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: regression diagnostics with complex survey data


From   Steven Samuels <sjsamuels@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: regression diagnostics with complex survey data
Date   Wed, 27 Jul 2011 07:10:49 -0400


Sara-

In addition:

•  After -svy: reg-, you can do a valid Ramsey RESET test by hand (it's what -ovtest- does -reg-) . Generate the 2nd, 3rd,  & 4th powers of the fitted values by hand;  add them to the regressors of the original -svy: reg- ; and -test- that they are all zero.

• -linktest- is another specification test that works after -svy: reg-

• For finding outliers and high leverage points, nothing in Stata is better than  -mmregress- (-findit-), although it does not take weights. -mmregress- is a robust regression program that can detect clumps of  outliers that ordinary LS regression would miss. (Here "robust" means "insensitive to outliers", the original meaning of "robust regression".)  You might generate and  plot the high residuals and robust distances identified by -mmregress- and weight the symbols by the probability weights.

However, -mmregress- uses a single robust SD to standardize residuals.  So, it can give misleading results with certain patterns of heteroskedastic data (e.g. funnel-shaped). You would need to  first plot the standardized robust residual against predicted values, before deciding if a particular point identified as outlier by -mmregress- truly is one. 

Steve
sjsamuels@gmail.com


On Jul 26, 2011, at 1:53 PM, Sara Head wrote:

Jaoa,

This is very helpful, I'm embarrassed I missed them in earlier
searches! Thank you for sending!!

Sara

On Tue, Jul 26, 2011 at 1:12 PM, Joao Ricardo F. Lima
<jricardofl@gmail.com> wrote:
> Hi Sara,
> 
> from the UCLA website:
> 
> http://www.ats.ucla.edu/stat/stata/faq/svycollin.htm
> 
> and statalist archives:
> 
> http://www.stata.com/statalist/archive/2011-02/msg01254.html
> 
> HTH,
> 
> Joao Lima
> 
> 2011/7/26 Sara Head <sara.head@gmail.com>:
>> Hi Statalist,
>> 
>> I am working on a mediation analysis using complex survey data from
>> the 2007 Bangladesh Demographic and Health Survey. I’m using Stata/IC
>> 11.1.
>> 
>> I account for complex design with the svyset command, and -estat
>> effects, deff- shows  design effects are all less than 2. I am now
>> trying to run regression diagnostics with my most-final model, but
>> Stata’s svy post estimation commands do not support leverage, dfit,
>> cooksd, dfbeta, or vif (among others).
>> 
>> Although there are several thorough articles and dissertations
>> discussing regressions diagnostics for complex data (google the
>> authors Li & Valliant, Liao & Valliant, and Valliant), these are all
>> over my head and not directed at syntax.
>> 
>> Q1: Could someone point me towards any programs or commands for
>> considering influential data, multicollinearity with complex survey
>> data?
>> 
>> Q2: Since  design effects are still relatively low, what are
>> Statalisters’ opinions about conducting diagnostics without accounting
>> for design?
>> 
>> Many thanks
>> 
>> *
>> *   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/
>> 
> 
> 
> 
> --
> ---------------------------------------------------------
> João Ricardo F. de Lima
> Pesquisador A - Economia Aplicada
> Embrapa Semiárido
> e-mail: joao.ricardo@cpatsa.embrapa.br
> fone: +55 (87) 3862-1711 Ramal: 167
> home-page: www.cpatsa.embrapa.br
> ----------------------------------------------------------
> 
> *
> *   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/
> 



-- 
Sara Head, MPH
PhD Candidate, Rollins School of Public Health
Emory University, Atlanta, Georgia
sara.head@gmail.com, 502-553-9159

*
*   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