topic exact binomial CI in Statistical Procedures

exact binomial CI

SSN_Ravi — Mon, 03 Apr 2017 16:44:16 GMT

Hi,

I am trying to calculate binomial CI for my efficacy data using below code. But it is throwing error when we have only one category(resp='0') in response variable for a particular category of by variable. As i am using LEVEL=2 option its throwing error. Let me know how to handle this.

proc freq data=dsin1;
by viswin leg_sort ;
tables &G_A_RESPVAR/ binomial (level=2);
exact binomial;
ods output OneWayFreqs=freqs
BinomialProp=Exttst(where=(name1 in("XL_BIN", "XU_BIN")));
run;

Thanks,

Ravi.

Re: exact binomial CI

ballardw — Mon, 03 Apr 2017 17:23:27 GMT

It really helps to show the code and error from the log. Best is to post it in a code box using the {i} menu icon in the forum as the main message windows will reformat text and many error messages include indicators as to the issue that get moved.

In general all confidence intervals require some variability in the data or you get a 0 width interval. It may be that you want to exclude the specific case(s) with only one response level from the analysis

Re: exact binomial CI

SSN_Ravi — Mon, 03 Apr 2017 20:00:30 GMT

Here is the error message

ERROR: The BINOMIAL LEVEL= value is greater than the number of levels for variable RESPSD1.

Re: exact binomial CI

ballardw — Mon, 03 Apr 2017 21:37:59 GMT

Maybe if we create a similar example it will help:

data example;
   input x ;
datalines;
1 
1 
1 
1 
1 
;
proc freq data=example;
   tables x /binomial(level=2);
   exact binomial;
run;

Generates that error. It is an error. In your case it may well only occur for some combinations of your by variables. Find out which combination (or combinations). Then using a WHERE clause do not use the analysis for those combinations. Note that for those combinations you will output similar to that generated by:

proc freq data=example;
   tables x /binomial(level=1);
   exact binomial;
run;

So run the code with the correct level only for those combinations.

Suppose you determine that the combination viswin=3 leg_sort=5 is the only combination with this problem then:

proc freq data=dsin1;
  by  viswin  leg_sort ;
  where viswin ne 3 and leg_sort ne 5;
  tables &G_A_RESPVAR/ binomial (level=2);
  exact binomial;
  ods output OneWayFreqs=freqs
             BinomialProp=Exttst(where=(name1 in("XL_BIN", "XU_BIN")));
run;

proc freq data=dsin1;
  by  viswin  leg_sort ;
  where viswin = 3 and leg_sort = 5;
  tables &G_A_RESPVAR/ binomial (level=1);
  exact binomial;
  ods output OneWayFreqs=freqs1Level
             BinomialProp=Exttst1Level(where=(name1 in("XL_BIN", "XU_BIN")));
run;

Of course if you have many combinations of the by variables with this issue then you may want to rethink things in general.