Statalist


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

st: autocorrelations due to clusters


From   "Dalhia Mani" <dalhia.mani@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   st: autocorrelations due to clusters
Date   Tue, 7 Oct 2008 18:41:47 -0500

Hi, thanks for the previous responses from this list.

I have a question about how I can control for autocorrelation.  Let me
first describe my data a little bit. I have a large number of firms,
and for each firm I can distinguish whether or not the firm is
"independant" or resides within a cohesive cluster of firms (operates
as a business group).  I can also identify which business group
particular firms belong to (although not all firms belong to a
business group).  I can also distinguish the position of firms (and
clusters of firms) in the overall network.  Some are isolated, some
are core, some are peripheral in the overall network - for instance
some isolate clusters have ties wthin the cluster but no ties to the
rest of the network.  The combination of the clustering dimension
(whether or not a firm belongs to a cohesive cluster) and the position
in the overall network (core, periphery, isolate) leads me to
distinguish between isolate firms, isolate clusters, peripheral firms,
peripheral clusters, core firms, and core clusters.  I have a dummy
variable for each of these network types.

My problem is that I expect that the errors for firms in a cluster
(irrespective of the position in the network of the cluster) are
correlated.  Therefore, based on advice from members of this group, I
decided to use "regress x y1 y2, robust cluster(group)" to correct for
this.  However, I discovered that I get nonsense statistics when I run
this command if I construct the variable "group" such that it
identifies the group number for each cluster firm and is blank for
every "independent" firm - the procedure appears to kick out all the
independent firms in my sample - the output says "dropped" in lieu of
the coefficients for isolate firms, peripheral firms, and core firms
(all firms for which the "group" variable is blank).

Is there a way to deal with this issue? I want the procedure to
correct for autocorrelation within cluster firms, but I want the
procedure to leave alone the independent firms in the sample.  Or is
there some other way of achieving I want to do instead of using
"robust cluster()."  Any help will be appreciated.

thanks
dalhia
*
*   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