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]

RE: RE: st: Robust Standard Errors in Paneldatasets

From   Amy Dunbar <[email protected]>
To   "[email protected]" <[email protected]>
Subject   RE: RE: st: Robust Standard Errors in Paneldatasets
Date   Tue, 26 Oct 2010 14:55:43 +0000

Thank you, Kit.  I have a better understanding of time indicators now.

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Christopher Baum
Sent: Tuesday, October 26, 2010 10:36 AM
To: [email protected]
Subject: re: RE: st: Robust Standard Errors in Paneldatasets

Amy wrote

Kit Baum wrote: "None of what you have found deals with clustering."  When I followed up on Kit's -xtivreg2_ suggestion, I found the following in the help for ivreg2:

cluster(varname1 varname2) provides 2-way cluster-robust SEs and statistics as proposed by Cameron, Gelbach and Miller (2006) and Thompson (2009).  "Two-way cluster-robust" means the SEs and statistics are robust to arbitrary within-group correlation in two distinct non-nested categories defined by varname1 and varname2.  A typical application would be panel data where one "category" is the panel and the other "category" is time; the resulting SEs are robust to arbitrary within-panel autocorrelation (clustering on panel id) and to arbitrary contemporaneous cross-panel correlation (clustering on time).  

In Petersen, Mitchell A. 2009. Estimating Standard Errors in Finance Panel Data Sets: Comparing Approaches. Review of Financial Studies 22 (1), Petersen provides a link to his web site (  On his web page he states:  "The routines currently written into Stata allow you to cluster by only one variable (e.g. one dimension such as firm or time). Papers by Thompson (2006) and by Cameron, Gelbach and Miller (2006) suggest a way to account for multiple dimensions at the same time. This approach allows for correlations among different firms in the same year and different years in the same firm, for example. See their papers and mine for more details and caveats. I have written a Stata ado file to implement this estimation procedure." 

The help file above indicates that -ivreg2- does deal with both, so I'm not sure what I am missing.  If I'm correct, -ivreg2- came out in 2008, so maybe Petersen wrote his paper before -ivreg2-, but his website doesn't mention -ivreg2-. 

Also see Gow, I., G. Ormazabal, and D. Taylor. 2010. Correcting for Cross-Sectional and Time-Series Dependence in Accounting Research. The Accounting Review 85 (2):483.  This paper references Petersen's Stata code.

It's still not clear to me when it's ok to deal with time effects (cross-sectional correlation) parametrically by including a time indicator variable and just correct for time-series dependence (serial correlation) with cluster (firm) or vice versa.  The -ivreg2- help states, "Users should be aware of the asymptotic requirements for the consistency of the chosen VCE,"   so when T is short, is the best option the parametric option?  - ivreg2- has a small sample correction option, so when would that be appropriate as opposed to including a time indicator variable? 

Thank you for considering my question.

The Petersen piece was published in 2009, but substantially completed  in 2005:
so it is not surprising that it is not up to date with respect to changes made in -ivreg2-(SSC)  in 2008.

Including time dummies allows for proper specification of the intercept if indeed it is time-varying. It does nothing to allow for correlation of errors across firms (this for some reason is a common misconception). When you use one-way clustering by panel (which is what xtreg, fe does with -robust-) you are still assuming that the errors are independent across panels.  Using the -small- option just adjusts for d.f., and if you have a large panel, the difference between t and z will be negligible. Including time indicators deals with specification; -small- only affects the VCE. If your intercept should be time-varying, omitting time indicators would be a misspecification. Ignoring cross-panel correlation could cause the VCE to be biased, but if T is small, your two-way clustering results may not be that much of an improvement. In my BOS'10 and UKSUG2010 presentations with Mark Schaffer and Austin Nichols, we discuss some of these issues.


Kit Baum   |   Boston College Economics & DIW Berlin   |
                              An Introduction to Stata Programming  |
   An Introduction to Modern Econometrics Using Stata  |

*   For searches and help try:

*   For searches and help try:

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