st: Boxtid r(430) flat region resulting in a missing likelihood

[sela <sela103@yahoo.de>]

Dear Stata Users,

I'm building a logistic regression modell and therefore am using the
-boxtid- command in order to test for linearity in the logit. I'm using the
following variables:
change: 0 , 1 (dependent variable)
retro: -2 until 2 (5 point scale, independent variable)
polint: 1 until 5 (9 point scale, independent variable)
use: -10 until 10 (21 point scale, independent variable)

when doing the following command
    boxtid logit change retro polint use [pw= cggesow]
the nonlinearity test for both retro and polint is significant.

This would be the output:


Logistic regression                               Number of obs   =       
710
                                                       Wald chi2(4)    =     
15.33
                                                       Prob > chi2     =    
0.0041
Log pseudolikelihood = -275.28524         Pseudo R2       =     0.0597

------------------------------------------------------------------------------
                  |               Robust
    change    |      Coef.   Std. Err.      z    P>|z|     [95% Conf.
Interval]
-------------+----------------------------------------------------------------
    Iretr__1 |   -.045204   .0565584    -0.80   0.424    -.1560563   
.0656484
    Iretr_p1 |   .0188079    .040473     0.46   0.642    -.0605176   
.0981334
    Iuse___1 |   2.302325   75.66987     0.03   0.976    -146.0079   
150.6125
    Iuse__p1 |   -.970768    17.1763    -0.06   0.955    -34.63569   
32.69415
       _cons |   -1.93799    .255246    -7.59   0.000    -2.438263  
-1.437717
------------------------------------------------------------------------------
retro    |  -.5394969   .2083624     -2.589   Nonlin. dev. 4.721   (P =
0.030)
      p1 |   3.535742   2.306073      1.533
------------------------------------------------------------------------------
polint   |  -.1599401   .2681029     -0.597   Nonlin. dev. 10.404  (P =
0.001)
      p1 |  -290.0341          .          .
------------------------------------------------------------------------------
use      | -.1236776   .1175078     -1.053   Nonlin. dev. 1.632   (P =
0.201)
      p1 |  -.2366999   2.137676     -0.111
------------------------------------------------------------------------------
Deviance:  539.095.



I then used exp(x) to change both polint and retro (so it's now polint_new =
exp(polint) and retro_new = exp(retro) ) but this caused the error message
"r(430) flat region resulting in a missing likelihood".

When doing the -boxtid- command only bivariate or also when excluding one
independent variable respectively, it does present results.

Is my idea of using exp(x) to transform the variables wrong? What remedy
would help me avoiding the flat region (and: what exactly does Stata mean by
flat region?)?

I hope someone wants to help me and can give me some advice since I've been
playing with those variables for quite sometime...

Thank you!

--
View this message in context: http://statalist.1588530.n2.nabble.com/Boxtid-r-430-flat-region-resulting-in-a-missing-likelihood-tp6212352p6212352.html
Sent from the Statalist mailing list archive at Nabble.com.
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/