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: nbreg vs zinb
From
Muhammad Anees <[email protected]>
To
[email protected]
Subject
Re: st: nbreg vs zinb
Date
Fri, 16 Mar 2012 18:51:44 +0500
Yes, I assume the picture is now not too blurr to be difficult to understand.
Anees
On Fri, Mar 16, 2012 at 4:39 PM, Simon Falck <[email protected]> wrote:
> Dear Anees,
>
> I took your advice and reread the Long/Freese chapter on models for count outcomes, in the second version from 2006 (see full reference below). I wrote down some findings, which perhaps can be useful as guidance in choosing count data model: NBRM or ZINB.
>
> The Long/Freese example in chapter 8.7, "Comparisons among count models" (p 405-414), illustrates a similar situation as me, the results from AIC/BIC/Voung test are ambiguous about what model is preferred. Long/Freese write (p. 407): "...the NBRM and ZIB do about equally well. From these results we might prefer the NBR because it is simpler". This suggests that I could choose NBRM over ZINB. However, Long/Freese also propose to make a formal LR-test. If I write the LR-test for the NBRM/ZINB using a "scalar" syntax, I end up "LR test comparing NBRM to ZINB: 8.898 Prob>=0.001", suggesting that the ZINB significantly improves the fit over the NBRM. Given the similarity between the Long/Freese example and my model(s) result(s), my interpretation is that model should be chosen upon the formal LR-test. Alternatively, results from both the NBRM and ZINB should be presented in the results section. In my case, the results (coefficients) from the NBRM and ZINB are not too differ!
en!
> t, so the latter can be considered as fair.
>
> /Simon
>
> Reference: J. Scott Long and Jeremy Freese (2006) Regression Models for Categorical Dependent Variables Using Stata, Second Edition. Stata Press
>
>
> -----Original Message-----
> From: [email protected] [mailto:[email protected]] On Behalf Of Muhammad Anees
> Sent: den 14 mars 2012 19:34
> To: [email protected]
> Subject: Re: st: nbreg vs zinb
>
> Thanks Clive,
>
> I am sure it is my post who you pointed at. Being a non-native English student, I would be really careful for the next time.
>
> The complete reference is as:
>
> LONG, J.S. and Freese, J., 2001. Regression Models for Categorical Dependent Variables Using Stata, Stata Corporation
>
> Thanks Again
> Anees
>
> On Wed, Mar 14, 2012 at 11:03 PM, <[email protected]> wrote:
>> You've now been on the list long enough not to scatter about names and years that aren't fully referenced. Read the Statalist FAQ again if you're not sure. Going by most of your posts, using a spellchecker wouldn't go amiss, either...
>>
>> C
>>
>> -----Original Message-----
>> From: Muhammad Anees <[email protected]>
>> Sender: [email protected]
>> Date: Wed, 14 Mar 2012 22:42:20
>> To: <[email protected]>
>> Reply-To: [email protected]: Re: st: nbreg vs zinb
>>
>> I would prefer NB model. You? I suggest read the Long and Freese notes
>> carefully.
>>
>> Anees
>>
>> On Wed, Mar 14, 2012 at 8:23 PM, Simon Falck <[email protected]> wrote:
>>> Dear Statlist,
>>>
>>> I´m regressing the number of start-ups, using count data.
>>>
>>> The data is clearly not Poisson.
>>>
>>> Hence, I am using nbreg and zinb models. I have some concern regarding what is the most appropriate model in my situation. Both seem relevant.
>>>
>>> Using the Long/Freese command, accordingly: countfit $dept $xlist,
>>> maxcount(10) nbreg zinb
>>>
>>> The Tests and Fit Statistics indicate that both nb and zinb may be preferred if BIC, AIC and Voung is consulted, although p>0.05.
>>>
>>> ---------------------------------------------------------------------
>>> ---- NBRM BIC= -2071.543 AIC= 2.714
>>> Prefer Over Evidence
>>> ---------------------------------------------------------------------
>>> ----
>>> vs ZINB BIC= -2046.607 dif= -24.935 NBRM
>>> ZINB Very strong
>>> AIC= 2.720 dif= -0.005
>>> NBRM ZINB
>>> Vuong= 1.292 prob= 0.098
>>> ZINB NBRM p=0.098
>>>
>>> What would be your suggestion regarding choice of model in my situation?
>>>
>>> Thanks in advance,
>>> /Simon
>>>
>>> *
>>> * 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/
>>
>>
>>
>> --
>>
>> Best
>> ---------------------------
>> Muhammad Anees
>> Assistant Professor/Programme Coordinator COMSATS Institute of
>> Information Technology Attock 43600, Pakistan
>> http://www.aneconomist.com
>>
>> *
>> * 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/
>
>
>
> --
>
> Best
> ---------------------------
> Muhammad Anees
> Assistant Professor/Programme Coordinator COMSATS Institute of Information Technology Attock 43600, Pakistan http://www.aneconomist.com
>
> *
> * 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/
--
Best
---------------------------
Muhammad Anees
Assistant Professor/Programme Coordinator
COMSATS Institute of Information Technology
Attock 43600, Pakistan
http://www.aneconomist.com
*
* 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/