How to estimate Relative Risk using a Multinomial Logistic Regression

Reply
New Contributor
Posts: 2

How to estimate Relative Risk using a Multinomial Logistic Regression

Hello SAS friends,

 

Please bear with me, I am a coding beginner and am not qualified to invoke macros. I use SAS Studio, University Edition.

 

I have a categorical response variable with 4 levels (DENV_CNTL2, levels 1, 3, 4, 5).
I have a continuous predictor variable (RBP4_ADJC).

Additional predictor variables include (Age, continuous; Sex, binary; Fever, binary).

I would like to run a multinomial logistic regression first with only 1 continuous predictor variable. I would like to run subsequent models with the additional predictor variables (categorical and continuous).

I have read that it's possible to estimate relative risk with PROC LOGISTIC using the %NLEstimate macro. I am having trouble writing the f or fdata parameters of this macro based on my variables. I have also never used a macro.

The code is below, and the output is attached.

PROC LOGISTIC DATA=Suze.ECData;
CLASS DENV_CNTL2 (ref = 1) / param = ref;
MODEL DENV_CNTL2=RBP4_ADJC / RL link=glogit;
STORE OUT=NLE;
RUN;

%NLEstimate(INSTORE=NLE, f=..... ???)

 

Thank you kindly!

 

Suze

Super User
Posts: 21,572

Re: How to estimate Relative Risk using a Multinomial Logistic Regression

[ Edited ]

If something is new, find a fully worked example, ensure you can replicate the results and then try and apply it to your data. 

Here’s the one for what you’re looking for. 

http://support.sas.com/kb/57/798.html

 

Calling a macro is the same as using a function in Excel. You put the right things in the function and you get results. 

Learning to use one is very straightforward, writing them is a different story. There no qualifications required to run macros besides a willingness to try it. 

 


suze152 wrote:

Hello SAS friends,

 

Please bear with me, I am a coding beginner and am not qualified to invoke macros. I use SAS Studio, University Edition.

 

I have a categorical response variable with 4 levels (DENV_CNTL2, levels 1, 3, 4, 5).
I have a continuous predictor variable (RBP4_ADJC).

Additional predictor variables include (Age, continuous; Sex, binary; Fever, binary).

I would like to run a multinomial logistic regression first with only 1 continuous predictor variable. I would like to run subsequent models with the additional predictor variables (categorical and continuous).

I have read that it's possible to estimate relative risk with PROC LOGISTIC using the %NLEstimate macro. I am having trouble writing the f or fdata parameters of this macro based on my variables. I have also never used a macro.

The code is below, and the output is attached.

PROC LOGISTIC DATA=Suze.ECData;
CLASS DENV_CNTL2 (ref = 1) / param = ref;
MODEL DENV_CNTL2=RBP4_ADJC / RL link=glogit;
STORE OUT=NLE;
RUN;

%NLEstimate(INSTORE=NLE, f=..... ???)

 

Thank you kindly!

 

Suze


 

New Contributor
Posts: 2

Re: How to estimate Relative Risk using a Multinomial Logistic Regression

Dear Reeza,

 

Thank you for your reply and helpful link. I have been following your suggested link and the example in the "Using PROC LOGISTIC and the NLEstimate macro" section.

I have not been able to find an example using a continuous predictor variable. How will a continuous predictor variable change the application of the NLEstimate macro?

I would like some guidance on how to write the datalines for the data set specified in the fdata=fd NLEstimate macro parameter. Please see the "shownames" in the screenshot attached. 

I am also unsure of my degrees of freedom parameter in the NLEstimate macro.

 

data fd;
length label f $32767;
infile datalines delimiter='|';
input label f;
datalines;
P(3)/P(1)| exp(b_p1+b_p4)
P(4)/P(1)| exp(b_p2+b_p5)
P(5)/P(1)| exp(b_p3+b_p6)
P(3)/P(4)| exp(b_p1+b_p4-b_p2-b_p5)
P(3)/P(5)| exp(b_p1+b_p4-b_p3-b_p6)
P(4)/P(5)| exp(b_p2+b_p5-b_p3-b_p6);

%NLEstimate(instore=NLE, fdata=fd, df=?)

 

Thank you!

 

Suze

 

 

Ask a Question
Discussion stats
  • 2 replies
  • 154 views
  • 0 likes
  • 2 in conversation