Lecture_20Video_20W11D2_20-_20Interactions
Introduction
Today's topic: Interaction in Regression Analysis
Considered the hardest part of the class.
Upcoming topics will cover practical applications and logistic regression.
Interactions in Regression
Definition
Interaction occurs when the effect of one predictor variable on the response variable depends on the value of another predictor.
Contrast with additive model: No interaction means the effects of predictors add up independently.
Example of Interaction
Variables: Blood pressure (response), Age, Gender (predictors)
No Interaction: Effect of age on blood pressure is the same for males and females (parallel lines).
With Interaction: Effect of age on blood pressure is different for males and females (non-parallel lines).
Study Case: Weight Loss Program
Study Design
Participants: Males and females were divided into treatment and control groups.
Initial average weight was equal in both groups.
Observations
After treatment, weight changes differ:
Males: Some weight loss.
Females: More significant weight loss.
Conclusion: Interaction exists between treatment effect and gender; the effect of the treatment depends on gender.
Mathematical Representation of Interaction
Model Structure
Include cross product term in the model for interaction.
Example predictors: Gender (X1), Treatment (X2).
Coefficients:
Beta0: Average weight for females in control group (both X1 and X2 are 0).
Beta1: Difference in average weight between males and females in the control group.
Beta2: Difference in average weight between treatment and control for females.
Beta3: Difference in differences, quantifying the interaction effect.
Importance of Interaction Terms
Determines how the effect of one predictor varies at different levels of another predictor.
If Beta3 is significant, it confirms interaction is present.
Interaction with Categorical and Continuous Variables
Categorical (e.g., Gender) and Continuous (e.g., Age)
Separate equations for each category:
Plugging in values provides distinct slopes for effects on the response variable.
If slopes are equal, indicates no interaction.
Continuous Variables Interaction
Collect terms involving predictors to find slopes:
Example: Y = Beta0 + Beta1X1 + Beta2X2 + Beta3X1X2
Interpretation of slopes depends on the interacting variables.
Importance of Sign and Magnitude of Interaction
Interpreting the Signs
Positive or negative signs of interaction terms indicate the type of relationship among predictors and response.
Positive interaction: Effect of one variable increases with the value of another.
Negative interaction: Effect of one variable decreases with increasing value of another.
Practical Application
Coding in R for Interaction Terms
To include interaction in models:
Use multiplication (X1*X2) or colon (X1:X2) syntax.
Model summary provides coefficients and significance levels for interpretation.
Conclusion
Understanding interaction is crucial for accurately modeling relationships in regression analysis.
Next session: Discuss practical coding and work through practice problems from the Framingham data.