Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Interpreting Probit Model (Missing 1st category coefficient within a categorical variable)

    Hi, I am wondering what model would be best to test knowledge of a disease (q142 is binary: do you know about disease X, yes or no) and comparing it to how it varies across 3 regions.

    I know I cannot use the probit model since the region is not binary.

    Please let me know what other model might be best to find the associations between my covariates and q142. Thank you!
    Click image for larger version

Name:	screenshot or probit model.PNG
Views:	1
Size:	35.1 KB
ID:	1496381

    Last edited by Cosmi Ca; 02 May 2019, 21:36.

  • #2
    This is exactly how it's supposed to work. When you have a three-level categorical variable, such as region2, if you tried to put all three level indicators into the model, in every observation exactly one of the three would be 1 and the others would be zero. So the sum of the three would always be three. That linear relationship would make these three variable colinear with the constant term of the model, and the model would be unidentifiable. So to make the model identifiable some constraint has to be imposed. The simplest constraint is to set one of the indicator coefficients to zero (which is equivalent to leaving that indicator out of the model altogether). This is such a common situation that it is done automatically by Stata (and by every other statistical package as far as I know.) If you are interested in the predicted outcome probabilities, you can't read them off of the -probit- output anyway because the coefficients are in the probit metric, not the probability metric. All you need to do is follow your -probit- regression with
    Code:
    margins region2
    and Stata will show you the predicted outcome probabilities in all three regions, adjusted for the distributions of all the other variables in the model.

    This is omission of one level of a categorical variable is explained in every basic textbook that covers regression. I suggest you consult one and read the chapter on regression with categorical predictors carefully for a fuller explanation and understanding.

    Comment

    Working...
    X