between car weight (X, in pounds) and gas mileage (Y, in miles per gallon) using a random sample of $n = 12$ cars.
• Sample Data/Output:   - Sample Slope ($b$): $-0.00714$   - Standard Error ($SE_b$): $0.00056$
• Confidence Interval (95%) Calculation:   - Step 1: T-interval for the slope of the regression line between weight and gas mileage.   - Step 2: Conditions met (linear scatter plot, random sample of 12, 12 is less than $10\%$ of all cars, normal residuals, similar variability in residual plot).   - Step 3:     - $df = 12 - 2 = 10$     - $t^<em>$ (for $95\%$ confidence and $df = 10$) is found using invT(0.025, 10).     - Interval: $-0.00714 \pm (t^</em>) \times 0.00056$     - Resulting Interval: $-0.00834$ to $-0.00594$
• Interpreting the Result: "I am $95\%$ confident that the interval from $-0.00834$ to $-0.00594\,mpg/lb$ captures the slope of the true regression line relating gas mileage to weight."
• Practical Meaning: For every one-pound increase in car weight, the gas mileage is predicted to decrease between $0.00594$ and $0.00834\,mpg$.

Special Considerations: Zero in the Interval

• If a confidence interval for slope contains zero (e.g., from $-0.5$ to $+0.2$), it suggests that there may be no linear relationship between the variables in the population.
• A slope of zero represents a horizontal line, meaning changes in X do not predict changes in Y.
• If the entire interval is negative, there is strong evidence of a negative relationship.
• If the entire interval is positive, there is strong evidence of a positive relationship.

Kelly’s Case Study: Car Price vs. Mileage

• Scenario: Kelly examines $n = 10$ cars to predict Price (Y) based on Mileage (X).
• Details:   - Sample size: $10$   - $df = 10 - 2 = 8$   - Slope ($b$): $-0.2083$   - Standard Error ($SE_b$): $0.0684$
• 99% Confidence Interval Calculation:   - $t^*$ for $99\%$ confidence and $df = 8$ is approximately $3.3554$   - Interval: $-0.2083 \pm 3.3554 \times 0.0684$   - Resulting Interval: $-0.4378$ to $+0.0212$
• Conclusion: Since the interval contains zero, there is not convincing evidence of a linear relationship. The price could decrease by $44\,\text{cents}$ per mile, increase by $2\,\text{cents}$ per mile, or not change at all (slope of zero).

Significance Tests for Slope

• Purpose: To determine if there is statistically significant evidence of a linear relationship between two variables.
• Step 1: Hypotheses and Naming:   - Null Hypothesis ($H_0$): $\beta = \beta_0$ (Usually $\beta = 0$, meaning no relationship).   - Alternative Hypothesis ($H_a$): \beta > 0, \beta < 0, or $\beta \neq 0$.
• Step 2: Conditions and Sampling Distribution:   - Same conditions as the T-interval.   - Assume $H_0$ is true, so the sampling distribution is centered at zero (if $\beta = 0$).
• Step 3: Calculating Test Statistic and P-value:   - Test Statistic ($t$): $t = \frac{b - \beta_0}{SE_b}$   - P-value: Use tcdf(lower, upper, df) to find the area under the T-curve corresponding to the alternative hypothesis.
• Step 4: Conclusion:   - If $P < \alpha$: Reject $H_0$, stating there is evidence of a relationship.   - If $P > \alpha$: Fail to reject $H_0$, stating there is lack of evidence of a relationship.

Example Case Study: Baby Walking Age and Temperature

• Scenario: Researchers studied $n = 16$ babies to see if the average daily temperature in their city affects the age (in days) at which they first walk.
• Hypotheses:   - $H_0: \beta = 0$ (Temperature doesn't affect walking age).   - H_a: \beta < 0 (In warmer cities, babies walk sooner).
• Data:   - Sample Slope ($b$): $-0.0555$   - Standard Error ($SE_b$): $0.0906$   - $df = 16 - 2 = 14$
• Calculations:   - $t = \frac{-0.0555 - 0}{0.0906} = -0.6126$   - P-value: tcdf(-99, -0.6126, 14) yields approximately $0.2769$.
• Conclusion: With a P-value of $0.2769$, which is greater than standard alpha levels ($0.01$ or $0.05$), we fail to reject the null. There is no statistically significant evidence that a negative linear relationship exists. The observed relationship is likely due to chance.

Final Tips for the AP Exam

• Computer Output Reading: You are rarely required to calculate $SE_b$ or the slope manually. You must be able to identify them in a table. The slope is usually in the row labeled with the X-variable name, and the Standard Error is in the column labeled "SE Coeff" or "Standard Error" next to the slope.
• Multiple Choice: Often tests the ability to construct the formula part of the interval ($b \pm t^* \times SE_b$). Remember $df = n-2$.
• Free Response: Focus on interpreting the results (contextualizing the slope and the confidence interval).