Proc surveymeans

[deleted_user]

Can anyone explain why output from proc surveymeans can result in a ratio, when added to the confidence interval, exceeds 100% ? i.e., 97.0±10.8=107.8

Ratio=0.970
Standard Error=0.055
Confidence Interval=(1.96)*(Standard Error)=1.96*0.055=0.108

Code:

proc surveymeans data=la ratio nobs;
by race_eth;
var d24u
strata frame;
cluster sridnum;
weight weight;
ratio d24u students;
ods output ratio=laoutput;
run;

Log:

NOTE: The BY statement provides completely separate analyses of the BY groups. It does not provide a statistically valid subpopulation or domain analysis, where the total number of units in the subpopulation is not known with certainty. If you want a domain analysis, you should use the DOMAIN statement.

NOTE: Only one cluster in a stratum for variable(s) d24u, students. The estimate of variance for d24u, students will omit this stratum.
Only one cluster in a stratum for ratio(s) d24u/students. The estimate of variance for d24u/students, will omit this stratum.
NOTE: The above message was for the following by-group:
race_eth=Bl


Thanks.

[deleted_user]

The short answer is because you are attempting to interpret it that way.

How you should interpret it is that you need to go as low as 86.2 to have that confidence but common sense (or your own constraint) tells you not to go above 100. But confidence interval formulation doesn't know intrinsically that the ratio can't go above 100, that is your own interpretation/constraint.

If you want to impose that then report it as:

97.0{+3.0/-10.8}

You can't expect the calculation to understand arbitrary constraints.

Ike

[deleted_user]

Thanks Ike.