RE: st: multinomial multilevel analysis in Stata

[Cameron McIntosh <cnm100@hotmail.com>]

It seems to me that you would need a Beta regression or another appropriate technique for dealing with your fractional (percentage) outcome variable, but I don't know if any of these have been integrated into multilevel frameworks. Anyone else? Thus, log-transforming the outcome might be a better option, and having zero values is not prohibitive:

 

Young, K.H., & Young, L.Y. (1975). Estimation of Regressions Involving Logarithmic Transformation of Zero Values in the Dependent Variable. The American Statistician, 29(3), 118-120. 

 

Newson, R. (2003). Stata tip 1: The eform() option of regress. The Stata Journal, 3(4), 445.

 

Cam

 

----------------------------------------
> Date: Fri, 14 Oct 2011 11:18:59 -0700
> From: jiarongchen2002@yahoo.com
> Subject: Re: st: multinomial multilevel analysis in Stata
> To: statalist@hsphsun2.harvard.edu
>
> Thank you, Maarten. If I keep the outcome variable as a continuous variable, what would be a better way? 1. to treat it as it is, 2. log of the outcome, since its' distribution is skewed. But what to do with it since there are lot of cases have zero value for the outcome? 3. standardize the outcome.
> Any thoughts would be appreciated.
> Rosie
>
> ----- Original Message -----
> From: Maarten Buis <maartenlbuis@gmail.com>
> To: statalist@hsphsun2.harvard.edu
> Cc:
> Sent: Friday, October 14, 2011 11:42 AM
> Subject: Re: st: multinomial multilevel analysis in Stata
>
> On Fri, Oct 14, 2011 at 5:22 PM, Rosie Chen <jiarongchen2002@yahoo.com> wrote:
> > (1) For example, suppose the outcome variable measures the percentage of budget purchasing luxurious items within a person's monthly salary, and it is on a continuous scale. I can recode it into a three-category variable, 1 indicating no purchase at all, 2 indicating larger than zero but smaller than 20%, and 3 indicating 20% or more. Should I treat it as an ordinal outcome or multinomial outcome?
>
> You should not recode it at all but model the continuous variable as
> is. With recoding you just loose information and loosing information
> is bad.
>
> > (3) Do I need to include weighting variable for each level of data, if there are 3 levels in the model? It seems to me to be common to use level 1 weights, but I haven't seen a lot of studies using weights for all levels.
>
> I believe you can do so with -gllamm-.
>
> Hope this helps,
> Maarten
>
> --------------------------
> Maarten L. Buis
> Institut fuer Soziologie
> Universitaet Tuebingen
> Wilhelmstrasse 36
> 72074 Tuebingen
> Germany
>
>
> http://www.maartenbuis.nl
> --------------------------
>
> *
> * 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/ 		 	   		  
*
*   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/