calculate the 95% CI for continuous and categorical variables

[twix17]

Hi everyone,

I need help to calculate the 95% CI for the variables  day_diff and result for the table below.

subject	date2	date1	day_diff	result
1001	22-Sep-22	4-Mar-22	202	1
1002	14-Dec-22	4-Mar-22	285	1
1003	26-Sep-22	5-Mar-22	205	1
1004	24-Aug-23	29-Sep-22	329	1
1005	21-Jun-23	15-Dec-22	188	1
1006	31-Oct-23	16-Dec-22	319	1
1007	15-Jun-23	16-Dec-22	181	0
1008	22-Jun-23	16-Dec-22	188	1
1009	25-May-23	16-Dec-22	160	0
1010	1-Nov-23	19-Jan-23	286	1
1011	15-Jan-24	19-Jan-23	361	1
1012	18-Jan-24	19-Jan-23	364	0
1013	22-May-23	9-Feb-23	102	1
1014	18-Jan-24	9-Feb-23	343	0
1015	6-Jun-23	30-Nov-22	188	1
1016	25-Apr-23	7-Dec-22	139	1
1017	11-Jul-23	12-Jan-23	180	1

Thanks

[ballardw]

You can calculate the upper and lower limits of continuous numeric variables in Proc MEANS, SUMMARY, UNIVARIATE, REPORT and TABULATE (and a few other places depending on what you are doing for analysis). So do you need a new data set? A report?

Do you need this calculated for some group of values or over the data set as a whole?

Proc means data=yourdatasetname clm ;
   var day_dif result;
run;

Will provide  confidence limits of the mean over the entire data set.

Confidence intervals for categorical variables is pretty much a non-starter as there isn't any standard definition of such.

[twix17]

Hello,

Thanks for your response. But here is what's i am trying to create bilateral asymptotic or exact confidence intervals for binomial distributions at the 95% level.

Thanks again

[ballardw]

---

@twix17 wrote:

Hello,

Thanks for your response. But here is what's i am trying to create bilateral asymptotic or exact confidence intervals for binomial distributions at the 95% level.

Thanks again

---

Quite a bit different than your topic statement

Proc freq data=yourdataset;

   tables  result/ binomial (cl=exact  );  /* there are multiple options for what type of confidence limits you want to use. Look at the proc freq syntax online help for specifics and limitations*/

run;

You may have to specify a LEVEL='1' or Level='0' option to control which level is used for reporting. The more skewed your proportion may be (further from 50%) the more likely  this may be to look funny for your interpretation if you do not specify.

[PaigeMiller]

Result does not appear to be normally distributed. I don't know about day_diff, if that is normally distributed, but you can look into that. If day_diff is not normally distributed, you might want to look into non-parametric tests. For result, I would think that a binomial distribution test/CI would work, you can do this in PROC FREQ.

Also, I assume you are talking about a 95% confidence interval for the mean, but since you didn't really say that, would you please state that my assumption is correct?

[twix17]

Hello,

Thanks for your response. But here is what's i am trying to create bilateral asymptotic or exact confidence intervals for binomial distributions at the 95% level.

Thank you.

[twix17]

Hi @PaigeMiller 

Yes, I want 95% confidence interval for the mean for the continuous variable.

Thank you.