Regression adjust

[Sara Crowe]

Hello
I am doing logistic regression adjust, but it only works for some outcomes, even though they are all binary.
Below I have posted the code and table for a succesful regression for one symptom, and below the line I tried for another symptom and it shows these fail messages. What am I doing wrong?


. teffects ra (any_skin_symptoms ib(first).age ib(first).smoking ib(first).tobacco ib(first).alcohol, logit) (types_of_sanit
> ation_worker)

Iteration 0: EE criterion = 4.529e-26
Iteration 1: EE criterion = 9.946e-34

Treatment-effects estimation Number of obs = 678
Estimator : regression adjustment
Outcome model : logit
Treatment model: none

	Robust
any_skin_symptoms	Coefficient	std. err.	z	P>z	[95% conf.	interval]
ATE
types_of_sanitation_worker
(Collector vs Sweeper)	.1002326	.0368271	2.72	0.006	.0280528	.1724124
(Transporter/Waste vehicle driver vs Sweeper)	.0204006	.0287798	0.71	0.478	-.0360068	.076808
(Pickers of dumping site vs Sweeper)	.0344529	.0387662	0.89	0.374	-.0415275	.1104333
POmean
types_of_sanitation_worker
Sweeper	.0643657	.0170494	3.78	0.000	.0309495	.0977818

__________________________________________________ __________________________________________________ _
. teffects ra (suffered_from_infection ib(first).age ib(first).smoking ib(first).tobacco ib(first).alcohol, logit) (types_of
> _sanitation_worker)
outcome model: perfect predictions detected; the model, as specified, is not identified
1.types_of_sanitation_worker#1.smoking != 0 (n=1) predicts failure perfectly
2.types_of_sanitation_worker#2.age != 0 (n=2) predicts failure perfectly
2.types_of_sanitation_worker#1.smoking != 0 (n=1) predicts success perfectly
4.types_of_sanitation_worker#2.age != 0 (n=2) predicts success perfectly
4.types_of_sanitation_worker#1.smoking != 0 (n=1) predicts success perfectly
4.types_of_sanitation_worker#1.alcohol != 0 (n=1) predicts failure perfectly
r(322);

[Andrew Musau]

outcome model: perfect predictions detected; the model, as specified, is not identified

This is the key line.The following link explains the problem of perfect prediction: https://stats.oarc.ucla.edu/other/mu...ith-the-issue/

[Sara Crowe]

Thank you very much! I will try applying the suggested solutions!