Statalist


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

Re: st: suggested references about the variables to include in zero-inflated portion of zinb?


From   Steven Samuels <sjhsamuels@earthlink.net>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: suggested references about the variables to include in zero-inflated portion of zinb?
Date   Sun, 26 Oct 2008 12:24:47 -0400

That's a very important point, Maarten. Thank you. David Freedman (2006) suggested that analysts compare 'robust' and conventional standard errors. If they are close, he recommended the use of the conventional version. If they are not close, he stressed that the whole model may be faulty--bringing all estimates into question. Hence my emphasis on a "good" model.

I'd like to amend one point in my previous post-that too many zero values are cause for suspicion. I spoke to a friend who is familiar with diagnostic psychological scales. She pointed out that the fewer the items in a scale, the more likely zeros would be. In her experience, mean scores were often small and lumps at zero were common. I've found the same an eight binary item brief screen for mental illness--most people had no "positive" items.

-Steve

Reference: D.A. Freedman. “On the so-called ‘Huber Sandwich Estimator’ and ‘robust’ standard errors.” The American Statistician vol. 60 (2006) pp. 299–302. Preprint at: http:// www.stat.berkeley.edu/~census/mlesan.pdf


On Oct 26, 2008, at 11:48 AM, Maarten buis wrote:

--- Steven Samuels <sjhsamuels@earthlink.net> wrote:
1. The reviewer's original opinion is not correct. If your target
parameter is the mean score, then OLS may give a consistent estimate,
even if the data are skew and non-normal. The proviso is that you
have a good prediction model for the mean.  However with OLS,
standard errors will be incorrect. The fix is easy: -reg- with a -
robust- option will give standard errors that are model-free.

The one proviso here is that -robust- typically requires more data than
the standard standard errors, as the asymptotics typically starts to
kick in later. This should not come as a surprise, with standard
standard errors you use information from your assumption about the
distribution, while with -robust- you explicitly exclude that
information and thus you require more information from another source,
which can only mean more data.

-- Maarten

-----------------------------------------
Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands

visiting address:
Buitenveldertselaan 3 (Metropolitan), room N515

+31 20 5986715

http://home.fsw.vu.nl/m.buis/
-----------------------------------------



*
*   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 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index