Variables are perfectly predicted the outcome

[Hm Saleh]

Hello guys,

I attached my LOGIT result. My results failed.

Because I had variables that perfectly predicted the outcome, I move to -firthlogit-
But -firthlogit- does not run, and takes too much time. So I am not sure if -firthlogit- is the solution or no.

Can anyone help to figure out why my results failed?
How can this problem be solved?

Thank you,

[Carlo Lazzaro]

Hm:
set aside the unavoidable fact that the MLE machinery cannot move forward maximizing the log function if there's no variation in the relationship between outcome and predictors, taking a glance to your attachment (btw: as we know, the FAQ encourage to share what the poster typed and what Stata gave her/him back in return via CODE delimiters, whereas non-Stata attachments should be avoided) you coded a too demanding model (there's an overkill of age categories, just to report what stroke me the most).
That said, two (hopefully constructive) comments follow:
1) when deciding what to plug in the right-hand side of your regression equation, just spend 10 minutes of your time thinking over how effectively you could disseminate your results to an audience with an average smattering if statistics related to your reserach field;
2) giving a fair and true view of the data generating process does not necessarily mean to create a 1000-predictor regression.

[Hm Saleh]

So, there is no solution!

[Joseph Coveney]

So, there is no solution!

Most of your problem is not perfect prediction, but rather that you have too many empty cells in your categorical × categorical × categorical interaction terms.

Start by treating age as a continuous predictor. If necessary, then do the same for education attainment and household income, using the midpoints of range categories for the latter. If you're concerned about missing nonlinear relationships, then see this:

help mkspline
search bspline