Predict 95% interval

[saza]

Was told to Predict, with 95% confidence, the values between which the true mean cancer mortality lies for all counties with an exposure index of 5.2.  

code was: 

libname cancer '/home/u59291263/Cancer';
run;

proc contents data=cancer.cancer;run;

proc reg data= cancer.cancer;
model exposure= mortality/ clm cli;
ID mortality;
run;

data work.newobs;
set cancer.cancer;
exposure= 5.2;
run;

proc reg data= work.newobs;
model exposure= mortality/ clm cli;
ID exposure;
run;

 But I am getting 5.2 as all my answers 

[PaigeMiller]

https://blogs.sas.com/content/iml/2014/02/17/the-missing-value-trick-for-scoring-a-regression-model....

[PaigeMiller · Accepted solution]

By the way, I just noticed this

Shouldn't your PROC REG have 

model mortality = exposure/ clm cli;

where exposure predicts mortality (instead of the way you wrote it, where mortality predicts exposure)?

[Rick_SAS]

I think you want to predict mortality (the response) from exposure (the independent variable). You have those variables reversed in your MODEL statement.

Try this:

data ScoreObs;
exposure= 5.2;
run;

data All;
set cancer ScoreObs(in=score);
NewData = (Score=1);
run;

proc reg data= All plots=none;
model mortality = exposure / clm cli;
output out=ScoreOut Pred=Pred lclm=LowerM uclm=UpperM lcl=LowerI ucl=UpperI;
run;

proc print data=ScoreOut;
where NewData=1;
run;

[Rick_SAS]

I assume that this is a homework problem, but I feel compelled to mention that linear regression with PROC REG is probably not the best way to measure mortality, which is a rate in the interval [0, 1]  (or 0 to 100 if you are measuring percentages).  A linear regression could predict a negative mortality rate or a mortality rate that is greater than 100%. Neither of these predictions makes sense, which is why we don't linear regression is not the best way to predict bounded quantities.