Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: RE: nlogit with dummies (follow up)


From   "Mentzakis, Emmanouil" <e.mentzakis@abdn.ac.uk>
To   <statalist@hsphsun2.harvard.edu>
Subject   st: RE: nlogit with dummies (follow up)
Date   Fri, 27 Jul 2007 20:34:19 +0100

Bidisha, 

I dont know if I understand correctly but when you say 

>I used interaction of all individual specific variables with the
choices as I thought that is the procedure to >do Nlogit (to interact
every variable with choices)

Do you mean that you interacted independent variables with the
dependent? Interacting your variables with choice (which I assume is
your dependent variable) will create variation within the alternatives
but is wrong. Hence, Stata probably will not tell you that there is
something wrong but there is. 

When I said interact the demographics with one of the attributes, I
meant with one of the other independent variables that vary within
alternatives.

As Rodrigo suggest:
> something like "married (dummy) times income (in dollars)". 

Maybe this is what creates the problem swith the Ses. 

Finally, if I ve got it all wrong and choice is not your dependent
variable and the variable that you interact with, then, appologies. 

Cheers
Manos  

-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu
[mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Rodrigo A.
Alfaro
Sent: 26 July 2007 22:42
To: statalist@hsphsun2.harvard.edu
Cc: lexshb@nottingham.ac.uk
Subject: st: nlogit with dummies (follow up)

Bidisha,

I guess that you have missing SEs because there are too much dummy
variables in your model and there are many missing observations on the
dataset... 
right? The latter could be 'solved' by creating imputed datasets (see
-ice- or http://www.multiple-imputation.com for more information). About
the dummies I don't know if you are technically able to identify the
model using mostly dummies. You should check your output and descriptive
with someone in your school or give us more info about what happens with
your data.

Rodrigo.
PS: Reply to the list, then you could have more feedback from other
users.

----- Original Message -----
From: "Bidisha Sayema" <lexshb@nottingham.ac.uk>
To: "Rodrigo A. Alfaro" <raalfaroa@gmail.com>
Sent: Thursday, July 26, 2007 10:36 AM
Subject: RE: st: NLOGIT

Many thanks indeed. I was able to run a nested logit but lot of my
variables came without SE or Z values. I think this is due to the use of
lot of variables (in bottom stage 20 and top stage 25 variables) where
most are dummies and there is no character specific variable. I however
dropped 1 bottom and 1 top category (with all interactions) from my
model but it seems that it was not sufficient. I used interaction of all
individual specific variables with the choices as I thought that is the
procedure to do Nlogit (to interact every variable with choices) and
also thought it would be useful to know about the sign and significance
of those coefficient. It would be appreciated if you could suggest me of
any way to capture the missing SEs.

Regards
Bidisha
*by MNL I meant Multinomial Logit.

-----Original Message-----
From: Rodrigo A. Alfaro [mailto:raalfaroa@gmail.com]
Sent: 26 July 2007 06:51
To: lexshb@nottingham.ac.uk
Subject: Fw: st: NLOGIT

> Check that
>
> (1) Your Stata is updated. -nlogit- was fixed for missing
observations.
> (2) Your covariates are not all dummies. Manos suggested to get some 
> multiplicative variables: something like "married (dummy) times income

> (in dollars)". Also, if ageskl is age times dummy-skilled and age 
> times dummy-unskilled why you have both?
> (3) Drop the nolog in the process... maybe there is some information 
> that we/you want to know, such as "not concave", "variable xyz dropped

> due to collinearity", etc.
> (4) Run a SIMPLE version of the model... I meant really simple.
> You should play with fraction of your dataset, and less covariates.
>
> Rodrigo.
>
>
> ----- Original Message -----
> From: "Neil Shephard" <nshephard@gmail.com>
> To: <statalist@hsphsun2.harvard.edu>
> Sent: Tuesday, July 24, 2007 8:13 AM
> Subject: Re: st: NLOGIT
>
>
>> MNL == Multiple Nested Logit?
>>
>> You've given us the command that you use, but not shown what the 
>> error's Stata is reporting are.  These are likely to be informative 
>> in helping resolve your problem.
>>
>> From the FAQ
>> (http://www.stata.com/support/faqs/res/statalist.html#advice)...
>>
>>    * Say exactly what you typed and exactly what Stata typed (or did)

>> in response. N.B. exactly! If you can, reproduce the error with one 
>> of Stata's provided datasets or a simple concocted dataset that you 
>> include in your posting.
>>
>>      Don't send examples like this, which gives no information:
>>          o The command -fiddle- worked yesterday, but it is not 
>> working today.
>>
>>    or like this, which gives very little information:
>>
>>        * I am using -frog-, but I just get the message "invalid
syntax".
>>
>>    This example shows better style, setting aside the fact that its 
>> answer can be found in the online help:
>>
>>        * I am using Stata 9 to try to draw a plot using -scatter-.
>> Here is an example showing my problem with the auto data:
>>
>>          . scatter mpg weight, s(p)
>>
>>        I get lots of very small points. Why doesn't Stata use plus 
>> symbols?
>>
>> Neil
>>
>> On 7/24/07, Bidisha Sayema <lexshb@nottingham.ac.uk> wrote:
>>> Dear Statalisters,
>>>
>>> I have done a MNL of employment status (4 choices: wage employed, 
>>> unemployed etc) with individual specific controls eg sex, education,

>>> age, race etc and in the next stage with only the wage employed run 
>>> another MNL of occupational choices (3 choices: skilled, unskilled
>>> etc) with controls like the first stage (not the same though). But I

>>> also want to do a NLOGIT while incorporating all the decisions, so 
>>> there will be 6 choices in the bottom set (3 choices out of wage 
>>> employed, 3 remaining categories of employment status) along with 
>>> the first set of 4 alternatives. I constructed the data set and the 
>>> tree is running fine but I was unable to run NLOGIT and was facing 
>>> error message. Is it due to the use of common individual specific 
>>> variables in both stages? But I need to have demographic and 
>>> education variables in both stages. Any suggestion is highly 
>>> appreciated. A simple version of the model is as
>>> follows:
>>>
>>> nlogit choice (occchoice=ageprof ageskl ageuskl agesemp ageunemp 
>>> agenemp
>>> /*
>>> */ age2prof age2skl age2uskl age2semp age2unemp age2nemp)
>>> (empstat=ageemp2 ageemp3 ageemp4 /*
>>> */ age2emp2 age2emp3 age2emp4 marriedemp2 marriedemp3 marriedemp4),
>>> group(caseno) nolog
>>>
>>> Best Regards
>>> Bidisha

*
*   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/

*
*   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/



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