Afternoon all,

I have a question about the excellent parmby module, provided by Roger Newson.

As a simple example, I'm running the following command:
Code:
clear all
frames reset
sysuse auto
proportion rep78

As expected, the citype on proportion defaults to logit and yields the following results:
Code:
Proportion estimation             Number of obs   =         69

--------------------------------------------------------------
             |                                   Logit
             | Proportion   Std. Err.     [95% Conf. Interval]
-------------+------------------------------------------------
       rep78 |
          1  |   .0289855   .0201966      .0070794    .1110924
          2  |    .115942   .0385422       .058317    .2173648
          3  |   .4347826   .0596787      .3214848    .5553295
          4  |   .2608696   .0528625      .1695907    .3788629
          5  |   .1594203   .0440694      .0895793     .267702
--------------------------------------------------------------
Running the same code in parmby...
Code:
parmby "proportion rep78", frame(output1, replace)
frame output1: list, noobs clean
...yields a different set of confidence intervals:
Code:
    parmseq      parm    estimate      stderr   dof           t           p        min95       max95  
          1   1.rep78   .02898551   .02019662    68    1.435166   .15582355   -.01131623   .06928724  
          2   2.rep78   .11594203   .03854218    68   3.0081856   .00368254    .03903231   .19285175  
          3   3.rep78   .43478261   .05967869    68   7.2853911   4.333e-10    .31569563   .55386958  
          4   4.rep78   .26086957    .0528625    68   4.9348699   5.465e-06    .15538409   .36635504  
          5   5.rep78   .15942029   .04406936    68   3.6174863    .0005668    .07148126   .24735932
Parmby returns confidence intervals for the proportion as if I had run the proportion command with the argument citype(normal).

I have attempted the following which have not solved the problem:
  • explicitly added citype(logit) to the parmby command
  • adjusted the available parmby confidence interval options
  • run commands with set trace on to ensure that no unseen error is occurring
  • read Mr Newson's paper on confidence intervals to see if there's anything obvious I'm missing
So, specific questions:
  • Has anyone experienced this or can explain why it's happening?
  • Is this expected behaviour for parmby?
  • Does anyone know of a way of "forcing" parmby to pick up the logit transformed CIs, rather than the normal/t-distribution CIs?
Thanks for considering my questions.

Warren