Independent variables being omitted in multivariable logistic regression analysis

[Kritika Jain]

Hello,

I am working on a quantitative project for my MSc dissertation.
I have approximately 70 variables and approximately 518,000 observations.
I have coded for the missing data for all the variables.
When I run the command "logistic" for the dependent and independent variable, without any covariates, it provides the results.
However, upon including covariates (multivariable regression), few of the independent variables and covariates are "omitted".

What could be the reason for the same?

Thank you for your help.

Regards
Kritika Jain

[Clyde Schechter]

There are two reasons a variable can be omitted from a logistic regression.

One possibility is that the variable is colinear with one or more other variables in the model. Stata will automatically omit as many of those variables as needed until no colinearity remains.

The other possibility is that the variable is a perfect predictor of the outcome. For example if the outcome is always 1 (or always 0) when x = 1 (or when x = 0), then the variable is omitted from the model. Another form of perfect prediction can arise with continuous variables. If there is some threshold value, call it t, such that the outcome is always 1 (or always 0) when x > t (or when x < t) then the variable is also omitted from the model.

Stata never omits a variable without telling you why. If you look at the very top of the output of your -logistic- command, before the regression table begins, you will see one or messages telling you about the variables that are omitted and explaining why.

[Kritika Jain]

Yes, for many of the variables, the outcome does not vary with the exposure. The exposure and non-exposed groups have no cases of the outcome.
Thank you for solving the query.

Regards
Kritika