Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: when to cluster


From   ag_guerreroserdan@yahoo.com
To   stata listserve <statalist@hsphsun2.harvard.edu>
Subject   st: when to cluster
Date   Thu, 14 May 2009 06:33:50 -0700 (PDT)

Dear all, 


Someone might have pose this question already but browsing the responses in the Stata archive I did no find a response to my problem. 


I have a question on when to use cluster in a regression analysis. I am doing a Difference in Difference model and I want also to control for various individual and household characteristics. 

Yi= b0+ b1 After+ b3Treatment + b4After*Treatment + Xis + His + ui


My treatment variable is at the Provincial level (only 18). I have individual observations that I have pooled from two different cross-sectional suveys, so this is not panel but I am following the same cohort in each survey. I have some questions: 

If my treatement is at the province level, do I also need to cluster to adjust standard errors? I thought it was necessary to correct for intraclass correlation but in a this UCLA course, they mention that only if the interclass correlation is high (above .3 or .4), the ones I get, are very small around .014 

 http://www.gseis.ucla.edu/courses/ed230bc1/notes3/cluster.html

Do I still need to use the cluster option? 

What are the implications if I cluster and do not really need to?

Do I also have to correct for inter-class correlation also (FE)? When I use both options, my t statistics jump up alot, more than 30, seems very strange!

thanks for your advice, 

Gaby 


 



      
*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index