1
$\begingroup$

Why do we need to use a base category (normalised to zero) when working with multinomial logit models?

Why do we need to report conditional logit coefficients?

Wouldn't marginal effects give us a better idea of the probability of belonging to a certain class?

Improve this question
$\endgroup$
1
  • 1
    $\begingroup$ One way to think about it: In a standard logit model, you don't have two equations, one for the "0" group and one for the "1" group; you just model the "1" group relative to the "0" group. $\endgroup$
    – Charlie
    Commented Jul 9, 2013 at 16:16

1 Answer 1

Reset to default
1
$\begingroup$

You need a base category in order to interpret the log odds ratio. These are always in comparison with a reference category.

I am not sure about your other two questions: What do you mean by "reporting conditional log coefficient"?

You need the estimated coefficients of your output to calculate the probability of belonging to a certain class.

The marginal effects are not a probability of belonging to a class, but refer to the marginal probability change if a variable is increased by one unit. These marginal probability effects can be discrete or based on derivatives, in which case they only hold for small changes in your independent variable.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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