[Limdep Nlogit List] correcting standard errors for clustered survey data

[William Greene]

Peter.
The 'cluster' correction is a different thing from the 'robust' estimator, I would argue.
The issue is methodological.  My point is the one that you make in your (1). About
the cluster estimator, while I am a little skeptical of it, I have observed in applications
that it seems to do what it is argued to be attempting to do.  So, I would agree with
your point 1.  On 2, you should be able to use ;CLUSTER = spec. where spec is either
a number indicating a fixed cluster size or a variable name that contains a group 
identifier such as the cluster number.
/Bill Greene

************************************************
Professor William Greene
Department of Economics
Stern School of Business
New York University
44 West 4th St., Rm. 7-78
New York, NY   10012
Ph. 212.998.0876
Fax. 212.995.4218
URL. http://www.stern.nyu.edu/~wgreene
Email. wgreene at stern.nyu.edu
************************************************

----- Original Message -----
From: Peter Glick <pjg4 at cornell.edu>
Date: Thursday, June 22, 2006 11:02 am
Subject: [Limdep Nlogit List] correcting standard errors for clustered	survey data

> Dear Professor Greene (and list members),
> 
> I would like adjust my multinomial logit standard errors for 
> sample design 
> (correlations of errors within clusters).  I was hoping there 
> would be a 
> general command for this in Limdep (I have version 8)  but there 
> seems not 
> to be.  Assuming this is correct, it would be very helpful to have 
> a few 
> points clarified:
> 
> 1.   The manual (under 'Robust Standard Errors') warns against 
> using the 
> Huber White robust covariance matrix for nonlinear models because 
> the very 
> need for this correction, e.g.,  heteroskedasticity, in effect 
> means the 
> parameters themselves are not consistent.  I assume that this does 
> not 
> refer to intra-cluster error correlations (violation of the 
> independence 
> assumption), since if it did, parameter estimates from just about 
> all 
> nonlinear models using standard household surveys would be 
> inconsistent.  Am I correct?
> 
> 2.   Despite the warning, the manual notes that for the discrete 
> choice 
> models that allow choice based sampling, one can obtain the White 
> robust 
> standard errors using: "; Wts = ONE ; Choice Based".   But I 
> believe this 
> would yield the standard sandwich estimator which would not be the 
> appropriate one for correcting for cluster correlations of the 
> errors, i.e, 
> it assumes independence.  Is this correct?   If so, I would need 
> to 
> construct the appropriate matrix using the scores, etc.  Has 
> anyone done 
> this and could they provide code?
> 
> Many thanks..
> 
> Peter Glick
> Cornell University
> _______________________________________________
> Limdep site list
> Limdep at limdep.itls.usyd.edu.au
> http://limdep.itls.usyd.edu.au
>