Re: st: Beta values in QREG

[Maarten Buis <maartenlbuis@gmail.com>]

I suspect that the real problem would more often be the standard
deviation. By standardizing a variable you change the unit of that
variable to have a standard deviation of 1. If the shape of the
distribution makes you doubt the usefulness of the mean as a measure
of central tendency, then I would be doubly worried about the standard
deviation as a measure variability.

More generally there is a "clash of logics" between standardizing
(where you use the moments) and quantile regression (where you use the
percentiles).

-- Maarten


On Mon, Jun 24, 2013 at 3:36 PM, Scott Holupka <scott.holupka@jhu.edu> wrote:
> Thanks.  I didn't want to ask for coding help until I was sure it was really
> a coding problem and not a more fundamental problem with the approach.  So
> you're first comment answers my more general question.  If I'm understanding
> it correctly, your point is why I would want to standardize values using a
> measure of central tendency if I already believe my measures are not
> normally-distributed, which, after all, is the reason for using
> quantile/median regression.
>
> Scott
>
>
>
> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Maarten Buis
> Sent: Saturday, June 22, 2013 3:34 PM
> To: statalist@hsphsun2.harvard.edu
> Subject: Re: st: Beta values in QREG
>
> On Fri, Jun 21, 2013 at 10:03 PM, Scott Holupka wrote:
>> Does anyone know if there is a statistical reason why the Stata
>> quantile regression program "qreg" does not provide an option for
>> producing beta values?  I know a question about beta values in qreg
>> was raised just a few months ago, and the one response suggested that
>> there might be a statistical reason why the option wasn't available,
>> but I didn't see anything more definitive.
>
> The logic behind quantile regression is all about avoiding the first (and
> second) moments and replacing those by the more robust quantiles.
> So substantively it would not make much sense to bring those moments back in
> a quantile regression by standardizing your variables. If you thought that
> your data was so problematic that you needed the more robust quantile
> regression, then it would be weird to use the non-robust standard deviation
> to define the scale of your variable. It is technically possible, but I
> would not recommend it.
>
>> I did try standardizing all of my variables and re-running QREG, as
>> had been previously suggested, but the results between the
>> unstandardized and standardized models seem so different I'm not sure
>> if I did something wrong or if there's a more fundamental reason why the
> results don't line up.
>
> In order for us to be able to judge that we would need to see what you have
> done. We cannot spot errors in code you don't show us.
>
> -- Maarten
>
> ---------------------------------
> Maarten L. Buis
> WZB
> Reichpietschufer 50
> 10785 Berlin
> Germany
>
> http://www.maartenbuis.nl
> ---------------------------------
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/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/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/



-- 
---------------------------------
Maarten L. Buis
WZB
Reichpietschufer 50
10785 Berlin
Germany

http://www.maartenbuis.nl
---------------------------------
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/