Missing values when I run regression analysis

[Gaurav Mohan]

Hi, I am running the following program

Regression analysis for predictors of ICU admission(icu) against patient variables such as (age), (race), and other patient parameters(insur los hospday i.loc cancel orgdys___0 orgdys___1 etc).

------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Code: logistic icu age race i.insur los hospday i.loc cancel orgdys___0 orgdys___1 orgdys___2 orgdys___3 orgdys___4 orgdys___5 sirs___0 sirs___1 sirs___2 sirs___3 flu hypoten abx ams qsofa___0 qsofa___1 qsofa___3 t2dm htn chf cad esrd copd afib cld smok alc i.bmi art

Result:

------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

I do not understand why I have so many missing values for p-value and CI etc


However, when I run a subset of the patient parameters such as the following example, I get meaningful results.


------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Code: logistic icu age race i.insur los hospday i.loc cancel orgdys___0 orgdys___1 orgdys___2 orgdys___3 orgdys___4 orgdys___5 sirs___0 sirs___1 sirs___2 sirs___3

Results:

------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Can someone please help me understand why the former code generates so many missing values in the results?

[Nick Cox]

Count how many parameters you are trying to estimate. With 80 or so observations wanting to fit such a very complicated model to a comparatively tiny dataset is doomed to failure. Sorry, but there is no way to make the message appear as good news.

[Leonardo Guizzetti]

Your first model had too many parameters to estimate and not enough data that it is simply inestimable. Also, your second model has too many parameters such that you have about 4 observations per variable. I’m surprised you could estimate it, but if it were me, I would simplify the model greatly.

Edit: crossed with #2