Study Notes on Hypothesis Testing and Linear Regression

12.1 Hypothesis Testing: Linear Regression Edition

Learning Targets

• I can construct a confidence interval, check the conditions for, and run a hypothesis test for the slope (𝛽) of the true regression line.

Hypothesis Testing Notes

• This section may appear in multiple-choice questions (MC) but has rarely been seen in free-response questions (FRQ).

Notation and Provided Information

• Sampling distributions for simple linear regression:

  • Random Variable for Slope:

  • Notation: $b$

  • Parameters of Sampling Distribution:

  • Mean: ext{mean}_ ext{error} = eta

  • Variance: $ext{var}(b)$

  • Standard Error of Sample Statistic: S = rac{ ext{Σ}(x - ar x)^2}{n - 1}

    • Where $ar x$ refers to the mean of $x$ values and $n$ is the sample size.

Population vs. Sample

• Definitions:

  • Population Regression Line:

  • A regression line calculated from every value in the population. It is known as the true regression line.

  • Equation: ar y = eta0 + eta1 x where:

    • $ar y$ = mean $y$ for a given $x$

    • $eta_0$ = population y-intercept

    • $eta_1$ = population slope

  • Sample Regression Line:

  • A regression line calculated from a sample. It is known as the estimated regression line.

  • Equation: $ilde y = b<em>0 + b</em>1 x$ where:

    • $ ilde y$ = estimated mean $y$ for a given $x$

    • $b_0$ = sample y-intercept

    • $b_1$ = sample slope

Old Faithful Example

• Observational Data:

  • Data Points: Duration and interval between eruptions of Old Faithful (263 eruptions in a month).

  • Population Least-Squares Line: Shown in blue on the scatterplot representing all eruption data.

Slope Sampling Distribution

• Sampling Distribution of the Slope:

  • Select a Simple Random Sample (SRS) of $n$ observations $(x, y)$ from a population.

  • Least-squares regression model: ar y = eta0 + eta1 x

  • Mean of the sampling distribution of $b1$: ext{mean}(b1) = eta_1

  • Standard Deviation of Sampling Distribution:

  • Exists if the 10% condition is satisfied: n < 0.10N

  • Interpretation: The slopes of sample regression lines differ from the population regression line by the standard deviation, typically around 1.42.

Experiments and Observational Studies

• Study Inquiry: Does seat location impact student scores?

  • 30 students randomly assigned to seats and their exam scores recorded.

  • Questions regarding random assignment significance.

Regression Analysis and Interpretations

• Slope of the Least-Squares Regression Line (LSRL) Interpretation:

  • Negative slope indicates potential causation link.

  • Conditions checked: Linear relationship, Independence, Normality, Equal Standard Deviation (LINER condition).

• Formula:

  • $ext{Non-Confidence Ranges:} ext{statistic } ext{±} ( ext{critical value}) imes ( ext{standard deviation})$

• A 95% confidence interval for the unknown slope B₁ is formulated as:

  • $b<em>1 ext{± } t^* imes SE</em>{b_1}$

Significance Tests for the Slope

• Null Hypothesis: $H_0: B = ext{hypothesized slope}$

• Formulation of the t-test statistic:

  • $t = rac{b<em>1 - ext{hypothesized slope}}{SE</em>b}$

  • P-value calculated for achieving or exceeding test statistic.

Various Examples for Contextual Application

GPA and ACT Score Analysis

• Small sample investigates the relationship between GPA and ACT scores through regression analysis.

Balloon Box Volume Study

• This study examines the correlation between the volume of balloons and various factors such as inflation method and material composition using linear regression techniques.

• Investigating how height of a balloon box correlates with the number of balloons required for filling, utilizing power models for predictive analysis.

Exponential Progression: Moore's Law

• Modeling growth in the number of transistors, hinting at exponential relationships through transformed data.

Conclusion/Reflections on Statistical Practices

• Importance of clearly stating hypothesis, defining parameters, recognizing patterns in data through visual representations such as residual plots, clarifying interpretations, and continuously aligning statistical findings to real-world implications.