st: problem with testparm

[Subramanian Swaminathan <ssubra@yahoo.com>]

Dear listers,

I have a problem in interpreting the results of the
stata output based on 
posson regression and the post-estimation command
'testparm'. I fitted a 
poisson regression model for a count data in which
'pos' is the dependent 
variable and 'prepost' and 'group' are the two
independent variables 
(indicator variables) with 'pmonths' as exposure
variable. I fitted the 
model with the interaction term prepost*group. The
following are the stata 
output.

. xi:poisson  pos i.prepost*i.group, exposure(pmonths)
irr
i.prepost         _Iprepost_0-2       (naturally
coded; _Iprepost_0 omitted)
i.group           _Igroup_1-3         (naturally
coded; _Igroup_1 omitted)
i.pre~t*i.group   _IpreXgro_#_#       (coded as above)

Iteration 0:   log likelihood = -147.60498
Iteration 1:   log likelihood = -147.45214
Iteration 2:   log likelihood =   -147.452
Iteration 3:   log likelihood =   -147.452

Poisson regression                               
Number of obs   = 
27
                                                  LR
chi2(8)      = 
228.58
                                                  Prob
> chi2     = 
0.0000
Log likelihood =   -147.452                      
Pseudo R2       = 
0.4367

------------------------------------------------------------------------------
         pos |        IRR   Std. Err.      z    P>|z| 
   [95% Conf. 
Interval]
-------------+----------------------------------------------------------------
 _Iprepost_1 |   2.701299   .3762209     7.14   0.000 
   2.055996 
3.549138
 _Iprepost_2 |   2.805195   .3723271     7.77   0.000 
   2.162644 
3.638656
   _Igroup_2 |   1.552784   .2320507     2.94   0.003 
   1.158527 
2.081211
   _Igroup_3 |   1.464632   .2245875     2.49   0.013 
   1.084438 
1.978118
_IpreXgr~1_2 |   .3413611   .0701716    -5.23   0.000 
    .228159 
.510729
_IpreXgr~1_3 |   .1402834   .0363008    -7.59   0.000 
   .0844779 
.2329537
_IpreXgr~2_2 |   .5064036   .0927443    -3.72   0.000 
   .3536756 
.7250842
_IpreXgr~2_3 |   .1425926    .033276    -8.35   0.000 
   .0902521 
.2252873
     pmonths | (exposure)
------------------------------------------------------------------------------

. testparm _IpreXgro_1_2 _IpreXgro_1_3, equal

 ( 1) - [pos]_IpreXgro_1_2 + [pos]_IpreXgro_1_3 = 0

              chi2(  1) =   11.23
         Prob > chi2 =    0.0008

. testparm _IpreXgro_2_2 _IpreXgro_2_3, equal

 ( 1) - [pos]_IpreXgro_2_2 + [pos]_IpreXgro_2_3 = 0

              chi2(  1) =   30.44
         Prob > chi2 =    0.0000

Comparison of the 95% CI for the parameters'
_IpreXgr~1_2 with _IpreXgr~1_3 
suggest that the two estimates do not differ
significantly. Whereas 
application of 'testparm' indicated that they differ
significantly. My 
questions are:

(1) Can I use the 95%CI (overlapping or
non-overlapping) for the parameters 
for comparing the difference in estimates?
(2) Can I use the testparm for comparing estimates
after fitting poisson 
regression?
(3) Which one of the tests is most appropriate
(comparison of CI or 
testparm)?
(3) If both can be used, should the conclusion be
same?

I would greatly appreciate the listers advice in
clarifying my problem.

thanking you in advance and with best wishes

Subramanian Swaminathan
Asst. Director
Vector Control Research Centre
(Indian Council of Medical Research)
Indira Nagar
Pondicherry - 605 006
INDIA
Subramanian Swaminathan
Vector Control Research Centre
(Indian Council of Medical Research)
Indira Nagar
Pondicherry - 605 006
INDIA 


__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 
*
*   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/