Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: st: RE: Missing standard errors in multinomial logit


From   "Verkuilen, Jay" <JVerkuilen@gc.cuny.edu>
To   <statalist@hsphsun2.harvard.edu>
Subject   RE: st: RE: Missing standard errors in multinomial logit
Date   Mon, 11 May 2009 16:10:22 -0400

Look at your constant for the second option. It's -36. That's bad and I
think it's what's causing your covariance matrix to be trouble. 

Something's up. I'd suspect you have some near perfect prediction. You
have A LOT of variables and interactions in this kind of model are
notorious for creating trouble. Start by paring out the vast majority of
your variables and add them in blocks. That will give you a much better
idea of the troublemaker. 

JV



-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu
[mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of
serena.trucchi@unito.it
Sent: Monday, May 11, 2009 8:26 AM
To: statalist@hsphsun2.harvard.edu
Subject: RE: st: RE: Missing standard errors in multinomial logit

I obtain similar results also without the robust option..
Moreover the standard errors of the estimated coefficient
for the constant are missing: what does it means?
The estimated variance and covariance (listed through the
command vce) for this coefficient are zero.
Thanks
Seerena


> ------------------ Messaggio originale -------------------
> Oggetto: RE: st: RE: Missing standard errors in
> multinomial logit
> Da:      "jverkuilen" <jverkuilen@gc.cuny.edu>
> Data:    Lun, 11 Maggio 2009, 2:13 pm
> A:       statalist@hsphsun2.harvard.edu
> ----------------------------------------------------------
>
> You asked for clustered robust SEs. You get SEs but no
> overall chi square it looks like (coeffs are a little
> garbled on my screen). This is normal. When Stata doesn't
> report something it is usually a sign that it shouldn't be
> reported. In this case there isn't a valid model chi
> square to report; note the switch to pseudo-likelihood.
>
> One way to screen SEs for a sign of trouble is to see
> whether they are proportional to 1/sqrt(n) (assuming an IV
> and DV that have variances near 1).
>
> -----Original Message-----
> From: serena.trucchi@unito.it
> To: statalist@hsphsun2.harvard.edu
> Sent: 5/11/2009 5:10 AM
> Subject: Re: st: RE: Missing standard errors in
> multinomial logit
>
> I estimate:
> mlogit choice sex eta yearsedu y000 unempl y000papa
> lavoramammaspread affitto_r td* reg_*, vce(cluster nquest)
> and the output is:
>
> Multinomial logistic regression                   Number
> of obs   =       9130
>                                                              Wald
> chi2(32)
>
>
> =
>
>
>
>
>
>
>
>
>
> .
>                                                              Prob
>>
> chi2
>
>
>
>
> =
>
>
>
>
>
>
>
>
>
> .
> Log pseudolikelihood = -3185.8196                 Pseudo
> R2       =     0.5029
>
>                               (Std. Err. adjusted for 5719
> clusters in nquest)
>
------------------------------------------------------------------------
------
>              |               Robust
>       choice |      Coef.   Std. Err.      z    P>|z|
> [95% Conf. Interval]
>
-------------+----------------------------------------------------------
------
> Indep.couple |
>          sex |  -1.259999   .0892111   -14.12   0.000
> -1.43485   -1.085149
>          eta |    .369897   .0155086    23.85   0.000
> .3395007    .4002934
>     yearsedu |  -.1782479   .0150689   -11.83   0.000
> -.2077824   -.1487133
>         y000 |   .0325322   .0062735     5.19   0.000
> .0202364    .0448281
>       unempl |  -1.387266   .1695346    -8.18   0.000
> -1.719548   -1.054985
>     y000papa |     .03222   .0059338     5.43   0.000
> .0205899    .0438501
>  lavoramamma |   23.84664   1.046417    22.79   0.000
> 21.7957    25.89758
>       spread |   31.92757   10.33789     3.09   0.002
> 11.66567    52.18947
>    affitto_r |   -17.1902   8.191793    -2.10   0.036
> -33.24582   -1.134579
>       td1995 |  -.7098379   .1494154    -4.75   0.000
> -1.002687    -.416989
>       td1998 |  -.5893725   .1931463    -3.05   0.002
> -.9679324   -.2108126
>       td2000 |  -.6813418   .2790856    -2.44   0.015
> -1.228339   -.1343441
>       td2002 |  -.9640511    .256836    -3.75   0.000
> -1.46744   -.4606618
>       reg_NO |   1.055877   .3113467     3.39   0.001
> .4456483    1.666105
>       reg_NE |   .8436263   .3178704     2.65   0.008
> .2206117    1.466641
>       reg_CE |   .6105833   .2383532     2.56   0.010
> .1434196    1.077747
>        _cons |  -33.96454          .        .       .
>       .           .
>
-------------+----------------------------------------------------------
------
> Indep.single |
>          sex |  -.2966189   .1449906    -2.05   0.041
> -.5807953   -.0124424
>          eta |    .251174   .0260973     9.62   0.000
> .2000242    .3023237
>     yearsedu |  -.0148035   .0222393    -0.67   0.506
> -.0583918    .0287848
>         y000 |   .0728665   .0113887     6.40   0.000
> .0505449     .095188
>       unempl |  -1.143601   .3043513    -3.76   0.000
> -1.740119   -.5470836
>     y000papa |  -.1020295   .0105506    -9.67   0.000
> -.1227083   -.0813508
>  lavoramamma |   25.67025   1.589592    16.15   0.000
> 22.5547    28.78579
>       spread |   30.10601    15.7503     1.91   0.056
> -.7640025    60.97603
>    affitto_r |   .6206542    11.4264     0.05   0.957
> -21.77468    23.01599
>       td1995 |  -.9829992   .2635091    -3.73   0.000
> -1.499468   -.4665307
>       td1998 |  -.5831464   .3118048    -1.87   0.061
> -1.194273    .0279797
>       td2000 |  -.4690283   .4415194    -1.06   0.288
> -1.33439    .3963338
>       td2002 |  -.4568875   .3842724    -1.19   0.234
> -1.210048    .2962725
>       reg_NO |    1.52762   .4547395     3.36   0.001
> .6363469    2.418893
>       reg_NE |   .9431163   .4625978     2.04   0.041
> .0364414    1.849791
>       reg_CE |   .8397561   .3691794     2.27   0.023
> .1161779    1.563334
>        _cons |  -36.50551          .        .       .
>       .           .
>
------------------------------------------------------------------------
------
> (choice==Co-reside is the base outcome)
>
>
>>
>> ------------------ Messaggio originale
>> -------------------
>> Oggetto: st: RE: Missing standard errors in multinomial
>> logit
>> Da:      "Verkuilen, Jay" <JVerkuilen@gc.cuny.edu>
>> Data:    Dom, 10 Maggio 2009, 12:29 am
>> A:       statalist@hsphsun2.harvard.edu
>> ----------------------------------------------------------
>>
>> serena.trucchi@unito.it wrote:
>>
>>>>I am trying to estimate choices about living
>>>> arrangements
>> by a multinomial logit (mlogit).
>> If I use more than one dummy variable (ex regional
>> dummies, or sex and a dummy for employee)  stata fails
>> to
>> compute standard errors of the coefficient of the
>> constant
>> term (they are missing in the output table and zero in
>> the
>> variance/covariance matrix).
>> How can I solve my problem?<<
>>
>> When Stata fails to compute standard errors this is
>> usually a sign of a
>> larger problem with your model. I'd guess that you have
>> near perfect
>> prediction---common when you enter dummies into a
>> multinomial logit---so
>> Stata is able to generate an estimate but it's a nearly
>> useless one. But
>> it could be some other serio
> [truncated by sender]
>
> *
> *   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/
>
>

*
*   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/

*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index