17.2 Marginal Effects in Different Contexts
- Linear Regression Models
For a simple linear regression:
E[Y|X]=β0+β1X,
the marginal effect is constant and equal to β1. This makes interpretation straightforward.
- Logit and Probit Models
In logistic regression, the expected value of Y is modeled as:
E[Y|X]=P(Y=1|X)=11+e−β0−β1X.
The marginal effect is given by:
∂E[Y|X]∂X=β1P(Y=1|X)(1−P(Y=1|X)).
Unlike linear models, the effect varies with X, requiring evaluation at specific values (e.g., means or percentiles).
- Interaction Effects and Nonlinear Terms
When models include interactions (e.g., X1X2) or transformations (e.g., log(X)), marginal effects become more complex. For example, in:
E[Y|X]=β0+β1X+β2X2,
the marginal effect of X is:
∂E[Y|X]∂X=β1+2β2X.
This means the marginal effect depends on the value of X.