2
$\begingroup$

If you are creating a regression model where the response variable is a numerical value, but one of the variables is a dummy (binary), can you use OLS-method?

Do you only use logistic regression if your response variable is binary?

$\endgroup$
  • 4
    $\begingroup$ Yes, you should only use logistic regression if your response variable is binary. If your response is categorical, you could use multinomial logistic regression. If your response is continuous, you should find another method (such as OLS). $\endgroup$ – Frank Nov 12 '18 at 15:07
  • $\begingroup$ @Frank - you might want to expand that into an answer, since it is! $\endgroup$ – jbowman Nov 12 '18 at 16:43
  • $\begingroup$ A relevant prior question: stats.stackexchange.com/questions/104573/… $\endgroup$ – Frank Nov 12 '18 at 17:43
2
$\begingroup$

In short:

  1. Yes, if your response variable is continuous, even if one of the X variables is binary, you can use OLS.
  2. Yes, you should only use logistic regression if your response variable is binary.

Why is this? To your first question, OLS handles binary X variables just fine. For simplicity, consider a model with two X variables. (Assume that they aren't perfectly correlated.) $$ y_i = \alpha + \beta_1 x_{1i} + \beta_2 x_{2i} + \varepsilon_i $$ If $x_{1i}$ is binary, then we interpret $\beta_1$ as follows: holding $x_{2i}$ constant, it's the predicted change in $y_i$ from observing $x_{i1} = 1$ instead of 0. (This holds whether $x_{2i}$ is continuous or binary, by the way.)

To your second question, binary logistic regression is designed specifically to model a binary response. Recall that the underlying model is the following: for a vector $x_i$, the probability that $y_i = 1$ is $$ P(y_i = 1 \mid x_i) = \frac{\exp(\beta' x_i)}{1 + \exp(\beta' x_i)}. $$ This doesn't generalize to continuous $y$. (It does generalize to categorical $y$; this is called multinomial logistic regression.)

| cite | improve this answer | |
$\endgroup$
0
$\begingroup$

Left hand side (LHS) variables (the y) under OLS can take on any value. Under logistic regression the (predicted) LHS variable is bounded to min=0, max=1.

You can use OLS for binary LHS variables. However, you will likely end up predicting values smaller zero or greater one. If you want to avoid this, use logistic regression.

Please note that if you want to use interactions of right hand side variables (the X), logistic regression is not a good choice. In this case OLS is often used ("linear probability model"), e.g. if you want to study the effect of a treatment (modelled as interaction term(s) in X) on a binary outcome.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Do you have a reference (or argument) for your claim about interactions? $\endgroup$ – kjetil b halvorsen Nov 12 '18 at 20:41

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.