Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

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

st: RE: spatwmat for panel data?

From   Gordon Hughes <[email protected]>
To   [email protected]
Subject   st: RE: spatwmat for panel data?
Date   Thu, 13 Jan 2011 10:54:02 +0000

Before you start to look for Stata routines or other software you should be more specific about the model that you are trying to estimate.

It cannot possibly make sense to estimate a model with spatial correlations between 44,000 separate units and there is no software that would enable you to do this - think of the size of the spatial weighting or correlation matrix. What is more likely is that your data is clustered in some sense - i.e. the spatial correlations are between groups of individuals on whom you have observations. This is a routine feature of large surveys and you can investigate the options in Stata for estimating models with cluster-adjusted standard errors - there are a number of -xt..- panel estimators with this feature or you can look at the estimators for survey data.

A further consideration is whether your interest lies in the spatial aspects of your data or merely in adjusting standard errors to allow for spatial correlation. If it is the latter, then the -xtscc- module provides Driscoll-Kraay robust standard errors. On the other hand, if it is the former then you may need to write your own routines in Mata. There is a rapidly growing literature on the econometrics of spatial panel data - see papers by Baltagi, Pesaran, Elhorst and many others - but relatively few packaged routines that implement the methods which are described in the literature.

Routines to estimate models with spatial autocorrelation or spatial lags are primarily designed for cross-section data and tend to assume that you have or can calculate a spatial weighting matrix across all units, which is not realistic in your case. This is why the distinction between clusters and spatial units is so important.


Gordon Hughes


Professor Gordon Hughes
Department of Economics
University of Edinburgh
31 Buccleuch Place
Edinburgh EH8 9JT
United Kingdom

E-mail : [email protected]

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index