Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


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

Re: st: problem with Margin command and interaction graph in fixed effect logistic regression


From   Nahla Betelmal <nahlaib@gmail.com>
To   "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu>
Subject   Re: st: problem with Margin command and interaction graph in fixed effect logistic regression
Date   Thu, 31 Oct 2013 16:51:46 +0000

Hi, thanks for your reply Alfonso. It is the nature of logistic fixed
effect not to report the constant. It does not report a constant
because it uses conditional ML to estimate the model, and the method
“conditions” the constant out of the likelihood function.  Implicitly
there is a constant, but it’s just not estimated or reported.

the command I used is clogit Y i.DV i.DV#c.IVA  i.DV#c.IVB i.DV#c.IVC,
group (i.fyear) or
which will be the same as if I used xtlogit, fe or

So we can not do anything to get the constant reported in the
regression, I was wondering if there is a way to get around that.
Maybe just plot the slops by specifying an expression of estimated
parameters that excludes or unifies the constant. I do not know really
how to do it , and I do hope that one of the members give insight into
this issue.

Many thanks again

Nahla

On 31 October 2013 16:15, Alfonso Sánchez-Peñalver
<alfonso.statalist@gmail.com> wrote:
> Hi Nahla,
>
> I believe you are right in thinking that margins is not reporting the margins because you are estimating the model without a constant. I tried the following simple example:
>
> ------------------ begin code -----------------------
> sysuse cancer.dta, clear
> logit died i.drug##c.age studytime, noconst
> margins drug, at(studytime=(10(5)30))
> ------------------- end code -----------------------
>
> I get the following results:
>
> Predictive margins      Number  of obs   =      48
> Model VCE    : OIM
>
> Expression   : Pr(died), predict()
>
> 1._at        : studytime       =          10
>
> 2._at        : studytime       =          15
>
> 3._at        : studytime       =          20
>
> 4._at        : studytime       =          25
>
> 5._at        : studytime       =          30
>
>
> Delta-method
> Margin   Std. Err.      z       P>z     [95% Conf.      Interval]
>
> _at#drug
> 1 1            .  (not estimable)
> 1 2            .  (not estimable)
> 1 3            .  (not estimable)
> 2 1            .  (not estimable)
> 2 2            .  (not estimable)
> 2 3            .  (not estimable)
> 3 1            .  (not estimable)
> 3 2            .  (not estimable)
> 3 3            .  (not estimable)
> 4 1            .  (not estimable)
> 4 2            .  (not estimable)
> 4 3            .  (not estimable)
> 5 1            .  (not estimable)
> 5 2            .  (not estimable)
> 5 3            .  (not estimable)
>
> I then do the -logit- estimtion without the -nonconst- option, and run the same -margins- command and get:
>
>
> Predictive margins                                Number of obs   =         48
> Model VCE    : OIM
>
> Expression   : Pr(died), predict()
>
> 1._at        : studytime       =          10
>
> 2._at        : studytime       =          15
>
> 3._at        : studytime       =          20
>
> 4._at        : studytime       =          25
>
> 5._at        : studytime       =          30
>
> ------------------------------------------------------------------------------
>              |            Delta-method
>              |     Margin   Std. Err.      z    P>|z|     [95% Conf. Interval]
> -------------+----------------------------------------------------------------
>     _at#drug |
>         1 1  |   .9465649   .0518676    18.25   0.000     .8449063    1.048224
>         1 2  |   .4488331    .149976     2.99   0.003     .1548854    .7427807
>         1 3  |   .6152748    .182977     3.36   0.001     .2566464    .9739032
>         2 1  |   .9333187   .0662422    14.09   0.000     .8034864    1.063151
>         2 2  |   .3948395   .1327874     2.97   0.003     .1345811    .6550979
>         2 3  |   .5688943   .1510893     3.77   0.000     .2727648    .8650238
>         3 1  |   .9171489   .0877723    10.45   0.000     .7451183    1.089179
>         3 2  |   .3431901   .1379768     2.49   0.013     .0727605    .6136197
>         3 3  |   .5217416   .1263294     4.13   0.000     .2741405    .7693427
>         4 1  |   .8975889   .1181116     7.60   0.000     .6660946    1.129083
>         4 2  |   .2948301   .1565933     1.88   0.060    -.0120872    .6017474
>         4 3  |   .4745012   .1173863     4.04   0.000     .2444282    .7045742
>         5 1  |   .8741791   .1585532     5.51   0.000     .5634206    1.184938
>         5 2  |   .2504701    .177148     1.41   0.157    -.0967337    .5976738
>         5 3  |   .4278449   .1277973     3.35   0.001     .1773668    .6783231
> ------------------------------------------------------------------------------
>
> So clearly, the noconst is the issue here. You have not explained which command you are using for your fixed effects logistic estimation, so I am not sure why it is running the estimation without a constant. Could you be a little more specific about the command you are using?
>
> Best,
>
> Alfonso
>
> On Oct 31, 2013, at 10:29 AM, Nahla Betelmal <nahlaib@gmail.com> wrote:
>
>> Dear Statalist members,
>>
>> I am facing difficulty in getting the margins and plotting the
>> interaction graph in a fixed-effect logistic regression. I believe
>> this is due to the fact that fixed effect logistic regression does not
>> report an intercept. I wonder if there is a way to get around this to
>> produce the graph.
>>
>> In the regression there is three interactions ( Dummy variable with
>> Continuous variable in each interaction)
>>
>>
>> Conditional (fixed-effects) logistic regression   Number of obs   =       9941
>>
>>           Wald chi2(34)   =     681.92
>>
>>         Prob > chi2     =     0.0000
>> Log pseudolikelihood = -3651.0306                               Pseudo
>> R2       =     0.1458
>>
>>                                              (Std. Err. adjusted for
>> 1220 clusters in firm)
>> ---------------------------------------------------------------------------------------------
>>                            |               Robust
>>                         To |      Coef.   Std. Err.      z    P>|z|
>>  [95% Conf. Interval]
>> ----------------------------+----------------------------------------------------------------
>>                   1.DV |   .1529853   .1576329     0.97   0.332
>> -.1559694    .4619401
>>                            |
>>            DVr#c.IVA |
>>                         0  |   .1643871   .0829615     1.98   0.048
>>  .0017855    .3269886
>>                         1  |   .2475583    .091258     2.71   0.007
>>   .068696    .4264206
>>                            |
>>            DV#c.IVB |
>>                         0  |   .1423918   .1179659     1.21   0.227
>>  -.088817    .3736007
>>                         1  |   .4909195   .1660735     2.96   0.003
>>  .1654214    .8164176
>>                            |
>>           DVr#c.IVC |
>>                         0  |   .5103285    .104731     4.87   0.000
>>  .3050596    .7155974
>>                         1  |   .5350398   .1325409     4.04   0.000
>>  .2752644    .7948152
>>
>>
>> I tried to get each interaction alone first, but did not work
>>
>> . margins DV, at( IVA=(-1.1(0.1)1.7))
>> default predict option not appropriate with margins
>> r(322);
>>
>>
>> I tried all together
>>
>> . margins OC_2year, at((mean) IVA (mean) IVB (mean) IVC)
>> default predict option not appropriate with margins
>> r(322);
>>
>>
>> Many thanks in advance
>>
>> Nahla Betelmal
>> *
>> *   For searches and help try:
>> *   http://www.stata.com/help.cgi?search
>> *   http://www.stata.com/support/faqs/resources/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/faqs/resources/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/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index