Re: st: Using xtmixed with a regional-level DV

[Atul Teckchandani <atulteck@hotmail.com>]

Dear Joerg,

I have aggregate data on the number of associations in a region, and I have 100+ regions. Furthermore, the total number of associations can be broken down into six different categories of associations. As such, I can run a model predicting the effects of each association category on the DV.

I also have data on diversity and participation for each association category. I was hoping to somehow demonstrate that it is not the type of association, but its level of diversity and/or participation that affects the DV. 

I was advised that HLM or multi-level modeling would be the appropriate method. However, your response suggests otherwise.

Can you recommend the appropriate empirical method to use for this purpose?

Thank you,

Atul

>I am not sure if I understand what exactly the multilevel set-up is.
>Do you have associations nested in regions? Or do you have just
>aggregate data for the number of associations in a given region? If
>the latter is true, this wouldn't be a multilevel design. If the
>former is true then you cannot do what you want which is regressing a
>level-2 outcome on level-1 predictors. It makes no sense because in
>each region the outcome variable would be a constant.
>Also, the "# of new firms" sounds like a count variable that is
>bounded at zero and most likely will be skewed. One is usually better
>off modeling these kind of data using a (possibly overdispersed)
>Poisson model instead of a linear model.
>J.
 
> From: atulteck@hotmail.com
> To: statalist@hsphsun2.harvard.edu
> Subject: Using xtmixed with a regional-level DV
> Date: Mon, 1 Oct 2012 22:56:05 -0700
> 
> Hi,
> 
> I am trying to do a regional-level analysis predicting the # of new firms created in a given region with the key independent variables being counts of six different types of voluntary associations in that region (e.g., business association, religious association, civic association, etc.).
> I also have data on the level of diversity of the members in each type of voluntary association, and on the typical level of participation of members in each type of voluntary association. The diversity and participation data is for all associations of a specific type (i.e., all business associations have the same diversity value and participation value, regardless of region where they are located).
> 
> I would like to show how the diversity and participation values also affect the DV. Based on my (limited) understanding, I believe I am looking to do multilevel modeling. The issue I am facing is that all of the multilevel modeling examples I have found have an individual-level DV with individual-level data as the “fixed effects” portion and then incorporate regional-level information as part of the “random effects.”
> 
> However, I want to do the opposite. I am predicting something at the regional-level, but also have data at the association level.
> I believe my fixed-effects equation would be as follows:
> xtmixed newfirms biz_assoc religious_assoc civic_assoc …
> 
> What would my random-effects equation look like when using the xtmixed command with this type of data?
> 
> Thank you in advance for the assistance!
> 
> Sincerely,
> 
> Atul Teckchandani
> Assistant Professor of Management
> Mihaylo College of Business and Economics
> California State University, Fullerton 		 	   		  
*
*   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/