st: Adjust command ?error

[Fred Wolfe <fwolfe@arthritis-research.org>]

As a follow-up to Kaleb Michauds posting on an seeming problem with 
-adjust-, this grew out of a further analysis of a manuscript we had just 
submitted where we subsequently obtained a counterintuitive result using 
adjust. We are working at the level of adjusted probabilities of 0.02 to 
0.03, where such a difference (probability of illness) is very important. 
To try to understand what was happening, we worked through a series of 
adjustments using and not using adjust. We found an inconsistency between 
manual adjustments (as per Stata manuals) and -adjust- adjustments only 
following logistic regression and adjust using the -pr- option, but not the 
-xb- option. We thought there was problem in the conversion from xb to pr 
within adjust (actually, in the xb value that adjust converts). Adjust is a 
complicated program and we couldn't see any clear errors in the code. We 
think there is an error in the program, but we might be wrong. As our 
manuscript has a table in it that relies on adjust, this is an issue of 
some importance to us (and others).

Thanks,

Fred


At 07:01 PM 1/21/2003 -0600, you wrote:

> There appears to be a problem with either my interpretation of what output 
> the -adjust- command should give or in the actual program itself.  It only 
> shows itself when using the pr (probability) option after a logistic 
> command (used in this example).
>
> Below I've used a sample Stata data set (census), created a binary 
> variable (longer) and then did a logistic using a single dependent 
> variable (medage).  I then use -predict- with the pr option to create the 
> p1 variable.  Using -adjust- without setting any values and using the pr 
> option, I generate the a1 variable and the output from -adjust- is by a 
> small categorical variable (region).  Lastly, I display the average values 
> of p1 and a1 by region (using -tabstat-) and they are exactly the same, 
> but they differ from the output from -adjust- using the -pr- option (but 
> not the xb option).
>
> Though the numbers seem similar, I am getting much larger and important 
> differences (~0.005) with my dataset of >10K observations.  Also, doing 
> these examples with the xb option gives output that are exactly the same - 
> the problem seems limited to the pr option.
>
> I look forward to any advice you can give.
> Sincerely,
> Kaleb Michaud
> Arthritis Research Center
>
>
> . use census
> (1980 Census data by state)
>
> . gen longer = length(state)
>
> . recode longer min/8=0 9/max=1
> (50 changes made)
>
> . logistic longer medage
>
> Logit estimates                                   Number of obs   =         50
>                                                   LR 
> chi2(1)      =       0.58
>                                                   Prob > 
> chi2     =     0.4481
> Log likelihood = -32.915529                       Pseudo R2       =     0.0087
>
> ------------------------------------------------------------------------------
>       longer | Odds Ratio   Std. Err.      z    P>|z|     [95% Conf. 
> Interval]
> -------------+----------------------------------------------------------------
>       medage 
> |   1.143314   .2048402     0.75   0.455     .8047506    1.624314
> ------------------------------------------------------------------------------
>
> . predict p1 if e(sample), pr
>
> . adjust, by(region) pr gen(a1)
>
> ------------------------------------------------------------------------------
>      Dependent variable: longer     Command: logistic
>        Created variable: a1
>     Variable left as is: medage
> ------------------------------------------------------------------------------
>
> ----------------------
> Census    |
> region    |         pr
> ----------+-----------
>        NE |    .433217
>   N Cntrl |     .37812
>     South |    .381077
>      West |    .339917
> ----------------------
>      Key:  pr  =  Probability
>
> . tabstat p1 a1 medage, by(region)
>
> Summary statistics: mean
>   by categories of: region (Census region)
>
>  region |        p1        a1    medage
> --------+------------------------------
>      NE |  .4335186  .4335186  31.23333
> N Cntrl |  .3783487  .3783487    29.525
>   South |  .3821012  .3821012  29.61875
>    West |  .3418868  .3418868  28.28462
> --------+------------------------------
>   Total |       .38       .38     29.54
> ---------------------------------------
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/

---------------------------------------------------------------------------- 
------------------------
Fred Wolfe                                                          Tel 
(316) 263-2125
National Data Bank for Rheumatic Diseases         Fax (316) 263-0761
Wichita, 
Kansas                                    fwolfe@arthritis-research.org
---------------------------------------------------------------------------- 
-------------------------

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/