Confidence Intervals and Wilcox Signed Rank Test

[Leslie Muller]

Hello - could you assist me with the Stata command/process of commands for producing confidence intervals with a Wilcox Signed Rank Test? I'm using the commands:
mean variable1
mean variable2
signrank variable1=variable2

I get a test statistic and p value, but no confidence intervals. I looked up signrank in Stata help and do not see any options for producing confidence intervals either.

Any suggestions would be greatly appreciated. Thank you.

Leslie Muller

[Carlo Lazzaro]

Leslie:
welcome to the list.
The following link may be useful: http://www.stata-journal.com/sjpdf.h...clenum=snp15_6

[Nick Cox]

The name Wilcox here is likely to be a typo for Wilcoxon arising from R's bizarre choice of function name. Oddly, however, at least two people called Wilcox have also worked in nonparametric or robust statistics.

[Leslie Muller]

That's very helpful! Thank you!
Leslie

[Leslie Muller]

Carlo,
I have one more question for you. I did the SomersD test, but am not getting a confidence interval that makes sense in the context of my difference in means. The estimate for absdiff here is not anywhere near equal to the difference in the means of the two variables that I'm comparing (difference is 5.581, not found in my output below; SomersD absdiff is 0.212). Hence, reporting the confidence interval doesn't make any sense next to the actual difference. In my code/output below I first show the means of my two variables of interest, then execute the SomersD code:

.mean totalvisits_ed_pre

Mean estimation Number of obs = 339
--------------------------------------------------------------------
| Mean Std. Err. [95% Conf. Interval]
-------------------+------------------------------------------------
totalvisits_ed_pre | 14.90265 .5064196 13.90652 15.89879
--------------------------------------------------------------------

. mean totalvisits_ed_post
Mean estimation Number of obs = 339

---------------------------------------------------------------------
| Mean Std. Err. [95% Conf. Interval]
--------------------+------------------------------------------------
totalvisits_ed_post | 9.321534 .4895261 8.358632 10.28444
---------------------------------------------------------------------

.gen signdiff=sign(totalvisits_ed_pre-totalvisits_ed_post)
. gen absdiff=abs(totalvisits_ed_pre-totalvisits_ed_post)
. somersd signdiff absdiff if absdiff!=0,transf(z)
Somers' D with variable: signdiff
Transformation: Fisher's z
Valid observations: 327

Symmetric 95% CI for transformed Somers' D
------------------------------------------------------------------------------
| Jackknife
signdiff | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
absdiff | .2121719 .0814114 2.61 0.009 .0526085 .3717354
------------------------------------------------------------------------------

Asymmetric 95% CI for untransformed Somers' D
Somers_D Minimum Maximum
absdiff .20904447 .05255998 .35550873

************************************************** ***

Thank you again!
Leslie

[Carlo Lazzaro]

Leslie:
I would say that you're expecting the wrong answer from Somers'D statistics.
See, for instance, pages 323-324 of the reference quoted in my previous reply.
However, another approach that may be useful when the requirements of the -ttest- are violated, is a bootstrap (or bootstrapped) -ttest-; see, if interested, -bootstrap- entry in Stata .pdf manual.