Probit and Logit Models Flashcards
1
Q
The LPM model controls for heteroskedasticity. Show that for such a model there is always heteroskedasticity.
A
The dependent variable eg. BCOVID has a Bernoulli distribution, with variance P*(1-P). P depends on X1 and X2.
The error term in a binary model is either 1-Xb with probability P or Xb with probability (1-P), hence the variance of the error term is:
P(1- Xb)+(1-P)Xb.
In an LPM model, there is, therefore, always heteroskedasticity.
In a standard linear regression model with a continuous dependent variable, the variance is, say, sigma2, which does not necessarily depend on X (only if there is heteroskedasticity).
