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: mlogtest after mlogit


From   Nick Cox <[email protected]>
To   "'[email protected]'" <[email protected]>
Subject   RE: st: mlogtest after mlogit
Date   Mon, 24 Oct 2011 19:28:47 +0100

I had success with this toy script which attempts to simulate an -mlogit- model with 9 response categories, 25 predictors and using -if- to select observations. 

I have -mlogtest- 1.7.6 from spost9_ado from http://www.indiana.edu/~jslsoc/stata

set obs 1000
gen y = ceil(9 * runiform()) 
tab y 
forval i = 1/ 25 { 
gen x`i' = rnormal()
} 
gen foo = _n  > 500
mlogit y x* if foo, b(3)
mlogtest, c

In principle what you typed looks legal and naturally your -mlogit- command worked. You too could test the commands with that script above. If you don't get problems, it looks like a problem with your data and model combined, and in the absence of solutions from Statalist members, you need to seek specific help from the authors of -mlogtest-. If you do get problems, you would need to say what they are. 

Nick 
[email protected] 

Chiara Mussida

Dear Nick,
thanks, I did not get your point before. Here there is the mlogit
command I typed:

 mlogit transition  male_unmarried female_married female_unmarried age
agesq ncomp child  northw northe centre Ubenef  edu1 edu2 health city
qu1nolav qu2nolav qu3nolav nopersincnolav noothersineq qu1ot qu2ot
qu3ot if age>=15 & age<=64, b(3)


covariates: 3 interaction variables (gender*martal status). age and
individual age squared, number of household components as continuous
variables. child: dummy variable for the presence of children, then 8
dummies (labels are clear).

qu1nolav qu2nolav qu3nolav are quintiles for not labour income
distribution (qu4 is the base)
nopersincnolav noothersineq dummies for absence of not labour income
and income from other household memebers, respectively
qu1ot qu2ot qu3ot quintiles for income of other household members
distribution (qu4 is the base)

The multinomial logit estimates refer to individuals aged between 15 and 64.





On 24 October 2011 15:34, Nick Cox <[email protected]> wrote:
> The exact -mlogit- command you typed is what anyone wanting to diagnose your problem would want to see. No one is asking for "the file", whatever that is.
>
> As said, there are different versions of -mlogtest-, as -findit mlogtest- reveals. Naming the authors does not discriminate between them.
>
> I've now tried and failed twice to elicit from you what should have been provided in your original post. At this point I leave you to whatever help other people can provide given your incomplete question.
>
> Nick
> [email protected]
>
> Chiara Mussida
>
> I cannot share the file with all statalister!
>
> Sure mlogtest is typewritten and I did not gave you this information,
> not for being frammentary but because I engenously thought it was not
> relevato: i'm so sorry! Now I got it: the program is typewritten by J.
> Scott Long and Jeremy Freese.
>
>
> On 24 October 2011 14:13, Nick Cox <[email protected]> wrote:
>> You still don't give the exact -mlogit- command you typed.
>>
>> If (a) you can reproduce this problem with a dataset accessible to all, and/or (b) you can spell out exact commands visible to all, it stands a better chance of being solved. Trying to solve the problem without (a) is difficult enough, but we still lack even the second, (b).
>>
>> In addition, on checking I find that you have not explained that -mlogtest- is a user-written command (versions in STB and various websites) and you have not explained which version you are using. The authors of -mlogtest- are not active members of Statalist, so you are reliant on people understanding a user-written command they have not written.
>>
>> Please re-read the Statalist FAQ before posting. By posting fragmentary questions with incomplete information you are making any attempts to help you frustratingly difficult.
>>
>> Nick
>> [email protected]
>>
>> Chiara Mussida
>>
>> Thanks Nick. I'm pleased to clarify the points you raised:
>> 1) yes, the categories are 9 at the end, because the outcome
>> transition=0 is replaceb completely by the other categories.
>> 2)I didn't do anything between mlogit  - stata command for multinomial
>> logit estimates - and the related mlogtest. Nonetheless mlogtest works
>> only after mlogit.
>>
>> Chiara
>>
>> On 24 October 2011 13:20, Nick Cox <[email protected]> wrote:
>>
>>> I don't understand Muhammad's first comment, which requires a change of the data, or indeed why Chiara even tried it out.
>>>
>>> In broad terms, Stata [sic] is having difficulties reconciling different information.
>>>
>>> That needs to be explored by looking at what you have said to Stata, through input of data and commands. There is little information in your posts to let anyone help you.
>>>
>>> First off, you say that there are 9 categories, but in principle there could be 10, according to your -generate- and -replace- statements, but there could be fewer than that depending on values of variables such as -UE- and on whether later -replace- commands overwrite the results of earlier commands.
>>>
>>> Second, and more important: What was the -mlogit- command? Did you do anything to your data between -mlogit- and -mlogtest-?
>>>
>>> Nick
>>> [email protected]
>>>
>>>
>>> Muhammad Anees
>>>
>>> Your test is offered by -wald- option after -mlogtest-. Hope this
>>> gives you what you needed.
>>>
>>> On Mon, Oct 24, 2011 at 3:42 PM, Chiara Mussida <[email protected]> wrote:
>>>
>>>> Thanks, unfortanately the problem remains also with recode
>>>
>>> On 24 October 2011 12:13, Muhammad Anees <[email protected]> wrote:
>>>
>>>>> try with -recode 0=1 if UE==1- for all your categories, the problem
>>>>> seems to be because -replace- might not have created the relevant
>>>>> categories.
>>>
>>> On Mon, Oct 24, 2011 at 2:59 PM, Chiara Mussida <[email protected]> wrote:
>>>
>>>>>> I run I mlogit with a dependent variables for 9 outcome categories. As
>>>>>> postestimation checks I tried to carry out a test for the equality of
>>>>>> coefs across categories: mlogtest, c
>>>>>>
>>>>>> before despalying the output of thus check, STATA (version 12) says:
>>>>>>  mlogtest, c
>>>>>>
>>>>>> Problem determining number of categories.
>>>>>>
>>>>>> what does it mean? my dependent variable, transition, is built as follows:
>>>>>> ge transition=0
>>>>>> replace transition=1 if UE==1
>>>>>> replace transition=2 if UN==1
>>>>>> replace transition=3 if UU==1
>>>>>> replace transition=4 if EU==1
>>>>>> replace transition=5 if EN==1
>>>>>> replace transition=6 if EE==1
>>>>>> replace transition=7 if NE==1
>>>>>> replace transition=8 if NU==1
>>>>>> replace transition=9 if NN==1
>>>>>>
>>>>>> where each outcome is a transition or permanence in each labour market
>>>>>> state (namely, Unemployment, Employment, Non labour force)

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