Bookmark and Share

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


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

Re: st: Adjusted Standard Error in QIC Program


From   Joerg Luedicke <joerg.luedicke@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Adjusted Standard Error in QIC Program
Date   Fri, 25 May 2012 07:28:56 -0700

The way QIC is defined in Hardin & Hilbe (2003)*, you use the "naive"
variance estimates from a model assuming independent observations and
the robust variance estimates from a model with specified working
correlation. This is what the user-written program -qic- (version
1.1.9, from SSC) is doing, it fits two models and then uses the
variance estimates from both models (plus the quasi-likelihood from
the second model) to calculate the QIC. Thus, the difference in your
results is most likely due to fitting the same model with (via -qic-)
and without robust standard errors, and you should perhaps go with the
robust standard errors.

Have a look at Hardin & Hilbe (2003), p. 139ff. for information about
the QIC, and type in Stata -help vce_option- and follow the link to
the manual for robust variance estimation.

J.

*Hardin JW & Hilbe JM (2003): Generalized Estimating Equations. Boca
Raton, FL: Chapman & Hall/CRC.



On Thu, May 24, 2012 at 5:29 PM, Kenneth Shermock <kshermo1@jhmi.edu> wrote:
> I ran the user-developed QIC program today on a dataset and got puzzling
> results.  First the output when using the xtgee command:
> xtgee out70250day group initgluccat unstablesocialhistory
> intx_unstablesocialhistory, family(binomial) link(logit) corr(ar1) force
>
> For our main variable of interest:
> Coeff=3D -0.50045
> Std err =3D 0.2235
> P=0.025
>
> I expected the results from the QIC program to be the same, but they are
> not.
>
> Output when using QIC program:
> qic out70250day group initgluccat unstablesocialhistory
> intx_unstablesocialhistory, i(mrn) t(date)family(binomial) link(logit)
> corr(ar1) force
>
> For our main variable of interest:
>
> Coeff=3D -0.50045
> Std err =3D 0.3062
> P=0.102
>
> I'm wondering if there is some glitch or if these results seem plausible.
> The only difference I see between the QIC and xtgee models is the QIC
> model "adjusts the std error for clustering" on one of my variables.  What
> is this adjustment and is it plausible that it has such a profound effect
> on model estimates?
>
>
> Best regards,
>
> Ken
>
> Kenneth M. Shermock, PharmD, PhD
> Director, Center for Pharmaceutical Outcomes and Policy
> The Johns Hopkins Hospital
>
> Core Faculty
> The Armstrong Institute for Patient Safety and Quality
> Johns Hopkins Medicine
>
> 600 North Wolfe Street
> Carnegie 180
> Baltimore, MD  21287
> 410-502-7674 (Desk)
> 410-955-0287 (Fax)
> kenneth@jhmi.edu
>
>
> http://emailcharter.org <http://emailcharter.org/>
> NNTO: No Need to Open (Entire Message Contained in Signature Line)
> NNTR: No Need to Reply
> RR: Reply Requested
>
>
>
>
>
> *
> *   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/

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