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: Assumptions for continuous predictor in negative binomial regression model


From   Nick Cox <[email protected]>
To   "[email protected]" <[email protected]>
Subject   Re: st: Assumptions for continuous predictor in negative binomial regression model
Date   Fri, 28 Mar 2014 10:54:20 +0000

Working backwards, I am of a view that categorizing a continuous
variable is usually a very bad idea, but I am very open to polynomial,
fractional polynomial or spline representations if plain or vanilla
linear representations are inadequate.

I detect a flavour of assumptions being correct or not in your
writing. At some level, all assumptions are wrong, but it's then a
case of how much wrong and does that bite hard in a way that impedes
or defeats your purpose, and most of all, can you do better?

The main way I know of to check model assumptions is through looking
at residuals. For various reasons, some good, this appears far less
common with count regression models than with the more usual kind.
Nick
[email protected]


On 28 March 2014 10:30,  <[email protected]> wrote:
> Thanks Nick and Marteen!
> I do now have a nice graph, but how exactly can I verify whether the assumption is now met or not? Is there a test or another nice graphical option to verify if the underlying assumption is met for a certain variable? ( and so to decide whether I can keep it as a continuous variable in the model or if I have to e.g. categorize it).
>
> Thanks!
> Isabel
>
> -----Original Message-----
> From: [email protected] [mailto:[email protected]] On Behalf Of Maarten Buis
> Sent: Freitag, 28. März 2014 11:11
> To: [email protected]
> Subject: Re: st: Assumptions for continuous predictor in negative binomial regression model
>
> On Fri, Mar 28, 2014 at 10:47 AM,  Isabel Lechner wrote:
>> I could find quite a lot of assumptions concerning a negative binomial regression model in general, but what I couldn't find was if there is assumptions about including a continuous variable as a predictor? e.g in logistic regression,  it requires that the independent continuous variable is linearly related to the log odds of the outcome.
>>
>> 1. Is this the same for the negative binomial regression? Or are there any other assumptions concerning the inclusion of a continuous predictor variable?
>>
>> 2. I tried to graphically show this linear relationship of the independent continuous variable and the log odds of the outcome for logistic regression, but I didn't get a satisfying result! Any advice on that? I'm sure there must be a pretty easy way in stata, but I couldn't figure it out!
>
> Nick answered your second question.
>
> Your first question: yes it assumes that that the effect of your variable is linear with respect to the log(count). An equivalent way of thinking about that is: a unit increase in your variable is associated with the same percentage increase or decrease in the expected count regardless of where one starts.
>
> -- Maarten
>
> ---------------------------------
> Maarten L. Buis
> WZB
> Reichpietschufer 50
> 10785 Berlin
> Germany
>
> http://www.maartenbuis.nl
> ---------------------------------
>
> *
> *   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