Notes on Simple Regression and Correlation Analysis
Introduction to Simple Regression and Correlation Analysis
- Focus on understanding relationships between two variables
- Key Objectives:
- Joint distribution of two variables
- Linear relationships between continuous variables
Chapter Overview
- Introduction
- The scatterplot
- Pearson’s correlation coefficient
- Regression: linear relationships
- Coefficient of determination
- Graphical representation of residuals
6.1 Introduction
- Multivariate Dataset: Observations on two or more variables
- Bivariate Dataset: Paired observations (x1, y1), (x2, y2), … , (xn, yn)
- Example Variables: Length and mass of men, IQ and academic performance, Diameter and height of trees
- Dependent and Independent Variables:
- Changes in Y explained by changes in X
- Example: Study hours (x) vs. Test marks (y)
6.2 Scatterplot
- Definition: Graphical representation of independent (x) vs dependent variable (y)
- First step to assess relationship potential between two continuous variables
- Example: Advertisement expenditure (x) vs. Passengers (y)
- Increasing x correlates with increasing y, indicating a potential linear relationship
6.3 Pearson’s Correlation Coefficient
- Correlation defines linear relationship between two variables
- Coefficient (r): Measures linear relationship degree
- Defined as the covariance between x and y
- Properties of Correlation Coefficient:
- Range: -1 ≤ r ≤ 1
- Positive r: Positive relationship
- Negative r: Negative relationship
- r = 1: Perfect positive relation
- r = -1: Perfect negative relation
- r = 0: No linear relation
- Example with strong positive correlations: r ≈ 0.993
6.5 Regression
- Purpose: Determine relationships and forecast values
- Estimation model:
- Theoretical: E(y)=a+βx+ϵ
- Estimated: y^=a+bx
- Interpretation of coefficients:
- a: Mean of y when x=0 (y-intercept)
- b: Change in y with a unit increase in x (gradient)
- Method of Least Squares:
- Minimizes vertical distances between actual observations and regression line
6.6 Coefficient of Determination
- Definition (R²): Indication of how well the regression curve fits the data
- Characteristics:
- Range: 0 ≤ R² ≤ 1
- R² = 1: Perfect fit
- R² = 0: No fit at all
- Example with R² = 0.993 indicating a very good fit
6.7 Graphical Representation of Residuals
- Residual: Difference between observed and predicted values
- Importance: Indicates fit quality of regression line
- Residual Plot: Scatter plot of residuals against independent variable
- Patterns to identify:
- No pattern: Good linear fit
- Non-linear pattern: Indicates non-linearity
- Increasing scatter: Indicates erratic behavior
- Outliers: Points greatly differing from regression line
Practical Considerations
- Ensure appropriate variable relationships are analyzed; avoid irrelevant correlations
- Handle predictions carefully, especially extrapolation, as it can lead to errors
- Non-linear data may necessitate transformation or non-linear regression methods
Calculator Use for Correlation and Regression
- Input data via specific modes for statistical calculations
- Calculate correlation coefficients and regression parameters using calculator functions
Self-Evaluation Exercises
- Application of learned concepts through construction of scatter plots, correlation, regression line fitting, and interpretation
- Example: Analyze Johan's heavy object tossing performance, reflecting on results and predicted values
- Calculation and explanation of coefficients and residuals for understanding model effectiveness.