Review:
Linear Probability Model
overall review score: 2.5
⭐⭐⭐
score is between 0 and 5
The linear probability model (LPM) is a simple regression technique used in binary outcome modeling, where the dependent variable takes on only two possible values (e.g., 0 or 1). It models the probability of the event occurring as a linear function of independent variables. While easy to implement and interpret, it has certain limitations, such as potentially predicting probabilities outside the [0,1] range and assuming constant effects across all values of predictors.
Key Features
- Uses ordinary least squares (OLS) for estimation
- Models binary dependent variables directly as a linear function
- Easy to interpret coefficients as changes in probability
- Computationally simple and fast to compute
- Prone to issues like heteroskedasticity and predicted probabilities outside [0,1]
- Often used as a baseline or introductory model before more complex alternatives
Pros
- Simplicity and ease of implementation
- Transparent interpretation of coefficients
- Computational efficiency for large datasets
- Helpful for pedagogical purposes and initial analysis
Cons
- Can produce predicted probabilities less than 0 or greater than 1
- Assumes constant impact of predictors across all probability levels
- Less appropriate than logistic or probit models for binary outcomes
- Issues with heteroskedasticity complicate inference
- Potential for biased estimates if the assumptions are violated