Re: st: Wald chi-square and ml with multiple equations

["E. Paul Wileyto" <epw@mail.med.upenn.edu>]

Nobody answered, so I dug around and got an answer myself.  Here it is 
for anyone who is curious about it.

Any ml procedure will, by default, report the Wald test for the first 
equation.  From the call to -zinb-, we see a table entry for the Wald 
chi2 with 9 degrees of freedom.  I want an overall test that covers 
equations 1 and 2, with 18 degrees of freedom.  If I were writing the 
routine, I could add an option to the ml max statement (or model 
statement for non-interactive) specifying waldtest(-2), which would make 
the table report the test for the first two equations.  However, you 
cannot mess around with the other people's code to do this, especially 
stock Stata commands.

The alternative is to use -test- afterwards, and the most efficient way 
to call it is:

test [#2]
test [#1], a

The first call will  report a test for equation 2, and the second call, 
with the "a" option will give a cumulative test of equation 1 and prior 
test.

Results look like this:

. test [#2]

( 1)  [inflate]drug = 0
( 2)  [inflate]sex = 0
( 3)  [inflate]wk_1 = 0
( 4)  [inflate]wk_2 = 0
( 5)  [inflate]wk_3 = 0
( 6)  [inflate]wk_4 = 0
( 7)  [inflate]wk_5 = 0
( 8)  [inflate]wk_6 = 0
( 9)  [inflate]wk_7 = 0

          chi2(  9) =   52.06
        Prob > chi2 =    0.0000

. test [#1],a

( 1)  [inflate]drug = 0
( 2)  [inflate]sex = 0
( 3)  [inflate]wk_1 = 0
( 4)  [inflate]wk_2 = 0
( 5)  [inflate]wk_3 = 0
( 6)  [inflate]wk_4 = 0
( 7)  [inflate]wk_5 = 0
( 8)  [inflate]wk_6 = 0
( 9)  [inflate]wk_7 = 0
(10)  [cignums]drug = 0
(11)  [cignums]sex = 0
(12)  [cignums]wk_1 = 0
(13)  [cignums]wk_2 = 0
(14)  [cignums]wk_3 = 0
(15)  [cignums]wk_4 = 0
(16)  [cignums]wk_5 = 0
(17)  [cignums]wk_6 = 0
(18)  [cignums]wk_7 = 0

          chi2( 18) =  137.09
        Prob > chi2 =    0.0000





E. Paul Wileyto wrote:
> When I am using ml with a multiple equation model, for instance 
> <zinb>, the Wald chi-square seems only to reflect the first equation.  
> Is this correct?  Is there any way to specify that the Wald chi-square 
> should refer to equations 1 and 2?
>
> Example, from our smoking data..
>
> . zinb cignums drug sex wk_*, inflate(drug sex wk_*) vce(cluster subject)
>
> Zero-inflated negative binomial regression        Number of obs   
> =      61187
>                                                  Nonzero obs     
> =      25293
>                                                  Zero obs        
> =      35894
>
> Inflation model      = logit                      Wald chi2(9)    
> =      14.20
> Log pseudolikelihood = -140745.3                  Prob > chi2     
> =     0.1152
>
>                             (Std. Err. adjusted for 1108 clusters in 
> subject)
> ------------------------------------------------------------------------------ 
>
>             |               Robust
>             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. 
> Interval]
> -------------+---------------------------------------------------------------- 
>
> cignums      |
>        drug |  -.0456683   .3742062    -0.12   0.903    -.7790989    
> .6877623
>         sex |  -.6529265   .3798595    -1.72   0.086    -1.397437    
> .0915844
>        wk_1 |   .3787082   .2037435     1.86   0.063    -.0206218    
> .7780381
>        wk_2 |   .1070914   .2334659     0.46   0.646    -.3504934    
> .5646762
>        wk_3 |   .0697002    .194177     0.36   0.720    -.3108797    
> .4502802
>        wk_4 |   .0677434   .1862907     0.36   0.716    -.2973796    
> .4328664
>        wk_5 |   .0437185   .0350697     1.25   0.213    -.0250169    
> .1124539
>        wk_6 |     .00758   .0215772     0.35   0.725    -.0347106    
> .0498706
>        wk_7 |   .0017718   .0107169     0.17   0.869     -.019233    
> .0227766
>       _cons |   3.153936   .3848694     8.19   0.000     2.399606    
> 3.908266
> -------------+---------------------------------------------------------------- 
>
> inflate      |
>        drug |   1.237415   .2633714     4.70   0.000     .7212164    
> 1.753613
>         sex |   -.645983   .2127249    -3.04   0.002    -1.062916   
> -.2290497
>        wk_1 |   1.210853   .2290275     5.29   0.000     .7619678    
> 1.659739
>        wk_2 |   1.012303    .190464     5.31   0.000         .639    
> 1.385605
>        wk_3 |    .839768   .1582912     5.31   0.000     .5295229    
> 1.150013
>        wk_4 |   .6872417   .1480842     4.64   0.000     .3970019    
> .9774815
>        wk_5 |   .4740839   .1026852     4.62   0.000     .2728246    
> .6753432
>        wk_6 |   .2748173   .0671609     4.09   0.000     .1431844    
> .4064501
>        wk_7 |   .1148675   .0404778     2.84   0.005     .0355325    
> .1942025
>       _cons |  -1.463269   .5311934    -2.75   0.006    -2.504389   
> -.4221496
> -------------+---------------------------------------------------------------- 
>
>    /lnalpha |    1.38432   .1631759     8.48   0.000     1.064501    
> 1.704139
> -------------+---------------------------------------------------------------- 
>
>       alpha |   3.992112   .6514166                      2.899393    
> 5.496653
> ------------------------------------------------------------------------------ 
>
>
>
> Thanks
>
> Paul

--
E. Paul Wileyto, Ph.D.
Assistant Professor of Biostatistics
Tobacco Use Research Center
School of Medicine, U. of Pennsylvania
3535 Market Street, Suite 4100
Philadelphia, PA  19104-3309

215-746-7147
Fax: 215-746-7140
epw@mail.med.upenn.edu



*
*   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/