Number of Observations in Domain v. Number of Observations Used

[_maldini_]

I am building sequential logistic regression models using PROC SURVEYLOGISTIC (weighted data). I have an exposure variable, an outcome variable and several covariates. As you would expect, the more covariates I add to the model, the lower the "Number of Observations Used" (See image #1).

I'm trying to create a table (above), that displays the number of events and the weighted percentage for each level of the exposure variable, for each model (each model has a different number of observations). 

My understanding is that the difference between the number of observations in the domain and the number of observations used is a function of missing values, non-positive weights and possibly other factors. 

The log (See image #3 below) says nothing of missing values or non-positive weights.

I can get the number of events and weighted percentage by using PROC SURVEGYFREQ (not shown) with a DOMAIN statement, but that uses the Number of Observations in Domain, not the Number of Observations Used.

My syntax for the DOMAIN variable is intended to eliminate any missing values: 

flag_1 = 
	(
	cann_use_status ne . and 
	ever_told_mi in(1,2) and
	age_yrs ne . and
 	gender ne . and
	race_all ne . and
 	educ_gtet_20 ne . and
 	hh_income ne . 
	); 

How can I get the number of events (i.e., number of "Yes" in ever_told_mi) and the weighted percentage using the Number of Observations Used, not the Number of Observations in Domain? 

Sorry for the long and complicated question. Thanks for your help.

[ballardw]

I'm really not sure what your question is. You state: "The log (See image #3 below) says nothing of missing values or non-positive weights."

However the third Note line clearly states that. I am not going to retype the text.

Hint: Post LOGS as text, copy the text from the log, on the forum open a text box and paste the text. It should be easier than creating an image. Plus when we want to make a suggestion to code or point out a syntax error it is much easier.

If you want a result for domain when it is missing then add the option to the domain statement.

From the documentation:

When determining levels of a DOMAIN variable, an observation with missing values for this DOMAIN variable is excluded, unless you specify the MISSING option.

---

@_maldini_ wrote:

I am building sequential logistic regression models using PROC SURVEYLOGISTIC (weighted data). I have an exposure variable, an outcome variable and several covariates. As you would expect, the more covariates I add to the model, the lower the "Number of Observations Used" (See image #1).

I'm trying to create a table (above), that displays the number of events and the weighted percentage for each level of the exposure variable, for each model (each model has a different number of observations). 

My understanding is that the difference between the number of observations in the domain and the number of observations used is a function of missing values, non-positive weights and possibly other factors. 

The log (See image #3 below) says nothing of missing values or non-positive weights.

I can get the number of events and weighted percentage by using PROC SURVEGYFREQ (not shown) with a DOMAIN statement, but that uses the Number of Observations in Domain, not the Number of Observations Used.

My syntax for the DOMAIN variable is intended to eliminate any missing values: 

flag_1 = 
	(
	cann_use_status ne . and 
	ever_told_mi in(1,2) and
	age_yrs ne . and
 	gender ne . and
	race_all ne . and
 	educ_gtet_20 ne . and
 	hh_income ne . 
	); 

How can I get the number of events (i.e., number of "Yes" in ever_told_mi) and the weighted percentage using the Number of Observations Used, not the Number of Observations in Domain? 

Sorry for the long and complicated question. Thanks for your help.

---