AP Statistics Unit 2: Exploring Two-Variable Data Notes
Introduction to Exploring Two-Variable Data
- General Concept: This unit focuses on taking a single set of data and collecting two variables from every individual in that set to analyze the relationship or association between them.
- Examples of Two-Variable Data Collections:
* Measuring the length and weight of a single sample of frogs.
* Examining a sample of hospital patients focusing on their age and the duration of their hospital stay.
* Tracking a sample of children to record their age alongside their television watching habits.
- Analytical Divisions: The unit is split into two primary sections:
1. Analyzing the relationship between two categorical variables.
2. Analyzing the relationship between two quantitative variables.
- Analytical Methods: For both sections, researchers utilize graphs and sample statistics to determine if patterns are random or indicative of a significant relationship.
Analyzing Two Categorical Variables
- Organization Tool: The most effective way to organize categorical data is through a two-way table.
- Example Study: A sample of students was asked two questions:
1. What is your mode of transportation to school (Bus, Walk, Self-Drive, or Parent-Drive)?
2. Were you tardy at least once in the last week (Yes/No)?
Types of Sample Statistics for Categorical Variables
- Marginal Relative Frequencies:
* Definition: These are proportions calculated using the row or column totals found in the margins of the table, divided by the grand total.
* Examples: The proportion of all students who were tardy; the proportion of all students who rode the bus.
- Joint Relative Frequencies:
* Definition: The proportion involving a specific combination of one category from each variable (the intersection of a row and a column).
* Keyword: Typically associated with the word "and."
* Example: The proportion of students who were tardy and rode the bus.
- Conditional Relative Frequencies:
* Definition: The proportion of individuals in a specific category given that they belong to a certain sub-category of the other variable.
* Mechanism: The denominator is limited to the sub-total of the condition provided.
* Example 1: "Given a student is tardy, what proportion ride the bus?" (Denominator = Total Tardy Students).
* Example 2: "Given they walk to school, what proportion are tardy?" (Denominator = Total Walking Students).
- Importance of Proportions: Relative frequencies (percentages/proportions) are used instead of raw counts because sample sizes across categories are often unequal, making counts misleading for comparison.
Visualizing Associations in Categorical Data
- Segmented Bar Graphs: These are used to display conditional relative frequencies.
* Each bar represents a condition (e.g., a mode of transportation).
* The bar is segmented to show the proportions of the other variable (e.g., Tardy vs. Not Tardy) within that condition.
- Association vs. Independence:
* Association: Exists when the variables are not independent; knowing the value of one variable changes the probability/proportion of the other.
* Independence: Exists when one variable has no impact on the other; knowing the mode of transport does not change the likelihood of being tardy.
Determining Association: Numerical Comparison
- Case 1: No Association (Independence):
* Marginal Probability: 34.6% of all students were tardy.
* Conditionals:
* Bus: 34%
* Parent: 34%
* Drive Self: 31/90≈34%
* Walk: 35.3%
* Conclusion: Since the conditional probabilities are approximately equal to the marginal probability, there is no association. Arrival mode does not influence tardiness.
- Case 2: Clear Association:
* Marginal Probability: 65/188≈34%
* Conditionals:
* Bus: 4/35≈11%
* Parent Drive: 5/29≈17%
* Drive Self: 48/90≈53%
* Walk: 8/34≈24%
* Conclusion: Because the conditions change the probability (e.g., driving oneself significantly increases tardiness risk), a clear association exists.
Analyzing Two Quantitative Variables
- Primary Tool: The Scatter Plot is used to represent the relationship between two measurable, numerical variables.
- Requirements: Both variables must be quantitative.
- Axis Definitions:
* X-axis: Explanatory Variable (used to predict or explain the response).
* Y-axis: Response Variable (the outcome being measured).
- Direction:
* Positive: As the explanatory variable increases, the response variable increases.
* Negative: As the explanatory variable increases, the response variable decreases.
- Form:
* The general shape of the data points. Common forms include linear, logarithmic, exponential, or parabolic (curved).
- Strength:
* How closely the points adhere to the form. Described as strong, moderate, or weak.
- Unusual Features:
* Includes outliers, clusters, or significant gaps in data.
- Context: Always use the specific variables and units (e.g., "As house size in square feet increases, there is a tendency for price in thousands of dollars to increase").
The Correlation Coefficient (r)
- Definition: A numerical value that quantifies the strength and direction of a linear relationship between two quantitative variables.
- Rules for Correlation:
* Only applies to linear data (do not use if a curve is present).
* Variables must be quantitative (cannot be used for eye color, etc.).
* It has no units.
- Scale of r:
* Range: [−1,1].
* Values close to −1 or 1: Strong relationship.
* Value close to 0: Weak relationship.
* 1 or −1: Points form a perfect straight line.
- Correlation vs. Causation: Even a very strong correlation (r≈1) does not imply that the explanatory variable causes the response variable to change.
Correlation Examples from Transcript
- Weight vs. Daily Steps: r=0.062 (Positive, but extremely weak; essentially horizontal).
- House Size vs. Price: High positive r (Strong, dots clearly forming a line).
- I.Q. vs. Frog Dissection Time: r=−0.621 (Moderate-strong negative relationship; smarter students tend to finish faster, but with significant scatter).
- Tree Diameter vs. Tree Height: r=0.773 (Fairly strong positive relationship, despite some clustering).
Linear Regression Models
- Purpose: To make predictions for the response variable (y) based on the explanatory variable (x).
- Equation Structure: y^=a+bx
* y^: Predicted response value (pronounced "y-hat").
* a: Y-intercept.
* b: Slope.
* x: Explanatory variable.
- Prediction Limits:
* Interpolation: Making predictions within the range of the observed x data (reliable).
* Extrapolation: Making predictions outside the observed data range (unreliable and not recommended).
- Mathematical Directionality: The equation is strictly for predicting y from x. You cannot plug in an actual y value to solve for x.
Residuals and the Least Squares Line
- Definition of a Residual: The difference between the actual observed value and the predicted value.
* Formula: Residual=y−y^
* Memory Aid: "A" comes before "P" (Actual minus Predicted).
- Properties:
* Vertical distance from a point to the line.
* Positive residual: Point is above the line.
* Negative residual: Point is below the line.
* The sum of all residuals in a dataset is always 0.
- Least Squares Regression Line (LSRL): The specific line that minimizes the sum of the squared residuals: min∑(y−y^)2.
- Residual Plot: A scatter plot of residuals against the explanatory variable (x).
* Good Fit: No discernible pattern (random scatter around the zero line). This indicates a linear model is appropriate.
* Poor Fit: A visible curve or pattern in the residual plot indicates the original data was not linear.
Regression Statistics and Interpretations
- Slope (b): b=r×sxsy
- Y-Intercept (a): a=yˉ−bxˉ
- Variables: sx, sy (standard deviations of x and y); xˉ, yˉ (means of x and y).
Interpretation of Terms
- Y-Intercept (a): The predicted value of the response variable when the explanatory variable is zero. (Often lacks contextual meaning due to extrapolation).
- Slope (b): For every one-unit increase in the explanatory variable (x), we predict the response variable (y) to change by b units.
- Coefficient of Determination (r2): The percentage of the variation in the response variable (y) that is explained by the variation in the explanatory variable (x) through the linear regression model.
* Example: r2=87% means 87% of the variation in house price is explained by house size.
- Standard Deviation of the Residuals (s): The typical distance that an actual observed value is from its predicted value.
* It measures the "average error" of the model's predictions.
* Context matters: Being off by 5lb is small for an elephant but large for a rabbit.
Computer Regression Output
- In AP Statistics, calculations are often secondary to interpretation. Data is frequently provided in a table.
- Key Columns:
* Coeff/Variable: Identifies the Constant (Intercept) and the Explanatory Variable (Slope).
* S: Standard deviation of residuals.
* R-Sq: Coefficient of determination (r2).
- Values to Ignore (until later units): SE (Standard Error), T-stat, and P-value.
Outliers and Influential Points
- Outlier: A point that does not fit the overall trend.
- Specific Types:
* Outlier in the Y-direction: Has a large residual. Typically weakens the correlation (r).
* Outlier in the X-direction (High Leverage Point): Has a x-value far from the mean of the other x-values. These have the potential to significantly anchor/pull the line.
* Influential Point: A point that, if removed, substantially changes the model (slope, intercept, or correlation).
- Patterns: An outlier in both x and y might fit the existing trend (e.g., a very large, expensive house). Adding such a point strengthens the correlation because it extends the linear pattern.
- High Leverage Effects: High leverage points often have small residuals because the LSRL is "pulled" toward them to minimize the squared distance, but they significantly alter the slope.