Re: st: What analysis to use -count variables

[Ronan Conroy <rconroy@rcsi.ie>]

On 28 Feb 2011, at 22:04, Marlis Gonzalez Fernandez wrote:

> I am working on the relationship between two count variables and are trying to figure out the best way to analyze.
> 
> The dependent var is the number of times a particular stage of swallowing occurs and we want to determine if # of chewing cycles  predicts this variable.
> 
> So both vars are count (# of st per swallow and number of chews per swallow)
> 

Sounds to me also as if you have possibly more than one observation per participant, which needs to be factored in using -svy-. 

> I thought that the best way to analyze was using nbreg (assumptions not met for poisson regression)

-gnbreg- is useful here, because it allows you to model the shape of the overdispersion parameter, which may, for example, be a function of the total number of chews or swallows since the start of the experiment. 

My reservation about the data is that, if it does represent multiple observations per participant, it may show time-dependent features that mean that observations are not independent. The phenomenon you are interested in my have a refractory period, so that its occurrence in one swallow may make it less likely to occur in the next. Or more likely – I don't know the underlying physiology here. 

Primarily, constantly looking at model fit and thinking about what variables ought to be in the model are the essential elements – same as ever. 

Ronán Conroy
rconroy@rcsi.ie
Associate Professor
Division of Population Health Sciences
Royal College of Surgeons in Ireland
Beaux Lane House
Dublin 2


*
*   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/