I'm trying to write down the equation describing a linear probability model.

If I was writing out the equation for an OLS model with continuous y with observation unit i , I would write:

$$y_i = \beta_0 + \beta_1 x_{i1} + \beta_2 x_{i2} + \epsilon_i $$

If I were specifying the model for a probit model, I would write: $$ Pr(y_i =1) = \Phi(\beta_0 + \beta_1 x_{i1} + \beta_2 x_{i2}) $$ or something thereabouts.

To be clear, a linear probability model is just an OLS model in which Y is a 0/1 indicator variable for some particular outcome. I know this is kind of a pedantic question, but how do you formally indicate in the first equation that $y_i$ is an indicator variable for an outcome just so that it's clear from the equation what you are doing?

Alternatively, is this a silly question and you just write the first equation and clarify the meaning of $y_i$ in the text?

edit: Maybe I'm encouraging too much analysis here. Simply put, what would be the reasonable equation to write down in a basic empirical driven paper to say "this is the form of the model and the relevant data items" where you were running a linear probability model?

  • $\begingroup$ Estimating equation is a term with a technical meaning. What you have here seem to be (partially specified) models. $\endgroup$ Sep 24 '20 at 13:28
  • $\begingroup$ Sorry, that's right. I'll edit the question. $\endgroup$
    – WildGunman
    Sep 24 '20 at 13:30
  • $\begingroup$ Somehow your Wikipedia reference successfully manages to write down this model: what aspect of that reference, then, requires clarification? $\endgroup$
    – whuber
    Sep 24 '20 at 13:40
  • $\begingroup$ @whuber I'm not sure that it does though. It writes $E[Y|X]$, which doesn't make it clear that Y is an indicator variable and then it writes $Pr(Y=1|X)$ which is a statement of probability rather than a factual statement of Y as an indicator variable. It strikes me that the correct thing to do is to explicitly declare Y as an indicator variable. Tell me why I'm wrong about this interpretation? $\endgroup$
    – WildGunman
    Sep 24 '20 at 13:52
  • 1
    $\begingroup$ @wildgunman The expectation of a 0/1 variable is the same as the probability its 1 $\endgroup$
    – Glen_b
    Sep 25 '20 at 5:36

You could write both regression models (and all others) the exact same way: $Y | X= x \sim p(y|x)$, for some conditional distributions $p(y|x)$. The case of binary $Y$ implies that these distributions are Bernoulli.

OLS is an estimation procedure, not a model. The distributions $p(y|x)$ are not specified by stating "OLS."

On the other hand, OLS is a good estimation procedure to use when the $p(y|x)$ are normal, homoscedastic, and have a conditional mean function that is a straight line.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.